aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--bootstrap/bin/no_dot_erlang.bootbin5607 -> 5905 bytes
-rw-r--r--bootstrap/bin/start.bootbin5607 -> 5905 bytes
-rw-r--r--bootstrap/bin/start_clean.bootbin5607 -> 5905 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/file.beambin14080 -> 14052 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/file_io_server.beambin15040 -> 15412 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/file_server.beambin5348 -> 5104 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/kernel.app7
-rw-r--r--bootstrap/lib/kernel/ebin/raw_file_io.beambin0 -> 1724 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/raw_file_io_compressed.beambin0 -> 2364 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/raw_file_io_deflate.beambin0 -> 2708 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/raw_file_io_delayed.beambin0 -> 5432 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/raw_file_io_inflate.beambin0 -> 4256 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/raw_file_io_list.beambin0 -> 2576 bytes
-rw-r--r--bootstrap/lib/kernel/ebin/raw_file_io_raw.beambin0 -> 396 bytes
-rw-r--r--erts/emulator/Makefile.in34
-rw-r--r--erts/emulator/beam/erl_lock_check.c4
-rw-r--r--erts/emulator/drivers/common/efile_drv.c4295
-rw-r--r--erts/emulator/drivers/common/erl_efile.h176
-rw-r--r--erts/emulator/drivers/common/gzio.c712
-rw-r--r--erts/emulator/drivers/common/gzio.h8
-rw-r--r--erts/emulator/drivers/common/inet_drv.c382
-rw-r--r--erts/emulator/drivers/unix/unix_efile.c1102
-rw-r--r--erts/emulator/drivers/win32/win_efile.c2058
-rw-r--r--erts/emulator/nifs/common/prim_file_nif.c1237
-rw-r--r--erts/emulator/nifs/common/prim_file_nif.h240
-rw-r--r--erts/emulator/nifs/unix/unix_prim_file.c957
-rw-r--r--erts/emulator/nifs/win32/win_prim_file.c1427
-rw-r--r--erts/preloaded/ebin/erl_prim_loader.beambin54704 -> 54268 bytes
-rw-r--r--erts/preloaded/ebin/init.beambin50348 -> 50472 bytes
-rw-r--r--erts/preloaded/ebin/prim_file.beambin43956 -> 27428 bytes
-rw-r--r--erts/preloaded/ebin/prim_inet.beambin75988 -> 77928 bytes
-rw-r--r--erts/preloaded/src/Makefile6
-rw-r--r--erts/preloaded/src/erl_prim_loader.erl20
-rw-r--r--erts/preloaded/src/init.erl5
-rw-r--r--erts/preloaded/src/prim_file.erl1991
-rw-r--r--erts/preloaded/src/prim_inet.erl51
-rw-r--r--lib/kernel/src/Makefile17
-rw-r--r--lib/kernel/src/file.erl40
-rw-r--r--lib/kernel/src/file_int.hrl33
-rw-r--r--lib/kernel/src/file_io_server.erl52
-rw-r--r--lib/kernel/src/file_server.erl137
-rw-r--r--lib/kernel/src/inet_int.hrl8
-rw-r--r--lib/kernel/src/kernel.app.src7
-rw-r--r--lib/kernel/src/raw_file_io.erl75
-rw-r--r--lib/kernel/src/raw_file_io_compressed.erl134
-rw-r--r--lib/kernel/src/raw_file_io_deflate.erl159
-rw-r--r--lib/kernel/src/raw_file_io_delayed.erl320
-rw-r--r--lib/kernel/src/raw_file_io_inflate.erl261
-rw-r--r--lib/kernel/src/raw_file_io_list.erl128
-rw-r--r--lib/kernel/src/raw_file_io_raw.erl25
50 files changed, 6244 insertions, 9864 deletions
diff --git a/bootstrap/bin/no_dot_erlang.boot b/bootstrap/bin/no_dot_erlang.boot
index a49298d260..2b909efb58 100644
--- a/bootstrap/bin/no_dot_erlang.boot
+++ b/bootstrap/bin/no_dot_erlang.boot
Binary files differ
diff --git a/bootstrap/bin/start.boot b/bootstrap/bin/start.boot
index a49298d260..2b909efb58 100644
--- a/bootstrap/bin/start.boot
+++ b/bootstrap/bin/start.boot
Binary files differ
diff --git a/bootstrap/bin/start_clean.boot b/bootstrap/bin/start_clean.boot
index a49298d260..2b909efb58 100644
--- a/bootstrap/bin/start_clean.boot
+++ b/bootstrap/bin/start_clean.boot
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/file.beam b/bootstrap/lib/kernel/ebin/file.beam
index 6d45887800..6afd377023 100644
--- a/bootstrap/lib/kernel/ebin/file.beam
+++ b/bootstrap/lib/kernel/ebin/file.beam
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/file_io_server.beam b/bootstrap/lib/kernel/ebin/file_io_server.beam
index 74a62d6388..4d13d35d76 100644
--- a/bootstrap/lib/kernel/ebin/file_io_server.beam
+++ b/bootstrap/lib/kernel/ebin/file_io_server.beam
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/file_server.beam b/bootstrap/lib/kernel/ebin/file_server.beam
index d609725fa8..f17b888898 100644
--- a/bootstrap/lib/kernel/ebin/file_server.beam
+++ b/bootstrap/lib/kernel/ebin/file_server.beam
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/kernel.app b/bootstrap/lib/kernel/ebin/kernel.app
index 68dcf6b930..55445f83c6 100644
--- a/bootstrap/lib/kernel/ebin/kernel.app
+++ b/bootstrap/lib/kernel/ebin/kernel.app
@@ -88,6 +88,13 @@
inet_udp,
inet_sctp,
pg2,
+ raw_file_io,
+ raw_file_io_compressed,
+ raw_file_io_deflate,
+ raw_file_io_delayed,
+ raw_file_io_inflate,
+ raw_file_io_list,
+ raw_file_io_raw,
seq_trace,
standard_error,
wrap_log_reader]},
diff --git a/bootstrap/lib/kernel/ebin/raw_file_io.beam b/bootstrap/lib/kernel/ebin/raw_file_io.beam
new file mode 100644
index 0000000000..734a051882
--- /dev/null
+++ b/bootstrap/lib/kernel/ebin/raw_file_io.beam
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/raw_file_io_compressed.beam b/bootstrap/lib/kernel/ebin/raw_file_io_compressed.beam
new file mode 100644
index 0000000000..caa3d9ac6b
--- /dev/null
+++ b/bootstrap/lib/kernel/ebin/raw_file_io_compressed.beam
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/raw_file_io_deflate.beam b/bootstrap/lib/kernel/ebin/raw_file_io_deflate.beam
new file mode 100644
index 0000000000..eb487b1fb0
--- /dev/null
+++ b/bootstrap/lib/kernel/ebin/raw_file_io_deflate.beam
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/raw_file_io_delayed.beam b/bootstrap/lib/kernel/ebin/raw_file_io_delayed.beam
new file mode 100644
index 0000000000..45b9586251
--- /dev/null
+++ b/bootstrap/lib/kernel/ebin/raw_file_io_delayed.beam
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/raw_file_io_inflate.beam b/bootstrap/lib/kernel/ebin/raw_file_io_inflate.beam
new file mode 100644
index 0000000000..f348054060
--- /dev/null
+++ b/bootstrap/lib/kernel/ebin/raw_file_io_inflate.beam
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/raw_file_io_list.beam b/bootstrap/lib/kernel/ebin/raw_file_io_list.beam
new file mode 100644
index 0000000000..283f3b892d
--- /dev/null
+++ b/bootstrap/lib/kernel/ebin/raw_file_io_list.beam
Binary files differ
diff --git a/bootstrap/lib/kernel/ebin/raw_file_io_raw.beam b/bootstrap/lib/kernel/ebin/raw_file_io_raw.beam
new file mode 100644
index 0000000000..b47a67055d
--- /dev/null
+++ b/bootstrap/lib/kernel/ebin/raw_file_io_raw.beam
Binary files differ
diff --git a/erts/emulator/Makefile.in b/erts/emulator/Makefile.in
index 9a09a0f6fa..15b65f1c71 100644
--- a/erts/emulator/Makefile.in
+++ b/erts/emulator/Makefile.in
@@ -634,6 +634,7 @@ GENERATE += $(TTF_DIR)/driver_tab.c
PRELOAD_BEAM = $(ERL_TOP)/erts/preloaded/ebin/otp_ring0.beam \
$(ERL_TOP)/erts/preloaded/ebin/erts_code_purger.beam \
$(ERL_TOP)/erts/preloaded/ebin/init.beam \
+ $(ERL_TOP)/erts/preloaded/ebin/prim_buffer.beam \
$(ERL_TOP)/erts/preloaded/ebin/prim_eval.beam \
$(ERL_TOP)/erts/preloaded/ebin/prim_inet.beam \
$(ERL_TOP)/erts/preloaded/ebin/prim_file.beam \
@@ -644,8 +645,7 @@ PRELOAD_BEAM = $(ERL_TOP)/erts/preloaded/ebin/otp_ring0.beam \
$(ERL_TOP)/erts/preloaded/ebin/erts_internal.beam \
$(ERL_TOP)/erts/preloaded/ebin/erl_tracer.beam \
$(ERL_TOP)/erts/preloaded/ebin/erts_literal_area_collector.beam \
- $(ERL_TOP)/erts/preloaded/ebin/erts_dirty_process_code_checker.beam \
- $(ERL_TOP)/erts/preloaded/ebin/prim_buffer.beam
+ $(ERL_TOP)/erts/preloaded/ebin/erts_dirty_process_code_checker.beam
ifeq ($(TARGET),win32)
# On windows the preloaded objects are in a resource object.
@@ -786,6 +786,9 @@ $(OBJDIR)/%.o: drivers/$(ERLANG_OSTYPE)/%.c
$(OBJDIR)/%.o: nifs/common/%.c
$(V_CC) $(CFLAGS) -DLIBSCTP=$(LIBSCTP) $(INCLUDES) -Inifs/common -Inifs/$(ERLANG_OSTYPE) -c $< -o $@
+$(OBJDIR)/%.o: nifs/$(ERLANG_OSTYPE)/%.c
+ $(V_CC) $(CFLAGS) $(INCLUDES) -Inifs/common -Inifs/$(ERLANG_OSTYPE) -I../etc/$(ERLANG_OSTYPE) -c $< -o $@
+
# ----------------------------------------------------------------------
# Specials
#
@@ -874,18 +877,17 @@ RUN_OBJS += \
LTTNG_OBJS = $(OBJDIR)/erlang_lttng.o
NIF_OBJS = \
$(OBJDIR)/erl_tracer_nif.o \
- $(OBJDIR)/zlib_nif.o \
- $(OBJDIR)/prim_buffer_nif.o
+ $(OBJDIR)/prim_buffer_nif.o \
+ $(OBJDIR)/prim_file_nif.o \
+ $(OBJDIR)/zlib_nif.o
ifeq ($(TARGET),win32)
DRV_OBJS = \
$(OBJDIR)/registry_drv.o \
- $(OBJDIR)/efile_drv.o \
$(OBJDIR)/inet_drv.o \
$(OBJDIR)/ram_file_drv.o \
$(OBJDIR)/ttsl_drv.o
OS_OBJS = \
- $(OBJDIR)/win_efile.o \
$(OBJDIR)/win_con.o \
$(OBJDIR)/dll_sys.o \
$(OBJDIR)/driver_tab.o \
@@ -894,7 +896,8 @@ OS_OBJS = \
$(OBJDIR)/sys_time.o \
$(OBJDIR)/sys_interrupt.o \
$(OBJDIR)/sys_env.o \
- $(OBJDIR)/dosmap.o
+ $(OBJDIR)/dosmap.o \
+ $(OBJDIR)/win_prim_file.o
else
OS_OBJS = \
@@ -902,14 +905,13 @@ OS_OBJS = \
$(OBJDIR)/sys_drivers.o \
$(OBJDIR)/sys_uds.o \
$(OBJDIR)/driver_tab.o \
- $(OBJDIR)/unix_efile.o \
+ $(OBJDIR)/elib_memmove.o \
$(OBJDIR)/gzio.o \
- $(OBJDIR)/elib_memmove.o
+ $(OBJDIR)/unix_prim_file.o
OS_OBJS += $(OBJDIR)/sys_float.o \
$(OBJDIR)/sys_time.o
DRV_OBJS = \
- $(OBJDIR)/efile_drv.o \
$(OBJDIR)/inet_drv.o \
$(OBJDIR)/ram_file_drv.o \
$(OBJDIR)/ttsl_drv.o
@@ -1137,6 +1139,7 @@ BEAM_SRC=$(wildcard beam/*.c)
DRV_COMMON_SRC=$(wildcard drivers/common/*.c)
DRV_OSTYPE_SRC=$(wildcard drivers/$(ERLANG_OSTYPE)/*.c)
NIF_COMMON_SRC=$(wildcard nifs/common/*.c)
+NIF_OSTYPE_SRC=$(wildcard nifs/$(ERLANG_OSTYPE)/*.c)
ALL_SYS_SRC=$(wildcard sys/$(ERLANG_OSTYPE)/*.c) $(wildcard sys/common/*.c)
# We use $(shell ls) here instead of wildcard as $(wildcard ) resolved at
# loadtime of the makefile and at that time these files are not generated yet.
@@ -1149,7 +1152,10 @@ ifeq ($(TARGET),win32)
#DEP_CC=$(EMU_CC)
DEP_CC=$(CC)
-DEP_FLAGS=-MM $(subst -O2,,$(CFLAGS)) $(INCLUDES) -I../etc/win32 -Idrivers/common -Idrivers/$(ERLANG_OSTYPE)
+DEP_FLAGS=-MM $(subst -O2,,$(CFLAGS)) $(INCLUDES) -I../etc/win32 \
+ -Idrivers/common -Idrivers/$(ERLANG_OSTYPE) \
+ -Inifs/common -Inifs/$(ERLANG_OSTYPE)
+
# ifeq (@MIXED_CYGWIN_VC@,yes)
# VC++ used for compiling. If __GNUC__ is defined we will include
# other headers then when compiling which will result in faulty
@@ -1169,7 +1175,9 @@ MG_FLAG=-MG
endif
DEP_CC=$(CC)
-DEP_FLAGS=-MM $(MG_FLAG) $(CFLAGS) $(INCLUDES) -Inifs/common -Idrivers/common -Idrivers/$(ERLANG_OSTYPE)
+DEP_FLAGS=-MM $(MG_FLAG) $(CFLAGS) $(INCLUDES) \
+ -Idrivers/common -Idrivers/$(ERLANG_OSTYPE) \
+ -Inifs/common -Inifs/$(ERLANG_OSTYPE)
SYS_SRC=$(ALL_SYS_SRC)
endif
@@ -1200,6 +1208,8 @@ $(TTF_DIR)/depend.mk: $(TTF_DIR)/GENERATED $(PRELOAD_SRC)
| $(SED_DEPEND) >> $(TTF_DIR)/depend.mk
$(V_at)$(DEP_CC) $(DEP_FLAGS) $(NIF_COMMON_SRC) \
| $(SED_DEPEND) >> $(TTF_DIR)/depend.mk
+ $(V_at)$(DEP_CC) $(DEP_FLAGS) -I../etc/$(ERLANG_OSTYPE) $(NIF_OSTYPE_SRC) \
+ | $(SED_DEPEND) >> $(TTF_DIR)/depend.mk
$(V_at)$(DEP_CC) $(DEP_FLAGS) $(SYS_SRC) \
| $(SED_DEPEND) >> $(TTF_DIR)/depend.mk
$(V_at)$(DEP_CC) $(DEP_FLAGS) $(TARGET_SRC) \
diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c
index 88716d478d..64950fc252 100644
--- a/erts/emulator/beam/erl_lock_check.c
+++ b/erts/emulator/beam/erl_lock_check.c
@@ -109,7 +109,6 @@ static erts_lc_lock_order_t erts_lock_order[] = {
{ "fun_tab", NULL },
{ "environ", NULL },
{ "release_literal_areas", NULL },
- { "efile_drv", "address" },
{ "drv_ev_state_grow", NULL, },
{ "drv_ev_state", "address" },
{ "safe_hash", "address" },
@@ -170,9 +169,6 @@ static erts_lc_lock_order_t erts_lock_order[] = {
{ "save_ops_lock", NULL },
#endif
#endif
-#ifdef USE_VM_PROBES
- { "efile_drv dtrace mutex", NULL },
-#endif
{ "mtrace_buf", NULL },
{ "os_monotonic_time", NULL },
{ "erts_alloc_hard_debug", NULL },
diff --git a/erts/emulator/drivers/common/efile_drv.c b/erts/emulator/drivers/common/efile_drv.c
deleted file mode 100644
index 4e1d2f0d7f..0000000000
--- a/erts/emulator/drivers/common/efile_drv.c
+++ /dev/null
@@ -1,4295 +0,0 @@
-/*
- * %CopyrightBegin%
- *
- * Copyright Ericsson AB 1996-2017. All Rights Reserved.
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- *
- * %CopyrightEnd%
- */
-/*
- * Purpose: Provides file and directory operations.
- *
- * This file is generic, and does the work of decoding the commands
- * and encoding the responses. System-specific functions are found in
- * the unix_efile.c and win_efile.c files.
- */
-
-/* Operations */
-
-#define FILE_OPEN 1 /* Essential for startup */
-#define FILE_READ 2
-#define FILE_LSEEK 3
-#define FILE_WRITE 4
-#define FILE_FSTAT 5 /* Essential for startup */
-#define FILE_PWD 6 /* Essential for startup */
-#define FILE_READDIR 7 /* Essential for startup */
-#define FILE_CHDIR 8
-#define FILE_FSYNC 9
-#define FILE_MKDIR 10
-#define FILE_DELETE 11
-#define FILE_RENAME 12
-#define FILE_RMDIR 13
-#define FILE_TRUNCATE 14
-#define FILE_READ_FILE 15 /* Essential for startup */
-#define FILE_WRITE_INFO 16
-#define FILE_LSTAT 19
-#define FILE_READLINK 20
-#define FILE_LINK 21
-#define FILE_SYMLINK 22
-#define FILE_CLOSE 23
-#define FILE_PWRITEV 24
-#define FILE_PREADV 25
-#define FILE_SETOPT 26
-#define FILE_IPREAD 27
-#define FILE_ALTNAME 28
-#define FILE_READ_LINE 29
-#define FILE_FDATASYNC 30
-#define FILE_FADVISE 31
-#define FILE_SENDFILE 32
-#define FILE_FALLOCATE 33
-#define FILE_CLOSE_ON_PORT_EXIT 34
-/* Return codes */
-
-#define FILE_RESP_OK 0
-#define FILE_RESP_ERROR 1
-#define FILE_RESP_DATA 2
-#define FILE_RESP_NUMBER 3
-#define FILE_RESP_INFO 4
-#define FILE_RESP_NUMERR 5
-#define FILE_RESP_LDATA 6
-#define FILE_RESP_N2DATA 7
-#define FILE_RESP_EOF 8
-#define FILE_RESP_FNAME 9
-#define FILE_RESP_ALL_DATA 10
-#define FILE_RESP_LFNAME 11
-
-/* Options */
-
-#define FILE_OPT_DELAYED_WRITE 0
-#define FILE_OPT_READ_AHEAD 1
-
-/* IPREAD variants */
-
-#define IPREAD_S32BU_P32BU 0
-
-/* Limits */
-
-#define FILE_SEGMENT_READ (256*1024)
-#define FILE_SEGMENT_WRITE (256*1024)
-
-/* Internal */
-
-/* Set to 1 to test having read_ahead implicitly for read_line */
-#define ALWAYS_READ_LINE_AHEAD 0
-
-
-/* Must not be possible to get from malloc()! */
-#define FILE_FD_INVALID ((Sint)(-1))
-
-#ifdef HAVE_CONFIG_H
-# include "config.h"
-#endif
-
-#include <ctype.h>
-#include <sys/types.h>
-#include <stdlib.h>
-
-/* Need (NON)BLOCKING macros for sendfile */
-#ifndef WANT_NONBLOCKING
-#define WANT_NONBLOCKING
-#endif
-
-#include "sys.h"
-
-#include "erl_driver.h"
-#include "erl_efile.h"
-#include "erl_threads.h"
-#include "gzio.h"
-#include "dtrace-wrapper.h"
-
-
-static ErlDrvSysInfo sys_info;
-
-/* For explanation of this var, see comment for same var in erl_async.c */
-static unsigned gcc_optimizer_hack = 0;
-
-#ifdef USE_VM_PROBES
-
-#define DTRACE_EFILE_BUFSIZ 128
-
-#define DTRACE_INVOKE_SETUP(op) \
- do { DTRACE3(efile_drv_int_entry, d->sched_i1, d->sched_i2, op); } while (0)
-#define DTRACE_INVOKE_SETUP_BY_NAME(op) \
- struct t_data *d = (struct t_data *) data ; \
- DTRACE_INVOKE_SETUP(op)
-#define DTRACE_INVOKE_RETURN(op) \
- do { DTRACE3(efile_drv_int_return, d->sched_i1, d->sched_i2, \
- op); } while (0) ; gcc_optimizer_hack++ ;
-
-/* Assign human-friendlier id numbers to scheduler & I/O worker threads */
-int dt_driver_idnum = 0;
-int dt_driver_io_worker_base = 5000;
-erts_mtx_t dt_driver_mutex;
-pthread_key_t dt_driver_key;
-
-typedef struct {
- int thread_num;
- Uint64 tag;
-} dt_private;
-
-dt_private *get_dt_private(int);
-#else /* USE_VM_PROBES */
-#define DTRACE_INVOKE_SETUP(op) do {} while (0)
-#define DTRACE_INVOKE_SETUP_BY_NAME(op) do {} while (0)
-#define DTRACE_INVOKE_RETURN(op) do {} while (0)
-#endif /* USE_VM_PROBES */
-
-/* #define TRACE 1 */
-#ifdef TRACE
-# define TRACE_C(c) do { putchar(c); fflush(stdout); } while (0)
-# define TRACE_S(s) do { fputs((s), stdout); fflush(stdout); } while (0)
-# define TRACE_F(args) do { printf args ;fflush(stdout); } while (0)
-#else
-# define TRACE_C(c) ((void)(0))
-# define TRACE_S(s) ((void)(0))
-# define TRACE_F(args) ((void)(0))
-#endif
-
-
-#define THRDS_AVAILABLE (sys_info.async_threads > 0)
-#ifdef HARDDEBUG /* HARDDEBUG in io.c is expected too */
-#define TRACE_DRIVER fprintf(stderr, "Efile: ")
-#else
-#define TRACE_DRIVER
-#endif
-#define MUTEX_INIT(m, p) do { IF_THRDS { TRACE_DRIVER; (m = driver_pdl_create(p)); } } while (0)
-#define MUTEX_LOCK(m) do { IF_THRDS { TRACE_DRIVER; driver_pdl_lock(m); } } while (0)
-#define MUTEX_UNLOCK(m) do { IF_THRDS { TRACE_DRIVER; driver_pdl_unlock(m); } } while (0)
-#define IF_THRDS if (THRDS_AVAILABLE)
-
-
-#define SENDFILE_FLGS_USE_THREADS (1 << 0)
-/**
- * On DARWIN sendfile can deadlock with close if called in
- * different threads. So until Apple fixes so that sendfile
- * is not buggy we disable usage of the async pool for
- * DARWIN. The testcase t_sendfile_crashduring reproduces
- * this error when using +A 10 and enabling SENDFILE_FLGS_USE_THREADS.
- */
-#if defined(__APPLE__) && defined(__MACH__)
-#define USE_THRDS_FOR_SENDFILE(DATA) 0
-#else
-#define USE_THRDS_FOR_SENDFILE(DATA) (DATA->flags & SENDFILE_FLGS_USE_THREADS)
-#endif /* defined(__APPLE__) && defined(__MACH__) */
-
-
-
-#if 0
-/* Experimental, for forcing all file operations to use the same thread. */
- static unsigned file_fixed_key = 1;
-# define KEY(desc) (&file_fixed_key)
-#else
-# define KEY(desc) (&(desc)->key)
-#endif
-
-#ifndef MAX
-# define MAX(x, y) (((x) > (y)) ? (x) : (y))
-#endif
-
-#ifdef FILENAMES_16BIT
-#ifdef USE_VM_PROBES
-#error 16bit characters in filenames and dtrace in combination is not supported.
-#endif
-# define FILENAME_BYTELEN(Str) filename_len_16bit(Str)
-# define FILENAME_COPY(To,From) filename_cpy_16bit((To),(From))
-# define FILENAME_CHARSIZE 2
-
- static int filename_len_16bit(char *str)
- {
- char *p = str;
- while(*p != '\0' || p[1] != '\0') {
- p += 2;
- }
- return (p - str);
- }
-
- static void filename_cpy_16bit(char *to, char *from)
- {
- while(*from != '\0' || from[1] != '\0') {
- *to++ = *from++;
- *to++ = *from++;
- }
- *to++ = *from++;
- *to++ = *from++;
- }
-
-#else
-# define FILENAME_BYTELEN(Str) strlen(Str)
-# define FILENAME_COPY(To,From) strcpy(To,From)
-# define FILENAME_CHARSIZE 1
-#endif
-
-#if (MAXPATHLEN+1)*FILENAME_CHARSIZE+1 > BUFSIZ
-# define RESBUFSIZE ((MAXPATHLEN+1)*FILENAME_CHARSIZE+1)
-#else
-# define RESBUFSIZE BUFSIZ
-#endif
-
-#define READDIR_CHUNKS (5)
-
-
-
-#if ALWAYS_READ_LINE_AHEAD
-#define DEFAULT_LINEBUF_SIZE 2048
-#else
-#define DEFAULT_LINEBUF_SIZE 512 /* Small, it's usually discarded anyway */
-#endif
-
-typedef unsigned char uchar;
-
-static ErlDrvData file_start(ErlDrvPort port, char* command);
-static int file_init(void);
-static void file_stop(ErlDrvData);
-static void file_output(ErlDrvData, char* buf, ErlDrvSizeT len);
-static ErlDrvSSizeT file_control(ErlDrvData, unsigned int command,
- char* buf, ErlDrvSizeT len,
- char **rbuf, ErlDrvSizeT rlen);
-static void file_timeout(ErlDrvData);
-static void file_outputv(ErlDrvData, ErlIOVec*);
-static void file_async_ready(ErlDrvData, ErlDrvThreadData);
-static void file_flush(ErlDrvData);
-
-#ifdef HAVE_SENDFILE
-static void file_ready_output(ErlDrvData data, ErlDrvEvent event);
-static void file_stop_select(ErlDrvEvent event, void* _);
-#endif /* HAVE_SENDFILE */
-
-
-enum e_timer {timer_idle, timer_again, timer_write};
-#ifdef HAVE_SENDFILE
-enum e_sendfile {sending, not_sending};
-#define SENDFILE_USE_THREADS (1 << 0)
-#endif /* HAVE_SENDFILE */
-
-struct t_data;
-
-typedef struct {
- SWord fd;
- ErlDrvPort port;
- unsigned int key; /* Async queue key */
- unsigned flags; /* Original flags from FILE_OPEN. */
- void (*invoke)(void *);
- struct t_data *d;
- void (*free)(void *);
- struct t_data *cq_head; /* Queue of incoming commands */
- struct t_data *cq_tail; /* -""- */
- enum e_timer timer_state;
-#ifdef HAVE_SENDFILE
- enum e_sendfile sendfile_state;
-#endif /* HAVE_SENDFILE */
- size_t read_bufsize;
- ErlDrvBinary *read_binp;
- size_t read_offset;
- size_t read_size;
- size_t write_bufsize;
- unsigned long write_delay;
- int write_error;
- Efile_error write_errInfo;
- ErlDrvPDL q_mtx; /* Mutex for the driver queue, known by the emulator. Also used for
- mutual exclusion when accessing field(s) below. */
- size_t write_buffered;
-#ifdef USE_VM_PROBES
- int idnum; /* Unique ID # for this driver thread/desc */
- char port_str[DTRACE_TERM_BUF_SIZE];
-#endif
-} file_descriptor;
-
-
-static int reply_error(file_descriptor*, Efile_error* errInfo);
-
-struct erl_drv_entry efile_driver_entry = {
- file_init,
- file_start,
- file_stop,
- file_output,
- NULL,
-#ifdef HAVE_SENDFILE
- file_ready_output,
-#else
- NULL,
-#endif /* HAVE_SENDFILE */
- "efile",
- NULL,
- NULL,
- file_control,
- file_timeout,
- file_outputv,
- file_async_ready,
- file_flush,
- NULL,
- NULL,
- ERL_DRV_EXTENDED_MARKER,
- ERL_DRV_EXTENDED_MAJOR_VERSION,
- ERL_DRV_EXTENDED_MINOR_VERSION,
- ERL_DRV_FLAG_USE_PORT_LOCKING,
- NULL,
- NULL,
-#ifdef HAVE_SENDFILE
- file_stop_select
-#else
- NULL
-#endif /* HAVE_SENDFILE */
-};
-
-
-
-static int thread_short_circuit;
-
-#define DRIVER_ASYNC(level, desc, f_invoke, data, f_free) \
-if (thread_short_circuit >= (level)) { \
- (*(f_invoke))(data); \
- file_async_ready((ErlDrvData)(desc), (data)); \
-} else { \
- driver_async((desc)->port, KEY(desc), (f_invoke), (data), (f_free)); \
-}
-
-
-
-struct t_pbuf_spec {
- Sint64 offset;
- size_t size;
-};
-
-struct t_pwritev {
- ErlDrvPort port;
- ErlDrvPDL q_mtx;
- size_t size;
- unsigned cnt;
- unsigned n;
- struct t_pbuf_spec specs[1];
-};
-
-struct t_preadv {
- ErlIOVec eiov;
- unsigned n;
- unsigned cnt;
- size_t size;
- Sint64 offsets[1];
-};
-
-#define READDIR_BUFSIZE (8*1024)*READDIR_CHUNKS
-#if READDIR_BUFSIZE < (1 + (2 + MAXPATHLEN)*FILENAME_CHARSIZE*READDIR_CHUNKS)
-# undef READDIR_BUFSIZE
-# define READDIR_BUFSIZE (1 + (2 + MAXPATHLEN)*FILENAME_CHARSIZE*READDIR_CHUNKS)
-#endif
-
-struct t_readdir_buf {
- struct t_readdir_buf *next;
- size_t n;
- char buf[READDIR_BUFSIZE];
-};
-
-struct t_data
-{
- struct t_data *next;
- int command;
- int level;
- void (*invoke)(void *);
- void (*free)(void *);
- void *data_to_free; /* used by FILE_CLOSE_ON_PORT_EXIT only */
- int again;
- int reply;
-#ifdef USE_VM_PROBES
- int sched_i1;
- Uint64 sched_i2;
- char sched_utag[DTRACE_EFILE_BUFSIZ+1];
-#endif
- int result_ok;
- Efile_error errInfo;
- int flags;
- SWord fd;
- int is_fd_unused;
- /**/
- Efile_info info;
- EFILE_DIR_HANDLE dir_handle; /* Handle to open directory. */
- ErlDrvBinary *bin;
- int drive;
- size_t n;
- /*off_t offset;*/
- /*size_t bytesRead; Bytes read from the file. */
- /**/
- union {
- struct {
- Sint64 offset;
- int origin;
- Sint64 location;
- } lseek;
- struct {
- ErlDrvPort port;
- ErlDrvPDL q_mtx;
- size_t size;
- size_t reply_size;
- } writev;
- struct t_pwritev pwritev;
- struct t_preadv preadv;
- struct {
- ErlDrvBinary *binp;
- size_t bin_offset;
- size_t bin_size;
- size_t size;
- } read;
- struct {
- ErlDrvBinary *binp; /* in - out */
- size_t read_offset; /* in - out */
- size_t read_size; /* in - out */
- size_t nl_pos; /* out */
- short nl_skip; /* out, 0 or 1 */
-#if !ALWAYS_READ_LINE_AHEAD
- short read_ahead; /* in, bool */
-#endif
- } read_line;
- struct {
- ErlDrvBinary *binp;
- int size;
- int offset;
- } read_file;
- struct {
- struct t_readdir_buf *first_buf;
- struct t_readdir_buf *last_buf;
- } read_dir;
- struct {
- Sint64 offset;
- Sint64 length;
- int advise;
- } fadvise;
-#ifdef HAVE_SENDFILE
- struct {
- ErlDrvPort port;
- ErlDrvPDL q_mtx;
- int out_fd;
- off_t offset;
- Uint64 nbytes;
- Uint64 written;
- } sendfile;
-#endif /* HAVE_SENDFILE */
- struct {
- Sint64 offset;
- Sint64 length;
- } fallocate;
- } c;
- char b[1];
-};
-
-#define EF_ALLOC(S) driver_alloc((S))
-#define EF_REALLOC(P, S) driver_realloc((P), (S))
-#define EF_SAFE_ALLOC(S) ef_safe_alloc((S))
-#define EF_SAFE_REALLOC(P, S) ef_safe_realloc((P), (S))
-#define EF_FREE(P) do { if((P)) driver_free((P)); } while(0)
-
-static void *ef_safe_alloc(Uint s)
-{
- void *p = EF_ALLOC(s);
- if (!p) erts_exit(ERTS_ERROR_EXIT, "efile drv: Can't allocate %lu bytes of memory\n", (unsigned long)s);
- return p;
-}
-
-/*********************************************************************
- * ErlIOVec manipulation functions.
- */
-
-/* char EV_CHAR_P(ErlIOVec *ev, int p, int q) */
-#define EV_CHAR_P(ev, p, q) \
- (((char *)(ev)->iov[q].iov_base) + (p))
-
-/* int EV_GET_CHAR(ErlIOVec *ev, char *p, int *pp, int *qp) */
-#define EV_GET_CHAR(ev, p, pp, qp) efile_ev_get_char(ev, p ,pp, qp)
-static int
-efile_ev_get_char(ErlIOVec *ev, char *p, size_t *pp, size_t *qp) {
- if (*pp + 1 <= ev->iov[*qp].iov_len) {
- *p = *EV_CHAR_P(ev, *pp, *qp);
- if (*pp + 1 < ev->iov[*qp].iov_len)
- *pp += 1;
- else {
- *qp += 1;
- *pp = 0;
- }
- return !0;
- }
- return 0;
-}
-
-/* Uint32 EV_UINT32(ErlIOVec *ev, int p, int q)*/
-#define EV_UINT32(ev, p, q) \
- ((Uint32) ((unsigned char *)(ev)->iov[q].iov_base)[p])
-
-/* int EV_GET_UINT32(ErlIOVec *ev, Uint32 *p, int *pp, int *qp) */
-#define EV_GET_UINT32(ev, p, pp, qp) efile_ev_get_uint32(ev, p, pp, qp)
-static int
-efile_ev_get_uint32(ErlIOVec *ev, Uint32 *p, size_t *pp, size_t *qp) {
- if (*pp + 4 <= ev->iov[*qp].iov_len) {
- *p = (EV_UINT32(ev, *pp, *qp) << 24)
- | (EV_UINT32(ev, *pp + 1, *qp) << 16)
- | (EV_UINT32(ev, *pp + 2, *qp) << 8)
- | (EV_UINT32(ev, *pp + 3, *qp));
- if (*pp + 4 < ev->iov[*qp].iov_len)
- *pp += 4;
- else {
- *qp += 1;
- *pp = 0;
- }
- return !0;
- }
- return 0;
-}
-
-/* Uint64 EV_UINT64(ErlIOVec *ev, int p, int q)*/
-#define EV_UINT64(ev, p, q) \
- ((Uint64) ((unsigned char *)(ev)->iov[q].iov_base)[p])
-
-/* int EV_GET_UINT64(ErlIOVec *ev, Uint64 *p, int *pp, int *qp) */
-#define EV_GET_UINT64(ev, p, pp, qp) efile_ev_get_uint64(ev, p, pp, qp)
-static int
-efile_ev_get_uint64(ErlIOVec *ev, Uint64 *p, size_t *pp, size_t *qp) {
- if (*pp + 8 <= ev->iov[*qp].iov_len) {
- *p = (EV_UINT64(ev, *pp, *qp) << 56)
- | (EV_UINT64(ev, *pp + 1, *qp) << 48)
- | (EV_UINT64(ev, *pp + 2, *qp) << 40)
- | (EV_UINT64(ev, *pp + 3, *qp) << 32)
- | (EV_UINT64(ev, *pp + 4, *qp) << 24)
- | (EV_UINT64(ev, *pp + 5, *qp) << 16)
- | (EV_UINT64(ev, *pp + 6, *qp) << 8)
- | (EV_UINT64(ev, *pp + 7, *qp));
- if (*pp + 8 < ev->iov[*qp].iov_len)
- *pp += 8;
- else {
- *qp += 1;
- *pp = 0;
- }
- return !0;
- }
- return 0;
-}
-
-/* int EV_GET_SINT64(ErlIOVec *ev, Uint64 *p, int *pp, int *qp) */
-#define EV_GET_SINT64(ev, p, pp, qp) efile_ev_get_sint64(ev, p, pp, qp)
-static int
-efile_ev_get_sint64(ErlIOVec *ev, Sint64 *p, size_t *pp, size_t *qp) {
- Uint64 *tmp = (Uint64*)p;
- return EV_GET_UINT64(ev, tmp, pp, qp);
-}
-
-#if 0
-
-static void ev_clear(ErlIOVec *ev) {
- ASSERT(ev);
- ev->size = 0;
- ev->vsize = 0;
- ev->iov = NULL;
- ev->binv = NULL;
-}
-
-/* Assumes that ->iov and ->binv were allocated with sys_alloc().
- */
-static void ev_free(ErlIOVec *ev) {
- if (! ev) {
- return;
- }
- if (ev->vsize > 0) {
- int i;
- ASSERT(ev->iov);
- ASSERT(ev->binv);
- for (i = 0; i < ev->vsize; i++) {
- if (ev->binv[i]) {
- driver_free_binary(ev->binv[i]);
- }
- }
- EF_FREE(ev->iov);
- EF_FREE(ev->binv);
- }
-}
-
-/* Copy the contents from source to dest.
- * Data in binaries is not copied, just the pointers;
- * and refc is incremented.
- */
-static ErlIOVec *ev_copy(ErlIOVec *dest, ErlIOVec *source) {
- int *ip;
- ASSERT(dest);
- ASSERT(source);
- if (source->vsize == 0) {
- /* Empty source */
- ev_clear(dest);
- return dest;
- }
- /* Allocate ->iov and ->binv */
- dest->iov = EF_ALLOC(sizeof(*dest->iov) * source->vsize);
- if (! dest->iov) {
- return NULL;
- }
- dest->binv = EF_ALLOC(sizeof(*dest->binv) * source->vsize);
- if (! dest->binv) {
- EF_FREE(dest->iov);
- return NULL;
- }
- dest->size = source->size;
- /* Copy one vector element at the time.
- * Use *ip as an alias for dest->vsize to improve readabiliy.
- * Keep dest consistent in every iteration by using
- * dest->vsize==*ip as loop variable.
- */
- for (ip = &dest->vsize, *ip = 0; *ip < source->vsize; (*ip)++) {
- if (source->iov[*ip].iov_len == 0) {
- /* Empty vector element */
- dest->iov[*ip].iov_len = 0;
- dest->iov[*ip].iov_base = NULL;
- dest->binv[*ip] = NULL;
- } else {
- /* Non empty vector element */
- if (source->binv[*ip]) {
- /* Contents in binary - copy pointers and increment refc */
- dest->iov[*ip] = source->iov[*ip];
- dest->binv[*ip] = source->binv[*ip];
- driver_binary_inc_refc(source->binv[*ip]);
- } else {
- /* Contents not in binary - allocate new binary and copy data */
- if (! (dest->binv[*ip] =
- driver_alloc_binary(source->iov[*ip].iov_len))) {
- goto failed;
- }
- sys_memcpy(dest->binv[*ip]->orig_bytes,
- source->iov[*ip].iov_base,
- source->iov[*ip].iov_len);
- dest->iov[*ip].iov_base = dest->binv[*ip]->orig_bytes;
- dest->iov[*ip].iov_len = source->iov[*ip].iov_len;
- }
- }
- }
- return dest;
- failed:
- ev_free(dest);
- return NULL;
-}
-
-#endif
-
-
-
-/*********************************************************************
- * Command queue functions
- */
-
-static void cq_enq(file_descriptor *desc, struct t_data *d) {
- ASSERT(d);
- if (desc->cq_head) {
- ASSERT(desc->cq_tail);
- ASSERT(!desc->cq_tail->next);
- desc->cq_tail = desc->cq_tail->next = d;
- } else {
- ASSERT(desc->cq_tail == NULL);
- desc->cq_head = desc->cq_tail = d;
- }
- d->next = NULL;
-}
-
-static struct t_data *cq_deq(file_descriptor *desc) {
- struct t_data *d = desc->cq_head;
- ASSERT(d || (!d && !desc->cq_tail));
- if (d) {
- ASSERT(!d->next || (d->next && desc->cq_tail != d));
- if ((desc->cq_head = d->next) == NULL) {
- ASSERT(desc->cq_tail == d);
- desc->cq_tail = NULL;
- }
- }
- return d;
-}
-
-
-/*********************************************************************
- * Driver entry point -> init
- */
-static int
-file_init(void)
-{
- char buf[21]; /* enough to hold any 64-bit integer */
- size_t bufsz = sizeof(buf);
- thread_short_circuit = (erl_drv_getenv("ERL_EFILE_THREAD_SHORT_CIRCUIT",
- buf,
- &bufsz) == 0
- ? atoi(buf)
- : 0);
- driver_system_info(&sys_info, sizeof(ErlDrvSysInfo));
-
- /* run initiation of efile_driver if needed */
- efile_init();
-
-#ifdef USE_VM_PROBES
- erts_mtx_init(&dt_driver_mutex, "efile_drv dtrace mutex", NIL,
- ERTS_LOCK_FLAGS_PROPERTY_STATIC | ERTS_LOCK_FLAGS_CATEGORY_IO);
- pthread_key_create(&dt_driver_key, NULL);
-#endif /* USE_VM_PROBES */
-
- return 0;
-}
-
-
-/*********************************************************************
- * Driver entry point -> start
- */
-static ErlDrvData
-file_start(ErlDrvPort port, char* command)
-
-{
- file_descriptor* desc;
-
- if ((desc = (file_descriptor*) EF_ALLOC(sizeof(file_descriptor)))
- == NULL) {
- errno = ENOMEM;
- return ERL_DRV_ERROR_ERRNO;
- }
- desc->fd = FILE_FD_INVALID;
- desc->port = port;
- desc->key = driver_async_port_key(port);
- desc->flags = 0;
- desc->invoke = NULL;
- desc->d = NULL;
- desc->free = NULL;
- desc->cq_head = NULL;
- desc->cq_tail = NULL;
- desc->timer_state = timer_idle;
-#ifdef HAVE_SENDFILE
- desc->sendfile_state = not_sending;
-#endif
- desc->read_bufsize = 0;
- desc->read_binp = NULL;
- desc->read_offset = 0;
- desc->read_size = 0;
- desc->write_delay = 0L;
- desc->write_bufsize = 0;
- desc->write_error = 0;
- MUTEX_INIT(desc->q_mtx, port); /* Refc is one, referenced by emulator now */
- desc->write_buffered = 0;
-#ifdef USE_VM_PROBES
- dtrace_drvport_str(port, desc->port_str);
- get_dt_private(0); /* throw away return value */
-#endif /* USE_VM_PROBES */
- return (ErlDrvData) desc;
-}
-
-static void do_close(int flags, SWord fd) {
- if (flags & EFILE_COMPRESSED) {
- erts_gzclose((ErtsGzFile)(fd));
- } else {
- efile_closefile((int) fd);
- }
-}
-
-static void invoke_close(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- DTRACE_INVOKE_SETUP(FILE_CLOSE);
- d->again = 0;
- do_close(d->flags, d->fd);
- DTRACE_INVOKE_RETURN(FILE_CLOSE);
-}
-
-static void free_data(void *data)
-{
- struct t_data *d = (struct t_data *) data;
-
- switch (d->command) {
- case FILE_OPEN:
- if (d->is_fd_unused && d->fd != FILE_FD_INVALID) {
- /* This is OK to do in scheduler thread because there can be no async op
- ongoing for this fd here, as we exited during async open.
- Ideally, this close should happen in an async thread too, but that would
- require a substantial rewrite, as we are here because of a dead port and
- cannot schedule async jobs for that port any more... */
- do_close(d->flags, d->fd);
- }
- break;
- case FILE_CLOSE_ON_PORT_EXIT:
- EF_FREE(d->data_to_free);
- break;
- }
-
- EF_FREE(data);
-}
-
-
-/*
- * Sends back an error reply to Erlang.
- */
-
-static void reply_posix_error(file_descriptor *desc, int posix_errno) {
- char response[256]; /* Response buffer. */
- char* s;
- char* t;
-
- /*
- * Contents of buffer sent back:
- *
- * +-----------------------------------------+
- * | FILE_RESP_ERROR | Posix error id string |
- * +-----------------------------------------+
- */
-
- TRACE_C('E');
-
- response[0] = FILE_RESP_ERROR;
- for (s = erl_errno_id(posix_errno), t = response+1; *s; s++, t++)
- *t = tolower(*s);
- driver_output2(desc->port, response, t-response, NULL, 0);
-}
-
-static void reply_Uint_posix_error(file_descriptor *desc, Uint num,
- int posix_errno) {
- char response[256]; /* Response buffer. */
- char* s;
- char* t;
-
- /*
- * Contents of buffer sent back:
- *
- * +----------------------------------------------------------------------+
- * | FILE_RESP_NUMERR | 64-bit number (big-endian) | Posix error id string |
- * +----------------------------------------------------------------------+
- */
-
- TRACE_C('N');
-
- response[0] = FILE_RESP_NUMERR;
-#if SIZEOF_VOID_P == 4
- put_int32(0, response+1);
-#else
- put_int32(num>>32, response+1);
-#endif
- put_int32((Uint32)num, response+1+4);
- for (s = erl_errno_id(posix_errno), t = response+1+4+4; *s; s++, t++)
- *t = tolower(*s);
- driver_output2(desc->port, response, t-response, NULL, 0);
-}
-
-#ifdef HAVE_SENDFILE
-static void reply_string_error(file_descriptor *desc, char* str) {
- char response[256]; /* Response buffer. */
- char* s;
- char* t;
-
- response[0] = FILE_RESP_ERROR;
- for (s = str, t = response+1; *s; s++, t++)
- *t = tolower(*s);
- driver_output2(desc->port, response, t-response, NULL, 0);
-}
-#endif
-
-static int reply_error(file_descriptor *desc,
- Efile_error *errInfo) /* The error codes. */
-{
- reply_posix_error(desc, errInfo->posix_errno);
- return 0;
-}
-
-static int reply_Uint_error(file_descriptor *desc, Uint num,
- Efile_error *errInfo) /* The error codes. */
-{
- reply_Uint_posix_error(desc, num, errInfo->posix_errno);
- return 0;
-}
-
-static int reply_ok(file_descriptor *desc) {
- char c = FILE_RESP_OK;
-
- driver_output2(desc->port, &c, 1, NULL, 0);
- return 0;
-}
-
-static int reply(file_descriptor *desc, int ok, Efile_error *errInfo) {
- if (!ok) {
- reply_error(desc, errInfo);
- } else {
- TRACE_C('K');
- reply_ok(desc);
- }
- return 0;
-}
-
-static int reply_Uint(file_descriptor *desc, Uint result) {
- char tmp[1+4+4];
-
- /*
- * Contents of buffer sent back:
- *
- * +-----------------------------------------------+
- * | FILE_RESP_NUMBER | 64-bit number (big-endian) |
- * +-----------------------------------------------+
- */
-
- TRACE_C('R');
-
- tmp[0] = FILE_RESP_NUMBER;
-#if SIZEOF_VOID_P == 4
- put_int32(0, tmp+1);
-#else
- put_int32(result>>32, tmp+1);
-#endif
- put_int32((Uint32)result, tmp+1+4);
- driver_output2(desc->port, tmp, sizeof(tmp), NULL, 0);
- return 0;
-}
-
-static int reply_Sint64(file_descriptor *desc, Sint64 result) {
- char tmp[1+4+4];
-
- /*
- * Contents of buffer sent back:
- *
- * +-----------------------------------------------+
- * | FILE_RESP_NUMBER | 64-bit number (big-endian) |
- * +-----------------------------------------------+
- */
-
- TRACE_C('R');
-
- tmp[0] = FILE_RESP_NUMBER;
- put_int64(result, tmp+1);
- driver_output2(desc->port, tmp, sizeof(tmp), NULL, 0);
- return 0;
-}
-
-#if 0
-static void reply_again(file_descriptor *desc) {
- char tmp[1];
- tmp[0] = FILE_RESP_AGAIN;
- driver_output2(desc->port, tmp, sizeof(tmp), NULL, 0);
-}
-#endif
-
-static void reply_ev(file_descriptor *desc, char response, ErlIOVec *ev) {
- char tmp[1];
- /* Data arriving at the Erlang process:
- * [Response, Binary0, Binary1, .... | BinaryN-1]
- */
- tmp[0] = response;
- driver_outputv(desc->port, tmp, sizeof(tmp), ev, 0);
-}
-
-static void reply_data(file_descriptor *desc,
- ErlDrvBinary *binp, size_t offset, size_t len) {
- char header[1+4+4];
- /* Data arriving at the Erlang process:
- * [?FILE_RESP_DATA, 64-bit length (big-endian) | Data]
- */
- header[0] = FILE_RESP_DATA;
-#if SIZEOF_SIZE_T == 4
- put_int32(0, header+1);
-#else
- put_int32(len>>32, header+1);
-#endif
- put_int32((Uint32)len, header+1+4);
- driver_output_binary(desc->port, header, sizeof(header),
- binp, offset, len);
-}
-
-static void reply_buf(file_descriptor *desc, char *buf, size_t len) {
- char header[1+4+4];
- /* Data arriving at the Erlang process:
- * [?FILE_RESP_DATA, 64-bit length (big-endian) | Data]
- */
- header[0] = FILE_RESP_DATA;
-#if SIZEOF_SIZE_T == 4
- put_int32(0, header+1);
-#else
- put_int32(len>>32, header+1);
-#endif
- put_int32((Uint32)len, header+1+4);
- driver_output2(desc->port, header, sizeof(header), buf, len);
-}
-
-static int reply_eof(file_descriptor *desc) {
- char c = FILE_RESP_EOF;
-
- driver_output2(desc->port, &c, 1, NULL, 0);
- return 0;
-}
-
-static void invoke_name(void *data, int (*f)(Efile_error *, char *))
-{
- struct t_data *d = (struct t_data *) data;
- char *name = (char *) d->b;
-
- d->again = 0;
- d->result_ok = (*f)(&d->errInfo, name);
-}
-
-static void invoke_mkdir(void *data)
-{
- DTRACE_INVOKE_SETUP_BY_NAME(FILE_MKDIR);
- invoke_name(data, efile_mkdir);
- DTRACE_INVOKE_RETURN(FILE_MKDIR);
-}
-
-static void invoke_rmdir(void *data)
-{
- DTRACE_INVOKE_SETUP_BY_NAME(FILE_RMDIR);
- invoke_name(data, efile_rmdir);
- DTRACE_INVOKE_RETURN(FILE_RMDIR);
-}
-
-static void invoke_delete_file(void *data)
-{
- DTRACE_INVOKE_SETUP_BY_NAME(FILE_DELETE);
- invoke_name(data, efile_delete_file);
- DTRACE_INVOKE_RETURN(FILE_DELETE);
-}
-
-static void invoke_chdir(void *data)
-{
- DTRACE_INVOKE_SETUP_BY_NAME(FILE_CHDIR);
- invoke_name(data, efile_chdir);
- DTRACE_INVOKE_RETURN(FILE_CHDIR);
-}
-
-static void invoke_fdatasync(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- int fd = (int) d->fd;
- DTRACE_INVOKE_SETUP(FILE_FDATASYNC);
-
- d->again = 0;
- d->result_ok = efile_fdatasync(&d->errInfo, fd);
- DTRACE_INVOKE_RETURN(FILE_FDATASYNC);
-}
-
-static void invoke_fsync(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- int fd = (int) d->fd;
- DTRACE_INVOKE_SETUP(FILE_FSYNC);
-
- d->again = 0;
- d->result_ok = efile_fsync(&d->errInfo, fd);
- DTRACE_INVOKE_RETURN(FILE_FSYNC);
-}
-
-static void invoke_truncate(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- int fd = (int) d->fd;
- DTRACE_INVOKE_SETUP(FILE_TRUNCATE);
-
- d->again = 0;
- d->result_ok = efile_truncate_file(&d->errInfo, &fd, d->flags);
- DTRACE_INVOKE_RETURN(FILE_TRUNCATE);
-}
-
-static void invoke_read(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- int status, segment;
- size_t size, read_size;
- DTRACE_INVOKE_SETUP(FILE_READ);
-
- segment = d->again && d->c.read.bin_size >= 2*FILE_SEGMENT_READ;
- if (segment) {
- size = FILE_SEGMENT_READ;
- } else {
- size = d->c.read.bin_size;
- }
- read_size = size;
- if (d->flags & EFILE_COMPRESSED) {
- read_size = erts_gzread((ErtsGzFile)d->fd,
- d->c.read.binp->orig_bytes + d->c.read.bin_offset,
- size);
- status = (read_size != (size_t) -1);
- if (!status) {
- d->errInfo.posix_errno = EIO;
- }
- } else {
- status = efile_read(&d->errInfo, d->flags, (int) d->fd,
- d->c.read.binp->orig_bytes + d->c.read.bin_offset,
- size,
- &read_size);
- }
- if ( (d->result_ok = status)) {
- ASSERT(read_size <= size);
- d->c.read.bin_offset += read_size;
- if (read_size < size || !segment) {
- d->c.read.bin_size = 0;
- d->again = 0;
- } else {
- d->c.read.bin_size -= read_size;
- }
- } else {
- d->again = 0;
- }
- DTRACE_INVOKE_RETURN(FILE_READ);
-}
-
-static void free_read(void *data)
-{
- struct t_data *d = (struct t_data *) data;
-
- driver_free_binary(d->c.read.binp);
- EF_FREE(d);
-}
-
-static void invoke_read_line(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- int status;
- size_t read_size = 0;
- int local_loop = (d->again == 0);
- DTRACE_INVOKE_SETUP(FILE_READ_LINE);
-
- do {
- size_t size = (d->c.read_line.binp)->orig_size -
- d->c.read_line.read_offset - d->c.read_line.read_size;
- if (size == 0) {
- /* Need more place */
- ErlDrvSizeT need = (d->c.read_line.read_size >= DEFAULT_LINEBUF_SIZE) ?
- d->c.read_line.read_size + DEFAULT_LINEBUF_SIZE : DEFAULT_LINEBUF_SIZE;
- ErlDrvBinary *newbin;
-#if !ALWAYS_READ_LINE_AHEAD
- /* Use read_ahead size if need does not exceed it */
- if (need < (d->c.read_line.binp)->orig_size &&
- d->c.read_line.read_ahead)
- need = (d->c.read_line.binp)->orig_size;
-#endif
- newbin = driver_alloc_binary(need);
- if (newbin == NULL) {
- d->result_ok = 0;
- d->errInfo.posix_errno = ENOMEM;
- d->again = 0;
- break;
- }
- memcpy(newbin->orig_bytes, (d->c.read_line.binp)->orig_bytes + d->c.read_line.read_offset,
- d->c.read_line.read_size);
- driver_free_binary(d->c.read_line.binp);
- d->c.read_line.binp = newbin;
- d->c.read_line.read_offset = 0;
- size = need - d->c.read_line.read_size;
- }
- if (d->flags & EFILE_COMPRESSED) {
- read_size = erts_gzread((ErtsGzFile)d->fd,
- d->c.read_line.binp->orig_bytes +
- d->c.read_line.read_offset + d->c.read_line.read_size,
- size);
- status = (read_size != (size_t) -1);
- if (!status) {
- d->errInfo.posix_errno = EIO;
- }
- } else {
- status = efile_read(&d->errInfo, d->flags, (int) d->fd,
- d->c.read_line.binp->orig_bytes +
- d->c.read_line.read_offset + d->c.read_line.read_size,
- size,
- &read_size);
- }
- if ( (d->result_ok = status)) {
- void *nl_ptr = memchr((d->c.read_line.binp)->orig_bytes +
- d->c.read_line.read_offset + d->c.read_line.read_size,'\n',read_size);
- ASSERT(read_size <= size);
- d->c.read_line.read_size += read_size;
- if (nl_ptr != NULL) {
- /* If found, we're done */
- d->c.read_line.nl_pos = ((char *) nl_ptr) -
- ((char *) ((d->c.read_line.binp)->orig_bytes)) + 1;
- if (d->c.read_line.nl_pos > 1 &&
- *(((char *) nl_ptr) - 1) == '\r') {
- --d->c.read_line.nl_pos;
- *(((char *) nl_ptr) - 1) = '\n';
- d->c.read_line.nl_skip = 1;
- } else {
- d->c.read_line.nl_skip = 0;
- }
- d->again = 0;
-#if !ALWAYS_READ_LINE_AHEAD
- if (!(d->c.read_line.read_ahead)) {
- /* Ouch! Undo buffering... */
- size_t too_much = d->c.read_line.read_size - d->c.read_line.nl_skip -
- (d->c.read_line.nl_pos - d->c.read_line.read_offset);
- d->c.read_line.read_size -= too_much;
- ASSERT(d->c.read_line.read_size >= 0);
- if (d->flags & EFILE_COMPRESSED) {
- Sint64 location = erts_gzseek((ErtsGzFile)d->fd,
- -((Sint64) too_much), EFILE_SEEK_CUR);
- if (location == -1) {
- d->result_ok = 0;
- d->errInfo.posix_errno = errno;
- }
- } else {
- Sint64 location;
- d->result_ok = efile_seek(&d->errInfo, (int) d->fd,
- -((Sint64) too_much), EFILE_SEEK_CUR,
- &location);
- }
- }
-#endif
- break;
- } else if (read_size == 0) {
- d->c.read_line.nl_pos =
- d->c.read_line.read_offset + d->c.read_line.read_size;
- d->c.read_line.nl_skip = 0;
- d->again = 0;
- break;
- }
- } else {
- d->again = 0;
- break;
- }
- } while (local_loop);
- DTRACE_INVOKE_RETURN(FILE_READ_LINE);
-}
-
-static void free_read_line(void *data)
-{
- struct t_data *d = (struct t_data *) data;
-
- driver_free_binary(d->c.read_line.binp);
- EF_FREE(d);
-}
-
-void read_file_zero_size(struct t_data* d);
-#define ZERO_FILE_CHUNK (64 * 1024)
-
-/* [ERL-327] Some special files like /proc/... have reported size 0 */
-void read_file_zero_size(struct t_data* d) {
- size_t total_read_size = 0;
- size_t allocated_size = ZERO_FILE_CHUNK; /* allocd in invoke_read_file */
- for (;;) {
- size_t read_result;
-
- /* Read until we hit EOF (read less than FILE_SEGMENT_READ) */
- d->result_ok = efile_read(&d->errInfo,
- EFILE_MODE_READ,
- (int) d->fd,
- (d->c.read_file.binp->orig_bytes +
- total_read_size),
- ZERO_FILE_CHUNK,
- &read_result);
- if (!d->result_ok) {
- break;
- }
-
- total_read_size += read_result;
- d->c.read_file.offset += read_result;
- if (read_result < ZERO_FILE_CHUNK) {
- break;
- }
-
- /* Grow before the next read call */
- allocated_size = total_read_size + ZERO_FILE_CHUNK;
- d->c.read_file.binp = driver_realloc_binary(d->c.read_file.binp,
- allocated_size);
- }
-
- /* Finalize the memory usage. Hopefully it was read fully on the first
- * go, so the binary allocation overhead becomes:
- * alloc ZERO_FILE_CHUNK (64kb) -> realloc real_size */
- if (allocated_size != total_read_size) {
- d->c.read_file.binp = driver_realloc_binary(d->c.read_file.binp,
- total_read_size);
- }
- d->again = 0;
-}
-
-static void invoke_read_file(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- size_t read_size;
- int chop;
- DTRACE_INVOKE_SETUP(FILE_READ_FILE);
-
- if (! d->c.read_file.binp) { /* First invocation only */
- int fd;
- Sint64 size;
-
- if (! (d->result_ok =
- efile_openfile(&d->errInfo, d->b,
- EFILE_MODE_READ, &fd, &size))) {
- goto done;
- }
- d->fd = fd;
- d->c.read_file.size = (int) size;
-
- /* For zero sized files allocate a reasonable chunk to attempt reading
- * anyway. Note: This will eat ZERO_FILE_CHUNK bytes for any 0 file
- * and free them immediately after (if the file was empty). */
- ERTS_ASSERT(size >= 0);
- d->c.read_file.binp = driver_alloc_binary(size != 0 ? (size_t)size
- : ZERO_FILE_CHUNK);
-
- if (size < 0 || size != d->c.read_file.size || !d->c.read_file.binp) {
- d->result_ok = 0;
- d->errInfo.posix_errno = ENOMEM;
- goto close;
- }
- d->c.read_file.offset = 0;
- }
- /* Invariant: d->c.read_file.size >= d->c.read_file.offset */
-
- if (d->c.read_file.size == 0) {
- read_file_zero_size(d);
- goto close;
- }
-
- read_size = (size_t) (d->c.read_file.size - d->c.read_file.offset);
- if (! read_size) goto close;
- chop = d->again && read_size >= FILE_SEGMENT_READ*2;
- if (chop) read_size = FILE_SEGMENT_READ;
- d->result_ok =
- efile_read(&d->errInfo,
- EFILE_MODE_READ,
- (int) d->fd,
- d->c.read_file.binp->orig_bytes + d->c.read_file.offset,
- read_size,
- &read_size);
- if (d->result_ok) {
- d->c.read_file.offset += read_size;
- if (chop) goto chop_done; /* again */
- }
- close:
- efile_closefile((int) d->fd);
- done:
- d->again = 0;
- chop_done:
- DTRACE_INVOKE_RETURN(FILE_READ_FILE);
-}
-
-static void free_read_file(void *data)
-{
- struct t_data *d = (struct t_data *) data;
-
- if (d->c.read_file.binp) driver_free_binary(d->c.read_file.binp);
- EF_FREE(d);
-}
-
-
-
-static void invoke_preadv(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- struct t_preadv *c = &d->c.preadv;
- ErlIOVec *ev = &c->eiov;
- size_t bytes_read_so_far = 0;
- unsigned char *p = (unsigned char *)ev->iov[0].iov_base + 4+4+8*c->cnt;
- DTRACE_INVOKE_SETUP(FILE_PREADV);
-
- while (c->cnt < c->n) {
- size_t read_size = ev->iov[1 + c->cnt].iov_len - c->size;
- size_t bytes_read = 0;
- int chop = d->again
- && bytes_read_so_far + read_size >= 2*FILE_SEGMENT_READ;
- if (chop) {
- ASSERT(bytes_read_so_far < FILE_SEGMENT_READ);
- read_size = FILE_SEGMENT_READ + FILE_SEGMENT_READ/2
- - bytes_read_so_far;
- }
- if ( (d->result_ok
- = efile_pread(&d->errInfo,
- (int) d->fd,
- c->offsets[c->cnt] + c->size,
- ((char *)ev->iov[1 + c->cnt].iov_base) + c->size,
- read_size,
- &bytes_read))) {
- bytes_read_so_far += bytes_read;
- if (chop && bytes_read == read_size) {
- c->size += bytes_read;
- goto done;
- }
- ASSERT(bytes_read <= read_size);
- ev->iov[1 + c->cnt].iov_len = bytes_read + c->size;
- ev->size += bytes_read + c->size;
- put_int64(bytes_read + c->size, p); p += 8;
- c->size = 0;
- c->cnt++;
- if (d->again
- && bytes_read_so_far >= FILE_SEGMENT_READ
- && c->cnt < c->n) {
- goto done;
- }
- } else {
- /* In case of a read error, ev->size will not be correct,
- * which does not matter since no read data is returned
- * to Erlang.
- */
- break;
- }
- }
- d->again = 0;
- done:
- DTRACE_INVOKE_RETURN(FILE_PREADV);
-}
-
-static void free_preadv(void *data) {
- struct t_data *d = data;
- int i;
- ErlIOVec *ev = &d->c.preadv.eiov;
-
- for(i = 0; i < ev->vsize; i++) {
- driver_free_binary(ev->binv[i]);
- }
- EF_FREE(d);
-}
-
-static void invoke_ipread(void *data)
-{
- struct t_data *d = data;
- struct t_preadv *c = &d->c.preadv;
- ErlIOVec *ev = &c->eiov;
- size_t bytes_read = 0;
- char buf[2*sizeof(Uint32)];
- Uint32 offset, size;
- DTRACE_INVOKE_SETUP(FILE_IPREAD);
-
- /* Read indirection header */
- if (! efile_pread(&d->errInfo, (int) d->fd, c->offsets[0],
- buf, sizeof(buf), &bytes_read)) {
- goto error;
- }
- if (bytes_read != sizeof(buf)) goto done; /* eof */
- size = get_int32(buf);
- offset = get_int32(buf+4);
- if (size > c->size) goto done; /* eof */
- c->n = 1;
- c->cnt = 0;
- c->size = 0;
- c->offsets[0] = offset;
- if (! (ev->binv[0] = driver_alloc_binary(3*8))) {
- d->errInfo.posix_errno = ENOMEM;
- goto error;
- }
- ev->vsize = 1;
- ev->iov[0].iov_len = 3*8;
- ev->iov[0].iov_base = ev->binv[0]->orig_bytes;
- ev->size = ev->iov[0].iov_len;
- put_int64(offset, ev->iov[0].iov_base);
- put_int64(size, ((char *)ev->iov[0].iov_base) + 2*8);
- if (size == 0) {
- put_int64(size, ((char *)ev->iov[0].iov_base) + 8);
- goto done;
- }
- if (! (ev->binv[1] = driver_alloc_binary(size))) {
- d->errInfo.posix_errno = ENOMEM;
- goto error;
- }
- ev->vsize = 2;
- ev->iov[1].iov_len = size;
- ev->iov[1].iov_base = ev->binv[1]->orig_bytes;
- /* Read data block */
- d->invoke = invoke_preadv;
- invoke_preadv(data);
- DTRACE_INVOKE_RETURN(FILE_IPREAD);
- return;
- error:
- d->result_ok = 0;
- d->again = 0;
- DTRACE_INVOKE_RETURN(FILE_IPREAD);
- return;
- done:
- d->result_ok = !0;
- d->again = 0;
- DTRACE_INVOKE_RETURN(FILE_IPREAD);
-}
-
-/* invoke_writev and invoke_pwritev are the only thread functions that
- * access non-thread data i.e the port queue and a mutex in the port
- * structure that is used to lock the port queue.
- *
- * The port will normally not be terminated until the port queue is
- * empty, but if the port is killed, i.e., exit(Port, kill) is called,
- * it will terminate regardless of the port queue state. When the
- * port is invalid driver_peekq() returns NULL and set the size to -1,
- * and driver_sizeq() returns -1.
- */
-
-static void invoke_writev(void *data) {
- struct t_data *d = (struct t_data *) data;
- SysIOVec *iov0;
- SysIOVec *iov;
- int iovlen;
- int iovcnt;
- size_t size;
- size_t p;
- int segment;
- DTRACE_INVOKE_SETUP(FILE_WRITE);
-
- segment = d->again && d->c.writev.size >= 2*FILE_SEGMENT_WRITE;
- if (segment) {
- size = FILE_SEGMENT_WRITE;
- } else {
- size = d->c.writev.size;
- }
-
- /* Copy the io vector to avoid locking the port que while writing,
- * also, both we and efile_writev might/will change the SysIOVec
- * when segmenting or due to partial write and we do not want to
- * tamper with the actual queue that we get from driver_peekq
- */
- MUTEX_LOCK(d->c.writev.q_mtx); /* Lock before accessing the port queue */
- iov0 = driver_peekq(d->c.writev.port, &iovlen);
-
- /* Calculate iovcnt */
- for (p = 0, iovcnt = 0;
- p < size && iovcnt < iovlen;
- p += iov0[iovcnt++].iov_len)
- ;
- iov = EF_SAFE_ALLOC(sizeof(SysIOVec)*iovcnt);
- memcpy(iov,iov0,iovcnt*sizeof(SysIOVec));
- MUTEX_UNLOCK(d->c.writev.q_mtx);
- /* Let go of lock until we deque from original vector */
-
- if (iovlen > 0) {
- ASSERT(iov[iovcnt-1].iov_len > p - size);
- iov[iovcnt-1].iov_len -= p - size;
- if (d->flags & EFILE_COMPRESSED) {
- int i, status = 1;
- for (i = 0; i < iovcnt; i++) {
- if (iov[i].iov_base && iov[i].iov_len > 0) {
- /* Just in case, I do not know what gzwrite does
- * with errno.
- */
- errno = EINVAL;
- status = erts_gzwrite((ErtsGzFile)d->fd,
- iov[i].iov_base,
- iov[i].iov_len) == iov[i].iov_len;
- if (! status) {
- d->errInfo.posix_errno =
- d->errInfo.os_errno = errno; /* XXX Correct? */
- break;
- }
- }
- }
- d->result_ok = status;
- } else {
- d->result_ok = efile_writev(&d->errInfo,
- d->flags, (int) d->fd,
- iov, iovcnt);
- }
- } else if (iovlen == 0) {
- d->result_ok = 1;
- }
- else { /* Port has terminated */
- d->result_ok = 0;
- d->errInfo.posix_errno = d->errInfo.os_errno = EINVAL;
- }
- EF_FREE(iov);
-
- if (! d->result_ok) {
- d->again = 0;
- MUTEX_LOCK(d->c.writev.q_mtx);
- driver_deq(d->c.writev.port, d->c.writev.size);
- MUTEX_UNLOCK(d->c.writev.q_mtx);
- } else {
- if (! segment) {
- d->again = 0;
- }
- d->c.writev.size -= size;
- TRACE_F(("w%lu", (unsigned long)size));
- MUTEX_LOCK(d->c.writev.q_mtx);
- driver_deq(d->c.writev.port, size);
- MUTEX_UNLOCK(d->c.writev.q_mtx);
- }
-
-
- DTRACE_INVOKE_RETURN(FILE_WRITE);
-}
-
-static void invoke_pwd(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- DTRACE_INVOKE_SETUP(FILE_PWD);
-
- d->again = 0;
- d->result_ok = efile_getdcwd(&d->errInfo,d->drive, d->b+1,
- RESBUFSIZE-1);
- DTRACE_INVOKE_RETURN(FILE_PWD);
-}
-
-static void invoke_readlink(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- char resbuf[RESBUFSIZE]; /* Result buffer. */
- DTRACE_INVOKE_SETUP(FILE_READLINK);
-
- d->again = 0;
- d->result_ok = efile_readlink(&d->errInfo, d->b, resbuf+1,
- RESBUFSIZE-1);
- if (d->result_ok != 0)
- FILENAME_COPY((char *) d->b + 1, resbuf+1);
- DTRACE_INVOKE_RETURN(FILE_READLINK);
-}
-
-static void invoke_altname(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- char resbuf[RESBUFSIZE]; /* Result buffer. */
- DTRACE_INVOKE_SETUP(FILE_ALTNAME);
-
- d->again = 0;
- d->result_ok = efile_altname(&d->errInfo, d->b, resbuf+1,
- RESBUFSIZE-1);
- if (d->result_ok != 0)
- FILENAME_COPY((char *) d->b + 1, resbuf+1);
- DTRACE_INVOKE_RETURN(FILE_ALTNAME);
-}
-
-static void invoke_pwritev(void *data) {
- struct t_data* const d = (struct t_data *) data;
- struct t_pwritev * const c = &d->c.pwritev;
- SysIOVec *iov0;
- SysIOVec *iov;
- int iovlen;
- int iovcnt;
- size_t p;
- int segment;
- size_t size, write_size, written;
- DTRACE_INVOKE_SETUP(FILE_PWRITEV);
-
- segment = d->again && c->size >= 2*FILE_SEGMENT_WRITE;
- if (segment) {
- size = FILE_SEGMENT_WRITE;
- } else {
- size = c->size;
- }
- d->result_ok = !0;
- p = 0;
- /* Lock the queue just for a while, we don't want it locked during write */
- MUTEX_LOCK(c->q_mtx);
- iov0 = driver_peekq(c->port, &iovlen);
- iov = EF_SAFE_ALLOC(sizeof(SysIOVec)*iovlen);
- memcpy(iov,iov0,sizeof(SysIOVec)*iovlen);
- MUTEX_UNLOCK(c->q_mtx);
-
- if (iovlen < 0)
- goto error; /* Port terminated */
- for (iovcnt = 0, written = 0;
- c->cnt < c->n && iovcnt < iovlen && written < size;
- c->cnt++) {
- int chop;
- write_size = c->specs[c->cnt].size;
- if (iov[iovcnt].iov_len - p < write_size) {
- goto error;
- }
- chop = segment && written + write_size >= 2*FILE_SEGMENT_WRITE;
- if (chop) {
- ASSERT(written < FILE_SEGMENT_WRITE);
- write_size = FILE_SEGMENT_WRITE + FILE_SEGMENT_WRITE/2
- - written;
- }
- d->result_ok = efile_pwrite(&d->errInfo, (int) d->fd,
- (char *)(iov[iovcnt].iov_base) + p,
- write_size,
- c->specs[c->cnt].offset);
- if (! d->result_ok) {
- d->again = 0;
- goto deq_error;
- }
- written += write_size;
- c->size -= write_size;
- if (chop) {
- c->specs[c->cnt].offset += write_size;
- c->specs[c->cnt].size -= write_size;
- /* Schedule out (d->again != 0) */
- break;
- }
- /* Move forward in buffer */
- p += write_size;
- ASSERT(iov[iovcnt].iov_len >= p);
- if (iov[iovcnt].iov_len == p) {
- /* Move to next iov[], we trust that it is not a
- * zero length vector, and thereby depend on that
- * such are not queued.
- */
- iovcnt++; p = 0;
- }
- }
- if (! segment) {
- if (c->cnt != c->n) {
- /* Mismatch between number of
- * pos/size specs vs number of queued buffers .
- */
- error:
- d->errInfo.posix_errno = EINVAL;
- d->result_ok = 0;
- d->again = 0;
- deq_error:
- MUTEX_LOCK(c->q_mtx);
- driver_deq(c->port, c->size);
- MUTEX_UNLOCK(c->q_mtx);
-
- goto done;
- } else {
- ASSERT(written == size);
- d->again = 0;
- }
- } else {
- ASSERT(written >= FILE_SEGMENT_WRITE);
- }
-
- MUTEX_LOCK(c->q_mtx);
- driver_deq(c->port, written);
- MUTEX_UNLOCK(c->q_mtx);
- done:
- EF_FREE(iov); /* Free our copy of the vector, nothing to restore */
-
- DTRACE_INVOKE_RETURN(FILE_PWRITEV);
-}
-
-static void invoke_flstat(void *data)
-{
- struct t_data *d = (struct t_data *) data;
-
- DTRACE3(efile_drv_int_entry, d->sched_i1, d->sched_i2,
- d->command == FILE_LSTAT ? FILE_LSTAT : FILE_FSTAT);
- d->again = 0;
- d->result_ok = efile_fileinfo(&d->errInfo, &d->info,
- d->b, d->command == FILE_LSTAT);
- DTRACE3(efile_drv_int_entry, d->sched_i1, d->sched_i2,
- d->command == FILE_LSTAT ? FILE_LSTAT : FILE_FSTAT);
- gcc_optimizer_hack++;
-}
-
-static void invoke_link(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- char *name = d->b;
- char *new_name;
- DTRACE_INVOKE_SETUP(FILE_LINK);
-
- d->again = 0;
- new_name = name+FILENAME_BYTELEN(name)+FILENAME_CHARSIZE;
- d->result_ok = efile_link(&d->errInfo, name, new_name);
- DTRACE_INVOKE_RETURN(FILE_LINK);
-}
-
-static void invoke_symlink(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- char *name = d->b;
- char *new_name;
- DTRACE_INVOKE_SETUP(FILE_SYMLINK);
-
- d->again = 0;
- new_name = name+FILENAME_BYTELEN(name)+FILENAME_CHARSIZE;
- d->result_ok = efile_symlink(&d->errInfo, name, new_name);
- DTRACE_INVOKE_RETURN(FILE_SYMLINK);
-}
-
-static void invoke_rename(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- char *name = d->b;
- char *new_name;
- DTRACE_INVOKE_SETUP(FILE_RENAME);
-
- d->again = 0;
- new_name = name+FILENAME_BYTELEN(name)+FILENAME_CHARSIZE;
- d->result_ok = efile_rename(&d->errInfo, name, new_name);
- DTRACE_INVOKE_RETURN(FILE_RENAME);
-}
-
-static void invoke_write_info(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- DTRACE_INVOKE_SETUP(FILE_WRITE_INFO);
-
- d->again = 0;
- d->result_ok = efile_write_info(&d->errInfo, &d->info, d->b);
- DTRACE_INVOKE_RETURN(FILE_WRITE_INFO);
-}
-
-static void invoke_lseek(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- int status;
- DTRACE_INVOKE_SETUP(FILE_LSEEK);
-
- d->again = 0;
- if (d->flags & EFILE_COMPRESSED) {
- int offset = (int) d->c.lseek.offset;
-
- if (offset != d->c.lseek.offset) {
- d->errInfo.posix_errno = EINVAL;
- status = 0;
- } else {
- d->c.lseek.location = erts_gzseek((ErtsGzFile)d->fd,
- offset, d->c.lseek.origin);
- if (d->c.lseek.location == -1) {
- d->errInfo.posix_errno = errno;
- status = 0;
- } else {
- status = 1;
- }
- }
- } else {
- status = efile_seek(&d->errInfo, (int) d->fd,
- d->c.lseek.offset, d->c.lseek.origin,
- &d->c.lseek.location);
- }
- d->result_ok = status;
- DTRACE_INVOKE_RETURN(FILE_LSEEK);
-}
-
-static void invoke_readdir(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- char *p = NULL;
- size_t file_bs;
- size_t n = 0, total = 0;
- struct t_readdir_buf *b = NULL;
- int res = 0;
- DTRACE_INVOKE_SETUP(FILE_READDIR);
-
- d->again = 0;
- d->errInfo.posix_errno = 0;
-
- do {
- total = READDIR_BUFSIZE;
- n = 1;
- b = EF_SAFE_ALLOC(sizeof(struct t_readdir_buf));
- b->next = NULL;
-
- if (d->c.read_dir.last_buf) {
- d->c.read_dir.last_buf->next = b;
- } else {
- d->c.read_dir.first_buf = b;
- }
- d->c.read_dir.last_buf = b;
-
- p = &b->buf[0];
- p[0] = FILE_RESP_LFNAME;
- file_bs = READDIR_BUFSIZE - n;
-
- do {
- res = efile_readdir(&d->errInfo, d->b, &d->dir_handle, p + n + 2, &file_bs);
-
- if (res) {
- put_int16((Uint16)file_bs, p + n);
- n += 2 + file_bs;
- file_bs = READDIR_BUFSIZE - n;
- }
- } while( res && ((total - n - 2) >= MAXPATHLEN*FILENAME_CHARSIZE));
-
- b->n = n;
- } while(res);
-
- d->result_ok = (d->errInfo.posix_errno == 0);
- DTRACE_INVOKE_RETURN(FILE_READDIR);
-}
-
-static void invoke_open(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- int status = 1; /* Status of open call. */
- DTRACE_INVOKE_SETUP(FILE_OPEN);
-
- d->again = 0;
- if ((d->flags & EFILE_COMPRESSED) == 0) {
- int fd;
- status = efile_openfile(&d->errInfo, d->b, d->flags, &fd, NULL);
- d->fd = fd;
- } else {
- char* mode = NULL;
-
- if (((d->flags & (EFILE_MODE_READ_WRITE)) == EFILE_MODE_READ_WRITE) ||
- (d->flags & EFILE_MODE_APPEND)) {
- status = 0;
- d->errInfo.posix_errno = EINVAL;
- } else {
- status = efile_may_openfile(&d->errInfo, d->b);
- if (status || (d->errInfo.posix_errno != EISDIR)) {
- mode = (d->flags & EFILE_MODE_READ) ? "rb" : "wb";
- d->fd = (SWord) erts_gzopen(d->b, mode);
- if ((ErtsGzFile)d->fd) {
- status = 1;
- } else {
- if (errno == 0) {
- errno = ENOMEM;
- }
- d->errInfo.posix_errno = errno;
- status = 0;
- }
- }
- }
- }
-
- d->result_ok = status;
- if (!status) {
- d->fd = FILE_FD_INVALID;
- }
- DTRACE_INVOKE_RETURN(FILE_OPEN);
-}
-
-static void invoke_fadvise(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- int fd = (int) d->fd;
- off_t offset = (off_t) d->c.fadvise.offset;
- off_t length = (off_t) d->c.fadvise.length;
- int advise = (int) d->c.fadvise.advise;
- DTRACE_INVOKE_SETUP(FILE_FADVISE);
-
- d->again = 0;
- d->result_ok = efile_fadvise(&d->errInfo, fd, offset, length, advise);
- DTRACE_INVOKE_RETURN(FILE_FADVISE);
-}
-
-#ifdef HAVE_SENDFILE
-static void invoke_sendfile(void *data)
-{
- struct t_data *d = (struct t_data *)data;
- int fd = d->fd;
- int out_fd = (int)d->c.sendfile.out_fd;
- Uint64 nbytes = d->c.sendfile.nbytes;
- int result = 0;
- d->again = 0;
-
- result = efile_sendfile(&d->errInfo, fd, out_fd, &d->c.sendfile.offset, &nbytes, NULL);
-
- d->c.sendfile.written += nbytes;
-
- if (result == 1 || (result == 0 && USE_THRDS_FOR_SENDFILE(d))) {
- d->result_ok = 0;
- } else if (result == 0 && (d->errInfo.posix_errno == EAGAIN
- || d->errInfo.posix_errno == EINTR)) {
- if ((d->c.sendfile.nbytes - nbytes) != 0) {
- d->result_ok = 1;
- if (d->c.sendfile.nbytes != 0)
- d->c.sendfile.nbytes -= nbytes;
- } else if (nbytes == 0 && d->c.sendfile.nbytes == 0) {
- d->result_ok = 1;
- } else
- d->result_ok = 0;
- } else {
- d->result_ok = -1;
- }
-}
-
-static void free_sendfile(void *data) {
- struct t_data *d = (struct t_data *)data;
- if (USE_THRDS_FOR_SENDFILE(d)) {
- SET_NONBLOCKING(d->c.sendfile.out_fd);
- } else {
- MUTEX_LOCK(d->c.sendfile.q_mtx);
- driver_deq(d->c.sendfile.port,1);
- MUTEX_UNLOCK(d->c.sendfile.q_mtx);
- driver_select(d->c.sendfile.port, (ErlDrvEvent)(long)d->c.sendfile.out_fd,
- ERL_DRV_USE_NO_CALLBACK|ERL_DRV_WRITE, 0);
- }
- EF_FREE(data);
-}
-
-static void file_ready_output(ErlDrvData data, ErlDrvEvent event)
-{
- file_descriptor* fd = (file_descriptor*) data;
-
- switch (fd->d->command) {
- case FILE_SENDFILE:
- driver_select(fd->d->c.sendfile.port, event,
- (int)ERL_DRV_WRITE,(int) 0);
- invoke_sendfile((void *)fd->d);
- file_async_ready(data, (ErlDrvThreadData)fd->d);
- break;
- default:
- break;
- }
-}
-
-static void file_stop_select(ErlDrvEvent event, void* _)
-{
-
-}
-
-static int flush_sendfile(file_descriptor *desc,void *_) {
- if (desc->sendfile_state == sending) {
- desc->d->result_ok = -1;
- desc->d->errInfo.posix_errno = ECONNABORTED;
- file_async_ready((ErlDrvData)desc,(ErlDrvThreadData)desc->d);
- }
- return 1;
-}
-#endif /* HAVE_SENDFILE */
-
-
-static void invoke_fallocate(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- int fd = (int) d->fd;
- Sint64 offset = d->c.fallocate.offset;
- Sint64 length = d->c.fallocate.length;
-
- d->again = 0;
- d->result_ok = efile_fallocate(&d->errInfo, fd, offset, length);
-}
-
-static void free_readdir(void *data)
-{
- struct t_data *d = (struct t_data *) data;
- struct t_readdir_buf *b1 = d->c.read_dir.first_buf;
-
- while (b1) {
- struct t_readdir_buf *b2 = b1;
- b1 = b1->next;
- EF_FREE(b2);
- }
- EF_FREE(d);
-}
-
-
-
-static void try_free_read_bin(file_descriptor *desc) {
- if ((desc->read_size == 0)
- && (desc->read_offset >= desc->read_binp->orig_size)) {
- ASSERT(desc->read_offset == desc->read_binp->orig_size);
- driver_free_binary(desc->read_binp);
- desc->read_binp = NULL;
- desc->read_offset = 0;
- desc->read_size = 0;
- }
-}
-
-
-
-static int try_again(file_descriptor *desc, struct t_data *d) {
- if (! d->again)
- return 0;
- if (desc->timer_state != timer_idle) {
- driver_cancel_timer(desc->port);
- }
- desc->timer_state = timer_again;
- desc->invoke = d->invoke;
- desc->d = d;
- desc->free = d->free;
- driver_set_timer(desc->port, 0L);
- return !0;
-}
-
-
-
-static void cq_execute(file_descriptor *desc) {
- struct t_data *d;
- register void *void_ptr; /* Soft cast variable */
- if (desc->timer_state == timer_again)
- return;
-#ifdef HAVE_SENDFILE
- if (desc->sendfile_state == sending)
- return;
-#endif
- if (! (d = cq_deq(desc)))
- return;
- TRACE_F(("x%i", (int) d->command));
- d->again = sys_info.async_threads == 0;
- DRIVER_ASYNC(d->level, desc, d->invoke, void_ptr=d, d->free);
-}
-
-static struct t_data *async_write(file_descriptor *desc, int *errp,
- int reply, Uint32 reply_size
-#ifdef USE_VM_PROBES
- ,Sint64 *dt_i1, Sint64 *dt_i2, Sint64 *dt_i3
-#endif
-) {
- struct t_data *d;
- if (! (d = EF_ALLOC(sizeof(struct t_data) - 1))) {
- if (errp) *errp = ENOMEM;
- return NULL;
- }
- TRACE_F(("w%lu", (unsigned long)desc->write_buffered));
- d->command = FILE_WRITE;
- d->fd = desc->fd;
- d->flags = desc->flags;
- d->c.writev.port = desc->port;
- d->c.writev.q_mtx = desc->q_mtx;
- d->c.writev.size = desc->write_buffered;
-#ifdef USE_VM_PROBES
- if (dt_i1 != NULL) {
- *dt_i1 = d->fd;
- *dt_i2 = d->flags;
- *dt_i3 = d->c.writev.size;
- }
-#endif
- d->reply = reply;
- d->c.writev.reply_size = reply_size;
- d->invoke = invoke_writev;
- d->free = free_data;
- d->level = 1;
- cq_enq(desc, d);
- desc->write_buffered = 0;
- return d;
-}
-
-static int flush_write(file_descriptor *desc, int *errp
-#ifdef USE_VM_PROBES
- , dt_private *dt_priv, char *dt_utag
-#endif
-) {
- int result = 0;
-#ifdef USE_VM_PROBES
- Sint64 dt_i1 = 0, dt_i2 = 0, dt_i3 = 0;
-#endif
- struct t_data *d = NULL;
-
- MUTEX_LOCK(desc->q_mtx);
- if (desc->write_buffered > 0) {
- if ((d = async_write(desc, errp, 0, 0
-#ifdef USE_VM_PROBES
- ,&dt_i1, &dt_i2, &dt_i3
-#endif
- )) == NULL) {
- result = -1;
- }
- }
- MUTEX_UNLOCK(desc->q_mtx);
-#ifdef USE_VM_PROBES
- if (d != NULL) {
- d->sched_i1 = dt_priv->thread_num;
- d->sched_i2 = dt_priv->tag;
- d->sched_utag[0] = '\0';
- if (dt_utag != NULL) {
- if (dt_utag[0] == '\0') {
- dt_utag = NULL;
- } else {
- strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1);
- d->sched_utag[sizeof(d->sched_utag) - 1] = '\0';
- }
- }
- DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++,
- dt_utag, FILE_WRITE,
- NULL, NULL, dt_i1, dt_i2, dt_i3, 0, desc->port_str);
- }
-#endif /* USE_VM_PROBES */
- return result;
-}
-
-static int check_write_error(file_descriptor *desc, int *errp) {
- if (desc->write_error) {
- if (errp) *errp = desc->write_errInfo.posix_errno;
- desc->write_error = 0;
- return -1;
- }
- return 0;
-}
-
-static int flush_write_check_error(file_descriptor *desc, int *errp
-#ifdef USE_VM_PROBES
- , dt_private *dt_priv, char *dt_utag
-#endif
- ) {
- int r;
- if ( (r = flush_write(desc, errp
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- )) != 0) {
- check_write_error(desc, NULL);
- return r;
- } else {
- return check_write_error(desc, errp);
- }
-}
-
-static struct t_data *async_lseek(file_descriptor *desc, int *errp, int reply,
- Sint64 offset, int origin
-#ifdef USE_VM_PROBES
- , Sint64 *dt_i1, Sint64 *dt_i2, Sint64 *dt_i3
-#endif
- ) {
- struct t_data *d;
- if (! (d = EF_ALLOC(sizeof(struct t_data)))) {
- *errp = ENOMEM;
- return NULL;
- }
- d->flags = desc->flags;
- d->fd = desc->fd;
- d->command = FILE_LSEEK;
- d->reply = reply;
- d->c.lseek.offset = offset;
- d->c.lseek.origin = origin;
-#ifdef USE_VM_PROBES
- if (dt_i1 != NULL) {
- *dt_i1 = d->fd;
- *dt_i2 = d->c.lseek.offset;
- *dt_i3 = d->c.lseek.origin;
- }
-#endif
- d->invoke = invoke_lseek;
- d->free = free_data;
- d->level = 1;
- cq_enq(desc, d);
- return d;
-}
-
-static void flush_read(file_descriptor *desc) {
- desc->read_offset = 0;
- desc->read_size = 0;
- if (desc->read_binp) {
- driver_free_binary(desc->read_binp);
- desc->read_binp = NULL;
- }
-}
-
-static int lseek_flush_read(file_descriptor *desc, int *errp
-#ifdef USE_VM_PROBES
- ,dt_private *dt_priv, char *dt_utag
-#endif
- ) {
- int r = 0;
- size_t read_size = desc->read_size;
-#ifdef USE_VM_PROBES
- Sint64 dt_i1 = 0, dt_i2 = 0, dt_i3 = 0;
-#endif
- struct t_data *d;
-
- flush_read(desc);
- if (read_size != 0) {
- if ((d = async_lseek(desc, errp, 0,
- -((ssize_t)read_size), EFILE_SEEK_CUR
-#ifdef USE_VM_PROBES
- , &dt_i1, &dt_i2, &dt_i3
-#endif
- )) == NULL) {
- r = -1;
- } else {
-#ifdef USE_VM_PROBES
- d->sched_i1 = dt_priv->thread_num;
- d->sched_i2 = dt_priv->tag;
- d->sched_utag[0] = '\0';
- if (dt_utag != NULL) {
- if (dt_utag[0] == '\0') {
- dt_utag = NULL;
- } else {
- strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1);
- d->sched_utag[sizeof(d->sched_utag) - 1] = '\0';
- }
- }
- DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++,
- dt_utag, FILE_LSEEK,
- NULL, NULL, dt_i1, dt_i2, dt_i3, 0, desc->port_str);
-#endif /* USE_VM_PROBES */
- }
- }
- return r;
-}
-
-
-/*********************************************************************
- * Driver entry point -> stop
- * The close has to be scheduled on async thread, so that currently active
- * async operation does not suddenly have the ground disappearing under their feet...
- */
-static void
-file_stop(ErlDrvData e)
-{
- file_descriptor* desc = (file_descriptor*)e;
-
- TRACE_C('p');
-
- IF_THRDS {
- flush_read(desc);
- if (desc->fd != FILE_FD_INVALID) {
- struct t_data *d = EF_SAFE_ALLOC(sizeof(struct t_data));
- d->command = FILE_CLOSE_ON_PORT_EXIT;
- d->reply = !0;
- d->fd = desc->fd;
- d->flags = desc->flags;
- d->invoke = invoke_close;
- d->free = free_data;
- d->level = 2;
- d->data_to_free = (void *) desc;
- cq_enq(desc, d);
- desc->fd = FILE_FD_INVALID;
- desc->flags = 0;
- cq_execute(desc);
- } else {
- EF_FREE(desc);
- }
- } else {
- if (desc->fd != FILE_FD_INVALID) {
- do_close(desc->flags, desc->fd);
- desc->fd = FILE_FD_INVALID;
- desc->flags = 0;
- }
- if (desc->read_binp) {
- driver_free_binary(desc->read_binp);
- }
- EF_FREE(desc);
- }
-}
-
-/*********************************************************************
- * Driver entry point -> ready_async
- */
-static void
-file_async_ready(ErlDrvData e, ErlDrvThreadData data)
-{
- file_descriptor *desc = (file_descriptor*)e;
- struct t_data *d = (struct t_data *) data;
- char header[5]; /* result code + count */
- char resbuf[RESBUFSIZE]; /* Result buffer. */
-#ifdef USE_VM_PROBES
- int sched_i1 = d->sched_i1, sched_i2 = d->sched_i2, command = d->command,
- result_ok = d->result_ok,
- posix_errno = d->result_ok ? 0 : d->errInfo.posix_errno;
- DTRACE_CHARBUF(sched_utag, DTRACE_EFILE_BUFSIZ+1);
-
- sched_utag[0] = '\0';
- if (DTRACE_ENABLED(efile_drv_return)) {
- strncpy(sched_utag, d->sched_utag, DTRACE_EFILE_BUFSIZ);
- sched_utag[DTRACE_EFILE_BUFSIZ] = '\0';
- }
-#endif /* USE_VM_PROBES */
-
- TRACE_C('r');
-
- if (try_again(desc, d)) {
- /* DTRACE TODO: what kind of probe makes sense here? */
- return;
- }
-
- switch (d->command)
- {
- case FILE_READ:
- if (!d->result_ok) {
- reply_error(desc, &d->errInfo);
- } else {
- size_t available_bytes =
- d->c.read.bin_offset + d->c.read.bin_size - desc->read_offset;
- if (available_bytes < d->c.read.size) {
- d->c.read.size = available_bytes;
- }
- TRACE_C('D');
- reply_data(desc, d->c.read.binp,
- desc->read_offset, d->c.read.size);
- desc->read_offset += d->c.read.size;
- desc->read_size =
- d->c.read.bin_offset + d->c.read.bin_size - desc->read_offset;
- try_free_read_bin(desc);
- }
- free_read(data);
- break;
- case FILE_READ_LINE:
- /* The read_line structure differs from the read structure.
- The data->read_offset and d->c.read_line.read_offset are copies, as are
- data->read_size and d->c.read_line.read_size
- The read_line function does not kniow in advance how large the binary has to be,
- why new allocation (but not reallocation of the old binary, for obvious reasons)
- may happen in the worker thread. */
- if (!d->result_ok) {
- reply_error(desc, &d->errInfo);
- } else {
- size_t len = d->c.read_line.nl_pos - d->c.read_line.read_offset;
- TRACE_C('L');
- reply_data(desc, d->c.read_line.binp,
- d->c.read_line.read_offset, len);
- desc->read_offset = d->c.read_line.read_offset + d->c.read_line.nl_skip + len;
- desc->read_size =
- d->c.read_line.read_size - d->c.read_line.nl_skip - len;
- if (desc->read_binp != d->c.read_line.binp) { /* New binary allocated */
- driver_free_binary(desc->read_binp);
- desc->read_binp = d->c.read_line.binp;
- driver_binary_inc_refc(desc->read_binp);
- }
-#if !ALWAYS_READ_LINE_AHEAD
- ASSERT(desc->read_bufsize > 0 || desc->read_size == 0);
- if (desc->read_bufsize == 0) {
- desc->read_offset = desc->read_binp->orig_size; /* triggers cleanup */
- }
-#endif
- try_free_read_bin(desc);
- }
- free_read_line(data);
- break;
- case FILE_READ_FILE:
- if (!d->result_ok)
- reply_error(desc, &d->errInfo);
- else {
- header[0] = FILE_RESP_ALL_DATA;
- TRACE_C('R');
- driver_output_binary(desc->port, header, 1,
- d->c.read_file.binp,
- 0, d->c.read_file.offset);
- }
- free_read_file(data);
- break;
- case FILE_WRITE:
- if (d->reply) {
- if (! d->result_ok) {
- reply_error(desc, &d->errInfo);
- } else {
- reply_Uint(desc, d->c.writev.reply_size);
- }
- } else {
- if (! d->result_ok) {
- desc->write_error = !0;
- desc->write_errInfo = d->errInfo;
- }
- }
- free_data(data);
- break;
- case FILE_LSEEK:
- if (d->reply) {
- if (d->result_ok)
- reply_Sint64(desc, d->c.lseek.location);
- else
- reply_error(desc, &d->errInfo);
- }
- free_data(data);
- break;
- case FILE_MKDIR:
- case FILE_RMDIR:
- case FILE_CHDIR:
- case FILE_DELETE:
- case FILE_FDATASYNC:
- case FILE_FSYNC:
- case FILE_TRUNCATE:
- case FILE_LINK:
- case FILE_SYMLINK:
- case FILE_RENAME:
- case FILE_WRITE_INFO:
- case FILE_FADVISE:
- case FILE_FALLOCATE:
- reply(desc, d->result_ok, &d->errInfo);
- free_data(data);
- break;
- case FILE_ALTNAME:
- case FILE_PWD:
- case FILE_READLINK:
- {
- int length;
- char *resbuf = d->b;
-
- if (!d->result_ok)
- reply_error(desc, &d->errInfo);
- else {
- resbuf[0] = FILE_RESP_FNAME;
- length = 1+FILENAME_BYTELEN((char*) resbuf+1);
- TRACE_C('R');
- driver_output2(desc->port, resbuf, 1, resbuf+1, length-1);
- }
- free_data(data);
- break;
- }
- case FILE_OPEN:
- if (!d->result_ok) {
- reply_error(desc, &d->errInfo);
- } else {
- ASSERT(d->is_fd_unused);
- desc->fd = d->fd;
- desc->flags = d->flags;
- d->is_fd_unused = 0;
- reply_Uint(desc, d->fd);
- }
- free_data(data);
- break;
- case FILE_FSTAT:
- case FILE_LSTAT:
- {
- if (d->result_ok) {
- resbuf[0] = FILE_RESP_INFO;
-
- put_int32(d->info.size_high, &resbuf[1 + ( 0 * 4)]);
- put_int32(d->info.size_low, &resbuf[1 + ( 1 * 4)]);
- put_int32(d->info.type, &resbuf[1 + ( 2 * 4)]);
-
- /* Note 64 bit indexing in resbuf here */
- put_int64(d->info.accessTime, &resbuf[1 + ( 3 * 4)]);
- put_int64(d->info.modifyTime, &resbuf[1 + ( 5 * 4)]);
- put_int64(d->info.cTime, &resbuf[1 + ( 7 * 4)]);
-
- put_int32(d->info.mode, &resbuf[1 + ( 9 * 4)]);
- put_int32(d->info.links, &resbuf[1 + (10 * 4)]);
- put_int32(d->info.major_device, &resbuf[1 + (11 * 4)]);
- put_int32(d->info.minor_device, &resbuf[1 + (12 * 4)]);
- put_int32(d->info.inode, &resbuf[1 + (13 * 4)]);
- put_int32(d->info.uid, &resbuf[1 + (14 * 4)]);
- put_int32(d->info.gid, &resbuf[1 + (15 * 4)]);
- put_int32(d->info.access, &resbuf[1 + (16 * 4)]);
-
-#define RESULT_SIZE (1 + (17 * 4))
- TRACE_C('R');
- driver_output2(desc->port, resbuf, RESULT_SIZE, NULL, 0);
-#undef RESULT_SIZE
- } else
- reply_error(desc, &d->errInfo);
- }
- free_data(data);
- break;
- case FILE_READDIR:
- if (!d->result_ok) {
- reply_error(desc, &d->errInfo);
- } else {
- struct t_readdir_buf *b1 = d->c.read_dir.first_buf;
- char op = FILE_RESP_LFNAME;
-
- TRACE_C('R');
- ASSERT(b1);
-
- while (b1) {
- struct t_readdir_buf *b2 = b1;
- char *p = &b1->buf[0];
- driver_output2(desc->port, p, 1, p + 1, b1->n - 1);
- b1 = b1->next;
- EF_FREE(b2);
- }
- driver_output2(desc->port, &op, 1, NULL, 0);
-
- d->c.read_dir.first_buf = NULL;
- d->c.read_dir.last_buf = NULL;
- }
- free_readdir(data);
- break;
- case FILE_CLOSE:
- if (d->reply) {
- TRACE_C('K');
- reply_ok(desc);
-#ifdef USE_VM_PROBES
- result_ok = 1;
-#endif
- }
- free_data(data);
- break;
- case FILE_PWRITEV:
- if (!d->result_ok) {
- reply_Uint_error(desc, d->c.pwritev.cnt, &d->errInfo);
- } else {
- reply_Uint(desc, d->c.pwritev.n);
- }
- free_data(data);
- break;
- case FILE_PREADV:
- if (!d->result_ok) {
- reply_error(desc, &d->errInfo);
- } else {
- reply_ev(desc, FILE_RESP_LDATA, &d->c.preadv.eiov);
- }
- free_preadv(data);
- break;
- case FILE_IPREAD:
- if (!d->result_ok) {
- reply_error(desc, &d->errInfo);
- } else if (!d->c.preadv.eiov.vsize) {
- reply_eof(desc);
- } else {
- reply_ev(desc, FILE_RESP_N2DATA, &d->c.preadv.eiov);
- }
- free_preadv(data);
- break;
-#ifdef HAVE_SENDFILE
- case FILE_SENDFILE:
- if (d->result_ok == -1) {
- if (d->errInfo.posix_errno == ECONNRESET ||
- d->errInfo.posix_errno == ENOTCONN ||
- d->errInfo.posix_errno == EPIPE)
- reply_string_error(desc,"closed");
- else
- reply_error(desc, &d->errInfo);
- desc->sendfile_state = not_sending;
- free_sendfile(data);
- } else if (d->result_ok == 0) {
- reply_Sint64(desc, d->c.sendfile.written);
- desc->sendfile_state = not_sending;
- free_sendfile(data);
- } else if (d->result_ok == 1) { /* If we are using select to send the rest of the data */
- desc->sendfile_state = sending;
- desc->d = d;
- driver_select(desc->port, (ErlDrvEvent)(long)d->c.sendfile.out_fd,
- ERL_DRV_USE|ERL_DRV_WRITE, 1);
- }
- break;
-#endif
- case FILE_CLOSE_ON_PORT_EXIT:
- /* See file_stop. However this is never invoked after the port is killed. */
- free_data(data);
- desc = NULL;
- /* This is it for this port, so just send dtrace and return, avoid doing anything to the freed data */
- DTRACE6(efile_drv_return, sched_i1, sched_i2, sched_utag,
- command, result_ok, posix_errno);
- return;
- default:
- abort();
- }
- DTRACE6(efile_drv_return, sched_i1, sched_i2, sched_utag,
- command, result_ok, posix_errno);
- if (desc->write_buffered != 0 && desc->timer_state == timer_idle ) {
- desc->timer_state = timer_write;
- driver_set_timer(desc->port, desc->write_delay);
- }
- cq_execute(desc);
-
-}
-
-
-/*********************************************************************
- * Driver entry point -> output
- */
-static void
-file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
-{
- file_descriptor* desc = (file_descriptor*)e;
- Efile_error errInfo; /* The error codes for the last operation. */
- Sint fd; /* The file descriptor for this port, if any,
- * -1 if none.
- */
- char* name; /* Points to the filename in buf. */
- int command;
- struct t_data *d = NULL;
-#ifdef USE_VM_PROBES
- char *dt_utag = NULL;
- char *dt_s1 = NULL, *dt_s2 = NULL;
- Sint64 dt_i1 = 0;
- Sint64 dt_i2 = 0;
- Sint64 dt_i3 = 0;
- Sint64 dt_i4 = 0;
- dt_private *dt_priv = get_dt_private(0);
-#endif /* USE_VM_PROBES */
-
- TRACE_C('o');
-
- fd = desc->fd;
- name = buf+1;
- command = *(uchar*)buf++;
-
- switch(command) {
-
- case FILE_MKDIR:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE);
-
- FILENAME_COPY(d->b, name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
-#endif
- d->command = command;
- d->invoke = invoke_mkdir;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
- case FILE_RMDIR:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE);
-
- FILENAME_COPY(d->b, name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
-#endif
- d->command = command;
- d->invoke = invoke_rmdir;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
- case FILE_DELETE:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE);
-
- FILENAME_COPY(d->b, name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
-#endif
- d->command = command;
- d->invoke = invoke_delete_file;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
- case FILE_RENAME:
- {
- char* new_name;
- int namelen = FILENAME_BYTELEN(name)+FILENAME_CHARSIZE;
- new_name = name+namelen;
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1
- + namelen
- + FILENAME_BYTELEN(new_name) + FILENAME_CHARSIZE);
-
- FILENAME_COPY(d->b, name);
- FILENAME_COPY(d->b + namelen, new_name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_s2 = d->b + namelen;
- dt_utag = buf + namelen + FILENAME_BYTELEN(new_name) + FILENAME_CHARSIZE;
-#endif
- d->flags = desc->flags;
- d->fd = fd;
- d->command = command;
- d->invoke = invoke_rename;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
- case FILE_CHDIR:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE);
-
- FILENAME_COPY(d->b, name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
-#endif
- d->command = command;
- d->invoke = invoke_chdir;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
- case FILE_PWD:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + RESBUFSIZE + 1);
-
- d->drive = *(uchar*)buf;
-#ifdef USE_VM_PROBES
- dt_utag = buf + 1;
-#endif
- d->command = command;
- d->invoke = invoke_pwd;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
- case FILE_READDIR:
- if (sys_info.async_threads > 0)
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) +
- FILENAME_CHARSIZE);
-
- FILENAME_COPY(d->b, name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
-#endif
- d->dir_handle = NULL;
- d->command = command;
- d->invoke = invoke_readdir;
- d->free = free_readdir;
- d->level = 2;
- d->c.read_dir.first_buf = NULL;
- d->c.read_dir.last_buf = NULL;
- goto done;
- }
- else
- {
- size_t resbufsize;
- size_t n = 0, total = 0;
- int res = 0;
- char resbuf[READDIR_BUFSIZE];
-
- EFILE_DIR_HANDLE dir_handle; /* Handle to open directory. */
-
- total = READDIR_BUFSIZE;
- errInfo.posix_errno = 0;
- dir_handle = NULL;
- resbuf[0] = FILE_RESP_LFNAME;
-
-#ifdef USE_VM_PROBES
- dt_s1 = name;
- dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
-#endif
- /* Fill the buffer with multiple directory listings before sending it to the
- * receiving process. READDIR_CHUNKS is minimum number of files sent to the
- * receiver.
- * Format for each driver_output2:
- * ------------------------------------
- * | Type | Len | Filename | ...
- * | 1 byte | 2 bytes | Len bytes | ...
- * ------------------------------------
- */
-
- do {
- n = 1;
- resbufsize = READDIR_BUFSIZE - n;
-
- do {
- res = efile_readdir(&errInfo, name, &dir_handle, resbuf + n + 2, &resbufsize);
-
- if (res) {
- put_int16((Uint16)resbufsize, resbuf + n);
- n += 2 + resbufsize;
- resbufsize = READDIR_BUFSIZE - n;
- }
- } while( res && ((total - n - 2) >= MAXPATHLEN*FILENAME_CHARSIZE));
-
- if (n > 1) {
- driver_output2(desc->port, resbuf, 1, resbuf + 1, n - 1);
- }
- } while(res);
-
- if (errInfo.posix_errno != 0) {
- reply_error(desc, &errInfo);
- return;
- }
-#ifdef USE_VM_PROBES
- if (dt_utag != NULL && dt_utag[0] == '\0') {
- dt_utag = NULL;
- }
-
- DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag,
- dt_utag, command, name, dt_s2,
- dt_i1, dt_i2, dt_i3, dt_i4, desc->port_str);
- DTRACE6(efile_drv_return, dt_priv->thread_num, dt_priv->tag++,
- dt_utag, command, 1, 0);
-#endif
- TRACE_C('R');
- driver_output2(desc->port, resbuf, 1, NULL, 0);
- return;
- }
- case FILE_OPEN:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(buf+4) +
- FILENAME_CHARSIZE);
-
- d->flags = get_int32((uchar*)buf);
- name = buf+4;
- FILENAME_COPY(d->b, name);
-#ifdef USE_VM_PROBES
- dt_i1 = d->flags;
- dt_s1 = d->b;
- dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
-#endif
- d->command = command;
- d->invoke = invoke_open;
- d->free = free_data;
- d->level = 2;
- d->is_fd_unused = 1;
- goto done;
- }
-
- case FILE_FDATASYNC:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data));
-
- d->fd = fd;
-#ifdef USE_VM_PROBES
- dt_utag = name;
- dt_i1 = fd;
-#endif
- d->command = command;
- d->invoke = invoke_fdatasync;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
- case FILE_FSYNC:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data));
-
- d->fd = fd;
-#ifdef USE_VM_PROBES
- dt_utag = name;
- dt_i1 = fd;
-#endif
- d->command = command;
- d->invoke = invoke_fsync;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
-
- case FILE_FSTAT:
- case FILE_LSTAT:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) +
- FILENAME_CHARSIZE);
-
- FILENAME_COPY(d->b, name);
- d->fd = fd;
-#ifdef USE_VM_PROBES
- dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
- if (command == FILE_LSTAT) {
- dt_s1 = d->b;
- } else {
- dt_i1 = fd;
- }
-#endif
- d->command = command;
- d->invoke = invoke_flstat;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
- case FILE_TRUNCATE:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data));
-
- d->flags = desc->flags;
- d->fd = fd;
-#ifdef USE_VM_PROBES
- dt_utag = name;
- dt_i1 = fd;
- dt_i2 = d->flags;
-#endif
- d->command = command;
- d->invoke = invoke_truncate;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
- case FILE_WRITE_INFO:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1
- + FILENAME_BYTELEN(buf + 9*4) + FILENAME_CHARSIZE);
-
- d->info.mode = get_int32(buf + 0 * 4);
- d->info.uid = get_int32(buf + 1 * 4);
- d->info.gid = get_int32(buf + 2 * 4);
- d->info.accessTime = get_int64(buf + 3 * 4);
- d->info.modifyTime = get_int64(buf + 5 * 4);
- d->info.cTime = get_int64(buf + 7 * 4);
-
- FILENAME_COPY(d->b, buf + 9*4);
-#ifdef USE_VM_PROBES
- dt_i1 = d->info.mode;
- dt_i2 = d->info.uid;
- dt_i3 = d->info.gid;
- dt_s1 = d->b;
- dt_utag = buf + 9 * 4 + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
-#endif
- d->command = command;
- d->invoke = invoke_write_info;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
- case FILE_READLINK:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 +
- MAX(RESBUFSIZE, (FILENAME_BYTELEN(name) +
- FILENAME_CHARSIZE)) + 1);
- FILENAME_COPY(d->b, name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
-#endif
- d->command = command;
- d->invoke = invoke_readlink;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
- case FILE_ALTNAME:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 +
- MAX(RESBUFSIZE, (FILENAME_BYTELEN(name) +
- FILENAME_CHARSIZE)) + 1);
- FILENAME_COPY(d->b, name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
-#endif
- d->command = command;
- d->invoke = invoke_altname;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
-
- case FILE_LINK:
- {
- char* new_name;
- int namelen = FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
-
- new_name = name+namelen;
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1
- + namelen
- + FILENAME_BYTELEN(new_name) + FILENAME_CHARSIZE);
-
- FILENAME_COPY(d->b, name);
- FILENAME_COPY(d->b + namelen, new_name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_s2 = d->b + namelen;
- dt_utag = buf + namelen + FILENAME_BYTELEN(dt_s2) + FILENAME_CHARSIZE;
-#endif
- d->flags = desc->flags;
- d->fd = fd;
- d->command = command;
- d->invoke = invoke_link;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
- case FILE_SYMLINK:
- {
- char* new_name;
- int namelen = FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
-
- new_name = name+namelen;
- d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1
- + namelen
- + FILENAME_BYTELEN(new_name) + FILENAME_CHARSIZE);
-
- FILENAME_COPY(d->b, name);
- FILENAME_COPY(d->b + namelen, new_name);
-#ifdef USE_VM_PROBES
- dt_s1 = d->b;
- dt_s2 = d->b + namelen;
- dt_utag = buf + namelen + FILENAME_BYTELEN(dt_s2) + FILENAME_CHARSIZE;
-#endif
- d->flags = desc->flags;
- d->fd = fd;
- d->command = command;
- d->invoke = invoke_symlink;
- d->free = free_data;
- d->level = 2;
- goto done;
- }
-
- case FILE_FADVISE:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data));
-
- d->fd = fd;
- d->command = command;
- d->invoke = invoke_fadvise;
- d->free = free_data;
- d->level = 2;
- d->c.fadvise.offset = get_int64((uchar*) buf);
- d->c.fadvise.length = get_int64(((uchar*) buf) + sizeof(Sint64));
- d->c.fadvise.advise = get_int32(((uchar*) buf) + 2 * sizeof(Sint64));
-#ifdef USE_VM_PROBES
- dt_i1 = d->fd;
- dt_i2 = d->c.fadvise.offset;
- dt_i3 = d->c.fadvise.length;
- dt_i4 = d->c.fadvise.advise;
- dt_utag = buf + 3 * sizeof(Sint64);
-#endif
- goto done;
- }
-
- case FILE_FALLOCATE:
- {
- d = EF_SAFE_ALLOC(sizeof(struct t_data));
-
- d->fd = fd;
- d->command = command;
- d->invoke = invoke_fallocate;
- d->free = free_data;
- d->level = 2;
- d->c.fallocate.offset = get_int64((uchar*) buf);
- d->c.fallocate.length = get_int64(((uchar*) buf) + sizeof(Sint64));
- goto done;
- }
-
- }
-
- /*
- * Ignore anything else -- let the caller hang.
- */
-
- return;
-
- done:
- if (d) {
-#ifdef USE_VM_PROBES
- d->sched_i1 = dt_priv->thread_num;
- d->sched_i2 = dt_priv->tag;
- d->sched_utag[0] = '\0';
- if (dt_utag != NULL) {
- if (dt_utag[0] == '\0') {
- dt_utag = NULL;
- } else {
- strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1);
- d->sched_utag[sizeof(d->sched_utag) - 1] = '\0';
- }
- }
- DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++,
- dt_utag, command, dt_s1, dt_s2,
- dt_i1, dt_i2, dt_i3, dt_i4, desc->port_str);
-#endif
- cq_enq(desc, d);
- }
-}
-
-/*********************************************************************
- * Driver entry point -> flush
- */
-static void
-file_flush(ErlDrvData e) {
- file_descriptor *desc = (file_descriptor *)e;
-#ifdef DEBUG
- int r;
-#endif
-#ifdef USE_VM_PROBES
- dt_private *dt_priv = get_dt_private(dt_driver_io_worker_base);
-#endif
-
- TRACE_C('f');
-
-#ifdef HAVE_SENDFILE
- flush_sendfile(desc, NULL);
-#endif
-
-#ifdef DEBUG
- r =
-#endif
- flush_write(desc, NULL
-#ifdef USE_VM_PROBES
- , dt_priv, (desc->d == NULL) ? NULL : desc->d->sched_utag
-#endif
- );
- /* Only possible reason for bad return value is ENOMEM, and
- * there is nobody to tell...
- */
-#ifdef DEBUG
- ASSERT(r == 0);
-#endif
- cq_execute(desc);
-}
-
-
-
-/*********************************************************************
- * Driver entry point -> control
- * Only debug functionality...
- */
-static ErlDrvSSizeT
-file_control(ErlDrvData e, unsigned int command,
- char* buf, ErlDrvSizeT len, char **rbuf, ErlDrvSizeT rlen) {
- file_descriptor *desc = (file_descriptor *)e;
- switch (command) {
- case 'K' :
- if (rlen < 4) {
- *rbuf = EF_ALLOC(4);
- }
- (*rbuf)[0] = ((desc->key) >> 24) & 0xFF;
- (*rbuf)[1] = ((desc->key) >> 16) & 0xFF;
- (*rbuf)[2] = ((desc->key) >> 8) & 0xFF;
- (*rbuf)[3] = (desc->key) & 0xFF;
- return 4;
- default:
- return 0;
- }
-}
-
-/*********************************************************************
- * Driver entry point -> timeout
- */
-static void
-file_timeout(ErlDrvData e) {
- file_descriptor *desc = (file_descriptor *)e;
- enum e_timer timer_state = desc->timer_state;
-#ifdef USE_VM_PROBES
- dt_private *dt_priv = get_dt_private(dt_driver_io_worker_base);
-#endif
-
- TRACE_C('t');
-
- desc->timer_state = timer_idle;
- switch (timer_state) {
- case timer_idle:
- ASSERT(0);
- break;
- case timer_again:
- ASSERT(desc->invoke);
- ASSERT(desc->free);
- driver_async(desc->port, KEY(desc), desc->invoke, desc->d, desc->free);
- break;
- case timer_write: {
-#ifdef DEBUG
- int r =
-#endif
- flush_write(desc, NULL
-#ifdef USE_VM_PROBES
- , dt_priv, (desc->d == NULL) ? NULL : desc->d->sched_utag
-#endif
- );
- /* Only possible reason for bad return value is ENOMEM, and
- * there is nobody to tell...
- */
- ASSERT(r == 0);
- cq_execute(desc);
- } break;
- } /* case */
-}
-
-
-
-/*********************************************************************
- * Driver entry point -> outputv
- */
-static void
-file_outputv(ErlDrvData e, ErlIOVec *ev) {
- file_descriptor* desc = (file_descriptor*)e;
- char command;
- size_t p, q;
- int err;
- struct t_data *d = NULL;
-#ifdef USE_VM_PROBES
- Sint64 dt_i1 = 0, dt_i2 = 0, dt_i3 = 0;
- Sint64 dt_i4 = 0;
- char *dt_utag = NULL;
- char *dt_s1 = NULL;
- dt_private *dt_priv = get_dt_private(dt_driver_io_worker_base);
-#endif
-
- TRACE_C('v');
-
- p = 0; q = 1;
- if (! EV_GET_CHAR(ev, &command, &p, &q)) {
- /* Empty command */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- /* 'command' contains the decoded command number,
- * 'p' and 'q' point out the next byte in the command:
- * ((char *)ev->iov[q].iov_base) + p;
- */
-
- TRACE_F(("%i", (int) command));
-
- switch (command) {
-
- case FILE_CLOSE: {
-#ifdef USE_VM_PROBES
- dt_utag = EV_CHAR_P(ev, p, q);
-#endif
- flush_read(desc);
- if (flush_write_check_error(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- if (desc->fd != FILE_FD_INVALID) {
- if (! (d = EF_ALLOC(sizeof(struct t_data)))) {
- reply_posix_error(desc, ENOMEM);
- } else {
- d->command = command;
- d->reply = !0;
- d->fd = desc->fd;
- d->flags = desc->flags;
-#ifdef USE_VM_PROBES
- dt_i1 = d->fd;
- dt_i2 = d->flags;
-#endif
- d->invoke = invoke_close;
- d->free = free_data;
- d->level = 2;
- cq_enq(desc, d);
- desc->fd = FILE_FD_INVALID;
- desc->flags = 0;
- }
- } else {
- reply_posix_error(desc, EBADF);
- }
- } goto done;
-
- case FILE_READ: {
- Uint32 sizeH, sizeL;
- size_t size, alloc_size;
-
- if (!EV_GET_UINT32(ev, &sizeH, &p, &q)
- || !EV_GET_UINT32(ev, &sizeL, &p, &q)) {
- /* Wrong buffer length to contain the read count */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-#ifdef USE_VM_PROBES
- dt_utag = EV_CHAR_P(ev, p, q);
-#endif
- if (flush_write_check_error(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
-#if ALWAYS_READ_LINE_AHEAD
- if (desc->read_bufsize == 0 && desc->read_binp != NULL && desc->read_size > 0) {
- /* We have allocated a buffer for line mode but should not really have a
- read-ahead buffer... */
- if (lseek_flush_read(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- }
-#endif
-#if SIZEOF_SIZE_T == 4
- if (sizeH != 0) {
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- size = sizeL;
-#else
- size = ((size_t)sizeH << 32) | sizeL;
-#endif
- if ((desc->fd == FILE_FD_INVALID)
- || (! (desc->flags & EFILE_MODE_READ)) ) {
- reply_posix_error(desc, EBADF);
- goto done;
- }
- if (size == 0) {
- reply_buf(desc, &command, 0);
- goto done;
- }
- if (desc->read_size >= size) {
- /* We already have all data */
- TRACE_C('D');
- reply_data(desc, desc->read_binp, desc->read_offset, size);
- desc->read_offset += size;
- desc->read_size -= size;
- try_free_read_bin(desc);
- goto done;
- }
- /* We may have some of the data
- */
- /* Justification for the following strange formula:
- * If the read request is for such a large block as more than
- * half the buffer size it may lead to a lot of unnecessary copying,
- * since the tail of the old buffer is copied to the head of the
- * new, and if the tail is almost half the buffer it is a lot
- * to copy. Therefore allocate the exact amount needed in
- * this case, giving no lingering tail. */
- alloc_size =
- size > (desc->read_bufsize>>1) ?
- size : desc->read_bufsize;
- if (! desc->read_binp) {
- /* Need to allocate a new binary for the result */
- if (! (desc->read_binp = driver_alloc_binary(alloc_size))) {
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- } else {
- /* We already have a buffer */
- if (desc->read_binp->orig_size - desc->read_offset < size) {
- /* Need to allocate a new binary for the result */
- ErlDrvBinary *binp;
- if (! (binp = driver_alloc_binary(alloc_size))) {
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- /* Move data we already have to the new binary */
- sys_memcpy(binp->orig_bytes,
- desc->read_binp->orig_bytes + desc->read_offset,
- desc->read_size);
- driver_free_binary(desc->read_binp);
- desc->read_offset = 0;
- desc->read_binp = binp;
- }
- }
- if (! (d = EF_ALLOC(sizeof(struct t_data)))) {
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- d->command = command;
- d->reply = !0;
- d->fd = desc->fd;
- d->flags = desc->flags;
- d->c.read.binp = desc->read_binp;
- d->c.read.bin_offset = desc->read_offset + desc->read_size;
- d->c.read.bin_size = desc->read_binp->orig_size - d->c.read.bin_offset;
- d->c.read.size = size;
-#ifdef USE_VM_PROBES
- dt_i1 = d->fd;
- dt_i2 = d->flags;
- dt_i3 = d->c.read.size;
-#endif
- driver_binary_inc_refc(d->c.read.binp);
- d->invoke = invoke_read;
- d->free = free_read;
- d->level = 1;
- cq_enq(desc, d);
- } goto done; /* case FILE_READ: */
-
- case FILE_READ_LINE: {
- /*
- * Icky little creature... We do mostly as ordinary file read, but with a few differences.
- * 1) We have to scan for proper newline sequence if there is a buffer already, we cannot know
- * in advance if the buffer contains a whole line without scanning.
- * 2) We do not know how large the buffer needs to be in advance. We give a default buffer,
- * but the worker may need to allocate a new one. Freeing the old and rereferencing a newly
- * allocated binary + dealing with offsets and lengts are done in file_async ready
- * for this OP.
- */
-#ifdef USE_VM_PROBES
- dt_utag = EV_CHAR_P(ev, p, q);
-#endif
- if (flush_write_check_error(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- if (ev->size != 1
-#ifdef USE_VM_PROBES
- + FILENAME_BYTELEN(dt_utag) + FILENAME_CHARSIZE
-#endif
- ) {
- /* Wrong command length */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- if ((desc->fd == FILE_FD_INVALID)
- || (! (desc->flags & EFILE_MODE_READ)) ) {
- reply_posix_error(desc, EBADF);
- goto done;
- }
- if (desc->read_size > 0) {
- /* look for '\n' in what we'we already got */
- void *nl_ptr = memchr(desc->read_binp->orig_bytes + desc->read_offset,'\n',desc->read_size);
- if (nl_ptr != NULL) {
- /* If found, we're done */
- int skip = 0;
- size_t size = ((char *) nl_ptr) -
- ((char *) (desc->read_binp->orig_bytes + desc->read_offset)) + 1;
- if (size > 1 &&
- *(((char *) nl_ptr) - 1) == '\r') {
- *(((char *) nl_ptr) - 1) = '\n';
- skip = 1;
- --size;
- }
- reply_data(desc, desc->read_binp, desc->read_offset, size);
- desc->read_offset += (size + skip);
- desc->read_size -= (size + skip);
- try_free_read_bin(desc);
- goto done;
- }
- }
- /* Now, it's up to the thread to work out the need for more buffers and such, it's
- no use doing it in this thread as we do not have the information required anyway.
- Even a NULL buffer could be handled by the thread, but code is simplified by us
- allocating it */
- if (! desc->read_binp) {
- int alloc_size = (desc->read_bufsize > DEFAULT_LINEBUF_SIZE) ? desc->read_bufsize :
- DEFAULT_LINEBUF_SIZE;
- /* Allocate a new binary for the result */
- if (! (desc->read_binp = driver_alloc_binary(alloc_size))) {
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- }
- if (! (d = EF_ALLOC(sizeof(struct t_data)))) {
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
-
- d->command = command;
- d->reply = !0;
- d->fd = desc->fd;
- d->flags = desc->flags;
- d->c.read_line.binp = desc->read_binp;
- d->c.read_line.read_offset = desc->read_offset;
- d->c.read_line.read_size = desc->read_size;
-#ifdef USE_VM_PROBES
- dt_i1 = d->fd;
- dt_i2 = d->flags;
- dt_i3 = d->c.read_line.read_offset;
-#endif
-#if !ALWAYS_READ_LINE_AHEAD
- d->c.read_line.read_ahead = (desc->read_bufsize > 0);
-#ifdef USE_VM_PROBES
- dt_i4 = d->c.read_line.read_ahead;
-#endif
-#endif
- driver_binary_inc_refc(d->c.read.binp);
- d->invoke = invoke_read_line;
- d->free = free_read_line;
- d->level = 1;
- cq_enq(desc, d);
- } goto done;
- case FILE_WRITE: { /* Dtrace: The dtrace user tag is not last in message,
- but follows the message tag directly.
- This is handled specially in prim_file.erl */
- ErlDrvSizeT skip = 1;
- ErlDrvSizeT size = ev->size - skip;
-
-#ifdef USE_VM_PROBES
- dt_utag = EV_CHAR_P(ev, p, q);
- skip += FILENAME_BYTELEN(dt_utag) + FILENAME_CHARSIZE;
- size = ev->size - skip;
-#endif
- if (lseek_flush_read(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- if (! (desc->flags & EFILE_MODE_WRITE)) {
- reply_posix_error(desc, EBADF);
- goto done;
- }
- if (size == 0) {
- reply_Uint(desc, size);
- goto done;
- }
- MUTEX_LOCK(desc->q_mtx);
- if (driver_enqv(desc->port, ev, skip)) {
- MUTEX_UNLOCK(desc->q_mtx);
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- desc->write_buffered += size;
- if (desc->write_buffered < desc->write_bufsize) {
- MUTEX_UNLOCK(desc->q_mtx);
- reply_Uint(desc, size);
- if (desc->timer_state == timer_idle) {
- desc->timer_state = timer_write;
- driver_set_timer(desc->port, desc->write_delay);
- }
- } else {
- if ((d = async_write(desc, &err, !0, size
-#ifdef USE_VM_PROBES
- , &dt_i1, &dt_i2, &dt_i3
-#endif
- )) == NULL) {
- MUTEX_UNLOCK(desc->q_mtx);
- reply_posix_error(desc, err);
- goto done;
- } else {
- MUTEX_UNLOCK(desc->q_mtx);
- }
- }
- } goto done; /* case FILE_WRITE */
-
- case FILE_PWRITEV: { /* Dtrace: The dtrace user tag is not last in message,
- but follows the message tag directly.
- This is handled specially in prim_file.erl */
- Uint32 i, j, n;
- size_t total;
-#ifdef USE_VM_PROBES
- char dt_tmp;
- int dt_utag_bytes = 1;
-
- dt_utag = EV_CHAR_P(ev, p, q);
- /* This will work for UTF-8, but not for UTF-16 - extra reminder here */
-#ifdef FILENAMES_16BIT
-#error 16bit characters in filenames and dtrace in combination is not supported.
-#endif
- while (EV_GET_CHAR(ev, &dt_tmp, &p, &q) && dt_tmp != '\0') {
- dt_utag_bytes++;
- }
-#endif
- if (ev->size < 1+4
-#ifdef USE_VM_PROBES
- + dt_utag_bytes
-#endif
- || !EV_GET_UINT32(ev, &n, &p, &q)) {
- /* Buffer too short to contain even the number of pos/size specs */
- reply_Uint_posix_error(desc, 0, EINVAL);
- goto done;
- }
- if (lseek_flush_read(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_Uint_posix_error(desc, 0, err);
- goto done;
- }
- if (flush_write_check_error(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_Uint_posix_error(desc, 0, err);
- goto done;
- }
- if (n == 0) {
- /* Trivial case - nothing to write */
- if (ev->size != 1+4) {
- reply_posix_error(desc, err);
- } else {
- reply_Uint(desc, 0);
- }
- goto done;
- }
- if (ev->size < 1+4+8*(2*n)
-#ifdef USE_VM_PROBES
- + dt_utag_bytes
-#endif
- ) {
- /* Buffer too short to contain even the pos/size specs */
- reply_Uint_posix_error(desc, 0, EINVAL);
- goto done;
- }
- d = EF_ALLOC(sizeof(struct t_data)
- + (n * sizeof(struct t_pbuf_spec)));
- if (! d) {
- reply_Uint_posix_error(desc, 0, ENOMEM);
- goto done;
- }
- d->command = command;
- d->reply = !0;
- d->fd = desc->fd;
- d->flags = desc->flags;
-#ifdef USE_VM_PROBES
- dt_i1 = d->fd;
- dt_i2 = d->flags;
-#endif
- d->c.pwritev.port = desc->port;
- d->c.pwritev.q_mtx = desc->q_mtx;
- d->c.pwritev.n = n;
- d->c.pwritev.cnt = 0;
- total = 0;
- j = 0;
- /* Create pos/size specs in the thread data structure
- * for all non-zero size binaries. Calculate total size.
- */
- for(i = 0; i < n; i++) {
- Uint32 sizeH, sizeL;
- size_t size;
- if ( !EV_GET_SINT64(ev, &d->c.pwritev.specs[i].offset, &p, &q)
- || !EV_GET_UINT32(ev, &sizeH, &p, &q)
- || !EV_GET_UINT32(ev, &sizeL, &p, &q)) {
- /* Misalignment in buffer */
- reply_Uint_posix_error(desc, 0, EINVAL);
- EF_FREE(d);
- goto done;
- }
-#if SIZEOF_SIZE_T == 4
- if (sizeH != 0) {
- reply_Uint_posix_error(desc, 0, EINVAL);
- EF_FREE(d);
- goto done;
- }
- size = sizeL;
-#else
- size = ((size_t)sizeH<<32) | sizeL;
-#endif
- if (size > 0) {
- total += size;
- d->c.pwritev.specs[j].size = size;
- j++;
- }
- }
- d->c.pwritev.size = total;
-#ifdef USE_VM_PROBES
- dt_i3 = d->c.pwritev.size;
-#endif
- if (j == 0) {
- /* Trivial case - nothing to write */
- EF_FREE(d);
- reply_Uint(desc, 0);
- } else {
- ErlDrvSizeT skip = 1 + 4 + 8 * (2*n)
-#ifdef USE_VM_PROBES
- + dt_utag_bytes
-#endif
- ;
- if (skip + total != ev->size) {
- /* Actual amount of data does not match
- * total of all pos/size specs
- */
- EF_FREE(d);
- reply_Uint_posix_error(desc, 0, EINVAL);
- } else {
- /* Enqueue the data */
- MUTEX_LOCK(desc->q_mtx);
- driver_enqv(desc->port, ev, skip);
- MUTEX_UNLOCK(desc->q_mtx);
- /* Execute the command */
- d->invoke = invoke_pwritev;
- d->free = free_data;
- d->level = 1;
- cq_enq(desc, d);
- }
- }
- } goto done; /* case FILE_PWRITEV: */
-
- case FILE_PREADV: { /* Dtrace: The dtrace user tag is not last in message,
- but follows the message tag directly.
- This is handled specially in prim_file.erl */
- register void * void_ptr;
- Uint32 i, n;
- ErlIOVec *res_ev;
-#ifdef USE_VM_PROBES
- char dt_tmp;
- int dt_utag_bytes = 1;
- /* This will work for UTF-8, but not for UTF-16 - extra reminder here */
-#ifdef FILENAMES_16BIT
-#error 16bit characters in filenames and dtrace in combination is not supported.
-#endif
- dt_utag = EV_CHAR_P(ev, p, q);
- while (EV_GET_CHAR(ev, &dt_tmp, &p, &q) && dt_tmp != '\0') {
- dt_utag_bytes++;
- }
-#endif
- if (lseek_flush_read(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- if (flush_write_check_error(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- if (ev->size < 1+8
-#ifdef USE_VM_PROBES
- + dt_utag_bytes
-#endif
- || !EV_GET_UINT32(ev, &n, &p, &q)
- || !EV_GET_UINT32(ev, &n, &p, &q)) {
- /* Buffer too short to contain even the number of pos/size specs */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- if (ev->size < 1+8+8*(2*n)
-#ifdef USE_VM_PROBES
- + dt_utag_bytes
-#endif
- ) {
- /* Buffer wrong length to contain the pos/size specs */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- /* Create the thread data structure with the contained ErlIOVec
- * and corresponding binaries for the response
- */
- d = EF_ALLOC(sizeof(*d)
- + (n * sizeof(*d->c.preadv.offsets))
- + ((1+n) * (sizeof(*res_ev->iov)
- + sizeof(*res_ev->binv))));
- if (! d) {
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- d->command = command;
- d->reply = !0;
- d->fd = desc->fd;
- d->flags = desc->flags;
-#ifdef USE_VM_PROBES
- dt_i1 = d->fd;
- dt_i2 = d->flags;
-#endif
- d->c.preadv.n = n;
- d->c.preadv.cnt = 0;
- d->c.preadv.size = 0;
- res_ev = &d->c.preadv.eiov;
- /* XXX possible alignment problems here for weird machines */
- res_ev->vsize = 1+d->c.preadv.n;
- res_ev->iov = void_ptr = &d->c.preadv.offsets[d->c.preadv.n];
- res_ev->binv = void_ptr = &res_ev->iov[res_ev->vsize];
- /* Read in the pos/size specs and allocate binaries for the results */
- for (i = 1; i < 1+n; i++) {
- Uint32 sizeH, sizeL;
- size_t size;
- if ( !EV_GET_SINT64(ev, &d->c.preadv.offsets[i-1], &p, &q)
- || !EV_GET_UINT32(ev, &sizeH, &p, &q)
- || !EV_GET_UINT32(ev, &sizeL, &p, &q)) {
- reply_posix_error(desc, EINVAL);
- break;
- }
-#if SIZEOF_SIZE_T == 4
- if (sizeH != 0) {
- reply_posix_error(desc, EINVAL);
- break;
- }
- size = sizeL;
-#else
- size = ((size_t)sizeH<<32) | sizeL;
-#endif
-#ifdef USE_VM_PROBES
- dt_i3 += size;
-#endif
- if (! (res_ev->binv[i] = driver_alloc_binary(size))) {
- reply_posix_error(desc, ENOMEM);
- break;
- } else {
- res_ev->iov[i].iov_len = size;
- res_ev->iov[i].iov_base = res_ev->binv[i]->orig_bytes;
- }
- }
- if (i < 1+n) {
- for (i--; i > 0; i--) {
- driver_free_binary(res_ev->binv[i]);
- }
- EF_FREE(d);
- goto done;
- }
- /* Allocate the header binary (index 0) */
- res_ev->binv[0] = driver_alloc_binary(4+4+8*n);
- if (! res_ev->binv[0]) {
- reply_posix_error(desc, ENOMEM);
- for (i = 1; i < 1+n; i++) {
- driver_free_binary(res_ev->binv[i]);
- }
- EF_FREE(d);
- goto done;
- }
- res_ev->iov[0].iov_len = 4+4+8*n;
- res_ev->iov[0].iov_base = res_ev->binv[0]->orig_bytes;
- /* Fill in the number of buffers in the header */
- put_int32(0, res_ev->iov[0].iov_base);
- put_int32(n, (char *)(res_ev->iov[0].iov_base) + 4);
- /**/
- res_ev->size = res_ev->iov[0].iov_len;
- if (n == 0) {
- /* Trivial case - nothing to read */
- reply_ev(desc, FILE_RESP_LDATA, res_ev);
- free_preadv(d);
- goto done;
- } else {
- d->invoke = invoke_preadv;
- d->free = free_preadv;
- d->level = 1;
- cq_enq(desc, d);
- }
- } goto done; /* case FILE_PREADV: */
-
- case FILE_LSEEK: {
- Sint64 offset; /* Offset for seek */
- Uint32 origin; /* Origin of seek. */
-
- if (ev->size < 1+8+4
- || !EV_GET_SINT64(ev, &offset, &p, &q)
- || !EV_GET_UINT32(ev, &origin, &p, &q)) {
- /* Wrong length of buffer to contain offset and origin */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-#ifdef USE_VM_PROBES
- dt_utag = EV_CHAR_P(ev, p, q);
-#endif
- if (lseek_flush_read(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- if (flush_write_check_error(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- if ((d = async_lseek(desc, &err, !0, offset, origin
-#ifdef USE_VM_PROBES
- , &dt_i1, &dt_i2, &dt_i3
-#endif
- )) == NULL) {
- reply_posix_error(desc, err);
- goto done;
- }
- } goto done;
-
- case FILE_READ_FILE: {
- char *filename;
- if (ev->size < 1+1) {
- /* Buffer contains empty name */
- reply_posix_error(desc, ENOENT);
- goto done;
- }
-#ifndef USE_VM_PROBES
- /* In the dtrace case, the iov has an extra element, the dtrace utag - we will need
- another test to see that
- the filename is in a single buffer: */
- if (ev->size-1 != ev->iov[q].iov_len-p) {
- /* Name not in one single buffer */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-#else
- if (((byte *)ev->iov[q].iov_base)[ev->iov[q].iov_len-1] != '\0') {
- /* Name not in one single buffer */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-#endif
- filename = EV_CHAR_P(ev, p, q);
- d = EF_ALLOC(sizeof(struct t_data) -1 + FILENAME_BYTELEN(filename) + FILENAME_CHARSIZE);
- if (! d) {
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- d->command = command;
- d->reply = !0;
- /* Copy name */
- FILENAME_COPY(d->b, filename);
-#ifdef USE_VM_PROBES
- {
- char dt_tmp;
-
- /* This will work for UTF-8, but not for UTF-16 - extra reminder here */
-#ifdef FILENAMES_16BIT
-#error 16bit characters in filenames and dtrace in combination is not supported.
-#endif
- while (EV_GET_CHAR(ev, &dt_tmp, &p, &q) && dt_tmp != '\0')
- ;
- dt_s1 = d->b;
- dt_utag = EV_CHAR_P(ev, p, q);
- }
-#endif
- d->c.read_file.binp = NULL;
- d->invoke = invoke_read_file;
- d->free = free_read_file;
- d->level = 2;
- cq_enq(desc, d);
- } goto done;
-
- case FILE_IPREAD: {
- /* This operation cheets by using invoke_preadv() and free_preadv()
- * plus its own invoke_ipread. Therefore the result format is
- * a bit awkward - the header binary contains one extra 64 bit
- * field that invoke_preadv() fortunately ignores,
- * and the first 64 bit field does not contain the number of
- * data binaries which invoke_preadv() also ignores.
- */
- register void * void_ptr;
- char mode;
- Sint64 hdr_offset;
- Uint32 max_size;
- ErlIOVec *res_ev;
- int vsize;
- if (! EV_GET_CHAR(ev, &mode, &p, &q)) {
- /* Empty command */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- if (mode != IPREAD_S32BU_P32BU) {
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- if (ev->size < 1+1+8+4
- || !EV_GET_SINT64(ev, &hdr_offset, &p, &q)
- || !EV_GET_UINT32(ev, &max_size, &p, &q)) {
- /* Buffer too short to contain
- * the header offset and max size spec */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-#ifdef USE_VM_PROBES
- dt_utag = EV_CHAR_P(ev, p, q);
-#endif
- if (lseek_flush_read(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- if (flush_write_check_error(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- /* Create the thread data structure with the contained ErlIOVec
- * and corresponding binaries for the response
- */
- vsize = 2;
- d = EF_ALLOC(sizeof(*d) +
- vsize*(sizeof(*res_ev->iov) + sizeof(*res_ev->binv)));
- if (! d) {
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- d->command = command;
- d->reply = !0;
- d->fd = desc->fd;
- d->flags = desc->flags;
- d->c.preadv.offsets[0] = hdr_offset;
- d->c.preadv.size = max_size;
-#ifdef USE_VM_PROBES
- dt_i1 = d->fd;
- dt_i2 = d->flags;
- dt_i3 = d->c.preadv.offsets[0];
- dt_i4 = d->c.preadv.size;
-#endif
- res_ev = &d->c.preadv.eiov;
- /* XXX possible alignment problems here for weird machines */
- res_ev->iov = void_ptr = d + 1;
- res_ev->binv = void_ptr = res_ev->iov + vsize;
- res_ev->size = 0;
- res_ev->vsize = 0;
- d->invoke = invoke_ipread;
- d->free = free_preadv;
- d->level = 1;
- cq_enq(desc, d);
- } goto done; /* case FILE_IPREAD: */
-
- case FILE_SETOPT: {
- char opt;
-
- if (ev->size < 1+1
- || !EV_GET_CHAR(ev, &opt, &p, &q)) {
- /* Buffer too short to contain even the option type */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-#ifdef USE_VM_PROBES
- dt_i1 = opt;
- dt_utag = EV_CHAR_P(ev, p, q);
-#endif
- switch (opt) {
- case FILE_OPT_DELAYED_WRITE: {
- Uint32 sizeH, sizeL, delayH, delayL;
- if (ev->size != 1+1+4*sizeof(Uint32)
-#ifdef USE_VM_PROBES
- + FILENAME_BYTELEN(dt_utag) + FILENAME_CHARSIZE
-#endif
- || !EV_GET_UINT32(ev, &sizeH, &p, &q)
- || !EV_GET_UINT32(ev, &sizeL, &p, &q)
- || !EV_GET_UINT32(ev, &delayH, &p, &q)
- || !EV_GET_UINT32(ev, &delayL, &p, &q)) {
- /* Buffer has wrong length to contain the option values */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-#if SIZEOF_SIZE_T == 4
- if (sizeH != 0) {
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- desc->write_bufsize = sizeL;
-#else
- desc->write_bufsize = ((size_t)sizeH << 32) | sizeL;
-#endif
-#if SIZEOF_LONG == 4
- if (delayH != 0) {
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- desc->write_delay = delayL;
-#else
- desc->write_delay = ((unsigned long)delayH << 32) | delayL;
-#endif
-#ifdef USE_VM_PROBES
- dt_i2 = desc->write_delay;
-#endif
- TRACE_C('K');
- reply_ok(desc);
- } goto done;
- case FILE_OPT_READ_AHEAD: {
- Uint32 sizeH, sizeL;
- if (ev->size != 1+1+2*sizeof(Uint32)
-#ifdef USE_VM_PROBES
- + FILENAME_BYTELEN(dt_utag)+FILENAME_CHARSIZE
-#endif
- || !EV_GET_UINT32(ev, &sizeH, &p, &q)
- || !EV_GET_UINT32(ev, &sizeL, &p, &q)) {
- /* Buffer has wrong length to contain the option values */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-#if SIZEOF_SIZE_T == 4
- if (sizeH != 0) {
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- desc->read_bufsize = sizeL;
-#else
- desc->read_bufsize = ((size_t)sizeH << 32) | sizeL;
-#endif
-#ifdef USE_VM_PROBES
- dt_i2 = desc->read_bufsize;
-#endif
- TRACE_C('K');
- reply_ok(desc);
- } goto done;
- default:
- reply_posix_error(desc, EINVAL);
- goto done;
- } /* case FILE_OPT_DELAYED_WRITE: */
- } ASSERT(0); goto done; /* case FILE_SETOPT: */
-
- case FILE_SENDFILE: {
-
-#ifdef HAVE_SENDFILE
- struct t_data *d;
- Uint32 out_fd, offsetH, offsetL, hd_len, tl_len;
- Uint64 nbytes;
- char flags;
-
- if (ev->size < 1 + 7 * sizeof(Uint32) + sizeof(char)
- || !EV_GET_UINT32(ev, &out_fd, &p, &q)
- || !EV_GET_CHAR(ev, &flags, &p, &q)
- || !EV_GET_UINT32(ev, &offsetH, &p, &q)
- || !EV_GET_UINT32(ev, &offsetL, &p, &q)
- || !EV_GET_UINT64(ev, &nbytes, &p, &q)
- || !EV_GET_UINT32(ev, &hd_len, &p, &q)
- || !EV_GET_UINT32(ev, &tl_len, &p, &q)) {
- /* Buffer has wrong length to contain all the needed values */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-
- if (hd_len != 0 || tl_len != 0) {
- /* We do not allow header, trailers */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-
-
- if (flags & SENDFILE_FLGS_USE_THREADS && !THRDS_AVAILABLE) {
- /* We do not allow use_threads flag on a system where
- no threads are available. */
- reply_posix_error(desc, EINVAL);
- goto done;
- }
-
- d = EF_SAFE_ALLOC(sizeof(struct t_data));
- d->fd = desc->fd;
- d->command = command;
- d->invoke = invoke_sendfile;
- d->free = free_sendfile;
- d->flags = flags;
- d->level = 2;
-
- d->c.sendfile.out_fd = (int) out_fd;
- d->c.sendfile.written = 0;
- d->c.sendfile.port = desc->port;
- d->c.sendfile.q_mtx = desc->q_mtx;
-
- #if SIZEOF_OFF_T == 4
- if (offsetH != 0) {
- reply_posix_error(desc, EINVAL);
- goto done;
- }
- d->c.sendfile.offset = (off_t) offsetL;
- #else
- d->c.sendfile.offset = ((off_t) offsetH << 32) | offsetL;
- #endif
-
- d->c.sendfile.nbytes = nbytes;
-
- if (USE_THRDS_FOR_SENDFILE(d)) {
- SET_BLOCKING(d->c.sendfile.out_fd);
- } else {
- /**
- * Write a place holder to queue in order to force file_flush
- * to be called before the driver is closed.
- */
- char tmp[1] = "";
- MUTEX_LOCK(d->c.sendfile.q_mtx);
- if (driver_enq(d->c.sendfile.port, tmp, 1)) {
- MUTEX_UNLOCK(d->c.sendfile.q_mtx);
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- MUTEX_UNLOCK(d->c.sendfile.q_mtx);
- }
-
- cq_enq(desc, d);
-#else
- reply_posix_error(desc, ENOTSUP);
-#endif
- goto done;
- } /* case FILE_SENDFILE: */
-
- } /* switch(command) */
-
- if (lseek_flush_read(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- }
- if (flush_write_check_error(desc, &err
-#ifdef USE_VM_PROBES
- , dt_priv, dt_utag
-#endif
- ) < 0) {
- reply_posix_error(desc, err);
- goto done;
- } else {
- /* Flatten buffer and send it to file_output(desc, buf, len) */
- int len = ev->size;
- char *buf = EF_ALLOC(len);
- if (! buf) {
- reply_posix_error(desc, ENOMEM);
- goto done;
- }
- driver_vec_to_buf(ev, buf, len);
- file_output((ErlDrvData) desc, buf, len);
- EF_FREE(buf);
- goto done;
- }
-
- done:
- if (d != NULL) {
-#ifdef USE_VM_PROBES
- /*
- * If d == NULL, then either:
- * 1). There was an error of some sort, or
- * 2). The command given to us is actually implemented
- * by file_output() instead.
- *
- * Case #1 is probably a TODO item, perhaps?
- * Case #2 we definitely don't want to activate a probe.
- */
- d->sched_i1 = dt_priv->thread_num;
- d->sched_i2 = dt_priv->tag;
- d->sched_utag[0] = '\0';
- if (dt_utag != NULL) {
- if (dt_utag[0] == '\0') {
- dt_utag = NULL;
- } else {
- strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1);
- d->sched_utag[sizeof(d->sched_utag) - 1] = '\0';
- }
- }
- DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++,
- dt_utag, command, dt_s1, NULL, dt_i1, dt_i2, dt_i3, dt_i4,
- desc->port_str);
-#endif
- }
- cq_execute(desc);
-}
-
-#ifdef USE_VM_PROBES
-dt_private *
-get_dt_private(int base)
-{
- dt_private *dt_priv = (dt_private *) pthread_getspecific(dt_driver_key);
-
- if (dt_priv == NULL) {
- dt_priv = EF_SAFE_ALLOC(sizeof(dt_private));
- erts_mtx_lock(&dt_driver_mutex);
- dt_priv->thread_num = (base + dt_driver_idnum++);
- erts_mtx_unlock(&dt_driver_mutex);
- dt_priv->tag = 0;
- pthread_setspecific(dt_driver_key, dt_priv);
- }
- return dt_priv;
-}
-#endif /* USE_VM_PROBES */
diff --git a/erts/emulator/drivers/common/erl_efile.h b/erts/emulator/drivers/common/erl_efile.h
deleted file mode 100644
index b7f063b4f2..0000000000
--- a/erts/emulator/drivers/common/erl_efile.h
+++ /dev/null
@@ -1,176 +0,0 @@
-/*
- * %CopyrightBegin%
- *
- * Copyright Ericsson AB 1997-2016. All Rights Reserved.
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- *
- * %CopyrightEnd%
- */
-/*
- * Defines the interfaces between the generic efile driver and its
- * operating-system dependent helpers.
- */
-
-#include "sys.h"
-#include "erl_driver.h"
-
-/*
- * Open modes for efile_openfile().
- */
-#define EFILE_MODE_READ 1
-#define EFILE_MODE_WRITE 2 /* Implies truncating file when used alone. */
-#define EFILE_MODE_READ_WRITE 3
-#define EFILE_MODE_APPEND 4
-#define EFILE_COMPRESSED 8
-#define EFILE_MODE_EXCL 16
-#define EFILE_NO_TRUNCATE 32 /* Special for reopening on VxWorks */
-#define EFILE_MODE_SYNC 64
-
-/*
- * Seek modes for efile_seek().
- */
-#define EFILE_SEEK_SET 0
-#define EFILE_SEEK_CUR 1
-#define EFILE_SEEK_END 2
-
-/*
- * File types returned by efile_fileinfo().
- */
-#define FT_DEVICE 1
-#define FT_DIRECTORY 2
-#define FT_REGULAR 3
-#define FT_SYMLINK 4
-#define FT_OTHER 5
-
-/*
- * Access attributes returned by efile_fileinfo() (the bits can be ORed
- * together).
- */
-#define FA_NONE 0
-#define FA_WRITE 1
-#define FA_READ 2
-
-/* Some OS'es (i.e. Windows) has filenames in wide charaqcters. That requires special handling */
-/* Note that we do *not* honor alignment in the communication to the OS specific driver, */
-/* which is not a problem on x86, but might be on other platforms. The OS specific efile */
-/* implementation is expected to align if needed */
-#ifdef __WIN32__
-#define FILENAMES_16BIT 1
-#endif
-
-/* We use sendfilev if it exist on solaris */
-#if !defined(HAVE_SENDFILE) && defined(HAVE_SENDFILEV)
-#define HAVE_SENDFILE
-#endif
-
-/*
- * An handle to an open directory. To be cast to the correct type
- * in the system-dependent directory functions.
- */
-
-typedef struct _Efile_Dir_Handle* EFILE_DIR_HANDLE;
-
-/*
- * Error information from the last call.
- */
-typedef struct _Efile_error {
- int posix_errno; /* Posix error number, as in <errno.h>. */
- int os_errno; /* Os-dependent error number (not used). */
-} Efile_error;
-
-/*
- * Describes what is returned by file:file_info/1.
- */
-
-typedef struct _Efile_info {
- Uint32 size_low; /* Size of file, lower 32 bits.. */
- Uint32 size_high; /* Size of file, higher 32 bits. */
- Uint32 type; /* Type of file -- one of FT_*. */
- Uint32 access; /* Access to file -- one of FA_*. */
- Uint32 mode; /* Access permissions -- bit field. */
- Uint32 links; /* Number of links to file. */
- Uint32 major_device; /* Major device or file system. */
- Uint32 minor_device; /* Minor device (for devices). */
- Uint32 inode; /* Inode number. */
- Uint32 uid; /* User id of owner. */
- Uint32 gid; /* Group id of owner. */
- Sint64 accessTime; /* Last time the file was accessed. */
- Sint64 modifyTime; /* Last time the file was modified. */
- Sint64 cTime; /* Creation time (Windows) or last
- * inode change (Unix).
- */
-} Efile_info;
-
-
-#ifdef HAVE_SENDFILE
-/*
- * Describes the structure of headers/trailers for sendfile
- */
-struct t_sendfile_hdtl {
- SysIOVec *headers;
- int hdr_cnt;
- SysIOVec *trailers;
- int trl_cnt;
-};
-#endif /* HAVE_SENDFILE */
-
-/*
- * Functions.
- */
-int efile_init(void);
-int efile_mkdir(Efile_error* errInfo, char* name);
-int efile_rmdir(Efile_error* errInfo, char* name);
-int efile_delete_file(Efile_error* errInfo, char* name);
-int efile_rename(Efile_error* errInfo, char* src, char* dst);
-int efile_chdir(Efile_error* errInfo, char* name);
-int efile_getdcwd(Efile_error* errInfo, int drive,
- char* buffer, size_t size);
-int efile_readdir(Efile_error* errInfo, char* name,
- EFILE_DIR_HANDLE* dir_handle,
- char* buffer, size_t *size);
-int efile_openfile(Efile_error* errInfo, char* name, int flags,
- int* pfd, Sint64* pSize);
-void efile_closefile(int fd);
-int efile_fdatasync(Efile_error* errInfo, int fd);
-int efile_fsync(Efile_error* errInfo, int fd);
-int efile_fileinfo(Efile_error* errInfo, Efile_info* pInfo,
- char *name, int info_for_link);
-int efile_write_info(Efile_error* errInfo, Efile_info* pInfo, char *name);
-int efile_write(Efile_error* errInfo, int flags, int fd,
- char* buf, size_t count);
-int efile_writev(Efile_error* errInfo, int flags, int fd,
- SysIOVec* iov, int iovcnt);
-int efile_read(Efile_error* errInfo, int flags, int fd,
- char* buf, size_t count, size_t* pBytesRead);
-int efile_seek(Efile_error* errInfo, int fd,
- Sint64 offset, int origin, Sint64* new_location);
-int efile_truncate_file(Efile_error* errInfo, int *fd, int flags);
-int efile_pwrite(Efile_error* errInfo, int fd,
- char* buf, size_t count, Sint64 offset);
-int efile_pread(Efile_error* errInfo, int fd,
- Sint64 offset, char* buf, size_t count, size_t* pBytesRead);
-int efile_readlink(Efile_error* errInfo, char *name,
- char* buffer, size_t size);
-int efile_altname(Efile_error* errInfo, char *name,
- char* buffer, size_t size);
-int efile_link(Efile_error* errInfo, char* old, char* new);
-int efile_symlink(Efile_error* errInfo, char* old, char* new);
-int efile_may_openfile(Efile_error* errInfo, char *name);
-int efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length,
- int advise);
-#ifdef HAVE_SENDFILE
-int efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
- off_t *offset, Uint64 *nbytes, struct t_sendfile_hdtl *hdtl);
-#endif /* HAVE_SENDFILE */
-int efile_fallocate(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length);
diff --git a/erts/emulator/drivers/common/gzio.c b/erts/emulator/drivers/common/gzio.c
index f60c781894..86c3b07cea 100644
--- a/erts/emulator/drivers/common/gzio.c
+++ b/erts/emulator/drivers/common/gzio.c
@@ -19,726 +19,16 @@
#include <unistd.h>
#endif
#include <ctype.h>
+
#include "erl_driver.h"
-#include "erl_efile.h"
#include "sys.h"
-#ifdef __WIN32__
-#ifndef HAVE_CONFLICTING_FREAD_DECLARATION
-#define HAVE_CONFLICTING_FREAD_DECLARATION
-#endif
-#define FILENAMES_16BIT 1
-#endif
-
-#ifdef STDC
-# define zstrerror(errnum) strerror(errnum)
-#else
-# define zstrerror(errnum) ""
-#endif
-
#include "gzio_zutil.h"
#include "erl_zlib.h"
#include "gzio.h"
-/********struct internal_state {int dummy;}; / * for buggy compilers */
-
-#define Z_BUFSIZE 4096
-
-#define ALLOC(size) driver_alloc(size)
-#define TRYFREE(p) {if (p) driver_free(p);}
-
static int gz_magic[2] = {0x1f, 0x8b}; /* gzip magic header */
-/* gzip flag byte */
-#define ASCII_FLAG 0x01 /* bit 0 set: file probably ascii text */
-#define HEAD_CRC 0x02 /* bit 1 set: header CRC present */
-#define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */
-#define ORIG_NAME 0x08 /* bit 3 set: original file name present */
-#define COMMENT 0x10 /* bit 4 set: file comment present */
-#define RESERVED 0xE0 /* bits 5..7: reserved */
-
-typedef struct gz_stream {
- z_stream stream;
- int z_err; /* error code for last stream operation */
- int z_eof; /* set if end of input file */
-#ifdef UNIX
- int file; /* .gz file descriptor */
-#else
- FILE *file; /* .gz file */
-#endif
- Byte *inbuf; /* input buffer */
- Byte *outbuf; /* output buffer */
- uLong crc; /* crc32 of uncompressed data */
- char *msg; /* error message */
- char *path; /* path name for debugging only */
- int transparent; /* 1 if input file is not a .gz file */
- char mode; /* 'w' or 'r' */
- int position; /* Position (for seek) */
- int (*destroy)(struct gz_stream*); /* Function to destroy
- * this structure. */
-} gz_stream;
-
-local ErtsGzFile gz_open (const char *path, const char *mode);
-local int get_byte (gz_stream *s);
-local void check_header (gz_stream *s);
-local int destroy (gz_stream *s);
-local uLong getLong (gz_stream *s);
-
-#ifdef UNIX
-/*
- * In Solaris 8 and earlier, fopen() and its friends cannot handle
- * file descriptors larger than 255. Therefore, we use read()/write()
- * on all Unix systems.
- */
-# define ERTS_GZWRITE(File, Buf, Count) write((File), (Buf), (Count))
-# define ERTS_GZREAD(File, Buf, Count) read((File), (Buf), (Count))
-#else
-/*
- * On all other operating systems, using fopen(), fread()/fwrite(), since
- * there is not guaranteed to exist any read()/write() (not part of
- * ANSI/ISO-C).
- */
-# define ERTS_GZWRITE(File, Buf, Count) fwrite((Buf), 1, (Count), (File))
-# define ERTS_GZREAD(File, Buf, Count) fread((Buf), 1, (Count), (File))
-#endif
-
-/*
- * Ripped from efile_drv.c
- */
-
-#ifdef FILENAMES_16BIT
-# define FILENAME_BYTELEN(Str) filename_len_16bit(Str)
-# define FILENAME_COPY(To,From) filename_cpy_16bit((To),(From))
-# define FILENAME_CHARSIZE 2
-
- static int filename_len_16bit(const char *str)
- {
- const char *p = str;
- while(*p != '\0' || p[1] != '\0') {
- p += 2;
- }
- return (p - str);
- }
-
- static void filename_cpy_16bit(char *to, const char *from)
- {
- while(*from != '\0' || from[1] != '\0') {
- *to++ = *from++;
- *to++ = *from++;
- }
- *to++ = *from++;
- *to++ = *from++;
- }
-
-#else
-# define FILENAME_BYTELEN(Str) strlen(Str)
-# define FILENAME_COPY(To,From) strcpy(To,From)
-# define FILENAME_CHARSIZE 1
-#endif
-
-/* ===========================================================================
- Opens a gzip (.gz) file for reading or writing. The mode parameter
- is as in fopen ("rb" or "wb"). The file is given either by file descriptor
- or path name (if fd == -1).
- gz_open return NULL if the file could not be opened or if there was
- insufficient memory to allocate the (de)compression state; errno
- can be checked to distinguish the two cases (if errno is zero, the
- zlib error is Z_MEM_ERROR).
-*/
-local ErtsGzFile gz_open (path, mode)
- const char *path;
- const char *mode;
-{
- int err;
- int level = Z_DEFAULT_COMPRESSION; /* compression level */
- char *p = (char*)mode;
- gz_stream *s;
- char fmode[80]; /* copy of mode, without the compression level */
- char *m = fmode;
-
- if (!path || !mode) return Z_NULL;
-
- s = (gz_stream *)ALLOC(sizeof(gz_stream));
- if (!s) return Z_NULL;
-
- erl_zlib_alloc_init(&s->stream);
- s->stream.next_in = s->inbuf = Z_NULL;
- s->stream.next_out = s->outbuf = Z_NULL;
- s->stream.avail_in = s->stream.avail_out = 0;
-#ifdef UNIX
- s->file = -1;
-#else
- s->file = NULL;
-#endif
- s->z_err = Z_OK;
- s->z_eof = 0;
- s->crc = crc32(0L, Z_NULL, 0);
- s->msg = NULL;
- s->transparent = 0;
- s->position = 0;
- s->destroy = destroy;
-
- s->path = (char*)ALLOC(FILENAME_BYTELEN(path)+FILENAME_CHARSIZE);
- if (s->path == NULL) {
- return s->destroy(s), (ErtsGzFile)Z_NULL;
- }
- FILENAME_COPY(s->path, path); /* do this early for debugging */
-
- s->mode = '\0';
- do {
- if (*p == 'r')
- s->mode = 'r';
- if (*p == 'w' || *p == 'a')
- s->mode = 'w';
- if (isdigit((int)*p)) {
- level = *p - '0';
- } else {
- *m++ = *p; /* Copy the mode */
- }
- } while (*p++ && m < fmode + sizeof(fmode) - 1);
- *m = '\0';
- if (s->mode == '\0')
- return s->destroy(s), (ErtsGzFile)Z_NULL;
-
- if (s->mode == 'w') {
- err = deflateInit2(&(s->stream), level,
- Z_DEFLATED, MAX_WBITS+16, DEF_MEM_LEVEL, 0);
- /* windowBits is passed < 0 to suppress zlib header */
-
- s->stream.next_out = s->outbuf = (Byte*)ALLOC(Z_BUFSIZE);
-
- if (err != Z_OK || s->outbuf == Z_NULL) {
- return s->destroy(s), (ErtsGzFile)Z_NULL;
- }
- } else {
- /*
- * It is tempting to use the built-in support in zlib
- * for handling GZIP headers, but unfortunately it
- * cannot handle multiple GZIP headers (which occur when
- * several GZIP files have been concatenated).
- */
-
- err = inflateInit2(&(s->stream), -MAX_WBITS);
- s->stream.next_in = s->inbuf = (Byte*)ALLOC(Z_BUFSIZE);
-
- if (err != Z_OK || s->inbuf == Z_NULL) {
- return s->destroy(s), (ErtsGzFile)Z_NULL;
- }
- }
- s->stream.avail_out = Z_BUFSIZE;
-
- errno = 0;
-#if defined(FILENAMES_16BIT)
- {
- FILE* efile_wfopen(const WCHAR* name, const WCHAR* mode);
- WCHAR wfmode[80];
- int i = 0;
- int j;
- for(j = 0; fmode[j] != '\0'; ++j) {
- wfmode[i++] = (WCHAR) fmode[j];
- }
- wfmode[i++] = L'\0';
- s->file = efile_wfopen((WCHAR *)path, wfmode);
- if (s->file == NULL) {
- return s->destroy(s), (ErtsGzFile)Z_NULL;
- }
- }
-#elif defined(UNIX)
- if (s->mode == 'r') {
- s->file = open(path, O_RDONLY);
- } else {
- s->file = open(path, O_WRONLY | O_CREAT | O_TRUNC, 0666);
- }
- if (s->file == -1) {
- return s->destroy(s), (ErtsGzFile)Z_NULL;
- }
-#else
- s->file = fopen(path, fmode);
- if (s->file == NULL) {
- return s->destroy(s), (ErtsGzFile)Z_NULL;
- }
-#endif
- if (s->mode == 'r') {
- check_header(s); /* skip the .gz header */
- }
- return (ErtsGzFile)s;
-}
-
-/* ===========================================================================
- Rewind a gzfile back to the beginning.
-*/
-
-local int gz_rewind (gz_stream *s)
-{
- TRYFREE(s->msg);
-
-#ifdef UNIX
- lseek(s->file, 0L, SEEK_SET);
-#else
- fseek(s->file, 0L, SEEK_SET);
-#endif
- inflateReset(&(s->stream));
- s->stream.next_in = Z_NULL;
- s->stream.next_out = Z_NULL;
- s->stream.avail_in = s->stream.avail_out = 0;
- s->z_err = Z_OK;
- s->z_eof = 0;
- s->crc = crc32(0L, Z_NULL, 0);
- s->msg = NULL;
- s->position = 0;
- s->stream.next_in = s->inbuf;
-
- s->stream.avail_out = Z_BUFSIZE;
-
- check_header(s); /* skip the .gz header */
- return 1;
-}
-
-/* ===========================================================================
- Opens a gzip (.gz) file for reading or writing.
-*/
-ErtsGzFile erts_gzopen (path, mode)
- const char *path;
- const char *mode;
-{
- return gz_open (path, mode);
-}
-
-
-/* ===========================================================================
- Read a byte from a gz_stream; update next_in and avail_in. Return EOF
- for end of file.
- IN assertion: the stream s has been successfully opened for reading.
-*/
-local int get_byte(s)
- gz_stream *s;
-{
- if (s->z_eof) return EOF;
- if (s->stream.avail_in == 0) {
-#ifdef UNIX
- ssize_t res;
- errno = 0;
- res = ERTS_GZREAD(s->file, s->inbuf, Z_BUFSIZE);
- if (res == 0) {
- s->stream.avail_in = 0;
- s->z_eof = 1;
- return EOF;
- } else if (res < 0) {
- s->stream.avail_in = 0;
- s->z_eof = 1;
- s->z_err = Z_ERRNO;
- return EOF;
- } else {
- s->stream.avail_in = (uInt) res;
- }
-#else
- errno = 0;
- s->stream.avail_in = ERTS_GZREAD(s->file, s->inbuf, Z_BUFSIZE);
- if (s->stream.avail_in == 0) {
- s->z_eof = 1;
- if (s->file && ferror(s->file))
- s->z_err = Z_ERRNO;
- return EOF;
- }
-#endif
- s->stream.next_in = s->inbuf;
- }
- s->stream.avail_in--;
- return *(s->stream.next_in)++;
-}
-
-/* ===========================================================================
- Check the gzip header of a gz_stream opened for reading. Set the stream
- mode to transparent if the gzip magic header is not present; set s->err
- to Z_DATA_ERROR if the magic header is present but the rest of the header
- is incorrect.
- IN assertion: the stream s has already been created sucessfully;
- s->stream.avail_in is zero for the first time, but may be non-zero
- for concatenated .gz files.
-*/
-local void check_header(s)
- gz_stream *s;
-{
- int method; /* method byte */
- int flags; /* flags byte */
- uInt len;
- int c;
-
- /* Check the gzip magic header */
- for (len = 0; len < 2; len++) {
- c = get_byte(s);
- if (c != gz_magic[len]) {
- if (len != 0) s->stream.avail_in++, s->stream.next_in--;
- if (c != EOF) {
- s->stream.avail_in++, s->stream.next_in--;
- s->transparent = 1;
- }
- s->z_err = s->stream.avail_in != 0 ? Z_OK : Z_STREAM_END;
- return;
- }
- }
- method = get_byte(s);
- flags = get_byte(s);
- if (method != Z_DEFLATED || (flags & RESERVED) != 0) {
- s->z_err = Z_DATA_ERROR;
- return;
- }
-
- /* Discard time, xflags and OS code: */
- for (len = 0; len < 6; len++) (void)get_byte(s);
-
- if ((flags & EXTRA_FIELD) != 0) { /* skip the extra field */
- len = (uInt)get_byte(s);
- len += ((uInt)get_byte(s))<<8;
- /* len is garbage if EOF but the loop below will quit anyway */
- while (len-- != 0 && get_byte(s) != EOF) ;
- }
- if ((flags & ORIG_NAME) != 0) { /* skip the original file name */
- while ((c = get_byte(s)) != 0 && c != EOF) ;
- }
- if ((flags & COMMENT) != 0) { /* skip the .gz file comment */
- while ((c = get_byte(s)) != 0 && c != EOF) ;
- }
- if ((flags & HEAD_CRC) != 0) { /* skip the header crc */
- for (len = 0; len < 2; len++) (void)get_byte(s);
- }
- s->z_err = s->z_eof ? Z_DATA_ERROR : Z_OK;
-}
-
- /* ===========================================================================
- * Cleanup then free the given gz_stream. Return a zlib error code.
- Try freeing in the reverse order of allocations.
- */
-local int destroy (s)
- gz_stream *s;
-{
- int err = Z_OK;
-
- if (!s) return Z_STREAM_ERROR;
-
- TRYFREE(s->msg);
-
- if (s->stream.state != NULL) {
- if (s->mode == 'w') {
- err = deflateEnd(&(s->stream));
- } else if (s->mode == 'r') {
- err = inflateEnd(&(s->stream));
- }
- }
-#ifdef UNIX
- if (s->file != -1 && close(s->file)) {
- err = Z_ERRNO;
- }
-#else
- if (s->file != NULL && fclose(s->file)) {
- err = Z_ERRNO;
- }
-#endif
- if (s->z_err < 0) err = s->z_err;
-
- TRYFREE(s->inbuf);
- TRYFREE(s->outbuf);
- TRYFREE(s->path);
- TRYFREE(s);
- return err;
-}
-
-/* ===========================================================================
- Reads the given number of uncompressed bytes from the compressed file.
- gzread returns the number of bytes actually read (0 for end of file).
-*/
-int
-erts_gzread(ErtsGzFile file, voidp buf, unsigned len)
-{
- gz_stream *s = (gz_stream*)file;
- Bytef *start = buf; /* starting point for crc computation */
- Byte *next_out; /* == stream.next_out but not forced far (for MSDOS) */
-
- if (s == NULL || s->mode != 'r') return Z_STREAM_ERROR;
-
- if (s->z_err == Z_DATA_ERROR || s->z_err == Z_ERRNO) return -1;
- if (s->z_err == Z_STREAM_END) return 0; /* EOF */
-
- s->stream.next_out = next_out = buf;
- s->stream.avail_out = len;
-
- while (s->stream.avail_out != 0) {
-
- if (s->transparent) {
- /* Copy first the lookahead bytes: */
- uInt n = s->stream.avail_in;
- if (n > s->stream.avail_out) n = s->stream.avail_out;
- if (n > 0) {
- zmemcpy(s->stream.next_out, s->stream.next_in, n);
- next_out += n;
- s->stream.next_out = next_out;
- s->stream.next_in += n;
- s->stream.avail_out -= n;
- s->stream.avail_in -= n;
- }
- if (s->stream.avail_out > 0) {
- s->stream.avail_out -= ERTS_GZREAD(s->file, next_out,
- s->stream.avail_out);
- }
- len -= s->stream.avail_out;
- s->stream.total_in += (uLong)len;
- s->stream.total_out += (uLong)len;
- if (len == 0) s->z_eof = 1;
- s->position += (int)len;
- return (int)len;
- }
- if (s->stream.avail_in == 0 && !s->z_eof) {
-#ifdef UNIX
- ssize_t res;
- errno = 0;
- res = ERTS_GZREAD(s->file, s->inbuf, Z_BUFSIZE);
- if (res == 0) {
- s->stream.avail_in = 0;
- s->z_eof = 1;
- return EOF;
- } else if (res < 0) {
- s->stream.avail_in = 0;
- s->z_eof = 1;
- s->z_err = Z_ERRNO;
- return EOF;
- } else {
- s->stream.avail_in = (uInt) res;
- }
-#else
- errno = 0;
- s->stream.avail_in = ERTS_GZREAD(s->file, s->inbuf, Z_BUFSIZE);
- if (s->stream.avail_in == 0) {
- s->z_eof = 1;
- if (s->file && ferror(s->file)) {
- s->z_err = Z_ERRNO;
- break;
- }
- }
-#endif
- s->stream.next_in = s->inbuf;
- }
- s->z_err = inflate(&(s->stream), Z_NO_FLUSH);
-
- if (s->z_err == Z_STREAM_END) {
- /* Check CRC and original size */
- s->crc = crc32(s->crc, start, (uInt)(s->stream.next_out - start));
- start = s->stream.next_out;
-
- if (getLong(s) != s->crc) {
- s->z_err = Z_DATA_ERROR;
- } else {
- (void)getLong(s);
- /* The uncompressed length returned by above getlong() may
- * be different from s->stream.total_out) in case of
- * concatenated .gz files. Check for such files:
- */
- check_header(s);
- if (s->z_err == Z_OK) {
- uLong total_in = s->stream.total_in;
- uLong total_out = s->stream.total_out;
-
- inflateReset(&(s->stream));
- s->stream.total_in = total_in;
- s->stream.total_out = total_out;
- s->crc = crc32(0L, Z_NULL, 0);
- }
- }
- }
- if (s->z_err != Z_OK || s->z_eof) break;
- }
- s->crc = crc32(s->crc, start, (uInt)(s->stream.next_out - start));
-
- s->position += (int)(len - s->stream.avail_out);
-
- return (int)(len - s->stream.avail_out);
-}
-
-/* ===========================================================================
- Writes the given number of uncompressed bytes into the compressed file.
- gzwrite returns the number of bytes actually written (0 in case of error).
-*/
-int
-erts_gzwrite(ErtsGzFile file, voidp buf, unsigned len)
-{
- gz_stream *s = (gz_stream*)file;
-
- if (s == NULL || s->mode != 'w') return Z_STREAM_ERROR;
-
- s->stream.next_in = buf;
- s->stream.avail_in = len;
-
- while (s->stream.avail_in != 0) {
-
- if (s->stream.avail_out == 0) {
-
- s->stream.next_out = s->outbuf;
- if (ERTS_GZWRITE(s->file, s->outbuf, Z_BUFSIZE) != Z_BUFSIZE) {
- s->z_err = Z_ERRNO;
- break;
- }
- s->stream.avail_out = Z_BUFSIZE;
- }
- s->z_err = deflate(&(s->stream), Z_NO_FLUSH);
- if (s->z_err != Z_OK) break;
- }
- s->position += (int)(len - s->stream.avail_in);
- return (int)(len - s->stream.avail_in);
-}
-
-/*
- * For use by Erlang file driver.
- *
- * XXX Limitations:
- * - SEEK_END is not allowed (length of file is not known).
- * - When writing, only forward seek is supported.
- */
-
-int
-erts_gzseek(ErtsGzFile file, int offset, int whence)
-{
- int pos;
- gz_stream* s = (gz_stream *) file;
-
- switch (whence) {
- case EFILE_SEEK_SET: whence = SEEK_SET; break;
- case EFILE_SEEK_CUR: whence = SEEK_CUR; break;
- case EFILE_SEEK_END: whence = SEEK_END; break;
- default:
- errno = EINVAL;
- return -1;
- }
-
- if (s == NULL) {
- errno = EINVAL;
- return -1;
- }
- if (s->z_err == Z_DATA_ERROR || s->z_err == Z_ERRNO) {
- errno = EIO;
- return -1;
- }
-
- switch (whence) {
- case SEEK_SET: pos = offset; break;
- case SEEK_CUR: pos = s->position+offset; break;
- case SEEK_END:
- default:
- errno = EINVAL; return -1;
- }
-
- if (pos == s->position) {
- return pos;
- }
-
- if (pos < s->position) {
- if (s->mode == 'w') {
- errno = EINVAL;
- return -1;
- }
- gz_rewind(s);
- }
-
- while (s->position < pos) {
- char buf[512];
- int n;
- int save_pos = s->position;
-
- n = pos - s->position;
- if (n > sizeof(buf))
- n = sizeof(buf);
-
- if (s->mode == 'r') {
- erts_gzread(file, buf, n);
- } else {
- memset(buf, '\0', n);
- erts_gzwrite(file, buf, n);
- }
- if (save_pos == s->position) break;
- }
-
- return s->position;
-}
-
-/* ===========================================================================
- Flushes all pending output into the compressed file. The parameter
- flush is as in the deflate() function.
- gzflush should be called only when strictly necessary because it can
- degrade compression.
-*/
-int
-erts_gzflush(ErtsGzFile file, int flush)
-{
- uInt len;
- int done = 0;
- gz_stream *s = (gz_stream*)file;
-
- if (s == NULL || s->mode != 'w') return Z_STREAM_ERROR;
-
- s->stream.avail_in = 0; /* should be zero already anyway */
-
- for (;;) {
- len = Z_BUFSIZE - s->stream.avail_out;
-
- if (len != 0) {
- if ((uInt)ERTS_GZWRITE(s->file, s->outbuf, len) != len) {
- s->z_err = Z_ERRNO;
- return Z_ERRNO;
- }
- s->stream.next_out = s->outbuf;
- s->stream.avail_out = Z_BUFSIZE;
- }
- if (done) break;
- s->z_err = deflate(&(s->stream), flush);
-
- /* deflate has finished flushing only when it hasn't used up
- * all the available space in the output buffer:
- */
- done = (s->stream.avail_out != 0 || s->z_err == Z_STREAM_END);
-
- if (s->z_err != Z_OK && s->z_err != Z_STREAM_END) break;
- }
-#ifndef UNIX
- fflush(s->file);
-#endif
- return s->z_err == Z_STREAM_END ? Z_OK : s->z_err;
-}
-
-/* ===========================================================================
- Reads a long in LSB order from the given gz_stream. Sets
-*/
-local uLong getLong (s)
- gz_stream *s;
-{
- uLong x = (uLong)get_byte(s);
- int c;
-
- x += ((uLong)get_byte(s))<<8;
- x += ((uLong)get_byte(s))<<16;
- c = get_byte(s);
- if (c == EOF) s->z_err = Z_DATA_ERROR;
- x += ((uLong)c)<<24;
- return x;
-}
-
-/* ===========================================================================
- Flushes all pending output if necessary, closes the compressed file
- and deallocates all the (de)compression state.
-*/
-int
-erts_gzclose(ErtsGzFile file)
-{
- int err;
- gz_stream *s = (gz_stream*)file;
-
- if (s == NULL) return Z_STREAM_ERROR;
-
- if (s->mode == 'w') {
- err = erts_gzflush (file, Z_FINISH);
- if (err != Z_OK) return s->destroy(s);
- }
- return s->destroy(s);
-}
-
-
/* ===========================================================================
Uncompresses the buffer given and returns a pointer to a binary.
If the buffer was not compressed with gzip, the buffer contents
diff --git a/erts/emulator/drivers/common/gzio.h b/erts/emulator/drivers/common/gzio.h
index ee0ebe7bd8..e331b5208b 100644
--- a/erts/emulator/drivers/common/gzio.h
+++ b/erts/emulator/drivers/common/gzio.h
@@ -20,13 +20,5 @@
#include "zlib.h"
-typedef struct erts_gzFile* ErtsGzFile;
-
-ErtsGzFile erts_gzopen (const char *path, const char *mode);
-int erts_gzread(ErtsGzFile file, voidp buf, unsigned len);
-int erts_gzwrite(ErtsGzFile file, voidp buf, unsigned len);
-int erts_gzseek(ErtsGzFile, int, int);
-int erts_gzflush(ErtsGzFile file, int flush);
-int erts_gzclose(ErtsGzFile file);
ErlDrvBinary* erts_gzinflate_buffer(char*, uLong);
ErlDrvBinary* erts_gzdeflate_buffer(char*, uLong);
diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c
index 95d61fcc5d..4294fb4f46 100644
--- a/erts/emulator/drivers/common/inet_drv.c
+++ b/erts/emulator/drivers/common/inet_drv.c
@@ -63,6 +63,20 @@
#include <sys/un.h>
#endif
+#ifdef HAVE_SENDFILE
+#if defined(__linux__) || (defined(__sun) && defined(__SVR4))
+ #include <sys/sendfile.h>
+#elif defined(__FreeBSD__) || defined(__DragonFly__)
+ /* Need to define __BSD_VISIBLE in order to expose prototype of sendfile */
+ #define __BSD_VISIBLE 1
+ #include <sys/socket.h>
+#endif
+#endif
+
+#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__)
+ #define __DARWIN__ 1
+#endif
+
/* All platforms fail on malloc errors. */
#define FATAL_MALLOC
@@ -701,6 +715,7 @@ static size_t my_strnlen(const char *s, size_t maxlen)
#define TCP_REQ_RECV 42
#define TCP_REQ_UNRECV 43
#define TCP_REQ_SHUTDOWN 44
+#define TCP_REQ_SENDFILE 45
/* UDP and SCTP requests */
#define PACKET_REQ_RECV 60 /* Common for UDP and SCTP */
/* #define SCTP_REQ_LISTEN 61 MERGED Different from TCP; not for UDP */
@@ -723,6 +738,7 @@ static size_t my_strnlen(const char *s, size_t maxlen)
#define TCP_ADDF_DELAYED_ECONNRESET 128 /* An ECONNRESET error occurred on send or shutdown */
#define TCP_ADDF_SHUTDOWN_WR_DONE 256 /* A shutdown(sock, SHUT_WR) or SHUT_RDWR was made */
#define TCP_ADDF_LINGER_ZERO 512 /* Discard driver queue on port close */
+#define TCP_ADDF_SENDFILE 1024 /* Send from an fd instead of the driver queue */
/* *_REQ_* replies */
#define INET_REP_ERROR 0
@@ -1235,6 +1251,21 @@ typedef struct {
inet_async_multi_op *multi_first;/* NULL == no multi-accept-queue, op is in ordinary queue */
inet_async_multi_op *multi_last;
MultiTimerData *mtd; /* Timer structures for multiple accept */
+#ifdef HAVE_SENDFILE
+ struct {
+ ErlDrvSizeT ioq_skip; /* The number of bytes in the queue at the time
+ * sendfile was issued, which must be sent
+ * before issuing the sendfile call itself. */
+ int dup_file_fd; /* The file handle to send from; this is
+ * duplicated when sendfile is issued to
+ * reduce (but not eliminate) the impact of a
+ * nasty race, so we have to remember to close
+ * it. */
+ Uint64 bytes_sent;
+ Uint64 offset;
+ Uint64 length;
+ } sendfile;
+#endif
} tcp_descriptor;
/* send function */
@@ -1245,6 +1276,8 @@ static int tcp_deliver(tcp_descriptor* desc, int len);
static int tcp_shutdown_error(tcp_descriptor* desc, int err);
+static int tcp_inet_sendfile(tcp_descriptor* desc);
+
static int tcp_inet_output(tcp_descriptor* desc, HANDLE event);
static int tcp_inet_input(tcp_descriptor* desc, HANDLE event);
@@ -1329,6 +1362,9 @@ static ErlDrvTermData am_ipv6_v6only;
static ErlDrvTermData am_netns;
static ErlDrvTermData am_bind_to_device;
#endif
+#ifdef HAVE_SENDFILE
+static ErlDrvTermData am_sendfile;
+#endif
static char str_eafnosupport[] = "eafnosupport";
static char str_einval[] = "einval";
@@ -3875,6 +3911,10 @@ static int inet_init()
INIT_ATOM(https);
INIT_ATOM(scheme);
+#ifdef HAVE_SENDFILE
+ INIT_ATOM(sendfile);
+#endif
+
/* add TCP, UDP and SCTP drivers */
add_driver_entry(&tcp_inet_driver_entry);
#ifdef HAVE_UDP
@@ -9270,6 +9310,13 @@ static void tcp_inet_stop(ErlDrvData e)
* will be freed through tcp_inet_stop later on. */
static void tcp_desc_close(tcp_descriptor* desc)
{
+#ifdef HAVE_SENDFILE
+ if(desc->tcp_add_flags & TCP_ADDF_SENDFILE) {
+ desc->tcp_add_flags &= ~TCP_ADDF_SENDFILE;
+ close(desc->sendfile.dup_file_fd);
+ }
+#endif
+
tcp_clear_input(desc);
tcp_clear_output(desc);
@@ -9608,6 +9655,60 @@ static ErlDrvSSizeT tcp_inet_ctl(ErlDrvData e, unsigned int cmd,
return ctl_reply(INET_REP_OK, NULL, 0, rbuf, rsize);
}
}
+
+ case TCP_REQ_SENDFILE: {
+#ifdef HAVE_SENDFILE
+ const ErlDrvSizeT required_len =
+ sizeof(desc->sendfile.dup_file_fd) +
+ sizeof(Uint64) * 2;
+
+ int raw_file_fd;
+
+ DEBUGF(("tcp_inet_ctl(%ld): SENDFILE\r\n", (long)desc->inet.port));
+
+ if (len != required_len) {
+ return ctl_error(EINVAL, rbuf, rsize);
+ } else if (!IS_CONNECTED(INETP(desc))) {
+ return ctl_error(ENOTCONN, rbuf, rsize);
+ }
+
+ sys_memcpy(&raw_file_fd, buf, sizeof(raw_file_fd));
+ buf += sizeof(raw_file_fd);
+
+ desc->sendfile.dup_file_fd = dup(raw_file_fd);
+
+ if(desc->sendfile.dup_file_fd == -1) {
+ return ctl_error(errno, rbuf, rsize);
+ }
+
+ desc->sendfile.offset = get_int64(buf);
+ buf += sizeof(Uint64);
+
+ desc->sendfile.length = get_int64(buf);
+ buf += sizeof(Uint64);
+
+ ASSERT(desc->sendfile.offset >= 0);
+ ASSERT(desc->sendfile.length >= 0);
+
+ desc->sendfile.ioq_skip = driver_sizeq(desc->inet.port);
+ desc->sendfile.bytes_sent = 0;
+
+ desc->inet.caller = driver_caller(desc->inet.port);
+ desc->tcp_add_flags |= TCP_ADDF_SENDFILE;
+
+ /* See if we can finish sending without selecting & rescheduling. */
+ tcp_inet_sendfile(desc);
+
+ if(desc->sendfile.length > 0) {
+ sock_select(INETP(desc), FD_WRITE, 1);
+ }
+
+ return ctl_reply(INET_REP_OK, NULL, 0, rbuf, rsize);
+#else
+ return ctl_error(ENOTSUP, rbuf, rsize);
+#endif
+ }
+
default:
DEBUGF(("tcp_inet_ctl(%ld): %u\r\n", (long)desc->inet.port, cmd));
return inet_ctl(INETP(desc), cmd, buf, len, rbuf, rsize);
@@ -9747,12 +9848,27 @@ static void tcp_inet_commandv(ErlDrvData e, ErlIOVec* ev)
static void tcp_inet_flush(ErlDrvData e)
{
tcp_descriptor* desc = (tcp_descriptor*)e;
- if (!(desc->inet.event_mask & FD_WRITE)) {
- /* Discard send queue to avoid hanging port (OTP-7615) */
- tcp_clear_output(desc);
+ int discard_output;
+
+ /* Discard send queue to avoid hanging port (OTP-7615) */
+ discard_output = !(desc->inet.event_mask & FD_WRITE);
+
+ discard_output |= desc->tcp_add_flags & TCP_ADDF_LINGER_ZERO;
+
+#ifdef HAVE_SENDFILE
+ /* The old file driver aborted when it was stopped during sendfile, so
+ * we'll clear the flag and discard all output. */
+ if(desc->tcp_add_flags & TCP_ADDF_SENDFILE) {
+ desc->tcp_add_flags &= ~TCP_ADDF_SENDFILE;
+ close(desc->sendfile.dup_file_fd);
+
+ discard_output = 1;
+ }
+#endif
+
+ if (discard_output) {
+ tcp_clear_output(desc);
}
- if (desc->tcp_add_flags & TCP_ADDF_LINGER_ZERO)
- tcp_clear_output(desc);
}
static void tcp_inet_process_exit(ErlDrvData e, ErlDrvMonitor *monitorp)
@@ -10647,7 +10763,9 @@ static int tcp_sendv(tcp_descriptor* desc, ErlIOVec* ev)
ev->size += h_len;
}
- if ((sz = driver_sizeq(ix)) > 0) {
+ sz = driver_sizeq(ix);
+
+ if ((desc->tcp_add_flags & TCP_ADDF_SENDFILE) || sz > 0) {
driver_enqv(ix, ev, 0);
if (sz+ev->size >= desc->high) {
DEBUGF(("tcp_sendv(%ld): s=%d, sender forced busy\r\n",
@@ -10741,8 +10859,9 @@ static int tcp_send(tcp_descriptor* desc, char* ptr, ErlDrvSizeT len)
inet_output_count(INETP(desc), len+h_len);
+ sz = driver_sizeq(ix);
- if ((sz = driver_sizeq(ix)) > 0) {
+ if ((desc->tcp_add_flags & TCP_ADDF_SENDFILE) || sz > 0) {
if (h_len > 0)
driver_enq(ix, buf, h_len);
driver_enq(ix, ptr, len);
@@ -10832,6 +10951,246 @@ static void tcp_inet_drv_input(ErlDrvData data, ErlDrvEvent event)
(void)tcp_inet_input((tcp_descriptor*)data, (HANDLE)event);
}
+#ifdef HAVE_SENDFILE
+static int tcp_sendfile_completed(tcp_descriptor* desc) {
+ ErlDrvTermData spec[LOAD_PORT_CNT + LOAD_TUPLE_CNT * 2 +
+ LOAD_ATOM_CNT * 2 + LOAD_UINT_CNT * 2];
+ Uint32 sent_low, sent_high;
+ int i;
+
+ desc->tcp_add_flags &= ~TCP_ADDF_SENDFILE;
+ close(desc->sendfile.dup_file_fd);
+
+ /* While we flushed the output queue prior to sending the file, we've
+ * deferred clearing busy status until now as there's no point in doing so
+ * while we still have a file to send.
+ *
+ * The watermark is checked since more data may have been added while we
+ * were sending the file. */
+
+ if (driver_sizeq(desc->inet.port) <= desc->low) {
+ if (IS_BUSY(INETP(desc))) {
+ desc->inet.caller = desc->inet.busy_caller;
+ desc->inet.state &= ~INET_F_BUSY;
+
+ set_busy_port(desc->inet.port, 0);
+
+ /* if we have a timer then cancel and send ok to client */
+ if (desc->busy_on_send) {
+ driver_cancel_timer(desc->inet.port);
+ desc->busy_on_send = 0;
+ }
+
+ inet_reply_ok(INETP(desc));
+ }
+ }
+
+ if (driver_sizeq(desc->inet.port) == 0) {
+ sock_select(INETP(desc), FD_WRITE, 0);
+ send_empty_out_q_msgs(INETP(desc));
+
+ if (desc->tcp_add_flags & TCP_ADDF_PENDING_SHUTDOWN) {
+ tcp_shutdown_async(desc);
+ }
+ }
+
+ sent_low = ((Uint64)desc->sendfile.bytes_sent >> 0) & 0xFFFFFFFF;
+ sent_high = ((Uint64)desc->sendfile.bytes_sent >> 32) & 0xFFFFFFFF;
+
+ i = LOAD_ATOM(spec, 0, am_sendfile);
+ i = LOAD_PORT(spec, i, desc->inet.dport);
+ i = LOAD_ATOM(spec, i, am_ok);
+ i = LOAD_UINT(spec, i, sent_low);
+ i = LOAD_UINT(spec, i, sent_high);
+ i = LOAD_TUPLE(spec, i, 3);
+ i = LOAD_TUPLE(spec, i, 3);
+
+ ASSERT(i == sizeof(spec)/sizeof(*spec));
+
+ return erl_drv_output_term(desc->inet.dport, spec, i);
+}
+
+static int tcp_sendfile_aborted(tcp_descriptor* desc, int socket_error) {
+ ErlDrvTermData spec[LOAD_PORT_CNT + LOAD_TUPLE_CNT * 2 + LOAD_ATOM_CNT * 3];
+ int i;
+
+ /* We don't clean up sendfile state here, as that's done in tcp_desc_close
+ * following normal error handling. All we do here is report the failure. */
+
+ i = LOAD_ATOM(spec, 0, am_sendfile);
+ i = LOAD_PORT(spec, i, desc->inet.dport);
+ i = LOAD_ATOM(spec, i, am_error);
+
+ switch (socket_error) {
+ case ECONNRESET:
+ case ENOTCONN:
+ case EPIPE:
+ i = LOAD_ATOM(spec, i, am_closed);
+ break;
+ default:
+ i = LOAD_ATOM(spec, i, error_atom(socket_error));
+ }
+
+ i = LOAD_TUPLE(spec, i, 2);
+ i = LOAD_TUPLE(spec, i, 3);
+
+ ASSERT(i == sizeof(spec)/sizeof(*spec));
+
+ return erl_drv_output_term(desc->inet.dport, spec, i);
+}
+
+static int tcp_inet_sendfile(tcp_descriptor* desc) {
+ ErlDrvPort ix = desc->inet.port;
+ int result = 0;
+ ssize_t n;
+
+ DEBUGF(("tcp_inet_sendfile(%ld) {s=%d\r\n", (long)ix, desc->inet.s));
+
+ /* If there was any data in the queue by the time sendfile was issued,
+ * we'll need to skip it first. Note that we don't clear busy status until
+ * we're finished sending the file. */
+ while (desc->sendfile.ioq_skip > 0) {
+ ssize_t bytes_to_send;
+ SysIOVec* iov;
+ int vsize;
+
+ ASSERT(driver_sizeq(ix) >= desc->sendfile.ioq_skip);
+
+ if ((iov = driver_peekq(ix, &vsize)) == NULL) {
+ ERTS_INTERNAL_ERROR("ioq empty when sendfile.ioq_skip > 0");
+ }
+
+ bytes_to_send = MIN(desc->sendfile.ioq_skip, iov[0].iov_len);
+ n = sock_send(desc->inet.s, iov[0].iov_base, bytes_to_send, 0);
+
+ if (!IS_SOCKET_ERROR(n)) {
+ desc->sendfile.ioq_skip -= n;
+ driver_deq(ix, n);
+ } else if (sock_errno() == ERRNO_BLOCK) {
+#ifdef __WIN32__
+ desc->inet.send_would_block = 1;
+#endif
+ goto done;
+ } else if (sock_errno() != EINTR) {
+ goto socket_error;
+ }
+ }
+
+ while (desc->sendfile.length > 0) {
+ /* For some reason the maximum ssize_t cannot be used as the max size.
+ * 1GB seems to work on all platforms */
+ const Sint64 SENDFILE_CHUNK_SIZE = ((1UL << 30) - 1);
+
+ ssize_t bytes_to_send = MIN(SENDFILE_CHUNK_SIZE, desc->sendfile.length);
+ off_t offset = desc->sendfile.offset;
+
+#if defined(__linux__)
+ n = sendfile(desc->inet.s, desc->sendfile.dup_file_fd, &offset,
+ bytes_to_send);
+#elif defined(__FreeBSD__) || defined(__DragonFly__) || defined(__DARWIN__)
+ {
+ off_t bytes_sent;
+ int error;
+
+ #if defined(__DARWIN__)
+ bytes_sent = bytes_to_send;
+
+ error = sendfile(desc->sendfile.dup_file_fd, desc->inet.s, offset,
+ &bytes_sent, NULL, 0);
+ n = bytes_sent;
+ #else
+ error = sendfile(desc->sendfile.dup_file_fd, desc->inet.s, offset,
+ bytes_to_send, NULL, &bytes_sent, 0);
+ n = bytes_sent;
+ #endif
+
+ if(error < 0) {
+ /* EAGAIN/EINTR report partial success by setting bytes_sent,
+ * so we have to skip error handling if nonzero, and skip EOF
+ * handling if zero, as it's possible that we didn't manage to
+ * send anything at all before being interrupted by a
+ * signal. */
+ if((errno != EAGAIN && errno != EINTR) || bytes_sent == 0) {
+ n = -1;
+ }
+ }
+ }
+#elif defined(__sun) && defined(__SVR4) && defined(HAVE_SENDFILEV)
+ {
+ sendfilevec_t sfvec[1];
+ size_t bytes_sent;
+ ssize_t error;
+
+ sfvec[0].sfv_fd = desc->sendfile.dup_file_fd;
+ sfvec[0].sfv_len = bytes_to_send;
+ sfvec[0].sfv_off = offset;
+ sfvec[0].sfv_flag = 0;
+
+ error = sendfilev(desc->inet.s, sfvec, 1, &bytes_sent);
+ n = bytes_sent;
+
+ if(error < 0) {
+ if(errno == EINVAL) {
+ /* On some solaris versions (I've seen it on SunOS 5.10),
+ * using a sfv_len larger than the filesize will result in
+ * a (-1 && errno == EINVAL). We translate this to a
+ * successful send of the data.*/
+ } else {
+ /* EAGAIN/EINTR behavior is identical to *BSD. */
+ if((errno != EAGAIN && errno != EINTR) || bytes_sent == 0) {
+ n = -1;
+ }
+ }
+ }
+ }
+#else
+ #error "Unsupported sendfile syscall; update configure test."
+#endif
+
+ if (n > 0) {
+ desc->sendfile.bytes_sent += n;
+ desc->sendfile.offset += n;
+ desc->sendfile.length -= n;
+ } else if (n == 0) {
+ /* EOF. */
+ desc->sendfile.length = 0;
+ break;
+ } else if (IS_SOCKET_ERROR(n) && sock_errno() != EINTR) {
+ if (sock_errno() != ERRNO_BLOCK) {
+ goto socket_error;
+ }
+
+#ifdef __WIN32__
+ desc->inet.send_would_block = 1;
+#endif
+ break;
+ }
+ }
+
+ if (desc->sendfile.length == 0) {
+ tcp_sendfile_completed(desc);
+ }
+
+ goto done;
+
+socket_error: {
+ int socket_errno = sock_errno();
+
+ DEBUGF(("tcp_inet_sendfile(%ld): send errno = %d (errno %d)\r\n",
+ (long)desc->inet.port, socket_errno, errno));
+
+ result = tcp_send_error(desc, socket_errno);
+ tcp_sendfile_aborted(desc, socket_errno);
+
+ goto done;
+ }
+
+done:
+ DEBUGF(("tcp_inet_sendfile(%ld) }\r\n", (long)desc->inet.port));
+ return result;
+}
+#endif /* HAVE_SENDFILE */
+
/* socket ready for ouput:
** 1. INET_STATE_CONNECTING => non block connect ?
** 2. INET_STATE_CONNECTED => write output
@@ -10892,7 +11251,14 @@ static int tcp_inet_output(tcp_descriptor* desc, HANDLE event)
async_ok(INETP(desc));
}
else if (IS_CONNECTED(INETP(desc))) {
- for (;;) {
+
+#ifdef HAVE_SENDFILE
+ if(desc->tcp_add_flags & TCP_ADDF_SENDFILE) {
+ return tcp_inet_sendfile(desc);
+ }
+#endif
+
+ for (;;) {
int vsize;
ssize_t n;
SysIOVec* iov;
diff --git a/erts/emulator/drivers/unix/unix_efile.c b/erts/emulator/drivers/unix/unix_efile.c
deleted file mode 100644
index 33e4d75ef7..0000000000
--- a/erts/emulator/drivers/unix/unix_efile.c
+++ /dev/null
@@ -1,1102 +0,0 @@
-/*
- * %CopyrightBegin%
- *
- * Copyright Ericsson AB 1997-2017. All Rights Reserved.
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- *
- * %CopyrightEnd%
- */
-/*
- * Purpose: Provides file and directory operations for Unix.
- */
-#ifdef HAVE_CONFIG_H
-# include "config.h"
-#endif
-#if defined(HAVE_POSIX_FALLOCATE) && !defined(__sun) && !defined(__sun__)
-#define _XOPEN_SOURCE 600
-#endif
-#if !defined(_GNU_SOURCE) && defined(HAVE_LINUX_FALLOC_H)
-#define _GNU_SOURCE
-#endif
-#include "sys.h"
-#include "erl_driver.h"
-#include "erl_efile.h"
-#include <utime.h>
-#ifdef HAVE_UNISTD_H
-#include <unistd.h>
-#endif
-#ifdef HAVE_SYS_UIO_H
-#include <sys/types.h>
-#include <sys/uio.h>
-#if defined(HAVE_SENDFILE) && (defined(__FreeBSD__) || defined(__DragonFly__))
-/* Need to define __BSD_VISIBLE in order to expose prototype of sendfile */
-#define __BSD_VISIBLE 1
-#include <sys/socket.h>
-#endif
-#endif
-#if defined(HAVE_SENDFILE) && (defined(__linux__) || (defined(__sun) && defined(__SVR4)))
-#include <sys/sendfile.h>
-#endif
-
-#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__)
-#define __DARWIN__ 1
-#endif
-
-#if defined(__DARWIN__) || defined(HAVE_LINUX_FALLOC_H) || defined(HAVE_POSIX_FALLOCATE)
-#include <fcntl.h>
-#endif
-
-#ifdef HAVE_LINUX_FALLOC_H
-#include <linux/falloc.h>
-#endif
-
-#ifdef SUNOS4
-# define getcwd(buf, size) getwd(buf)
-#endif
-
-/* Find a definition of MAXIOV, that is used in the code later. */
-#if defined IOV_MAX
-#define MAXIOV IOV_MAX
-#elif defined UIO_MAXIOV
-#define MAXIOV UIO_MAXIOV
-#else
-#define MAXIOV 16
-#endif
-
-
-/*
- * Macros for testing file types.
- */
-
-#define ISDIR(st) (S_ISDIR((st).st_mode))
-#define ISREG(st) (S_ISREG((st).st_mode))
-#define ISDEV(st) (S_ISCHR((st).st_mode) || S_ISBLK((st).st_mode))
-#define ISLNK(st) (S_ISLNK((st).st_mode))
-#ifdef NO_UMASK
-#define FILE_MODE 0644
-#define DIR_MODE 0755
-#else
-#define FILE_MODE 0666
-#define DIR_MODE 0777
-#endif
-
-#define IS_DOT_OR_DOTDOT(s) \
- (s[0] == '.' && (s[1] == '\0' || (s[1] == '.' && s[2] == '\0')))
-
-static int check_error(int result, Efile_error* errInfo);
-
-static int
-check_error(int result, Efile_error *errInfo)
-{
- if (result < 0) {
- errInfo->posix_errno = errInfo->os_errno = errno;
- return 0;
- }
- return 1;
-}
-
-int
-efile_init() {
- return 1;
-}
-
-int
-efile_mkdir(Efile_error* errInfo, /* Where to return error codes. */
- char* name) /* Name of directory to create. */
-{
-#ifdef NO_MKDIR_MODE
- return check_error(mkdir(name), errInfo);
-#else
- return check_error(mkdir(name, DIR_MODE), errInfo);
-#endif
-}
-
-int
-efile_rmdir(Efile_error* errInfo, /* Where to return error codes. */
- char* name) /* Name of directory to delete. */
-{
- if (rmdir(name) == 0) {
- return 1;
- }
- if (errno == ENOTEMPTY) {
- errno = EEXIST;
- }
- if (errno == EEXIST) {
- int saved_errno = errno;
- struct stat file_stat;
- struct stat cwd_stat;
-
- /*
- * The error code might be wrong if this is the current directory.
- */
-
- if (stat(name, &file_stat) == 0 && stat(".", &cwd_stat) == 0 &&
- file_stat.st_ino == cwd_stat.st_ino &&
- file_stat.st_dev == cwd_stat.st_dev) {
- saved_errno = EINVAL;
- }
- errno = saved_errno;
- }
- return check_error(-1, errInfo);
-}
-
-int
-efile_delete_file(Efile_error* errInfo, /* Where to return error codes. */
- char* name) /* Name of file to delete. */
-{
- if (unlink(name) == 0) {
- return 1;
- }
- if (errno == EISDIR) { /* Linux sets the wrong error code. */
- errno = EPERM;
- }
- return check_error(-1, errInfo);
-}
-
-/*
- *---------------------------------------------------------------------------
- *
- * Changes the name of an existing file or directory, from src to dst.
- * If src and dst refer to the same file or directory, does nothing
- * and returns success. Otherwise if dst already exists, it will be
- * deleted and replaced by src subject to the following conditions:
- * If src is a directory, dst may be an empty directory.
- * If src is a file, dst may be a file.
- * In any other situation where dst already exists, the rename will
- * fail.
- *
- * Results:
- * If the directory was successfully created, returns 1.
- * Otherwise the return value is 0 and errno is set to
- * indicate the error. Some possible values for errno are:
- *
- * EACCES: src or dst parent directory can't be read and/or written.
- * EEXIST: dst is a non-empty directory.
- * EINVAL: src is a root directory or dst is a subdirectory of src.
- * EISDIR: dst is a directory, but src is not.
- * ENOENT: src doesn't exist, or src or dst is "".
- * ENOTDIR: src is a directory, but dst is not.
- * EXDEV: src and dst are on different filesystems.
- *
- * Side effects:
- * The implementation of rename may allow cross-filesystem renames,
- * but the caller should be prepared to emulate it with copy and
- * delete if errno is EXDEV.
- *
- *---------------------------------------------------------------------------
- */
-
-int
-efile_rename(Efile_error* errInfo, /* Where to return error codes. */
- char* src, /* Original name. */
- char* dst) /* New name. */
-{
- if (rename(src, dst) == 0) {
- return 1;
- }
- if (errno == ENOTEMPTY) {
- errno = EEXIST;
- }
-#if defined (sparc)
- /*
- * SunOS 4.1.4 reports overwriting a non-empty directory with a
- * directory as EINVAL instead of EEXIST (first rule out the correct
- * EINVAL result code for moving a directory into itself). Must be
- * conditionally compiled because realpath() is only defined on SunOS.
- */
-
- if (errno == EINVAL) {
- char srcPath[MAXPATHLEN], dstPath[MAXPATHLEN];
- DIR *dirPtr;
- struct dirent *dirEntPtr;
-
-#ifdef PURIFY
- memset(srcPath, '\0', sizeof(srcPath));
- memset(dstPath, '\0', sizeof(dstPath));
-#endif
-
- if ((realpath(src, srcPath) != NULL)
- && (realpath(dst, dstPath) != NULL)
- && (strncmp(srcPath, dstPath, strlen(srcPath)) != 0)) {
- dirPtr = opendir(dst);
- if (dirPtr != NULL) {
- while ((dirEntPtr = readdir(dirPtr)) != NULL) {
- if ((strcmp(dirEntPtr->d_name, ".") != 0) &&
- (strcmp(dirEntPtr->d_name, "..") != 0)) {
- errno = EEXIST;
- closedir(dirPtr);
- return check_error(-1, errInfo);
- }
- }
- closedir(dirPtr);
- }
- }
- errno = EINVAL;
- }
-#endif /* sparc */
-
- if (strcmp(src, "/") == 0) {
- /*
- * Alpha reports renaming / as EBUSY and Linux reports it as EACCES,
- * instead of EINVAL.
- */
-
- errno = EINVAL;
- }
-
- /*
- * DEC Alpha OSF1 V3.0 returns EACCES when attempting to move a
- * file across filesystems and the parent directory of that file is
- * not writable. Most other systems return EXDEV. Does nothing to
- * correct this behavior.
- */
-
- return check_error(-1, errInfo);
-}
-
-int
-efile_chdir(Efile_error* errInfo, /* Where to return error codes. */
- char* name) /* Name of directory to make current. */
-{
- return check_error(chdir(name), errInfo);
-}
-
-
-int
-efile_getdcwd(Efile_error* errInfo, /* Where to return error codes. */
- int drive, /* 0 - current, 1 - A, 2 - B etc. */
- char* buffer, /* Where to return the current
- directory. */
- size_t size) /* Size of buffer. */
-{
- if (drive == 0) {
- if (getcwd(buffer, size) == NULL)
- return check_error(-1, errInfo);
-
-#ifdef SIMSPARCSOLARIS
- /* We get "host:" prepended to the dirname - remove!. */
- {
- int i = 0;
- int j = 0;
- while ((buffer[i] != ':') && (buffer[i] != '\0')) i++;
- if (buffer[i] == ':') {
- i++;
- while ((buffer[j++] = buffer[i++]) != '\0');
- }
- }
-#endif
- return 1;
- }
-
- /*
- * Drives other than 0 is not supported on Unix.
- */
-
- errno = ENOTSUP;
- return check_error(-1, errInfo);
-}
-
-int
-efile_readdir(Efile_error* errInfo, /* Where to return error codes. */
- char* name, /* Name of directory to open. */
- EFILE_DIR_HANDLE* p_dir_handle, /* Pointer to directory
- handle of
- open directory.*/
- char* buffer, /* Pointer to buffer for
- one filename. */
- size_t *size) /* in-out Size of buffer, length
- of name. */
-{
- DIR *dp; /* Pointer to directory structure. */
- struct dirent* dirp; /* Pointer to directory entry. */
-
- /*
- * If this is the first call, we must open the directory.
- */
-
- if (*p_dir_handle == NULL) {
- dp = opendir(name);
- if (dp == NULL)
- return check_error(-1, errInfo);
- *p_dir_handle = (EFILE_DIR_HANDLE) dp;
- }
-
- /*
- * Retrieve the name of the next file using the directory handle.
- */
-
- dp = *((DIR **)((void *)p_dir_handle));
- for (;;) {
- dirp = readdir(dp);
- if (dirp == NULL) {
- closedir(dp);
- return 0;
- }
- if (IS_DOT_OR_DOTDOT(dirp->d_name))
- continue;
- buffer[0] = '\0';
- strncat(buffer, dirp->d_name, (*size)-1);
- *size = strlen(dirp->d_name);
- return 1;
- }
-}
-
-int
-efile_openfile(Efile_error* errInfo, /* Where to return error codes. */
- char* name, /* Name of directory to open. */
- int flags, /* Flags to user for opening. */
- int* pfd, /* Where to store the file
- descriptor. */
- Sint64 *pSize) /* Where to store the size of the
- file. */
-{
- struct stat statbuf;
- int fd;
- int mode; /* Open mode. */
-
- switch (flags & (EFILE_MODE_READ|EFILE_MODE_WRITE)) {
- case EFILE_MODE_READ:
- mode = O_RDONLY;
- break;
- case EFILE_MODE_WRITE:
- if (flags & EFILE_NO_TRUNCATE)
- mode = O_WRONLY | O_CREAT;
- else
- mode = O_WRONLY | O_CREAT | O_TRUNC;
- break;
- case EFILE_MODE_READ_WRITE:
- mode = O_RDWR | O_CREAT;
- break;
- default:
- errno = EINVAL;
- return check_error(-1, errInfo);
- }
-
- if (flags & EFILE_MODE_APPEND) {
- mode &= ~O_TRUNC;
- mode |= O_APPEND;
- }
- if (flags & EFILE_MODE_EXCL) {
- mode |= O_EXCL;
- }
- if (flags & EFILE_MODE_SYNC) {
-#ifdef O_SYNC
- mode |= O_SYNC;
-#else
- errno = ENOTSUP;
- return check_error(-1, errInfo);
-#endif
- }
-
-#ifdef HAVE_FSTAT
- while (((fd = open(name, mode, FILE_MODE)) < 0) && (errno == EINTR));
- if (!check_error(fd, errInfo)) return 0;
-#endif
-
- if (
-#ifdef HAVE_FSTAT
- fstat(fd, &statbuf) < 0
-#else
- stat(name, &statbuf) < 0
-#endif
- ) {
- /* statbuf is undefined: if the caller depends on it,
- i.e. invoke_read_file(), fail the call immediately */
- if (pSize && flags == EFILE_MODE_READ) {
- check_error(-1, errInfo);
-#ifdef HAVE_FSTAT
- efile_closefile(fd);
-#endif
- return 0;
- }
- }
- else if (! ISREG(statbuf)) {
- struct stat nullstatbuf;
- /*
- * For UNIX only, here is some ugly code to allow
- * /dev/null to be opened as a file.
- */
- if ( (stat("/dev/null", &nullstatbuf) < 0)
- || (statbuf.st_ino != nullstatbuf.st_ino)
- || (statbuf.st_dev != nullstatbuf.st_dev) ) {
-#ifdef HAVE_FSTAT
- efile_closefile(fd);
-#endif
- errno = EISDIR;
- return check_error(-1, errInfo);
- }
- }
-
-#ifndef HAVE_FSTAT
- while (((fd = open(name, mode, FILE_MODE)) < 0) && (errno == EINTR));
- if (!check_error(fd, errInfo)) return 0;
-#endif
-
- *pfd = fd;
- if (pSize) *pSize = statbuf.st_size;
- return 1;
-}
-
-int
-efile_may_openfile(Efile_error* errInfo, char *name) {
- struct stat statbuf; /* Information about the file */
- int result;
-
- result = stat(name, &statbuf);
- if (!check_error(result, errInfo))
- return 0;
- if (!ISREG(statbuf)) {
- errno = EISDIR;
- return check_error(-1, errInfo);
- }
- return 1;
-}
-
-void
-efile_closefile(int fd)
-{
- close(fd);
-}
-
-int
-efile_fdatasync(Efile_error *errInfo, /* Where to return error codes. */
- int fd) /* File descriptor for file to sync data. */
-{
-#if defined(HAVE_FDATASYNC) && !defined(__DARWIN__)
- return check_error(fdatasync(fd), errInfo);
-#else
- return efile_fsync(errInfo, fd);
-#endif
-}
-
-int
-efile_fsync(Efile_error *errInfo, /* Where to return error codes. */
- int fd) /* File descriptor for file to sync. */
-{
-#ifdef NO_FSYNC
- undefined fsync /* XXX: Really? */
-#else
-#if defined(__DARWIN__) && defined(F_FULLFSYNC)
- return check_error(fcntl(fd, F_FULLFSYNC), errInfo);
-#else
- return check_error(fsync(fd), errInfo);
-#endif /* __DARWIN__ */
-#endif /* NO_FSYNC */
-}
-
-int
-efile_fileinfo(Efile_error* errInfo, Efile_info* pInfo,
- char* name, int info_for_link)
-{
- struct stat statbuf; /* Information about the file */
- int result;
-
- if (info_for_link) {
- result = lstat(name, &statbuf);
- } else {
- result = stat(name, &statbuf);
- }
- if (!check_error(result, errInfo)) {
- return 0;
- }
-
-#if SIZEOF_OFF_T == 4
- pInfo->size_high = 0;
-#else
- pInfo->size_high = (Uint32)(statbuf.st_size >> 32);
-#endif
- pInfo->size_low = (Uint32)statbuf.st_size;
-
-#ifdef NO_ACCESS
- /* Just look at read/write access for owner. */
-
- pInfo->access = ((statbuf.st_mode >> 6) & 07) >> 1;
-
-#else
- pInfo->access = FA_NONE;
- if (access(name, R_OK) == 0)
- pInfo->access |= FA_READ;
- if (access(name, W_OK) == 0)
- pInfo->access |= FA_WRITE;
-
-#endif
-
- if (ISDEV(statbuf))
- pInfo->type = FT_DEVICE;
- else if (ISDIR(statbuf))
- pInfo->type = FT_DIRECTORY;
- else if (ISREG(statbuf))
- pInfo->type = FT_REGULAR;
- else if (ISLNK(statbuf))
- pInfo->type = FT_SYMLINK;
- else
- pInfo->type = FT_OTHER;
-
- pInfo->accessTime = (Sint64)statbuf.st_atime;
- pInfo->modifyTime = (Sint64)statbuf.st_mtime;
- pInfo->cTime = (Sint64)statbuf.st_ctime;
-
- pInfo->mode = statbuf.st_mode;
- pInfo->links = statbuf.st_nlink;
- pInfo->major_device = statbuf.st_dev;
- pInfo->minor_device = statbuf.st_rdev;
- pInfo->inode = statbuf.st_ino;
- pInfo->uid = statbuf.st_uid;
- pInfo->gid = statbuf.st_gid;
-
- return 1;
-}
-
-int
-efile_write_info(Efile_error *errInfo, Efile_info *pInfo, char *name)
-{
- struct utimbuf tval;
-
- /*
- * On some systems chown will always fail for a non-root user unless
- * POSIX_CHOWN_RESTRICTED is not set. Others will succeed as long as
- * you don't try to chown a file to someone besides youself.
- */
-
- if (chown(name, pInfo->uid, pInfo->gid) && errno != EPERM) {
- return check_error(-1, errInfo);
- }
-
- if (pInfo->mode != -1) {
- mode_t newMode = pInfo->mode & (S_ISUID | S_ISGID |
- S_IRWXU | S_IRWXG | S_IRWXO);
- if (chmod(name, newMode)) {
- newMode &= ~(S_ISUID | S_ISGID);
- if (chmod(name, newMode)) {
- return check_error(-1, errInfo);
- }
- }
- }
-
- tval.actime = (time_t)pInfo->accessTime;
- tval.modtime = (time_t)pInfo->modifyTime;
-
- return check_error(utime(name, &tval), errInfo);
-}
-
-
-int
-efile_write(Efile_error* errInfo, /* Where to return error codes. */
- int flags, /* Flags given when file was
- opened. */
- int fd, /* File descriptor to write to. */
- char* buf, /* Buffer to write. */
- size_t count) /* Number of bytes to write. */
-{
- ssize_t written; /* Bytes written in last operation. */
-
- while (count > 0) {
- if ((written = write(fd, buf, count)) < 0) {
- if (errno != EINTR)
- return check_error(-1, errInfo);
- else
- written = 0;
- }
- ASSERT(written <= count);
- buf += written;
- count -= written;
- }
- return 1;
-}
-
-int
-efile_writev(Efile_error* errInfo, /* Where to return error codes */
- int flags, /* Flags given when file was
- * opened */
- int fd, /* File descriptor to write to */
- SysIOVec* iov, /* Vector of buffer structs.
- * The structs may be changed i.e.
- * due to incomplete writes */
- int iovcnt) /* Number of structs in vector */
-{
- int cnt = 0; /* Buffers so far written */
-
- ASSERT(iovcnt >= 0);
-
- while (cnt < iovcnt) {
- if ((! iov[cnt].iov_base) || (iov[cnt].iov_len <= 0)) {
- /* Empty buffer - skip */
- cnt++;
- } else { /* Non-empty buffer */
- ssize_t w; /* Bytes written in this call */
-#ifdef HAVE_WRITEV
- int b = iovcnt - cnt; /* Buffers to write */
- /* Use as many buffers as MAXIOV allows */
- if (b > MAXIOV)
- b = MAXIOV;
- if (b > 1) {
- do {
- w = writev(fd, &iov[cnt], b);
- } while (w < 0 && errno == EINTR);
- if (w < 0 && errno == EINVAL) {
- goto single_write;
- }
- } else
- single_write:
- /* Degenerated io vector - use regular write */
-#endif
- {
- do {
- size_t iov_len = iov[cnt].iov_len;
- size_t limit = 1024*1024*1024; /* 1GB */
- if (iov_len > limit) {
- iov_len = limit;
- }
- w = write(fd, iov[cnt].iov_base, iov_len);
- } while (w < 0 && errno == EINTR);
- ASSERT(w <= iov[cnt].iov_len ||
- (w == -1 && errno != EINTR));
- }
- if (w < 0) return check_error(-1, errInfo);
- /* Move forward to next buffer to write */
- for (; cnt < iovcnt && w > 0; cnt++) {
- if (iov[cnt].iov_base && iov[cnt].iov_len > 0) {
- if (w < iov[cnt].iov_len) {
- /* Adjust the buffer for next write */
- iov[cnt].iov_len -= w;
- iov[cnt].iov_base = ((char *)iov[cnt].iov_base) + w;
- w = 0;
- break;
- } else {
- w -= iov[cnt].iov_len;
- }
- }
- }
- ASSERT(w == 0);
- } /* else Non-empty buffer */
- } /* while (cnt< iovcnt) */
- return 1;
-}
-
-int
-efile_read(Efile_error* errInfo, /* Where to return error codes. */
- int flags, /* Flags given when file was opened. */
- int fd, /* File descriptor to read from. */
- char* buf, /* Buffer to read into. */
- size_t count, /* Number of bytes to read. */
- size_t *pBytesRead) /* Where to return number of
- bytes read. */
-{
- ssize_t n;
-
- for (;;) {
- if ((n = read(fd, buf, count)) >= 0)
- break;
- else if (errno != EINTR)
- return check_error(-1, errInfo);
- }
- *pBytesRead = (size_t) n;
- return 1;
-}
-
-
-/* pread() and pwrite() */
-/* Some unix systems, notably Solaris has these syscalls */
-/* It is especially nice for i.e. the dets module to have support */
-/* for this, even if the underlying OS dosn't support it, it is */
-/* reasonably easy to work around by first calling seek, and then */
-/* calling read(). */
-/* This later strategy however changes the file pointer, which pread() */
-/* does not do. We choose to ignore this and say that the location */
-/* of the file pointer is undefined after a call to any of the p functions*/
-
-
-int
-efile_pread(Efile_error* errInfo, /* Where to return error codes. */
- int fd, /* File descriptor to read from. */
- Sint64 offset, /* Offset in bytes from BOF. */
- char* buf, /* Buffer to read into. */
- size_t count, /* Number of bytes to read. */
- size_t *pBytesRead) /* Where to return
- number of bytes read. */
-{
-#if defined(HAVE_PREAD) && defined(HAVE_PWRITE)
- ssize_t n;
- off_t off = (off_t) offset;
- if (off != offset) {
- errno = EINVAL;
- return check_error(-1, errInfo);
- }
- for (;;) {
- if ((n = pread(fd, buf, count, offset)) >= 0)
- break;
- else if (errno != EINTR)
- return check_error(-1, errInfo);
- }
- *pBytesRead = (size_t) n;
- return 1;
-#else
- {
- int res = efile_seek(errInfo, fd, offset, EFILE_SEEK_SET, NULL);
- if (res) {
- return efile_read(errInfo, 0, fd, buf, count, pBytesRead);
- } else {
- return res;
- }
- }
-#endif
-}
-
-
-
-int
-efile_pwrite(Efile_error* errInfo, /* Where to return error codes. */
- int fd, /* File descriptor to write to. */
- char* buf, /* Buffer to write. */
- size_t count, /* Number of bytes to write. */
- Sint64 offset) /* where to write it */
-{
-#if defined(HAVE_PREAD) && defined(HAVE_PWRITE)
- ssize_t written; /* Bytes written in last operation. */
- off_t off = (off_t) offset;
- if (off != offset) {
- errno = EINVAL;
- return check_error(-1, errInfo);
- }
-
- while (count > 0) {
- if ((written = pwrite(fd, buf, count, offset)) < 0) {
- if (errno != EINTR)
- return check_error(-1, errInfo);
- else
- written = 0;
- }
- ASSERT(written <= count);
- buf += written;
- count -= written;
- offset += written;
- }
- return 1;
-#else /* For unix systems that don't support pread() and pwrite() */
- {
- int res = efile_seek(errInfo, fd, offset, EFILE_SEEK_SET, NULL);
-
- if (res) {
- return efile_write(errInfo, 0, fd, buf, count);
- } else {
- return res;
- }
- }
-#endif
-}
-
-
-int
-efile_seek(Efile_error* errInfo, /* Where to return error codes. */
- int fd, /* File descriptor to do the seek on. */
- Sint64 offset, /* Offset in bytes from the given
- origin. */
- int origin, /* Origin of seek (SEEK_SET, SEEK_CUR,
- SEEK_END). */
- Sint64 *new_location) /* Resulting new location in file. */
-{
- off_t off, result;
-
- switch (origin) {
- case EFILE_SEEK_SET: origin = SEEK_SET; break;
- case EFILE_SEEK_CUR: origin = SEEK_CUR; break;
- case EFILE_SEEK_END: origin = SEEK_END; break;
- default:
- errno = EINVAL;
- return check_error(-1, errInfo);
- }
- off = (off_t) offset;
- if (off != offset) {
- errno = EINVAL;
- return check_error(-1, errInfo);
- }
-
- errno = 0;
- result = lseek(fd, off, origin);
-
- /*
- * Note that the man page for lseek (on SunOs 5) says:
- *
- * "if fildes is a remote file descriptor and offset is
- * negative, lseek() returns the file pointer even if it is
- * negative."
- */
-
- if (result < 0 && errno == 0)
- errno = EINVAL;
- if (result < 0)
- return check_error(-1, errInfo);
- if (new_location) {
- *new_location = result;
- }
- return 1;
-}
-
-
-int
-efile_truncate_file(Efile_error* errInfo, int *fd, int flags)
-{
-#ifndef NO_FTRUNCATE
- off_t offset;
-
- return check_error((offset = lseek(*fd, 0, 1)) >= 0 &&
- ftruncate(*fd, offset) == 0 ? 1 : -1,
- errInfo);
-#else
- return 1;
-#endif
-}
-
-int
-efile_readlink(Efile_error* errInfo, char* name, char* buffer, size_t size)
-{
- int len;
- ASSERT(size > 0);
- len = readlink(name, buffer, size-1);
- if (len == -1) {
- return check_error(-1, errInfo);
- }
- buffer[len] = '\0';
- return 1;
-}
-
-int
-efile_altname(Efile_error* errInfo, char* name, char* buffer, size_t size)
-{
- errno = ENOTSUP;
- return check_error(-1, errInfo);
-}
-
-int
-efile_link(Efile_error* errInfo, char* old, char* new)
-{
- return check_error(link(old, new), errInfo);
-}
-
-int
-efile_symlink(Efile_error* errInfo, char* old, char* new)
-{
- return check_error(symlink(old, new), errInfo);
-}
-
-int
-efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset,
- Sint64 length, int advise)
-{
-#ifdef HAVE_POSIX_FADVISE
- return check_error(posix_fadvise(fd, offset, length, advise), errInfo);
-#else
- return check_error(0, errInfo);
-#endif
-}
-
-#ifdef HAVE_SENDFILE
-/* For some reason the maximum size_t cannot be used as the max size
- 3GB seems to work on all platforms */
-#define SENDFILE_CHUNK_SIZE ((1UL << 30) -1)
-
-/*
- * sendfile: The implementation of the sendfile system call varies
- * a lot on different *nix platforms so to make the api similar in all
- * we have to emulate some things in linux and play with variables on
- * bsd/darwin.
- *
- * All of the calls will split a command which tries to send more than
- * SENDFILE_CHUNK_SIZE of data at once.
- *
- * On platforms where *nbytes of 0 does not mean the entire file, this is
- * simulated.
- *
- * It could be possible to implement header/trailer in sendfile. Though
- * you would have to emulate it in linux and on BSD/Darwin some complex
- * calculations have to be made when using a non blocking socket to figure
- * out how much of the header/file/trailer was sent in each command.
- *
- * The semantics of the API is this:
- * Return value: 1 if all data was sent and the function does not need to
- * be called again. 0 if an error occures OR if there is more data which
- * has to be sent (EAGAIN or EINTR will be set appropriately)
- *
- * The amount of data written in a call is returned through nbytes.
- *
- */
-
-int
-efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
- off_t *offset, Uint64 *nbytes, struct t_sendfile_hdtl* hdtl)
-{
- Uint64 written = 0;
-#if defined(__linux__)
- ssize_t retval;
- do {
- /* check if *nbytes is 0 or greater than chunk size */
- if (*nbytes == 0 || *nbytes > SENDFILE_CHUNK_SIZE)
- retval = sendfile(out_fd, in_fd, offset, SENDFILE_CHUNK_SIZE);
- else
- retval = sendfile(out_fd, in_fd, offset, *nbytes);
- if (retval > 0) {
- written += retval;
- *nbytes -= retval;
- }
- } while (retval == SENDFILE_CHUNK_SIZE);
- if (written != 0) {
- /* -1 is not returned by the linux API so we have to simulate it */
- retval = -1;
- errno = EAGAIN;
- }
-#elif defined(__sun) && defined(__SVR4) && defined(HAVE_SENDFILEV)
- ssize_t retval;
- size_t len;
- sendfilevec_t fdrec;
- fdrec.sfv_fd = in_fd;
- fdrec.sfv_flag = 0;
- do {
- fdrec.sfv_off = *offset;
- len = 0;
- /* check if *nbytes is 0 or greater than chunk size */
- if (*nbytes == 0 || *nbytes > SENDFILE_CHUNK_SIZE)
- fdrec.sfv_len = SENDFILE_CHUNK_SIZE;
- else
- fdrec.sfv_len = *nbytes;
-
- retval = sendfilev(out_fd, &fdrec, 1, &len);
-
- if (retval == -1 && errno == EINVAL) {
- /* On some solaris versions (I've seen it on SunOS 5.10),
- using a sfv_len larger then a filesize will result in
- a -1 && errno == EINVAL return. We translate this so
- a successful send of the data.*/
- retval = len;
- }
-
- if (retval != -1 || errno == EAGAIN || errno == EINTR) {
- *offset += len;
- *nbytes -= len;
- written += len;
- }
- } while (len == SENDFILE_CHUNK_SIZE);
-#elif defined(__DARWIN__)
- int retval;
- off_t len;
- do {
- /* check if *nbytes is 0 or greater than chunk size */
- if(*nbytes > SENDFILE_CHUNK_SIZE)
- len = SENDFILE_CHUNK_SIZE;
- else
- len = *nbytes;
- retval = sendfile(in_fd, out_fd, *offset, &len, NULL, 0);
- if (retval != -1 || errno == EAGAIN || errno == EINTR) {
- *offset += len;
- *nbytes -= len;
- written += len;
- }
- } while (len == SENDFILE_CHUNK_SIZE);
-#elif defined(__FreeBSD__) || defined(__DragonFly__)
- off_t len;
- int retval;
- do {
- if (*nbytes > SENDFILE_CHUNK_SIZE)
- retval = sendfile(in_fd, out_fd, *offset, SENDFILE_CHUNK_SIZE,
- NULL, &len, 0);
- else
- retval = sendfile(in_fd, out_fd, *offset, *nbytes, NULL, &len, 0);
- if (retval != -1 || errno == EAGAIN || errno == EINTR) {
- *offset += len;
- *nbytes -= len;
- written += len;
- }
- } while(len == SENDFILE_CHUNK_SIZE);
-#endif
- *nbytes = written;
- return check_error(retval, errInfo);
-}
-#endif /* HAVE_SENDFILE */
-
-#ifdef HAVE_POSIX_FALLOCATE
-static int
-call_posix_fallocate(int fd, Sint64 offset, Sint64 length)
-{
- int ret;
-
- /*
- * On Linux and Solaris for example, posix_fallocate() returns
- * a positive error number on error and it does not set errno.
- * On FreeBSD however (9.0 at least), it returns -1 on error
- * and it sets errno.
- */
- do {
- ret = posix_fallocate(fd, (off_t) offset, (off_t) length);
- if (ret > 0) {
- errno = ret;
- ret = -1;
- }
- } while (ret != 0 && errno == EINTR);
-
- return ret;
-}
-#endif /* HAVE_POSIX_FALLOCATE */
-
-int
-efile_fallocate(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length)
-{
-#if defined HAVE_FALLOCATE
- /* Linux specific, more efficient than posix_fallocate. */
- int ret;
-
- do {
- ret = fallocate(fd, FALLOC_FL_KEEP_SIZE, (off_t) offset, (off_t) length);
- } while (ret != 0 && errno == EINTR);
-
-#if defined HAVE_POSIX_FALLOCATE
- /* Fallback to posix_fallocate if available. */
- if (ret != 0) {
- ret = call_posix_fallocate(fd, offset, length);
- }
-#endif
-
- return check_error(ret, errInfo);
-#elif defined F_PREALLOCATE
- /* Mac OS X specific, equivalent to posix_fallocate. */
- int ret;
- fstore_t fs;
-
- memset(&fs, 0, sizeof(fs));
- fs.fst_flags = F_ALLOCATECONTIG;
- fs.fst_posmode = F_VOLPOSMODE;
- fs.fst_offset = (off_t) offset;
- fs.fst_length = (off_t) length;
-
- ret = fcntl(fd, F_PREALLOCATE, &fs);
-
- if (-1 == ret) {
- fs.fst_flags = F_ALLOCATEALL;
- ret = fcntl(fd, F_PREALLOCATE, &fs);
-
-#if defined HAVE_POSIX_FALLOCATE
- /* Fallback to posix_fallocate if available. */
- if (-1 == ret) {
- ret = call_posix_fallocate(fd, offset, length);
- }
-#endif
- }
-
- return check_error(ret, errInfo);
-#elif defined HAVE_POSIX_FALLOCATE
- /* Other Unixes, use posix_fallocate if available. */
- return check_error(call_posix_fallocate(fd, offset, length), errInfo);
-#else
- errno = ENOTSUP;
- return check_error(-1, errInfo);
-#endif
-}
diff --git a/erts/emulator/drivers/win32/win_efile.c b/erts/emulator/drivers/win32/win_efile.c
deleted file mode 100644
index 2d366b5833..0000000000
--- a/erts/emulator/drivers/win32/win_efile.c
+++ /dev/null
@@ -1,2058 +0,0 @@
-/*
- * %CopyrightBegin%
- *
- * Copyright Ericsson AB 1997-2016. All Rights Reserved.
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- *
- * %CopyrightEnd%
- */
-/*
- * Purpose: Provides file and directory operations for Windows.
- */
-
-#include <windows.h>
-#ifdef HAVE_CONFIG_H
-# include "config.h"
-#endif
-#include "sys.h"
-#include <ctype.h>
-#include <wchar.h>
-#include "erl_efile.h"
-
-#define DBG_TRACE_MASK 0
-/* 1 = file name ops
- * 2 = file descr ops
- * 4 = errors
- * 8 = path name conversion
- */
-#if !DBG_TRACE_MASK
-# define DBG_TRACE(M,S)
-# define DBG_TRACE1(M,FMT,A)
-# define DBG_TRACE2(M,FMT,A,B)
-#else
-# define DBG_TRACE(M,S) do { if ((M)&DBG_TRACE_MASK) fwprintf(stderr, L"DBG_TRACE %d: %s\r\n", __LINE__, (WCHAR*)(S)); }while(0)
-# define DBG_TRACE1(M,FMT,A) do { if ((M)&DBG_TRACE_MASK) fwprintf(stderr, L"DBG_TRACE %d: " L##FMT L"\r\n", __LINE__, (A)); }while(0)
-# define DBG_TRACE2(M,FMT,A,B) do { if ((M)&DBG_TRACE_MASK) fwprintf(stderr, L"DBG_TRACE %d: " L##FMT L"\r\n", __LINE__, (A), (B)); }while(0)
-#endif
-
-/*
- * Microsoft-specific function to map a WIN32 error code to a Posix errno.
- */
-
-#define ISSLASH(a) ((a) == L'\\' || (a) == L'/')
-#define ISDIR(st) (((st).st_mode&S_IFMT) == S_IFDIR)
-#define ISREG(st) (((st).st_mode&S_IFMT) == S_IFREG)
-
-#define IS_DOT_OR_DOTDOT(s) \
- ((s)[0] == L'.' && ((s)[1] == L'\0' || ((s)[1] == L'.' && (s)[2] == L'\0')))
-
-#define FILE_SHARE_FLAGS (FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE)
-
-#ifndef INVALID_FILE_ATTRIBUTES
-#define INVALID_FILE_ATTRIBUTES ((DWORD) 0xFFFFFFFF)
-#endif
-
-#define TICKS_PER_SECOND (10000000ULL)
-#define EPOCH_DIFFERENCE (11644473600LL)
-
-#define FILETIME_TO_EPOCH(epoch, ft) \
- do { \
- ULARGE_INTEGER ull; \
- ull.LowPart = (ft).dwLowDateTime; \
- ull.HighPart = (ft).dwHighDateTime; \
- (epoch) = ((ull.QuadPart / TICKS_PER_SECOND) - EPOCH_DIFFERENCE); \
- } while(0)
-
-#define EPOCH_TO_FILETIME(ft, epoch) \
- do { \
- ULARGE_INTEGER ull; \
- ull.QuadPart = (((epoch) + EPOCH_DIFFERENCE) * TICKS_PER_SECOND); \
- (ft).dwLowDateTime = ull.LowPart; \
- (ft).dwHighDateTime = ull.HighPart; \
- } while(0)
-
-
-static int check_error(int result, Efile_error* errInfo);
-static int set_error(Efile_error* errInfo);
-static int set_os_errno(Efile_error* errInfo, DWORD os_errno);
-static int is_root_unc_name(const WCHAR *path);
-static int extract_root(WCHAR *name);
-static unsigned short dos_to_posix_mode(int attr, const WCHAR *name);
-
-
-struct wpath_tmp_buffer {
- struct wpath_tmp_buffer* next;
- WCHAR buffer[1];
-};
-
-typedef struct {
- Efile_error* errInfo;
- struct wpath_tmp_buffer* buf_list;
-}Efile_call_state;
-
-static void call_state_init(Efile_call_state* state, Efile_error* errInfo)
-{
- state->errInfo = errInfo;
- state->buf_list = NULL;
-}
-static WCHAR* wpath_tmp_alloc(Efile_call_state* state, size_t len)
-{
- size_t sz = offsetof(struct wpath_tmp_buffer, buffer)
- + (len+1)*sizeof(WCHAR);
- struct wpath_tmp_buffer* p = driver_alloc(sz);
- p->next = state->buf_list;
- state->buf_list = p;
- return p->buffer;
-}
-static void call_state_free(Efile_call_state* state)
-{
- while(state->buf_list) {
- struct wpath_tmp_buffer* next = state->buf_list->next;
- driver_free(state->buf_list);
- state->buf_list = next;
- }
-}
-static WCHAR* get_cwd_wpath_tmp(Efile_call_state* state)
-{
- WCHAR dummy;
- DWORD size = GetCurrentDirectoryW(0, &dummy);
- WCHAR* ret = NULL;
-
- if (size) {
- ret = wpath_tmp_alloc(state, size);
- if (!GetCurrentDirectoryW(size, ret)) {
- ret = NULL;
- }
- }
- return ret;
-}
-static WCHAR* get_full_wpath_tmp(Efile_call_state* state,
- const WCHAR* file,
- WCHAR** file_part,
- DWORD extra)
-{
- WCHAR dummy;
- DWORD size = GetFullPathNameW(file, 0, &dummy, NULL);
- WCHAR* ret = NULL;
-
- if (size) {
- int ok;
- ret = wpath_tmp_alloc(state, size + extra);
- if (file_part) {
- ok = (GetFullPathNameW(file, size, ret, file_part) != 0);
- }
- else {
- ok = (_wfullpath(ret, file, size) != NULL);
- }
- if (!ok) {
- ret = NULL;
- }
- }
- return ret;
-}
-
-static void ensure_wpath_max(Efile_call_state* state, WCHAR** pathp, size_t max);
-static int do_rmdir(Efile_call_state*, char* name);
-static int do_rename(Efile_call_state*, char* src, char* dst);
-static int do_readdir(Efile_call_state*, char* name, EFILE_DIR_HANDLE*, char* buffer, size_t *size);
-static int do_fileinfo(Efile_call_state*, Efile_info*, char* orig_name, int info_for_link);
-static char* do_readlink(Efile_call_state*, char* name, char* buffer, size_t size);
-static int do_altname(Efile_call_state*, char* orig_name, char* buffer, size_t size);
-
-
-static int errno_map(DWORD last_error) {
-
- switch (last_error) {
- case ERROR_SUCCESS:
- return 0;
- case ERROR_INVALID_FUNCTION:
- case ERROR_INVALID_DATA:
- case ERROR_INVALID_PARAMETER:
- case ERROR_INVALID_TARGET_HANDLE:
- case ERROR_INVALID_CATEGORY:
- case ERROR_NEGATIVE_SEEK:
- return EINVAL;
- case ERROR_DIR_NOT_EMPTY:
- return EEXIST;
- case ERROR_BAD_FORMAT:
- return ENOEXEC;
- case ERROR_PATH_NOT_FOUND:
- case ERROR_FILE_NOT_FOUND:
- case ERROR_NO_MORE_FILES:
- return ENOENT;
- case ERROR_TOO_MANY_OPEN_FILES:
- return EMFILE;
- case ERROR_ACCESS_DENIED:
- case ERROR_INVALID_ACCESS:
- case ERROR_CURRENT_DIRECTORY:
- case ERROR_SHARING_VIOLATION:
- case ERROR_LOCK_VIOLATION:
- case ERROR_INVALID_PASSWORD:
- case ERROR_DRIVE_LOCKED:
- return EACCES;
- case ERROR_INVALID_HANDLE:
- return EBADF;
- case ERROR_NOT_ENOUGH_MEMORY:
- case ERROR_OUTOFMEMORY:
- case ERROR_OUT_OF_STRUCTURES:
- return ENOMEM;
- case ERROR_INVALID_DRIVE:
- case ERROR_BAD_UNIT:
- case ERROR_NOT_READY:
- case ERROR_REM_NOT_LIST:
- case ERROR_DUP_NAME:
- case ERROR_BAD_NETPATH:
- case ERROR_NETWORK_BUSY:
- case ERROR_DEV_NOT_EXIST:
- case ERROR_BAD_NET_NAME:
- return ENXIO;
- case ERROR_NOT_SAME_DEVICE:
- return EXDEV;
- case ERROR_WRITE_PROTECT:
- return EROFS;
- case ERROR_BAD_LENGTH:
- case ERROR_BUFFER_OVERFLOW:
- return E2BIG;
- case ERROR_SEEK:
- case ERROR_SECTOR_NOT_FOUND:
- return ESPIPE;
- case ERROR_NOT_DOS_DISK:
- return ENODEV;
- case ERROR_GEN_FAILURE:
- return ENODEV;
- case ERROR_SHARING_BUFFER_EXCEEDED:
- case ERROR_NO_MORE_SEARCH_HANDLES:
- return EMFILE;
- case ERROR_HANDLE_EOF:
- case ERROR_BROKEN_PIPE:
- return EPIPE;
- case ERROR_HANDLE_DISK_FULL:
- case ERROR_DISK_FULL:
- return ENOSPC;
- case ERROR_NOT_SUPPORTED:
- return ENOTSUP;
- case ERROR_FILE_EXISTS:
- case ERROR_ALREADY_EXISTS:
- case ERROR_CANNOT_MAKE:
- return EEXIST;
- case ERROR_ALREADY_ASSIGNED:
- return EBUSY;
- case ERROR_NO_PROC_SLOTS:
- return EAGAIN;
- case ERROR_CANT_RESOLVE_FILENAME:
- return EMLINK;
- case ERROR_PRIVILEGE_NOT_HELD:
- return EPERM;
- case ERROR_ARENA_TRASHED:
- case ERROR_INVALID_BLOCK:
- case ERROR_BAD_ENVIRONMENT:
- case ERROR_BAD_COMMAND:
- case ERROR_CRC:
- case ERROR_OUT_OF_PAPER:
- case ERROR_READ_FAULT:
- case ERROR_WRITE_FAULT:
- case ERROR_WRONG_DISK:
- case ERROR_NET_WRITE_FAULT:
- return EIO;
- default: /* not to do with files I expect. */
- return EIO;
- }
-}
-
-static int
-check_error(int result, Efile_error* errInfo)
-{
- if (result < 0) {
- errInfo->posix_errno = errno;
- errInfo->os_errno = GetLastError();
- DBG_TRACE2(4, "ERROR os_error=%d errno=%d @@@@@@@@@@@@@@@@@@@@@@@@@@@@",
- errInfo->os_errno, errInfo->posix_errno);
- return 0;
- }
- return 1;
-}
-
-static void
-save_last_error(Efile_error* errInfo)
-{
- errInfo->posix_errno = errno;
- errInfo->os_errno = GetLastError();
- DBG_TRACE2(4, "ERROR os_error=%d errno=%d $$$$$$$$$$$$$$$$$$$$$$$$$$$$$",
- errInfo->os_errno, errInfo->posix_errno);
-}
-
-
-/*
- * Fills the provided error information structure with information
- * with the error code given by GetLastError() and its corresponding
- * Posix error number.
- *
- * Returns 0.
- */
-
-static int
-set_error(Efile_error* errInfo)
-{
- set_os_errno(errInfo, GetLastError());
- return 0;
-}
-
-
-static int
-set_os_errno(Efile_error* errInfo, DWORD os_errno)
-{
- errInfo->os_errno = os_errno;
- errInfo->posix_errno = errno_map(os_errno);
- DBG_TRACE2(4, "ERROR os_error=%d errno=%d ############################",
- errInfo->os_errno, errInfo->posix_errno);
- return 0;
-}
-
-int
-efile_init() {
- return 1;
-}
-
-/*
- * A writev with Unix semantics, but with Windows arguments
- */
-static int
-win_writev(Efile_error* errInfo,
- HANDLE fd, /* handle to file */
- FILE_SEGMENT_ELEMENT iov[], /* array of buffer pointers */
- DWORD *size) /* number of bytes to write */
-{
- OVERLAPPED ov;
- ov.Offset = 0L;
- ov.OffsetHigh = 0L;
- ov.hEvent = CreateEvent(NULL, TRUE, FALSE, NULL);
- if (ov.hEvent == NULL)
- return set_error(errInfo);
- if (! write_file_gather(fd, iov, *size, NULL, &ov))
- return set_error(errInfo);
- if (WaitForSingleObject(ov.hEvent, INFINITE) != WAIT_OBJECT_0)
- return set_error(errInfo);
- if (! GetOverlappedResult(fd, &ov, size, FALSE))
- return set_error(errInfo);
- return 1;
-}
-
-
-/* Check '*pathp' and convert it if needed to something that windows will accept.
- * Typically use UNC path with \\?\ prefix if absolute path is longer than 260.
- */
-static void ensure_wpath(Efile_call_state* state, WCHAR** pathp)
-{
- ensure_wpath_max(state, pathp, MAX_PATH);
-}
-
-static void ensure_wpath_max(Efile_call_state* state, WCHAR** pathp, size_t max)
-{
- WCHAR* path = *pathp;
- WCHAR* p;
- size_t len = wcslen(path);
- int unc_fixup = 0;
-
- if (path[0] == 0) {
- DBG_TRACE(8, L"Let empty path pass through");
- return;
- }
-
- DBG_TRACE1(8,"IN: %s", path);
-
- if (path[1] == L':' && ISSLASH(path[2])) { /* absolute path */
- if (len >= max) {
- WCHAR *src, *dst;
-
- *pathp = wpath_tmp_alloc(state, 4+len+1);
- dst = *pathp;
- wcscpy(dst, L"\\\\?\\");
- for (src=path,dst+=4; *src; src++) {
- if (*src == L'/') {
- if (dst[-1] != L'\\') {
- *dst++ = L'\\';
- }
- /*else ignore redundant slashes */
- }
- else
- *dst++ = *src;
- }
- *dst = 0;
- unc_fixup = 1;
- }
- }
- else if (!(ISSLASH(path[0]) && ISSLASH(path[1]))) { /* relative path */
- DWORD cwdLen = GetCurrentDirectoryW(0, NULL);
- DWORD absLen = cwdLen + 1 + len;
- if (absLen >= max) {
- WCHAR *fullPath = wpath_tmp_alloc(state, 4+4+absLen);
- DWORD fullLen;
-
- fullLen = GetFullPathNameW(path, 4 + absLen, fullPath+4, NULL);
- if (fullLen >= 4+absLen) {
- *pathp = path;
- DBG_TRACE2(8,"ensure_wpath FAILED absLen=%u %s", (int)absLen, path);
- return;
- }
- /* GetFullPathNameW can return paths longer than MAX_PATH without the \\?\ prefix.
- * At least seen on Windows 7. Go figure...
- */
- if (fullLen >= max && wcsncmp(fullPath+4, L"\\\\?\\", 4) != 0) {
- wcsncpy(fullPath, L"\\\\?\\", 4);
- *pathp = fullPath;
- }
- else {
- *pathp = fullPath + 4;
- }
- }
- }
-
- if (unc_fixup) {
- WCHAR* endp;
-
- p = *pathp;
- len = wcslen(p);
- endp = p + len;
- if (len > 4) {
- p += 4;
- while (*p) {
- if (p[0] == L'\\' && p[1] == L'.') {
- if (p[2] == L'\\' || !p[2]) { /* single dot */
- wmemmove(p, p+2, (&endp[1] - &p[2]));
- endp -= 2;
- }
- else if (p[2] == L'.' && (p[3] == L'\\' || !p[3])) { /* double dot */
- WCHAR* r;
- for (r=p-1; *r == L'\\'; --r)
- /*skip redundant slashes*/;
- for (; *r != L'\\'; --r)
- /*find start of prev directory*/;
- if (r < *pathp + 6)
- break;
- wmemmove(r, p+3, (&endp[1] - &p[3]));
- p = r;
- }
- else p += 3;
- }
- else ++p;
- }
- }
- }
- DBG_TRACE1(8,"OUT: %s", *pathp);
-}
-
-int
-efile_mkdir(Efile_error* errInfo, /* Where to return error codes. */
- char* name) /* Name of directory to create. */
-{
- Efile_call_state state;
- WCHAR* wname = (WCHAR*)name;
- int ret;
-
- DBG_TRACE(1, name);
- call_state_init(&state, errInfo);
- ensure_wpath_max(&state, &wname, 248); /* Yes, 248 limit for normal paths */
-
- ret = (int) CreateDirectoryW(wname, NULL);
- if (!ret)
- set_error(errInfo);
-
- call_state_free(&state);
- return ret;
-}
-
-int
-efile_rmdir(Efile_error* errInfo, /* Where to return error codes. */
- char* name) /* Name of directory to delete. */
-{
- Efile_call_state state;
- int ret;
-
- DBG_TRACE(1, name);
- call_state_init(&state, errInfo);
- ret = do_rmdir(&state, name);
- call_state_free(&state);
- return ret;
-}
-
-static int do_rmdir(Efile_call_state* state, char* name)
-{
- OSVERSIONINFO os;
- DWORD attr;
- WCHAR *wname = (WCHAR *) name;
- WCHAR *buffer = NULL;
-
- ensure_wpath(state, &wname);
-
- if (RemoveDirectoryW(wname) != FALSE) {
- return 1;
- }
- errno = errno_map(GetLastError());
- if (errno == EACCES) {
- attr = GetFileAttributesW(wname);
- if (attr != (DWORD) -1) {
- if ((attr & FILE_ATTRIBUTE_DIRECTORY) == 0) {
- /*
- * Windows 95 reports calling RemoveDirectory on a file as an
- * EACCES, not an ENOTDIR.
- */
-
- errno = ENOTDIR;
- goto end;
- }
-
- /*
- * Windows 95 reports removing a non-empty directory as
- * an EACCES, not an EEXIST. If the directory is not empty,
- * change errno so caller knows what's going on.
- */
-
- os.dwOSVersionInfoSize = sizeof(os);
- GetVersionEx(&os);
- if (os.dwPlatformId == VER_PLATFORM_WIN32_WINDOWS) {
- HANDLE handle;
- WIN32_FIND_DATAW data;
- int len = wcslen(wname);
-
- buffer = wpath_tmp_alloc(state, len + 4);
- wcscpy(buffer, wname);
- if (buffer[0] && buffer[len-1] != L'\\' && buffer[len-1] != L'/') {
- wcscat(buffer, L"\\");
- }
- wcscat(buffer, L"*.*");
- handle = FindFirstFileW(buffer, &data);
- if (handle != INVALID_HANDLE_VALUE) {
- while (1) {
- if ((wcscmp(data.cFileName, L".") != 0)
- && (wcscmp(data.cFileName, L"..") != 0)) {
- /*
- * Found something in this directory.
- */
-
- errno = EEXIST;
- break;
- }
- if (FindNextFileW(handle, &data) == FALSE) {
- break;
- }
- }
- FindClose(handle);
- }
- }
- }
- }
-
- if (errno == ENOTEMPTY) {
- /*
- * Posix allows both EEXIST or ENOTEMPTY, but we'll always
- * return EEXIST to allow easy matching in Erlang code.
- */
-
- errno = EEXIST;
- }
-
- end:
- save_last_error(state->errInfo);
- return 0;
-}
-
-int
-efile_delete_file(Efile_error* errInfo, /* Where to return error codes. */
- char* name) /* Name of file to delete. */
-{
- Efile_call_state state;
- int ret;
- DBG_TRACE(1, name);
- call_state_init(&state, errInfo);
- ret = do_delete_file(&state, name);
- call_state_free(&state);
- return ret;
-}
-
-static int do_delete_file(Efile_call_state* state, char* name)
-{
- DWORD attr;
- WCHAR *wname = (WCHAR *) name;
-
- ensure_wpath(state, &wname);
-
- if (DeleteFileW(wname) != FALSE) {
- return 1;
- }
-
- errno = errno_map(GetLastError());
- if (errno == EACCES) {
- attr = GetFileAttributesW(wname);
- if (attr != (DWORD) -1) {
- if (attr & FILE_ATTRIBUTE_DIRECTORY) {
- /*
- * Windows NT reports removing a directory as EACCES instead
- * of EPERM.
- */
-
- errno = EPERM;
- }
- }
- } else if (errno == ENOENT) {
- attr = GetFileAttributesW(wname);
- if (attr != (DWORD) -1) {
- if (attr & FILE_ATTRIBUTE_DIRECTORY) {
- /*
- * Windows 95 reports removing a directory as ENOENT instead
- * of EPERM.
- */
-
- errno = EPERM;
- }
- }
- } else if (errno == EINVAL) {
- /*
- * Windows NT reports removing a char device as EINVAL instead of
- * EACCES.
- */
-
- errno = EACCES;
- }
-
- return check_error(-1, state->errInfo);
-}
-
-/*
- *---------------------------------------------------------------------------
- *
- * Changes the name of an existing file or directory, from src to dst.
- * If src and dst refer to the same file or directory, does nothing
- * and returns success. Otherwise if dst already exists, it will be
- * deleted and replaced by src subject to the following conditions:
- * If src is a directory, dst may be an empty directory.
- * If src is a file, dst may be a file.
- * In any other situation where dst already exists, the rename will
- * fail.
- *
- * Some possible error codes:
- *
- * EACCES: src or dst parent directory can't be read and/or written.
- * EEXIST: dst is a non-empty directory.
- * EINVAL: src is a root directory or dst is a subdirectory of src.
- * EISDIR: dst is a directory, but src is not.
- * ENOENT: src doesn't exist, or src or dst is "".
- * ENOTDIR: src is a directory, but dst is not.
- * EXDEV: src and dst are on different filesystems.
- *
- * Side effects:
- * The implementation of rename may allow cross-filesystem renames,
- * but the caller should be prepared to emulate it with copy and
- * delete if errno is EXDEV.
- *
- *---------------------------------------------------------------------------
- */
-
-int
-efile_rename(Efile_error* errInfo, char* src, char* dst)
-{
- Efile_call_state state;
- int ret;
- DBG_TRACE(1, src);
- call_state_init(&state, errInfo);
- ret = do_rename(&state, src, dst);
- call_state_free(&state);
- return ret;
-}
-
-static int
-do_rename(Efile_call_state* state,
- char* src, /* Original name. */
- char* dst) /* New name. */
-{
- DWORD srcAttr, dstAttr;
- WCHAR *wsrc = (WCHAR *) src;
- WCHAR *wdst = (WCHAR *) dst;
-
- ensure_wpath(state, &wsrc);
- ensure_wpath(state, &wdst);
-
- if (MoveFileW(wsrc, wdst) != FALSE) {
- return 1;
- }
-
- errno = errno_map(GetLastError());
- srcAttr = GetFileAttributesW(wsrc);
- dstAttr = GetFileAttributesW(wdst);
- if (srcAttr == (DWORD) -1) {
- srcAttr = 0;
- }
- if (dstAttr == (DWORD) -1) {
- dstAttr = 0;
- }
-
- if (errno == EBADF) {
- errno = EACCES;
- return check_error(-1, state->errInfo);
- }
- if (errno == EACCES) {
- decode:
- if (srcAttr & FILE_ATTRIBUTE_DIRECTORY) {
- WCHAR *srcPath, *dstPath;
- WCHAR *srcRest, *dstRest;
- int size;
-
- srcPath = get_full_wpath_tmp(state, wsrc, &srcRest, 0);
- if (!srcPath) {
- save_last_error(state->errInfo);
- return 0;
- }
-
- dstPath = get_full_wpath_tmp(state, wdst, &dstRest, 0);
- if (!dstPath) {
- save_last_error(state->errInfo);
- return 0;
- }
-
- if (srcRest == NULL) {
- srcRest = srcPath + wcslen(srcPath);
- }
- if (_wcsnicmp(srcPath, dstPath, srcRest - srcPath) == 0) {
- /*
- * Trying to move a directory into itself.
- */
-
- errno = EINVAL;
- }
- if (extract_root(srcPath)) {
- /*
- * Attempt to move a root directory. Never allowed.
- */
- errno = EINVAL;
- }
-
- (void) extract_root(dstPath);
- if (dstPath[0] == L'\0') {
- /*
- * The filename was invalid. (Don't know why,
- * but play it safe.)
- */
- errno = EINVAL;
- }
- if (_wcsicmp(srcPath, dstPath) != 0) {
- /*
- * If src is a directory and dst filesystem != src
- * filesystem, errno should be EXDEV. It is very
- * important to get this behavior, so that the caller
- * can respond to a cross filesystem rename by
- * simulating it with copy and delete. The MoveFile
- * system call already handles the case of moving a
- * *file* between filesystems.
- */
-
- errno = EXDEV;
- }
- }
-
- /*
- * Other types of access failure is that dst is a read-only
- * filesystem, that an open file referred to src or dest, or that
- * src or dest specified the current working directory on the
- * current filesystem. EACCES is returned for those cases.
- */
-
- } else if (errno == EEXIST) {
- /*
- * Reports EEXIST any time the target already exists. If it makes
- * sense, remove the old file and try renaming again.
- */
-
- if (srcAttr & FILE_ATTRIBUTE_DIRECTORY) {
- if (dstAttr & FILE_ATTRIBUTE_DIRECTORY) {
- /*
- * Overwrite empty dst directory with src directory. The
- * following call will remove an empty directory. If it
- * fails, it's because it wasn't empty.
- */
-
- if (RemoveDirectoryW(wdst)) {
- /*
- * Now that that empty directory is gone, we can try
- * renaming again. If that fails, we'll put this empty
- * directory back, for completeness.
- */
-
- if (MoveFileW(wsrc, wdst) != FALSE) {
- return 1;
- }
-
- /*
- * Some new error has occurred. Don't know what it
- * could be, but report this one.
- */
-
- errno = errno_map(GetLastError());
- CreateDirectoryW(wdst, NULL);
- SetFileAttributesW(wdst, dstAttr);
- if (errno == EACCES) {
- /*
- * Decode the EACCES to a more meaningful error.
- */
-
- goto decode;
- }
- }
- } else { /* (dstAttr & FILE_ATTRIBUTE_DIRECTORY) == 0 */
- errno = ENOTDIR;
- }
- } else { /* (srcAttr & FILE_ATTRIBUTE_DIRECTORY) == 0 */
- if (dstAttr & FILE_ATTRIBUTE_DIRECTORY) {
- errno = EISDIR;
- } else {
- /*
- * Overwrite existing file by:
- *
- * 1. Rename existing file to temp name.
- * 2. Rename old file to new name.
- * 3. If success, delete temp file. If failure,
- * put temp file back to old name.
- */
-
- WCHAR *tempName;
- int result;
- WCHAR *rest;
-
- tempName = get_full_wpath_tmp(state, wdst, &rest, 14);
- if (!tempName || !rest) {
- save_last_error(state->errInfo);
- return 0;
- }
-
- *rest = L'\0';
- result = -1;
- if (GetTempFileNameW(tempName, L"erlr", 0, tempName) != 0) {
- /*
- * Strictly speaking, need the following DeleteFile and
- * MoveFile to be joined as an atomic operation so no
- * other app comes along in the meantime and creates the
- * same temp file.
- */
-
- DeleteFileW(tempName);
- if (MoveFileW(wdst, tempName) != FALSE) {
- if (MoveFileW(wsrc, wdst) != FALSE) {
- SetFileAttributesW(tempName, FILE_ATTRIBUTE_NORMAL);
- DeleteFileW(tempName);
- return 1;
- } else {
- DeleteFileW(wdst);
- MoveFileW(tempName, wdst);
- }
- }
-
- /*
- * Can't backup dst file or move src file. Return that
- * error. Could happen if an open file refers to dst.
- */
-
- errno = errno_map(GetLastError());
- if (errno == EACCES) {
- /*
- * Decode the EACCES to a more meaningful error.
- */
- goto decode;
- }
- }
- return result;
- }
- }
- }
- return check_error(-1, state->errInfo);
-}
-
-int
-efile_chdir(Efile_error* errInfo, /* Where to return error codes. */
- char* name) /* Name of directory to make current. */
-{
- /* We don't even try to handle long paths here
- * as current working directory is always limited to MAX_PATH
- * even if we use UNC paths and SetCurrentDirectoryW()
- */
- int success = check_error(_wchdir((WCHAR *) name), errInfo);
- if (!success && errInfo->posix_errno == EINVAL)
- /* POSIXification of errno */
- errInfo->posix_errno = ENOENT;
- return success;
-}
-
-int
-efile_getdcwd(Efile_error* errInfo, /* Where to return error codes. */
- int drive, /* 0 - current, 1 - A, 2 - B etc. */
- char* buffer, /* Where to return the current directory. */
- size_t size) /* Size of buffer. */
-{
- WCHAR *wbuffer = (WCHAR *) buffer;
- size_t wbuffer_size = size / 2;
- DBG_TRACE(1, L"#getdcwd#");
- if (_wgetdcwd(drive, wbuffer, wbuffer_size) == NULL) {
- return check_error(-1, errInfo);
- }
- DBG_TRACE1(8, "getdcwd OS=%s", wbuffer);
- if (wcsncmp(wbuffer, L"\\\\?\\", 4) == 0) {
- wmemmove(wbuffer, wbuffer+4, wcslen(wbuffer+4)+1);
- }
- for ( ; *wbuffer; wbuffer++)
- if (*wbuffer == L'\\')
- *wbuffer = L'/';
- DBG_TRACE1(8, "getdcwd ERLANG=%s", (WCHAR*)buffer);
- return 1;
-}
-
-int
-efile_readdir(Efile_error* errInfo, char* name, EFILE_DIR_HANDLE* dir_handle,
- char* buffer, size_t *size)
-{
- Efile_call_state state;
- int ret;
- DBG_TRACE(dir_handle?2:1, name);
- call_state_init(&state, errInfo);
- ret = do_readdir(&state, name, dir_handle, buffer, size);
- call_state_free(&state);
- return ret;
-}
-
-static int do_readdir(Efile_call_state* state,
- char* name, /* Name of directory to list */
- EFILE_DIR_HANDLE* dir_handle, /* Handle of opened directory or NULL */
- char* buffer, /* Buffer to put one filename in */
- size_t *size) /* in-out size of buffer/size of filename excluding zero
- termination in bytes*/
-{
- HANDLE dir; /* Handle to directory. */
- WIN32_FIND_DATAW findData; /* Data found by FindFirstFile() or FindNext(). */
- /* Alignment is not honored, this works on x86 because of alignment fixup by processor.
- Not perfect, but faster than alinging by hand (really) */
- WCHAR *wbuffer = (WCHAR *) buffer;
-
- /*
- * First time we must setup everything.
- */
-
- if (*dir_handle == NULL) {
- WCHAR *wname = (WCHAR *) name;
- WCHAR* wildcard;
- int length;
- WCHAR* s;
-
- ensure_wpath_max(state, &wname, MAX_PATH-2);
- length = wcslen(wname);
-
- wildcard = wpath_tmp_alloc(state, length+3);
-
- wcscpy(wildcard, wname);
- s = wildcard+length-1;
- if (*s != L'/' && *s != L'\\')
- *++s = L'\\';
- *++s = L'*';
- *++s = L'\0';
- DEBUGF(("Reading %ws\n", wildcard));
- dir = FindFirstFileW(wildcard, &findData);
- if (dir == INVALID_HANDLE_VALUE) {
- set_error(state->errInfo);
- return 0;
- }
- *dir_handle = (EFILE_DIR_HANDLE) dir;
-
- if (!IS_DOT_OR_DOTDOT(findData.cFileName)) {
- wcscpy(wbuffer, findData.cFileName);
- *size = wcslen(wbuffer)*2;
- return 1;
- }
- }
-
- /*
- * Retrieve the name of the next file using the directory handle.
- */
-
- dir = (HANDLE) *dir_handle;
-
- for (;;) {
- if (FindNextFileW(dir, &findData)) {
- if (IS_DOT_OR_DOTDOT(findData.cFileName))
- continue;
- wcscpy(wbuffer, findData.cFileName);
- *size = wcslen(wbuffer)*2;
- return 1;
- }
-
- if (GetLastError() == ERROR_NO_MORE_FILES) {
- state->errInfo->posix_errno = state->errInfo->os_errno = 0;
- }
- else {
- set_error(state->errInfo);
- }
- FindClose(dir);
- return 0;
- }
-}
-
-int
-efile_openfile(Efile_error* errInfo, char* name, int flags, int* pfd, Sint64* pSize)
-{
- Efile_call_state state;
- int ret;
- DBG_TRACE1(1, "openfile(%s)", name);
- call_state_init(&state, errInfo);
- ret = do_openfile(&state, name, flags, pfd, pSize);
- call_state_free(&state);
- return ret;
-}
-
-static
-int do_openfile(Efile_call_state* state, /* Where to return error codes. */
- char* name, /* Name of directory to open. */
- int flags, /* Flags to use for opening. */
- int* pfd, /* Where to store the file descriptor. */
- Sint64* pSize) /* Where to store the size of the file. */
-{
- Efile_error* errInfo = state->errInfo;
- BY_HANDLE_FILE_INFORMATION fileInfo; /* File information from a handle. */
- HANDLE fd; /* Handle to open file. */
- DWORD access; /* Access mode: GENERIC_READ, GENERIC_WRITE. */
- DWORD crFlags;
- DWORD flagsAndAttrs = FILE_ATTRIBUTE_NORMAL;
- WCHAR *wname = (WCHAR *) name;
-
- switch (flags & (EFILE_MODE_READ|EFILE_MODE_WRITE)) {
- case EFILE_MODE_READ:
- access = GENERIC_READ;
- crFlags = OPEN_EXISTING;
- break;
- case EFILE_MODE_WRITE:
- access = GENERIC_WRITE;
- crFlags = CREATE_ALWAYS;
- break;
- case EFILE_MODE_READ_WRITE:
- access = GENERIC_READ|GENERIC_WRITE;
- crFlags = OPEN_ALWAYS;
- break;
- default:
- errno = EINVAL;
- check_error(-1, errInfo);
- return 0;
- }
-
- if (flags & EFILE_MODE_SYNC) {
- flagsAndAttrs = FILE_FLAG_WRITE_THROUGH;
- }
-
- if (flags & EFILE_MODE_APPEND) {
- crFlags = OPEN_ALWAYS;
- }
- if (flags & EFILE_MODE_EXCL) {
- crFlags = CREATE_NEW;
- }
- ensure_wpath(state, &wname);
- fd = CreateFileW(wname, access,
- FILE_SHARE_FLAGS,
- NULL, crFlags, flagsAndAttrs, NULL);
-
- /*
- * Check for errors.
- */
-
- if (fd == INVALID_HANDLE_VALUE) {
- DWORD attr;
-
- set_error(errInfo);
-
- /*
- * If the error is EACESS, the reason could be that we tried to
- * open a directory. In that case, we'll change the error code
- * to EISDIR.
- */
- if (errInfo->posix_errno &&
- (attr = GetFileAttributesW(wname)) != INVALID_FILE_ATTRIBUTES &&
- (attr & FILE_ATTRIBUTE_DIRECTORY)) {
- errInfo->posix_errno = EISDIR;
- }
- return 0;
- }
-
- /*
- * Get and return the length of the open file.
- */
-
- if (!GetFileInformationByHandle(fd, &fileInfo))
- return set_error(errInfo);
- *pfd = (int) fd;
- if (pSize) {
- *pSize = (Sint64)
- (((Uint64)fileInfo.nFileSizeHigh << 32) |
- (Uint64)fileInfo.nFileSizeLow);
- }
- return 1;
-}
-
-int
-efile_may_openfile(Efile_error* errInfo, char *name)
-{
- Efile_call_state state;
- WCHAR *wname = (WCHAR *) name;
- DWORD attr;
- int ret;
-
- DBG_TRACE(1, name);
- call_state_init(&state, errInfo);
- ensure_wpath(&state, &wname);
- if ((attr = GetFileAttributesW(wname)) == INVALID_FILE_ATTRIBUTES) {
- errno = ENOENT;
- ret = check_error(-1, errInfo);
- }
- else if (attr & FILE_ATTRIBUTE_DIRECTORY) {
- errno = EISDIR;
- ret = check_error(-1, errInfo);
- }
- else ret = 1;
-
- call_state_free(&state);
- return ret;
-}
-
-void
-efile_closefile(fd)
-int fd; /* File descriptor for file to close. */
-{
- DBG_TRACE(2, L"");
- CloseHandle((HANDLE) fd);
-}
-
-FILE* efile_wfopen(const WCHAR* name, const WCHAR* mode)
-{
- Efile_call_state state;
- Efile_error dummy;
- FILE* f;
- call_state_init(&state, &dummy);
- ensure_wpath(&state, (WCHAR**)&name);
- f = _wfopen(name, mode);
- call_state_free(&state);
- return f;
-}
-
-int
-efile_fdatasync(errInfo, fd)
-Efile_error* errInfo; /* Where to return error codes. */
-int fd; /* File descriptor for file to sync. */
-{
- DBG_TRACE(2, L"");
- /* Not available in Windows, just call regular fsync */
- return efile_fsync(errInfo, fd);
-}
-
-int
-efile_fsync(errInfo, fd)
-Efile_error* errInfo; /* Where to return error codes. */
-int fd; /* File descriptor for file to sync. */
-{
- DBG_TRACE(2, L"");
- if (!FlushFileBuffers((HANDLE) fd)) {
- return check_error(-1, errInfo);
- }
- return 1;
-}
-
-int
-efile_fileinfo(Efile_error* errInfo, Efile_info* pInfo,
- char* orig_name, int info_for_link)
-{
- Efile_call_state state;
- int ret;
- DBG_TRACE(1, L"");
- call_state_init(&state, errInfo);
- ret = do_fileinfo(&state, pInfo, orig_name, info_for_link);
- call_state_free(&state);
- return ret;
-}
-
-static int
-do_fileinfo(Efile_call_state* state, Efile_info* pInfo,
- char* orig_name, int info_for_link)
-{
- Efile_error* errInfo = state->errInfo;
- HANDLE findhandle; /* Handle returned by FindFirstFile(). */
- WIN32_FIND_DATAW findbuf; /* Data return by FindFirstFile(). */
- WCHAR* name = NULL;
- WCHAR* win_path;
- int name_len;
- int drive; /* Drive for filename (1 = A:, 2 = B: etc). */
- WCHAR *worig_name = (WCHAR *) orig_name;
-
- ensure_wpath(state, &worig_name);
- /* Don't allow wildcards to be interpreted by system */
-
-
- /*
- * Move the name to a buffer and make sure to remove a trailing
- * slash, because it causes FindFirstFile() to fail on Win95.
- */
-
- name_len = wcslen(worig_name);
-
- name = wpath_tmp_alloc(state, name_len+1);
- wcscpy(name, worig_name);
- if (name_len > 2 && ISSLASH(name[name_len-1]) &&
- name[name_len-2] != L':') {
- name[name_len-1] = L'\0';
- }
-
- win_path = name;
- if (wcsncmp(name, L"\\\\?\\", 4) == 0) {
- win_path += 4;
- }
-
- if (wcspbrk(win_path, L"?*")) {
- enoent:
- errInfo->posix_errno = ENOENT;
- errInfo->os_errno = ERROR_FILE_NOT_FOUND;
- return 0;
- }
-
- /* Try to get disk from name. If none, get current disk. */
-
- if (win_path[1] != L':') {
- WCHAR* cwd_path = get_cwd_wpath_tmp(state);
- drive = 0;
- if (cwd_path[1] == L':') {
- drive = towlower(cwd_path[0]) - L'a' + 1;
- }
- } else if (*win_path && win_path[2] == L'\0') {
- /*
- * X: and nothing more is an error.
- */
- errInfo->posix_errno = ENOENT;
- errInfo->os_errno = ERROR_FILE_NOT_FOUND;
- return 0;
- } else {
- drive = towlower(*win_path) - L'a' + 1;
- }
-
- findhandle = FindFirstFileW(name, &findbuf);
- if (findhandle == INVALID_HANDLE_VALUE) {
- WCHAR* path = NULL;
-
- if (!(wcspbrk(name, L"./\\") &&
- (path = get_full_wpath_tmp(state, name, NULL, 0)) &&
- /* root dir. ('C:\') or UNC root dir. ('\\server\share\') */
- ((wcslen(path) == 3) || is_root_unc_name(path)) &&
- (GetDriveTypeW(path) > 1) ) ) {
-
- errInfo->posix_errno = ENOENT;
- errInfo->os_errno = ERROR_FILE_NOT_FOUND;
- return 0;
- }
-
- /*
- * Root directories (such as C:\ or \\server\share\ are fabricated.
- */
-
- findbuf.dwFileAttributes = FILE_ATTRIBUTE_DIRECTORY;
- findbuf.nFileSizeHigh = 0;
- findbuf.nFileSizeLow = 0;
- findbuf.cFileName[0] = L'\0';
-
- pInfo->links = 1;
- pInfo->cTime = pInfo->accessTime = pInfo->modifyTime = 0;
- } else {
- SYSTEMTIME SystemTime;
- FILETIME LocalFTime;
-
- /*first check if we are a symlink */
- if (!info_for_link && (findbuf.dwFileAttributes &
- FILE_ATTRIBUTE_REPARSE_POINT)){
- /*
- * given that we know this is a symlink,
- we should be able to find its target */
- WCHAR* target_name = (WCHAR*) do_readlink(state, (char *) name, NULL, 0);
- if (target_name) {
- FindClose(findhandle);
- return do_fileinfo(state, pInfo,
- (char *) target_name, info_for_link);
- }
- }
-
- /* number of links: */
- {
- HANDLE handle; /* Handle returned by CreateFile() */
- BY_HANDLE_FILE_INFORMATION fileInfo; /* from CreateFile() */
-
- /* We initialise nNumberOfLinks as GetFileInformationByHandle
- does not always initialise this field */
- fileInfo.nNumberOfLinks = 1;
- if (handle = CreateFileW(name, GENERIC_READ, FILE_SHARE_FLAGS, NULL,
- OPEN_EXISTING, 0, NULL)) {
- GetFileInformationByHandle(handle, &fileInfo);
- pInfo->links = fileInfo.nNumberOfLinks;
- CloseHandle(handle);
- } else {
- pInfo->links = 1;
- }
- }
-
- FILETIME_TO_EPOCH(pInfo->modifyTime, findbuf.ftLastWriteTime);
-
- if (findbuf.ftLastAccessTime.dwLowDateTime == 0 &&
- findbuf.ftLastAccessTime.dwHighDateTime == 0) {
- pInfo->accessTime = pInfo->modifyTime;
- } else {
- FILETIME_TO_EPOCH(pInfo->accessTime, findbuf.ftLastAccessTime);
- }
-
- if (findbuf.ftCreationTime.dwLowDateTime == 0 &&
- findbuf.ftCreationTime.dwHighDateTime == 0) {
- pInfo->cTime = pInfo->modifyTime;
- } else {
- FILETIME_TO_EPOCH(pInfo->cTime ,findbuf.ftCreationTime);
- }
- FindClose(findhandle);
- }
-
- pInfo->size_low = findbuf.nFileSizeLow;
- pInfo->size_high = findbuf.nFileSizeHigh;
-
- if (info_for_link && (findbuf.dwFileAttributes &
- FILE_ATTRIBUTE_REPARSE_POINT))
- pInfo->type = FT_SYMLINK;
- else if (findbuf.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY)
- pInfo->type = FT_DIRECTORY;
- else
- pInfo->type = FT_REGULAR;
-
- if (findbuf.dwFileAttributes & FILE_ATTRIBUTE_READONLY)
- pInfo->access = FA_READ;
- else
- pInfo->access = FA_READ|FA_WRITE;
-
- pInfo->mode = dos_to_posix_mode(findbuf.dwFileAttributes, name);
- pInfo->major_device = drive;
- pInfo->minor_device = 0;
- pInfo->inode = 0;
- pInfo->uid = 0;
- pInfo->gid = 0;
-
- return 1;
-}
-
-int
-efile_write_info(Efile_error* errInfo,
- Efile_info* pInfo,
- char* name)
-{
- Efile_call_state state;
- int ret;
- call_state_init(&state, errInfo);
- ret = do_write_info(&state, pInfo, name);
- call_state_free(&state);
- return ret;
-}
-
-static int
-do_write_info(Efile_call_state* state,
- Efile_info* pInfo,
- char* name)
-{
- Efile_error* errInfo = state->errInfo;
- SYSTEMTIME timebuf;
- FILETIME ModifyFileTime;
- FILETIME AccessFileTime;
- FILETIME CreationFileTime;
- HANDLE fd;
- DWORD attr;
- DWORD tempAttr;
- WCHAR *wname = (WCHAR *) name;
-
- DBG_TRACE(1, name);
-
- ensure_wpath(state, &wname);
-
- /*
- * Get the attributes for the file.
- */
-
- tempAttr = attr = GetFileAttributesW(wname);
- if (attr == 0xffffffff) {
- return set_error(errInfo);
- }
- if (pInfo->mode != -1) {
- if (pInfo->mode & _S_IWRITE) {
- /* clear read only bit */
- attr &= ~FILE_ATTRIBUTE_READONLY;
- } else {
- /* set read only bit */
- attr |= FILE_ATTRIBUTE_READONLY;
- }
- }
-
- /*
- * Construct all file times.
- */
-
- EPOCH_TO_FILETIME(ModifyFileTime, pInfo->modifyTime);
- EPOCH_TO_FILETIME(AccessFileTime, pInfo->accessTime);
- EPOCH_TO_FILETIME(CreationFileTime, pInfo->cTime);
-
- /*
- * If necessary, set the file times.
- */
-
- /*
- * If the has read only access, we must temporarily turn on
- * write access (this is necessary for native filesystems,
- * but not for NFS filesystems).
- */
-
- if (tempAttr & FILE_ATTRIBUTE_READONLY) {
- tempAttr &= ~FILE_ATTRIBUTE_READONLY;
- if (!SetFileAttributesW(wname, tempAttr)) {
- return set_error(errInfo);
- }
- }
-
- fd = CreateFileW(wname, GENERIC_READ|GENERIC_WRITE,
- FILE_SHARE_FLAGS,
- NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL);
- if (fd != INVALID_HANDLE_VALUE) {
- BOOL result = SetFileTime(fd, &CreationFileTime, &AccessFileTime, &ModifyFileTime);
- if (!result) {
- return set_error(errInfo);
- }
- CloseHandle(fd);
- }
-
- /*
- * If the file doesn't have the correct attributes, set them now.
- * (It could have been done before setting the file times, above).
- */
-
- if (tempAttr != attr) {
- if (!SetFileAttributesW(wname, attr)) {
- return set_error(errInfo);
- }
- }
- return 1;
-}
-
-
-int
-efile_pwrite(errInfo, fd, buf, count, offset)
-Efile_error* errInfo; /* Where to return error codes. */
-int fd; /* File descriptor to write to. */
-char* buf; /* Buffer to write. */
-size_t count; /* Number of bytes to write. */
-Sint64 offset; /* where to write it */
-{
- int res;
- DBG_TRACE(2, L"");
- res = efile_seek(errInfo, fd, offset, EFILE_SEEK_SET, NULL);
- if (res) {
- return efile_write(errInfo, EFILE_MODE_WRITE, fd, buf, count);
- } else {
- return res;
- }
-}
-
-/* position and read/write as a single atomic op */
-int
-efile_pread(errInfo, fd, offset, buf, count, pBytesRead)
-Efile_error* errInfo; /* Where to return error codes. */
-int fd; /* File descriptor to read from. */
-Sint64 offset; /* Offset in bytes from BOF. */
-char* buf; /* Buffer to read into. */
-size_t count; /* Number of bytes to read. */
-size_t* pBytesRead; /* Where to return number of bytes read. */
-{
- int res;
- DBG_TRACE(2, L"");
- res = efile_seek(errInfo, fd, offset, EFILE_SEEK_SET, NULL);
- if (res) {
- return efile_read(errInfo, EFILE_MODE_READ, fd, buf, count, pBytesRead);
- } else {
- return res;
- }
-}
-
-
-
-int
-efile_write(errInfo, flags, fd, buf, count)
-Efile_error* errInfo; /* Where to return error codes. */
-int flags; /* Flags given when file was opened. */
-int fd; /* File descriptor to write to. */
-char* buf; /* Buffer to write. */
-size_t count; /* Number of bytes to write. */
-{
- DWORD written; /* Bytes written in last operation. */
- OVERLAPPED overlapped;
- OVERLAPPED* pOverlapped = NULL;
-
- DBG_TRACE(2, L"");
- if (flags & EFILE_MODE_APPEND) {
- memset(&overlapped, 0, sizeof(overlapped));
- overlapped.Offset = 0xffffffff;
- overlapped.OffsetHigh = 0xffffffff;
- pOverlapped = &overlapped;
- }
- while (count > 0) {
- if (!WriteFile((HANDLE) fd, buf, count, &written, pOverlapped))
- return set_error(errInfo);
- buf += written;
- count -= written;
- }
- return 1;
-}
-
-int
-efile_writev(Efile_error* errInfo, /* Where to return error codes */
- int flags, /* Flags given when file was
- * opened */
- int fd, /* File descriptor to write to */
- SysIOVec* iov, /* Vector of buffer structs.
- * The structs are unchanged
- * after the call */
- int iovcnt) /* Number of structs in vector */
-{
- int cnt; /* Buffers so far written */
- OVERLAPPED overlapped;
- OVERLAPPED* pOverlapped = NULL;
-
- DBG_TRACE(2, L"");
- ASSERT(iovcnt >= 0);
-
- if (flags & EFILE_MODE_APPEND) {
- memset(&overlapped, 0, sizeof(overlapped));
- overlapped.Offset = 0xffffffff;
- overlapped.OffsetHigh = 0xffffffff;
- pOverlapped = &overlapped;
- }
- for (cnt = 0; cnt < iovcnt; cnt++) {
- if (iov[cnt].iov_base && iov[cnt].iov_len > 0) {
- /* Non-empty buffer */
- int p; /* Position in buffer */
- int w = iov[cnt].iov_len;/* Bytes written in this call */
- for (p = 0; p < iov[cnt].iov_len; p += w) {
- if (!WriteFile((HANDLE) fd,
- iov[cnt].iov_base + p,
- iov[cnt].iov_len - p,
- &w,
- pOverlapped))
- return set_error(errInfo);
- }
- }
- }
- return 1;
-}
-
-int
-efile_read(errInfo, flags, fd, buf, count, pBytesRead)
-Efile_error* errInfo; /* Where to return error codes. */
-int flags; /* Flags given when file was opened. */
-int fd; /* File descriptor to read from. */
-char* buf; /* Buffer to read into. */
-size_t count; /* Number of bytes to read. */
-size_t* pBytesRead; /* Where to return number of bytes read. */
-{
- DWORD nbytes = 0;
-
- DBG_TRACE(2, L"");
- if (!ReadFile((HANDLE) fd, buf, count, &nbytes, NULL))
- return set_error(errInfo);
-
- *pBytesRead = nbytes;
- return 1;
-}
-
-int
-efile_seek(errInfo, fd, offset, origin, new_location)
-Efile_error* errInfo; /* Where to return error codes. */
-int fd; /* File descriptor to do the seek on. */
-Sint64 offset; /* Offset in bytes from the given origin. */
-int origin; /* Origin of seek (SEEK_SET, SEEK_CUR,
- * SEEK_END).
- */
-Sint64* new_location; /* Resulting new location in file. */
-{
- LARGE_INTEGER off, new_loc;
-
- DBG_TRACE(2, L"");
- switch (origin) {
- case EFILE_SEEK_SET: origin = FILE_BEGIN; break;
- case EFILE_SEEK_CUR: origin = FILE_CURRENT; break;
- case EFILE_SEEK_END: origin = FILE_END; break;
- default:
- errno = EINVAL;
- check_error(-1, errInfo);
- break;
- }
-
- off.QuadPart = offset;
- if (! SetFilePointerEx((HANDLE) fd, off,
- new_location ? &new_loc : NULL, origin)) {
- return set_error(errInfo);
- }
- if (new_location) {
- *new_location = new_loc.QuadPart;
- DEBUGF(("efile_seek(offset=%ld, origin=%d) -> %ld\n",
- (long) offset, origin, (long) *new_location));
- } else {
- DEBUGF(("efile_seek(offset=%ld, origin=%d)\n", (long) offset, origin));
- }
- return 1;
-}
-
-int
-efile_truncate_file(errInfo, fd, flags)
-Efile_error* errInfo; /* Where to return error codes. */
-int *fd; /* File descriptor for file to truncate. */
-int flags;
-{
- DBG_TRACE(2, L"");
- if (!SetEndOfFile((HANDLE) (*fd)))
- return set_error(errInfo);
- return 1;
-}
-
-
-/*
- * is_root_unc_name - returns TRUE if the argument is a UNC name specifying
- * a root share. That is, if it is of the form \\server\share\.
- * This routine will also return true if the argument is of the
- * form \\server\share (no trailing slash) but Win32 currently
- * does not like that form.
- *
- * Forward slashes ('/') may be used instead of backslashes ('\').
- */
-
-static int
-is_root_unc_name(const WCHAR *path)
-{
- /*
- * If a root UNC name, path will start with 2 (but not 3) slashes
- */
-
- if ((wcslen(path) >= 5) /* minimum string is "//x/y" */
- && ISSLASH(path[0]) && ISSLASH(path[1]))
- {
- const WCHAR *p = path + 2;
-
- /*
- * find the slash between the server name and share name
- */
- while ( * ++ p )
- if ( ISSLASH(*p) )
- break ;
-
- if ( *p && p[1] )
- {
- /*
- * is there a further slash?
- */
- while ( * ++ p )
- if ( ISSLASH(*p) )
- break ;
-
- /*
- * just final slash (or no final slash)
- */
- if ( !*p || !p[1])
- return 1;
- }
- }
-
- return 0 ;
-}
-
-/*
- * Extracts the root part of an absolute filename (by modifying the string
- * pointed to by the name argument). The name can start
- * with either a driver letter (for example, C:\), or a UNC name
- * (for example, \\guinness\bjorn).
- *
- * If the name is invalid, the buffer will be modified to point to
- * an empty string.
- *
- * Returns: 1 if the name consists of just the root part, 0 if
- * the name was longer.
- */
-
-static int
-extract_root(WCHAR* name)
-{
- int len = wcslen(name);
-
- if (iswalpha(name[0]) && name[1] == L':' && ISSLASH(name[2])) {
- WCHAR c = name[3];
- name[3] = L'\0';
- return c == L'\0';
- } else if (len < 5 || !ISSLASH(name[0]) || !ISSLASH(name[1])) {
- goto error;
- } else { /* Try to find the end of the UNC name. */
- WCHAR* p;
- WCHAR c;
-
- /*
- * Find the slash between the server name and share name.
- */
-
- for (p = name + 2; *p; p++)
- if (ISSLASH(*p))
- break;
- if (*p == L'\0')
- goto error;
-
- /*
- * Find the slash after the share name.
- */
-
- for (p++; *p; p++)
- if (ISSLASH(*p))
- break;
- c = *p;
- *p = L'\0';
- return c == L'\0' || p[1] == L'\0';
- }
-
- error:
- *name = L'\0';
- return 1;
-}
-
-static unsigned short
-dos_to_posix_mode(int attr, const WCHAR *name)
-{
- register unsigned short uxmode;
- unsigned dosmode;
- register const WCHAR *p;
-
- dosmode = attr & 0xff;
- if ((p = name)[1] == L':')
- p += 2;
-
- /* check to see if this is a directory - note we must make a special
- * check for the root, which DOS thinks is not a directory
- */
-
- uxmode = (unsigned short)
- (((ISSLASH(*p) && !p[1]) || (dosmode & FILE_ATTRIBUTE_DIRECTORY) ||
- *p == L'\0') ? _S_IFDIR|_S_IEXEC : _S_IFREG);
-
- /* If attribute byte does not have read-only bit, it is read-write */
-
- uxmode |= (dosmode & FILE_ATTRIBUTE_READONLY) ?
- _S_IREAD : (_S_IREAD|_S_IWRITE);
-
- /* see if file appears to be executable - check extension of name */
-
- if (p = wcsrchr(name, L'.')) {
- if (!_wcsicmp(p, L".exe") ||
- !_wcsicmp(p, L".cmd") ||
- !_wcsicmp(p, L".bat") ||
- !_wcsicmp(p, L".com"))
- uxmode |= _S_IEXEC;
- }
-
- /* propagate user read/write/execute bits to group/other fields */
-
- uxmode |= (uxmode & 0700) >> 3;
- uxmode |= (uxmode & 0700) >> 6;
-
- return uxmode;
-}
-
-
-int
-efile_readlink(Efile_error* errInfo, char* name, char* buffer, size_t size)
-{
- Efile_call_state state;
- int ret;
- DBG_TRACE(1, name);
- call_state_init(&state, errInfo);
- ret = !!do_readlink(&state, name, buffer, size);
- call_state_free(&state);
- return ret;
-}
-
-/* If buffer==0, return buffer allocated by wpath_tmp_allocate
-*/
-static char*
-do_readlink(Efile_call_state* state, char* name, char* buffer, size_t size)
-{
- /*
- * load dll and see if we have CreateSymbolicLink at runtime:
- * (Vista only)
- */
- HINSTANCE hModule = NULL;
- WCHAR *wname = (WCHAR *) name;
- WCHAR *wbuffer = (WCHAR *) buffer;
- DWORD wsize = size / sizeof(WCHAR);
- char* ret = NULL;
-
- if ((hModule = LoadLibrary("kernel32.dll")) != NULL) {
- typedef DWORD (WINAPI * GETFINALPATHNAMEBYHANDLEPTR)(
- HANDLE hFile,
- LPCWSTR lpFilePath,
- DWORD cchFilePath,
- DWORD dwFlags);
-
- GETFINALPATHNAMEBYHANDLEPTR pGetFinalPathNameByHandle =
- (GETFINALPATHNAMEBYHANDLEPTR)GetProcAddress(hModule, "GetFinalPathNameByHandleW");
-
- if (pGetFinalPathNameByHandle != NULL) {
- DWORD fileAttributes;
- ensure_wpath(state, &wname);
- /* first check if file is a symlink; {error, einval} otherwise */
- fileAttributes = GetFileAttributesW(wname);
- if ((fileAttributes & FILE_ATTRIBUTE_REPARSE_POINT)) {
- DWORD success = 0;
- HANDLE h = CreateFileW(wname, GENERIC_READ, FILE_SHARE_FLAGS, NULL, OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL);
- int len;
- if(h != INVALID_HANDLE_VALUE) {
- if (!wbuffer) { /* dynamic allocation */
- WCHAR dummy;
- wsize = pGetFinalPathNameByHandle(h, &dummy, 0, 0);
- if (wsize) {
- wbuffer = wpath_tmp_alloc(state, wsize);
- }
- }
- if (wbuffer
- && (success = pGetFinalPathNameByHandle(h, wbuffer, wsize, 0))
- && success < wsize) {
- WCHAR* wp;
-
- /* GetFinalPathNameByHandle prepends path with "\\?\": */
- len = wcslen(wbuffer);
- wmemmove(wbuffer,wbuffer+4,len-3);
- if (len - 4 >= 2 && wbuffer[1] == L':' && wbuffer[0] >= L'A' &&
- wbuffer[0] <= L'Z') {
- wbuffer[0] = wbuffer[0] + L'a' - L'A';
- }
-
- for (wp=wbuffer ; *wp; wp++)
- if (*wp == L'\\')
- *wp = L'/';
- }
- CloseHandle(h);
- }
- if (success) {
- ret = (char*) wbuffer;
- } else {
- set_error(state->errInfo);
- }
- } else {
- errno = EINVAL;
- save_last_error(state->errInfo);
- }
- goto done;
- }
- }
- errno = ENOTSUP;
- save_last_error(state->errInfo);
-
-done:
- if (hModule)
- FreeLibrary(hModule);
- return ret;
-}
-
-
-int
-efile_altname(Efile_error* errInfo, char* orig_name, char* buffer, size_t size)
-{
- Efile_call_state state;
- int ret;
- DBG_TRACE(1, orig_name);
- call_state_init(&state, errInfo);
- ret = do_altname(&state, orig_name, buffer, size);
- call_state_free(&state);
- return ret;
-}
-
-static int
-do_altname(Efile_call_state* state, char* orig_name, char* buffer, size_t size)
-{
- WIN32_FIND_DATAW wfd;
- HANDLE fh;
- WCHAR* name;
- int name_len;
- WCHAR* full_path = NULL;
- WCHAR *worig_name = (WCHAR *) orig_name;
- WCHAR *wbuffer = (WCHAR *) buffer;
- int drive; /* Drive for filename (1 = A:, 2 = B: etc). */
-
- /* Don't allow wildcards to be interpreted by system */
-
- if (wcspbrk(worig_name, L"?*")) {
- enoent:
- state->errInfo->posix_errno = ENOENT;
- state->errInfo->os_errno = ERROR_FILE_NOT_FOUND;
- return 0;
- }
-
- /*
- * Move the name to a buffer and make sure to remove a trailing
- * slash, because it causes FindFirstFile() to fail on Win95.
- */
- ensure_wpath(state, &worig_name);
- name_len = wcslen(worig_name);
-
- name = wpath_tmp_alloc(state, name_len + 1);
- wcscpy(name, worig_name);
- if (name_len > 2 && ISSLASH(name[name_len-1]) &&
- name[name_len-2] != L':') {
- name[name_len-1] = L'\0';
- }
-
- /* Try to get disk from name. If none, get current disk. */
-
- if (name[1] != L':') {
- WCHAR* cwd_path = get_cwd_wpath_tmp(state);
- drive = 0;
- if (cwd_path[1] == L':') {
- drive = towlower(cwd_path[0]) - L'a' + 1;
- }
- } else if (*name && name[2] == L'\0') {
- /*
- * X: and nothing more is an error.
- */
- goto enoent;
- } else {
- drive = towlower(*name) - L'a' + 1;
- }
- fh = FindFirstFileW(name,&wfd);
- if (fh == INVALID_HANDLE_VALUE) {
- DWORD fff_error = GetLastError();
- if (!(wcspbrk(name, L"./\\") &&
- (full_path = get_full_wpath_tmp(state, name, NULL, 0)) &&
- /* root dir. ('C:\') or UNC root dir. ('\\server\share\') */
- ((wcslen(full_path) == 3) || is_root_unc_name(full_path)) &&
- (GetDriveTypeW(full_path) > 1) ) ) {
-
- set_os_errno(state->errInfo, fff_error);
- return 0;
- }
- /*
- * Root directories (such as C:\ or \\server\share\ are fabricated.
- */
- wcscpy(wbuffer,name);
- return 1;
- }
-
- wcscpy(wbuffer,wfd.cAlternateFileName);
- if (!*wbuffer) {
- wcscpy(wbuffer,wfd.cFileName);
- }
- FindClose(fh);
- return 1;
-}
-
-
-int
-efile_link(Efile_error* errInfo, char* old, char* new)
-{
- Efile_call_state state;
- WCHAR *wold = (WCHAR *) old;
- WCHAR *wnew = (WCHAR *) new;
- int ret;
- DBG_TRACE(1, old);
- call_state_init(&state, errInfo);
- ensure_wpath(&state, &wold);
- ensure_wpath(&state, &wnew);
- if(!CreateHardLinkW(wnew, wold, NULL)) {
- ret = set_error(errInfo);
- }
- else ret =1;
- call_state_free(&state);
- return ret;
-}
-
-int
-efile_symlink(Efile_error* errInfo, char* old, char* new)
-{
- Efile_call_state state;
- int ret;
- DBG_TRACE2(1, "symlink(%s <- %s)", old, new);
- call_state_init(&state, errInfo);
- ret = do_symlink(&state, old, new);
- call_state_free(&state);
- return ret;
-}
-
-static int
-do_symlink(Efile_call_state* state, char* old, char* new)
-{
- /*
- * Load dll and see if we have CreateSymbolicLink at runtime:
- * (Vista only)
- */
- HINSTANCE hModule = NULL;
- WCHAR *wold = (WCHAR *) old;
- WCHAR *wnew = (WCHAR *) new;
-
- DBG_TRACE(1, old);
- if ((hModule = LoadLibrary("kernel32.dll")) != NULL) {
- typedef BOOLEAN (WINAPI * CREATESYMBOLICLINKFUNCPTR) (
- LPCWSTR lpSymlinkFileName,
- LPCWSTR lpTargetFileName,
- DWORD dwFlags);
-
- CREATESYMBOLICLINKFUNCPTR pCreateSymbolicLink =
- (CREATESYMBOLICLINKFUNCPTR) GetProcAddress(hModule,
- "CreateSymbolicLinkW");
- /* A for MBCS, W for UNICODE... char* above implies 'W'! */
- if (pCreateSymbolicLink != NULL) {
- ensure_wpath(state, &wold);
- ensure_wpath(state, &wnew);
- {
- DWORD attr = GetFileAttributesW(wold);
- int flag = (attr != INVALID_FILE_ATTRIBUTES &&
- attr & FILE_ATTRIBUTE_DIRECTORY) ? 1 : 0;
- /* SYMBOLIC_LINK_FLAG_DIRECTORY = 1 */
- BOOLEAN success = pCreateSymbolicLink(wnew, wold, flag);
- FreeLibrary(hModule);
-
- if (success) {
- return 1;
- } else {
- return set_error(state->errInfo);
- }
- }
- } else
- FreeLibrary(hModule);
- }
- errno = ENOTSUP;
- return check_error(-1, state->errInfo);
-}
-
-int
-efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset,
- Sint64 length, int advise)
-{
- DBG_TRACE(2, L"");
- /* posix_fadvise is not available on Windows, do nothing */
- errno = ERROR_SUCCESS;
- return check_error(0, errInfo);
-}
-
-int
-efile_fallocate(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length)
-{
- DBG_TRACE(2, L"");
- /* No file preallocation method available in Windows. */
- errno = errno_map(ERROR_NOT_SUPPORTED);
- SetLastError(ERROR_NOT_SUPPORTED);
-
- return check_error(-1, errInfo);
-}
diff --git a/erts/emulator/nifs/common/prim_file_nif.c b/erts/emulator/nifs/common/prim_file_nif.c
new file mode 100644
index 0000000000..6874f41d75
--- /dev/null
+++ b/erts/emulator/nifs/common/prim_file_nif.c
@@ -0,0 +1,1237 @@
+/*
+ * %CopyrightBegin%
+ *
+ * Copyright Ericsson 2017. All Rights Reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * %CopyrightEnd%
+ */
+
+#define STATIC_ERLANG_NIF 1
+
+#include "erl_nif.h"
+#include "config.h"
+#include "sys.h"
+
+#ifdef VALGRIND
+# include <valgrind/memcheck.h>
+#endif
+
+#include "erl_driver.h"
+#include "prim_file_nif.h"
+
+/* NIF interface declarations */
+static int load(ErlNifEnv *env, void** priv_data, ERL_NIF_TERM load_info);
+static int upgrade(ErlNifEnv *env, void** priv_data, void** old_priv_data, ERL_NIF_TERM load_info);
+static void unload(ErlNifEnv *env, void* priv_data);
+
+static ErlNifResourceType *efile_resource_type;
+
+static ERL_NIF_TERM am_ok;
+static ERL_NIF_TERM am_error;
+static ERL_NIF_TERM am_continue;
+
+static ERL_NIF_TERM am_file_info;
+
+/* File modes */
+static ERL_NIF_TERM am_read;
+static ERL_NIF_TERM am_write;
+static ERL_NIF_TERM am_exclusive;
+static ERL_NIF_TERM am_append;
+static ERL_NIF_TERM am_sync;
+static ERL_NIF_TERM am_skip_type_check;
+
+/* enum efile_access_t; read and write are defined above.*/
+static ERL_NIF_TERM am_read_write;
+static ERL_NIF_TERM am_none;
+
+/* enum efile_advise_t */
+static ERL_NIF_TERM am_normal;
+static ERL_NIF_TERM am_random;
+static ERL_NIF_TERM am_sequential;
+static ERL_NIF_TERM am_will_need;
+static ERL_NIF_TERM am_dont_need;
+static ERL_NIF_TERM am_no_reuse;
+
+/* enum efile_filetype_t */
+static ERL_NIF_TERM am_device;
+static ERL_NIF_TERM am_directory;
+static ERL_NIF_TERM am_regular;
+static ERL_NIF_TERM am_symlink;
+static ERL_NIF_TERM am_other;
+
+/* enum efile_seek_t, 'eof' marker. */
+static ERL_NIF_TERM am_bof;
+static ERL_NIF_TERM am_cur;
+static ERL_NIF_TERM am_eof;
+
+static ERL_NIF_TERM read_info_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM set_permissions_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM set_owner_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM set_time_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM read_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM list_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM make_hard_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM make_soft_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM rename_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM make_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM del_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM del_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM get_device_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM get_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM set_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM read_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM get_handle_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM altname_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+static ERL_NIF_TERM open_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]);
+
+/* All file handle operations are passed through a wrapper that handles state
+ * transitions, marking it as busy during the course of the operation, and
+ * closing on completion if the owner died in the middle of an operation.
+ *
+ * This is pretty ugly but required as there's no way to tell when it's safe to
+ * asynchronously close a file; the event could have fired just before landing
+ * in a system call which will fail with EBADF at best or alias a newly opened
+ * fd at worst.
+ *
+ * The old driver got away with enqueueing the close operation on the same
+ * async queue as all of its other operations, but since dirty schedulers use a
+ * single global queue there's no natural way to schedule an asynchronous close
+ * "behind" other operations.
+ *
+ * The states may transition as follows:
+ *
+ * IDLE ->
+ * BUSY (file_handle_wrapper) |
+ * CLOSED (owner_death_callback)
+ *
+ * BUSY ->
+ * IDLE (file_handle_wrapper)
+ * CLOSED (close_nif_impl)
+ * CLOSE_PENDING (owner_death_callback)
+ *
+ * CLOSE_PENDING ->
+ * CLOSED (file_handle_wrapper)
+ */
+
+typedef ERL_NIF_TERM (*file_op_impl_t)(efile_data_t *d, ErlNifEnv *env,
+ int argc, const ERL_NIF_TERM argv[]);
+static ERL_NIF_TERM file_handle_wrapper(file_op_impl_t operation, ErlNifEnv *env,
+ int argc, const ERL_NIF_TERM argv[]);
+
+#define WRAP_FILE_HANDLE_EXPORT(name) \
+ static ERL_NIF_TERM name ## _impl (efile_data_t *d, ErlNifEnv *env, \
+ int argc, const ERL_NIF_TERM argv[]);\
+ static ERL_NIF_TERM name(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) { \
+ return file_handle_wrapper( name ## _impl , env, argc, argv); \
+ }
+
+WRAP_FILE_HANDLE_EXPORT(close_nif)
+WRAP_FILE_HANDLE_EXPORT(read_nif)
+WRAP_FILE_HANDLE_EXPORT(write_nif)
+WRAP_FILE_HANDLE_EXPORT(pread_nif)
+WRAP_FILE_HANDLE_EXPORT(pwrite_nif)
+WRAP_FILE_HANDLE_EXPORT(seek_nif)
+WRAP_FILE_HANDLE_EXPORT(sync_nif)
+WRAP_FILE_HANDLE_EXPORT(truncate_nif)
+WRAP_FILE_HANDLE_EXPORT(allocate_nif)
+WRAP_FILE_HANDLE_EXPORT(advise_nif)
+WRAP_FILE_HANDLE_EXPORT(get_handle_nif)
+WRAP_FILE_HANDLE_EXPORT(ipread_s32bu_p32bu_nif)
+
+static ErlNifFunc nif_funcs[] = {
+ /* File handle ops */
+ {"open_nif", 2, open_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"close_nif", 1, close_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"read_nif", 2, read_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"write_nif", 2, write_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"pread_nif", 3, pread_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"pwrite_nif", 3, pwrite_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"seek_nif", 3, seek_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"sync_nif", 2, sync_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"truncate_nif", 1, truncate_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"allocate_nif", 3, allocate_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"advise_nif", 4, advise_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+
+ /* Filesystem ops */
+ {"make_hard_link_nif", 2, make_hard_link_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"make_soft_link_nif", 2, make_soft_link_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"rename_nif", 2, rename_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"read_info_nif", 2, read_info_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"set_permissions_nif", 2, set_permissions_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"set_owner_nif", 3, set_owner_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"set_time_nif", 4, set_time_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"read_link_nif", 1, read_link_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"list_dir_nif", 1, list_dir_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"make_dir_nif", 1, make_dir_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"del_file_nif", 1, del_file_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"del_dir_nif", 1, del_dir_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"get_device_cwd_nif", 1, get_device_cwd_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"set_cwd_nif", 1, set_cwd_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"get_cwd_nif", 0, get_cwd_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+
+ /* These operations are equivalent to chained calls of other operations,
+ * but have been moved down to avoid excessive rescheduling. */
+ {"ipread_s32bu_p32bu_nif", 3, ipread_s32bu_p32bu_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+ {"read_file_nif", 1, read_file_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+
+ /* Internal ops. */
+ {"get_handle_nif", 1, get_handle_nif},
+ {"altname_nif", 1, altname_nif, ERL_NIF_DIRTY_JOB_IO_BOUND},
+};
+
+ERL_NIF_INIT(prim_file, nif_funcs, load, NULL, upgrade, unload)
+
+static void owner_death_callback(ErlNifEnv* env, void* obj, ErlNifPid* pid, ErlNifMonitor* mon);
+static void gc_callback(ErlNifEnv *env, void* data);
+
+static int load(ErlNifEnv *env, void** priv_data, ERL_NIF_TERM load_info)
+{
+ ErlNifResourceTypeInit callbacks;
+
+ am_ok = enif_make_atom(env, "ok");
+ am_error = enif_make_atom(env, "error");
+ am_continue = enif_make_atom(env, "continue");
+
+ am_read = enif_make_atom(env, "read");
+ am_write = enif_make_atom(env, "write");
+ am_exclusive = enif_make_atom(env, "exclusive");
+ am_append = enif_make_atom(env, "append");
+ am_sync = enif_make_atom(env, "sync");
+ am_skip_type_check = enif_make_atom(env, "skip_type_check");
+
+ am_read_write = enif_make_atom(env, "read_write");
+ am_none = enif_make_atom(env, "none");
+
+ am_normal = enif_make_atom(env, "normal");
+ am_random = enif_make_atom(env, "random");
+ am_sequential = enif_make_atom(env, "sequential");
+ am_will_need = enif_make_atom(env, "will_need");
+ am_dont_need = enif_make_atom(env, "dont_need");
+ am_no_reuse = enif_make_atom(env, "no_reuse");
+
+ am_device = enif_make_atom(env, "device");
+ am_directory = enif_make_atom(env, "directory");
+ am_regular = enif_make_atom(env, "regular");
+ am_symlink = enif_make_atom(env, "symlink");
+ am_other = enif_make_atom(env, "other");
+
+ am_file_info = enif_make_atom(env, "file_info");
+
+ am_bof = enif_make_atom(env, "bof");
+ am_cur = enif_make_atom(env, "cur");
+ am_eof = enif_make_atom(env, "eof");
+
+ callbacks.down = owner_death_callback;
+ callbacks.dtor = gc_callback;
+ callbacks.stop = NULL;
+
+ efile_resource_type = enif_open_resource_type_x(env, "efile", &callbacks,
+ ERL_NIF_RT_CREATE, NULL);
+
+ *priv_data = NULL;
+
+ return 0;
+}
+
+static void unload(ErlNifEnv *env, void* priv_data)
+{
+
+}
+
+static int upgrade(ErlNifEnv *env, void** priv_data, void** old_priv_data, ERL_NIF_TERM load_info)
+{
+ if(*old_priv_data != NULL) {
+ return -1; /* Don't know how to do that */
+ }
+ if(*priv_data != NULL) {
+ return -1; /* Don't know how to do that */
+ }
+ if(load(env, priv_data, load_info)) {
+ return -1;
+ }
+ return 0;
+}
+
+static ERL_NIF_TERM posix_error_to_tuple(ErlNifEnv *env, posix_errno_t posix_errno) {
+ ERL_NIF_TERM error = enif_make_atom(env, erl_errno_id(posix_errno));
+ return enif_make_tuple2(env, am_error, error);
+}
+
+static int get_file_data(ErlNifEnv *env, ERL_NIF_TERM opaque, efile_data_t **d) {
+ return enif_get_resource(env, opaque, efile_resource_type, (void **)d);
+}
+
+static ERL_NIF_TERM file_handle_wrapper(file_op_impl_t operation, ErlNifEnv *env,
+ int argc, const ERL_NIF_TERM argv[]) {
+
+ efile_data_t *d;
+
+ enum efile_state_t previous_state;
+ ERL_NIF_TERM result;
+
+ if(argc < 1 || !get_file_data(env, argv[0], &d)) {
+ return enif_make_badarg(env);
+ }
+
+ previous_state = erts_atomic32_cmpxchg_acqb(&d->state,
+ EFILE_STATE_BUSY, EFILE_STATE_IDLE);
+
+ if(previous_state == EFILE_STATE_IDLE) {
+ result = operation(d, env, argc - 1, &argv[1]);
+
+ previous_state = erts_atomic32_cmpxchg_relb(&d->state,
+ EFILE_STATE_IDLE, EFILE_STATE_BUSY);
+
+ ASSERT(previous_state != EFILE_STATE_IDLE);
+
+ if(previous_state == EFILE_STATE_CLOSE_PENDING) {
+ /* This is the only point where a change from CLOSE_PENDING is
+ * possible, and we're running synchronously, so we can't race with
+ * anything else here. */
+ erts_atomic32_set_acqb(&d->state, EFILE_STATE_CLOSED);
+ efile_close(d);
+ }
+ } else {
+ /* CLOSE_PENDING should be impossible at this point since it requires
+ * a transition from BUSY; the only valid state here is CLOSED. */
+ ASSERT(previous_state == EFILE_STATE_CLOSED);
+
+ result = posix_error_to_tuple(env, EINVAL);
+ }
+
+ return result;
+}
+
+static void owner_death_callback(ErlNifEnv* env, void* obj, ErlNifPid* pid, ErlNifMonitor* mon) {
+ efile_data_t *d = (efile_data_t*)obj;
+
+ (void)env;
+ (void)pid;
+ (void)mon;
+
+ for(;;) {
+ enum efile_state_t previous_state;
+
+ previous_state = erts_atomic32_cmpxchg_acqb(&d->state,
+ EFILE_STATE_CLOSED, EFILE_STATE_IDLE);
+
+ switch(previous_state) {
+ case EFILE_STATE_IDLE:
+ efile_close(d);
+ return;
+ case EFILE_STATE_CLOSE_PENDING:
+ case EFILE_STATE_CLOSED:
+ /* We're either already closed or managed to mark ourselves for
+ * closure in the previous iteration. */
+ return;
+ case EFILE_STATE_BUSY:
+ /* Schedule ourselves to be closed once the current operation
+ * finishes, retrying the [IDLE -> CLOSED] transition in case we
+ * narrowly passed the [BUSY -> IDLE] one. */
+ erts_atomic32_cmpxchg_nob(&d->state,
+ EFILE_STATE_CLOSE_PENDING, EFILE_STATE_BUSY);
+ break;
+ }
+ }
+}
+
+static void gc_callback(ErlNifEnv *env, void* data) {
+ efile_data_t *d = (efile_data_t*)data;
+
+ enum efile_state_t previous_state;
+
+ (void)env;
+
+ previous_state = erts_atomic32_cmpxchg_acqb(&d->state,
+ EFILE_STATE_CLOSED, EFILE_STATE_IDLE);
+
+ ASSERT(previous_state != EFILE_STATE_CLOSE_PENDING &&
+ previous_state != EFILE_STATE_BUSY);
+
+ if(previous_state == EFILE_STATE_IDLE) {
+ efile_close(d);
+ }
+}
+
+static ERL_NIF_TERM efile_filetype_to_atom(enum efile_filetype_t type) {
+ switch(type) {
+ case EFILE_FILETYPE_DEVICE: return am_device;
+ case EFILE_FILETYPE_DIRECTORY: return am_directory;
+ case EFILE_FILETYPE_REGULAR: return am_regular;
+ case EFILE_FILETYPE_SYMLINK: return am_symlink;
+ case EFILE_FILETYPE_OTHER: return am_other;
+ }
+
+ return am_other;
+}
+
+static ERL_NIF_TERM efile_access_to_atom(enum efile_access_t type) {
+ if(type & EFILE_ACCESS_READ && !(type & EFILE_ACCESS_WRITE)) {
+ return am_read;
+ } else if(type & EFILE_ACCESS_WRITE && !(type & EFILE_ACCESS_READ)) {
+ return am_write;
+ } else if(type & EFILE_ACCESS_READ_WRITE) {
+ return am_read_write;
+ }
+
+ return am_none;
+}
+
+static enum efile_modes_t efile_translate_modelist(ErlNifEnv *env, ERL_NIF_TERM list) {
+ enum efile_modes_t modes;
+ ERL_NIF_TERM head, tail;
+
+ modes = 0;
+
+ while(enif_get_list_cell(env, list, &head, &tail)) {
+ if(enif_is_identical(head, am_read)) {
+ modes |= EFILE_MODE_READ;
+ } else if(enif_is_identical(head, am_write)) {
+ modes |= EFILE_MODE_WRITE;
+ } else if(enif_is_identical(head, am_exclusive)) {
+ modes |= EFILE_MODE_EXCLUSIVE;
+ } else if(enif_is_identical(head, am_append)) {
+ modes |= EFILE_MODE_APPEND;
+ } else if(enif_is_identical(head, am_sync)) {
+ modes |= EFILE_MODE_SYNC;
+ } else if(enif_is_identical(head, am_skip_type_check)) {
+ modes |= EFILE_MODE_SKIP_TYPE_CHECK;
+ } else {
+ /* Modes like 'raw', 'ram', 'delayed_writes' etc are handled
+ * further up the chain. */
+ }
+
+ list = tail;
+ }
+
+ if(modes & (EFILE_MODE_APPEND | EFILE_MODE_EXCLUSIVE)) {
+ /* 'append' and 'exclusive' are documented as "open for writing." */
+ modes |= EFILE_MODE_WRITE;
+ } else if(!(modes & EFILE_MODE_READ_WRITE)) {
+ /* Defaulting to read if !(W|R) is undocumented, but specifically
+ * tested against in file_SUITE. */
+ modes |= EFILE_MODE_READ;
+ }
+
+ return modes;
+}
+
+static ERL_NIF_TERM open_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+ efile_data_t *d;
+
+ ErlNifPid controlling_process;
+ enum efile_modes_t modes;
+ ERL_NIF_TERM result;
+ efile_path_t path;
+
+ if(argc != 2 || !enif_is_list(env, argv[1])) {
+ return enif_make_badarg(env);
+ }
+
+ modes = efile_translate_modelist(env, argv[1]);
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_open(&path, modes, efile_resource_type, &d))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ result = enif_make_resource(env, d);
+ enif_release_resource(d);
+
+ enif_self(env, &controlling_process);
+
+ if(enif_monitor_process(env, d, &controlling_process, &d->monitor)) {
+ return posix_error_to_tuple(env, EINVAL);
+ }
+
+ return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM close_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ enum efile_state_t previous_state;
+
+ if(argc != 0) {
+ return enif_make_badarg(env);
+ }
+
+ previous_state = erts_atomic32_cmpxchg_acqb(&d->state,
+ EFILE_STATE_CLOSED, EFILE_STATE_BUSY);
+
+ ASSERT(previous_state == EFILE_STATE_CLOSE_PENDING ||
+ previous_state == EFILE_STATE_BUSY);
+
+ if(previous_state == EFILE_STATE_BUSY) {
+ enif_demonitor_process(env, d, &d->monitor);
+
+ if(!efile_close(d)) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ }
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM read_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ Sint64 bytes_read, block_size;
+ SysIOVec read_vec[1];
+ ErlNifBinary result;
+
+ if(argc != 1 || !enif_is_number(env, argv[0])) {
+ return enif_make_badarg(env);
+ }
+
+ if(!enif_get_int64(env, argv[0], &block_size) || block_size < 0) {
+ return posix_error_to_tuple(env, EINVAL);
+ }
+
+ if(!enif_alloc_binary(block_size, &result)) {
+ return posix_error_to_tuple(env, ENOMEM);
+ }
+
+ read_vec[0].iov_base = result.data;
+ read_vec[0].iov_len = result.size;
+
+ bytes_read = efile_readv(d, read_vec, 1);
+ ASSERT(bytes_read <= block_size);
+
+ if(bytes_read < 0) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ } else if(bytes_read == 0) {
+ enif_release_binary(&result);
+ return am_eof;
+ }
+
+ if(bytes_read < block_size && !enif_realloc_binary(&result, bytes_read)) {
+ ERTS_INTERNAL_ERROR("Failed to shrink read result.");
+ }
+
+ return enif_make_tuple2(env, am_ok, enif_make_binary(env, &result));
+}
+
+static ERL_NIF_TERM write_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ ErlNifIOVec vec, *input = &vec;
+ Sint64 bytes_written;
+ ERL_NIF_TERM tail;
+
+ if(argc != 1 || !enif_inspect_iovec(env, 64, argv[0], &tail, &input)) {
+ return enif_make_badarg(env);
+ }
+
+ bytes_written = efile_writev(d, input->iov, input->iovcnt);
+
+ if(bytes_written < 0) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ }
+
+ if(!enif_is_empty_list(env, tail)) {
+ ASSERT(bytes_written > 0);
+ return enif_make_tuple2(env, am_continue, tail);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM pread_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ Sint64 bytes_read, block_size, offset;
+ SysIOVec read_vec[1];
+ ErlNifBinary result;
+
+ if(argc != 2 || !enif_is_number(env, argv[0])
+ || !enif_is_number(env, argv[1])) {
+ return enif_make_badarg(env);
+ }
+
+ if(!enif_get_int64(env, argv[0], &offset) ||
+ !enif_get_int64(env, argv[1], &block_size) ||
+ (offset < 0 || block_size < 0)) {
+ return posix_error_to_tuple(env, EINVAL);
+ }
+
+ if(!enif_alloc_binary(block_size, &result)) {
+ return posix_error_to_tuple(env, ENOMEM);
+ }
+
+ read_vec[0].iov_base = result.data;
+ read_vec[0].iov_len = result.size;
+
+ bytes_read = efile_preadv(d, offset, read_vec, 1);
+
+ if(bytes_read < 0) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ } else if(bytes_read == 0) {
+ enif_release_binary(&result);
+ return am_eof;
+ }
+
+ if(bytes_read < block_size && !enif_realloc_binary(&result, bytes_read)) {
+ ERTS_INTERNAL_ERROR("Failed to shrink pread result.");
+ }
+
+ return enif_make_tuple2(env, am_ok, enif_make_binary(env, &result));
+}
+
+static ERL_NIF_TERM pwrite_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ ErlNifIOVec vec, *input = &vec;
+ Sint64 bytes_written, offset;
+ ERL_NIF_TERM tail;
+
+ if(argc != 2 || !enif_is_number(env, argv[0])
+ || !enif_inspect_iovec(env, 64, argv[1], &tail, &input)) {
+ return enif_make_badarg(env);
+ }
+
+ if(!enif_get_int64(env, argv[0], &offset) || offset < 0) {
+ return posix_error_to_tuple(env, EINVAL);
+ }
+
+ bytes_written = efile_pwritev(d, offset, input->iov, input->iovcnt);
+
+ if(bytes_written < 0) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ }
+
+ if(!enif_is_empty_list(env, tail)) {
+ ASSERT(bytes_written > 0);
+ return enif_make_tuple3(env, am_continue,
+ enif_make_int64(env, bytes_written), tail);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM seek_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ Sint64 new_position, offset;
+ enum efile_seek_t seek;
+
+ if(argc != 2 || !enif_get_int64(env, argv[1], &offset)) {
+ return enif_make_badarg(env);
+ }
+
+ if(enif_is_identical(argv[0], am_bof)) {
+ seek = EFILE_SEEK_BOF;
+ } else if(enif_is_identical(argv[0], am_cur)) {
+ seek = EFILE_SEEK_CUR;
+ } else if(enif_is_identical(argv[0], am_eof)) {
+ seek = EFILE_SEEK_EOF;
+ } else {
+ return enif_make_badarg(env);
+ }
+
+ if(!efile_seek(d, seek, offset, &new_position)) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ }
+
+ return enif_make_tuple2(env, am_ok, enif_make_uint64(env, new_position));
+}
+
+static ERL_NIF_TERM sync_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ int data_only;
+
+ if(argc != 1 || !enif_get_int(env, argv[0], &data_only)) {
+ return enif_make_badarg(env);
+ }
+
+ if(!efile_sync(d, data_only)) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM truncate_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ if(argc != 0) {
+ return enif_make_badarg(env);
+ }
+
+ if(!efile_truncate(d)) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM allocate_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ Sint64 offset, length;
+
+ if(argc != 2 || !enif_is_number(env, argv[0])
+ || !enif_is_number(env, argv[1])) {
+ return enif_make_badarg(env);
+ }
+
+ if(!enif_get_int64(env, argv[0], &offset) ||
+ !enif_get_int64(env, argv[1], &length) ||
+ (offset < 0 || length < 0)) {
+ return posix_error_to_tuple(env, EINVAL);
+ }
+
+ if(!efile_allocate(d, offset, length)) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM advise_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ enum efile_advise_t advise;
+ Sint64 offset, length;
+
+ if(argc != 3 || !enif_is_number(env, argv[0])
+ || !enif_is_number(env, argv[1])) {
+ return enif_make_badarg(env);
+ }
+
+ if(!enif_get_int64(env, argv[0], &offset) ||
+ !enif_get_int64(env, argv[1], &length) ||
+ (offset < 0 || length < 0)) {
+ return posix_error_to_tuple(env, EINVAL);
+ }
+
+ if(enif_is_identical(argv[2], am_normal)) {
+ advise = EFILE_ADVISE_NORMAL;
+ } else if(enif_is_identical(argv[2], am_random)) {
+ advise = EFILE_ADVISE_RANDOM;
+ } else if(enif_is_identical(argv[2], am_sequential)) {
+ advise = EFILE_ADVISE_SEQUENTIAL;
+ } else if(enif_is_identical(argv[2], am_will_need)) {
+ advise = EFILE_ADVISE_WILL_NEED;
+ } else if(enif_is_identical(argv[2], am_dont_need)) {
+ advise = EFILE_ADVISE_DONT_NEED;
+ } else if(enif_is_identical(argv[2], am_no_reuse)) {
+ advise = EFILE_ADVISE_NO_REUSE;
+ } else {
+ /* The tests check for EINVAL instead of badarg. Sigh. */
+ return posix_error_to_tuple(env, EINVAL);
+ }
+
+ if(!efile_advise(d, offset, length, advise)) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ }
+
+ return am_ok;
+}
+
+/* This undocumented function reads a pointer and then reads the data block
+ * described by said pointer. It was reverse-engineered from the old
+ * implementation so while all tests pass it may not be entirely correct. Our
+ * current understanding is as follows:
+ *
+ * Pointer layout:
+ *
+ * <<Size:1/integer-unit:32, Offset:1/integer-unit:32>>
+ *
+ * Where Offset is the -absolute- address to the data block.
+ *
+ * *) If we fail to read the pointer block in its entirety, we return eof.
+ * *) If the provided max_payload_size is larger than Size, we return eof.
+ * *) If we fail to read any data whatsoever at Offset, we return
+ * {ok, {Size, Offset, eof}}
+ * *) Otherwise, we return {ok, {Size, Offset, Data}}. Note that the size
+ * of Data may be smaller than Size if we encounter EOF before we could
+ * read the entire block.
+ *
+ * On errors we'll return {error, posix()} regardless of whether they
+ * happened before or after reading the pointer block. */
+static ERL_NIF_TERM ipread_s32bu_p32bu_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ Sint64 payload_offset, payload_size;
+
+ SysIOVec read_vec[1];
+ Sint64 bytes_read;
+
+ ErlNifBinary payload;
+
+ if(argc != 2 || !enif_is_number(env, argv[0])
+ || !enif_is_number(env, argv[1])) {
+ return enif_make_badarg(env);
+ }
+
+ {
+ Sint64 max_payload_size, pointer_offset;
+ unsigned char pointer_block[8];
+
+ if(!enif_get_int64(env, argv[0], &pointer_offset) ||
+ !enif_get_int64(env, argv[1], &max_payload_size) ||
+ (pointer_offset < 0 || max_payload_size >= 1u << 31)) {
+ return posix_error_to_tuple(env, EINVAL);
+ }
+
+ read_vec[0].iov_base = pointer_block;
+ read_vec[0].iov_len = sizeof(pointer_block);
+
+ bytes_read = efile_preadv(d, pointer_offset, read_vec, 1);
+
+ if(bytes_read < 0) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ } else if(bytes_read < sizeof(pointer_block)) {
+ return am_eof;
+ }
+
+ payload_size = (Uint32)get_int32(&pointer_block[0]);
+ payload_offset = (Uint32)get_int32(&pointer_block[4]);
+
+ if(payload_size > max_payload_size) {
+ return am_eof;
+ }
+ }
+
+ if(!enif_alloc_binary(payload_size, &payload)) {
+ return posix_error_to_tuple(env, ENOMEM);
+ }
+
+ read_vec[0].iov_base = payload.data;
+ read_vec[0].iov_len = payload.size;
+
+ bytes_read = efile_preadv(d, payload_offset, read_vec, 1);
+
+ if(bytes_read < 0) {
+ return posix_error_to_tuple(env, d->posix_errno);
+ } else if(bytes_read == 0) {
+ enif_release_binary(&payload);
+
+ return enif_make_tuple2(env, am_ok,
+ enif_make_tuple3(env,
+ enif_make_uint(env, payload_size),
+ enif_make_uint(env, payload_offset),
+ am_eof));
+ }
+
+ if(bytes_read < payload.size && !enif_realloc_binary(&payload, bytes_read)) {
+ ERTS_INTERNAL_ERROR("Failed to shrink ipread payload.");
+ }
+
+ return enif_make_tuple2(env, am_ok,
+ enif_make_tuple3(env,
+ enif_make_uint(env, payload_size),
+ enif_make_uint(env, payload_offset),
+ enif_make_binary(env, &payload)));
+}
+
+static ERL_NIF_TERM get_handle_nif_impl(efile_data_t *d, ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ if(argc != 0) {
+ return enif_make_badarg(env);
+ }
+
+ return efile_get_handle(env, d);
+}
+
+static ERL_NIF_TERM read_info_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_fileinfo_t info = {0};
+ efile_path_t path;
+ int follow_links;
+
+ if(argc != 2 || !enif_get_int(env, argv[1], &follow_links)) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_read_info(&path, follow_links, &info))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ /* #file_info as declared in file.hrl */
+ return enif_make_tuple(env, 14,
+ am_file_info,
+ enif_make_uint64(env, info.size),
+ efile_filetype_to_atom(info.type),
+ efile_access_to_atom(info.access),
+ enif_make_int64(env, MAX(EFILE_MIN_FILETIME, info.a_time)),
+ enif_make_int64(env, MAX(EFILE_MIN_FILETIME, info.m_time)),
+ enif_make_int64(env, MAX(EFILE_MIN_FILETIME, info.c_time)),
+ enif_make_uint(env, info.mode),
+ enif_make_uint(env, info.links),
+ enif_make_uint(env, info.major_device),
+ enif_make_uint(env, info.minor_device),
+ enif_make_uint(env, info.inode),
+ enif_make_uint(env, info.uid),
+ enif_make_uint(env, info.gid)
+ );
+}
+
+static ERL_NIF_TERM set_permissions_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t path;
+ Uint32 permissions;
+
+ if(argc != 2 || !enif_get_uint(env, argv[1], &permissions)) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_set_permissions(&path, permissions))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM set_owner_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t path;
+ Uint32 uid, gid;
+
+ if(argc != 3 || !enif_get_uint(env, argv[1], &uid)
+ || !enif_get_uint(env, argv[2], &gid)) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_set_owner(&path, uid, gid))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM set_time_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ Sint64 accessed, modified, created;
+ efile_path_t path;
+
+ if(argc != 4 || !enif_get_int64(env, argv[1], &accessed)
+ || !enif_get_int64(env, argv[2], &modified)
+ || !enif_get_int64(env, argv[3], &created)) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_set_time(&path, accessed, modified, created))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM read_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t path;
+ ERL_NIF_TERM result;
+
+ if(argc != 1) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_read_link(env, &path, &result))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM list_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t path;
+ ERL_NIF_TERM result;
+
+ if(argc != 1) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_list_dir(env, &path, &result))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM rename_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t existing_path, new_path;
+
+ if(argc != 2) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &existing_path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_marshal_path(env, argv[1], &new_path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_rename(&existing_path, &new_path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM make_hard_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t existing_path, new_path;
+
+ if(argc != 2) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &existing_path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_marshal_path(env, argv[1], &new_path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_make_hard_link(&existing_path, &new_path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM make_soft_link_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t existing_path, new_path;
+
+ if(argc != 2) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &existing_path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_marshal_path(env, argv[1], &new_path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_make_soft_link(&existing_path, &new_path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM make_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t path;
+
+ if(argc != 1) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_make_dir(&path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM del_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t path;
+
+ if(argc != 1) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_del_file(&path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM del_dir_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t path;
+
+ if(argc != 1) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_del_dir(&path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+static ERL_NIF_TERM get_device_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ ERL_NIF_TERM result;
+ int device_index;
+
+ if(argc != 1 || !enif_get_int(env, argv[0], &device_index)) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_get_device_cwd(env, device_index, &result))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM get_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+ ERL_NIF_TERM result;
+
+ if(argc != 0) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_get_cwd(env, &result))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return enif_make_tuple2(env, am_ok, result);
+}
+
+static ERL_NIF_TERM set_cwd_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t path;
+
+ if(argc != 1) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_set_cwd(&path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return am_ok;
+}
+
+/** @brief Reads an entire file into \c result, stopping after \c size bytes or
+ * EOF. It will read until EOF if size is 0. */
+static posix_errno_t read_file(efile_data_t *d, size_t size, ErlNifBinary *result) {
+ size_t initial_buffer_size;
+ ssize_t bytes_read;
+
+ if(size == 0) {
+ initial_buffer_size = 16 << 10;
+ } else {
+ initial_buffer_size = size;
+ }
+
+ if(!enif_alloc_binary(initial_buffer_size, result)) {
+ return ENOMEM;
+ }
+
+ bytes_read = 0;
+
+ for(;;) {
+ ssize_t block_bytes_read;
+ SysIOVec read_vec[1];
+
+ read_vec[0].iov_base = result->data + bytes_read;
+ read_vec[0].iov_len = result->size - bytes_read;
+
+ block_bytes_read = efile_readv(d, read_vec, 1);
+
+ if(block_bytes_read < 0) {
+ enif_release_binary(result);
+ return d->posix_errno;
+ }
+
+ bytes_read += block_bytes_read;
+
+ if(block_bytes_read < (result->size - bytes_read)) {
+ /* EOF */
+ break;
+ } else if(bytes_read == size) {
+ break;
+ }
+
+ if(!enif_realloc_binary(result, bytes_read * 2)) {
+ enif_release_binary(result);
+ return ENOMEM;
+ }
+ }
+
+ /* The file may have shrunk since we queried its size, so we have to do
+ * this even when the size is known. */
+ if(bytes_read < result->size && !enif_realloc_binary(result, bytes_read)) {
+ ERTS_INTERNAL_ERROR("Failed to shrink read_file result.");
+ }
+
+ return 0;
+}
+
+static ERL_NIF_TERM read_file_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_fileinfo_t info = {0};
+ efile_path_t path;
+ efile_data_t *d;
+
+ ErlNifBinary result;
+
+ if(argc != 1) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_read_info(&path, 1, &info))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_open(&path, EFILE_MODE_READ, efile_resource_type, &d))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ posix_errno = read_file(d, info.size, &result);
+ enif_release_resource(d);
+
+ if(posix_errno) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return enif_make_tuple2(env, am_ok, enif_make_binary(env, &result));
+}
+
+static ERL_NIF_TERM altname_nif(ErlNifEnv *env, int argc, const ERL_NIF_TERM argv[]) {
+ posix_errno_t posix_errno;
+
+ efile_path_t path;
+ ERL_NIF_TERM result;
+
+ if(argc != 1) {
+ return enif_make_badarg(env);
+ }
+
+ if((posix_errno = efile_marshal_path(env, argv[0], &path))) {
+ return posix_error_to_tuple(env, posix_errno);
+ } else if((posix_errno = efile_altname(env, &path, &result))) {
+ return posix_error_to_tuple(env, posix_errno);
+ }
+
+ return enif_make_tuple2(env, am_ok, result);
+}
diff --git a/erts/emulator/nifs/common/prim_file_nif.h b/erts/emulator/nifs/common/prim_file_nif.h
new file mode 100644
index 0000000000..cc9bc8f5c3
--- /dev/null
+++ b/erts/emulator/nifs/common/prim_file_nif.h
@@ -0,0 +1,240 @@
+/*
+ * %CopyrightBegin%
+ *
+ * Copyright Ericsson 2017. All Rights Reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * %CopyrightEnd%
+ */
+
+typedef int posix_errno_t;
+
+enum efile_modes_t {
+ EFILE_MODE_READ = (1 << 0),
+ EFILE_MODE_WRITE = (1 << 1), /* Implies truncating file when used alone. */
+ EFILE_MODE_APPEND = (1 << 2),
+ EFILE_MODE_EXCLUSIVE = (1 << 3),
+ EFILE_MODE_SYNC = (1 << 4),
+
+ EFILE_MODE_SKIP_TYPE_CHECK = (1 << 5), /* Special for device files on Unix. */
+ EFILE_MODE_NO_TRUNCATE = (1 << 6), /* Special for reopening on VxWorks. */
+
+ EFILE_MODE_READ_WRITE = EFILE_MODE_READ | EFILE_MODE_WRITE
+};
+
+enum efile_access_t {
+ EFILE_ACCESS_NONE = 0,
+ EFILE_ACCESS_READ = 1,
+ EFILE_ACCESS_WRITE = 2,
+ EFILE_ACCESS_READ_WRITE = EFILE_ACCESS_READ | EFILE_ACCESS_WRITE
+};
+
+enum efile_seek_t {
+ EFILE_SEEK_BOF,
+ EFILE_SEEK_CUR,
+ EFILE_SEEK_EOF
+};
+
+enum efile_filetype_t {
+ EFILE_FILETYPE_DEVICE,
+ EFILE_FILETYPE_DIRECTORY,
+ EFILE_FILETYPE_REGULAR,
+ EFILE_FILETYPE_SYMLINK,
+ EFILE_FILETYPE_OTHER
+};
+
+enum efile_advise_t {
+ EFILE_ADVISE_NORMAL,
+ EFILE_ADVISE_RANDOM,
+ EFILE_ADVISE_SEQUENTIAL,
+ EFILE_ADVISE_WILL_NEED,
+ EFILE_ADVISE_DONT_NEED,
+ EFILE_ADVISE_NO_REUSE
+};
+
+enum efile_state_t {
+ EFILE_STATE_IDLE = 0,
+ EFILE_STATE_BUSY = 1,
+ EFILE_STATE_CLOSE_PENDING = 2,
+ EFILE_STATE_CLOSED = 3
+};
+
+typedef struct {
+ Sint64 size; /* Size of file */
+ Uint32 type; /* Type of file -- one of EFILE_FILETYPE_*. */
+ Uint32 access; /* Access to file -- one of EFILE_ACCESS_*. */
+ Uint32 mode; /* Access permissions -- bit field. */
+ Uint32 links; /* Number of links to file. */
+ Uint32 major_device; /* Major device or file system. */
+ Uint32 minor_device; /* Minor device (for devices). */
+ Uint32 inode; /* Inode number. */
+ Uint32 uid; /* User id of owner. */
+ Uint32 gid; /* Group id of owner. */
+ Sint64 a_time; /* Last time the file was accessed. */
+ Sint64 m_time; /* Last time the file was modified. */
+ Sint64 c_time; /* Windows: creation time, Unix: last inode
+ * change. */
+} efile_fileinfo_t;
+
+/* The smallest value that can be converted freely between universal, local,
+ * and POSIX time, as required by read_file_info/2. Corresponds to
+ * {{1902,1,1},{0,0,0}} */
+#define EFILE_MIN_FILETIME -2145916800
+
+/* Initializes an efile_data_t; must be used in efile_open on success. */
+#define EFILE_INIT_RESOURCE(__d, __modes) do { \
+ erts_atomic32_init_acqb(&(__d)->state, EFILE_STATE_IDLE); \
+ (__d)->posix_errno = 0; \
+ (__d)->modes = __modes; \
+ } while(0)
+
+typedef struct {
+ erts_atomic32_t state;
+
+ posix_errno_t posix_errno;
+ enum efile_modes_t modes;
+
+ ErlNifMonitor monitor;
+} efile_data_t;
+
+typedef ErlNifBinary efile_path_t;
+
+/* @brief Translates the given "raw name" into the format expected by the APIs
+ * used by the underlying implementation. The result is transient and does not
+ * need to be released.
+ *
+ * This may change the structure of the path and its results should never be
+ * passed on to the user. Refer to the OS-specific implementation for details.
+ *
+ * @param path The term to translate; it must have been encoded with
+ * prim_file:internal_native2name for compatibility reasons. */
+posix_errno_t efile_marshal_path(ErlNifEnv *env, ERL_NIF_TERM path, efile_path_t *result);
+
+/* @brief Returns the underlying handle as an implementation-defined term.
+ *
+ * This is an internal function intended to support tests and tricky
+ * operations like sendfile(2). */
+ERL_NIF_TERM efile_get_handle(ErlNifEnv *env, efile_data_t *d);
+
+/* @brief Read until EOF or the given iovec has been filled.
+ *
+ * @return -1 on failure, or the number of bytes read on success. The return
+ * value will be 0 if no bytes could be read before EOF or the end of the
+ * iovec. */
+Sint64 efile_readv(efile_data_t *d, SysIOVec *iov, int iovlen);
+
+/* @brief Write the entirety of the given iovec.
+ *
+ * @return -1 on failure, or the number of bytes written on success. "Partial"
+ * failures will be reported with -1 and not the number of bytes we managed to
+ * write to disk before the failure. */
+Sint64 efile_writev(efile_data_t *d, SysIOVec *iov, int iovlen);
+
+/* @brief As \c efile_readv, but starting from a file offset. */
+Sint64 efile_preadv(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen);
+
+/* @brief As \c efile_writev, but starting from a file offset. */
+Sint64 efile_pwritev(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen);
+
+int efile_seek(efile_data_t *d, enum efile_seek_t seek, Sint64 offset, Sint64 *new_position);
+
+int efile_sync(efile_data_t *d, int data_only);
+
+int efile_advise(efile_data_t *d, Sint64 offset, Sint64 length, enum efile_advise_t advise);
+int efile_allocate(efile_data_t *d, Sint64 offset, Sint64 length);
+int efile_truncate(efile_data_t *d);
+
+posix_errno_t efile_open(const efile_path_t *path, enum efile_modes_t modes,
+ ErlNifResourceType *nif_type, efile_data_t **d);
+
+/** @brief Closes a file. The file must have entered the CLOSED state prior to
+ * calling this to prevent double close. */
+int efile_close(efile_data_t *d);
+
+/* **** **** **** **** **** **** **** **** **** **** **** **** **** **** **** */
+
+posix_errno_t efile_read_info(const efile_path_t *path, int follow_link, efile_fileinfo_t *result);
+
+/** @brief Sets the file times to the given values. Refer to efile_fileinfo_t
+ * for a description of each. */
+posix_errno_t efile_set_time(const efile_path_t *path, Sint64 a_time, Sint64 m_time, Sint64 c_time);
+
+/** @brief On Unix, this sets the file permissions according to the docs for
+ * file:write_file_info/2. On Windows it uses the "owner write permission" flag
+ * to toggle whether the file is read-only or not. */
+posix_errno_t efile_set_permissions(const efile_path_t *path, Uint32 permissions);
+
+/** @brief On Unix, this will set the owner/group to the given values. It will
+ * do nothing on other platforms. */
+posix_errno_t efile_set_owner(const efile_path_t *path, Uint32 owner, Uint32 group);
+
+/** @brief Resolves the final path of the given link. */
+posix_errno_t efile_read_link(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result);
+
+/** @brief Lists the contents of the given directory.
+ * @param result [out] A list of all the directory/file names contained in the
+ * given directory. */
+posix_errno_t efile_list_dir(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result);
+
+/** @brief Changes the name of an existing file or directory, from old_path
+ * to new_path.
+ *
+ * If old_path and new_path refer to the same file or directory, it does
+ * nothing and returns success. Otherwise if new_path already exists, it will
+ * be deleted and replaced by src subject to the following conditions:
+ *
+ * If old_path is a directory, new_path may be an empty directory.
+ * If old_path is a file, new_path may be a file.
+ *
+ * Neither of these are guaranteed to be atomic. In any other situation where
+ * new_path already exists, the rename will fail.
+ *
+ * Some possible error codes:
+ *
+ * - EACCES: Either paths or one of their parent directories can't be read
+ * and/or written.
+ * - EEXIST: new_path is a non-empty directory.
+ * - EINVAL: old_path is a root directory or new_path is a subdirectory
+ * of new_path.
+ * - EISDIR: new_path is a directory, but old_path is not.
+ * - ENOTDIR: old_path is a directory, but new_path is not.
+ * - ENOENT: old_path doesn't exist, or either path is "".
+ * - EXDEV: The paths are on different filesystems.
+ *
+ * The implementation of rename may allow cross-filesystem renames,
+ * but the caller should be prepared to emulate it with copy and
+ * delete if errno is EXDEV. */
+posix_errno_t efile_rename(const efile_path_t *old_path, const efile_path_t *new_path);
+
+posix_errno_t efile_make_hard_link(const efile_path_t *existing_path, const efile_path_t *new_path);
+posix_errno_t efile_make_soft_link(const efile_path_t *existing_path, const efile_path_t *new_path);
+posix_errno_t efile_make_dir(const efile_path_t *path);
+
+posix_errno_t efile_del_file(const efile_path_t *path);
+posix_errno_t efile_del_dir(const efile_path_t *path);
+
+posix_errno_t efile_get_cwd(ErlNifEnv *env, ERL_NIF_TERM *result);
+posix_errno_t efile_set_cwd(const efile_path_t *path);
+
+/** @brief A Windows-specific function for returning the working directory of a
+ * given device.
+ *
+ * @param device_index The drive index; 1 for A, 2 for B, etc.
+ * @param result [out] The working directory of the given device
+ */
+posix_errno_t efile_get_device_cwd(ErlNifEnv *env, int device_index, ERL_NIF_TERM *result);
+
+/** @brief A Windows-specific function for returning the 8.3-name of a given
+ * file or directory. */
+posix_errno_t efile_altname(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result);
diff --git a/erts/emulator/nifs/unix/unix_prim_file.c b/erts/emulator/nifs/unix/unix_prim_file.c
new file mode 100644
index 0000000000..57c8ef62e1
--- /dev/null
+++ b/erts/emulator/nifs/unix/unix_prim_file.c
@@ -0,0 +1,957 @@
+/*
+ * %CopyrightBegin%
+ *
+ * Copyright Ericsson 2017. All Rights Reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * %CopyrightEnd%
+ */
+
+#include "erl_nif.h"
+#include "config.h"
+#include "sys.h"
+
+#ifdef VALGRIND
+# include <valgrind/memcheck.h>
+#endif
+
+#include "prim_file_nif.h"
+
+#if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__)
+#define __DARWIN__ 1
+#endif
+
+#if defined(__DARWIN__) || defined(HAVE_LINUX_FALLOC_H) || defined(HAVE_POSIX_FALLOCATE)
+#include <fcntl.h>
+#endif
+
+#ifdef HAVE_LINUX_FALLOC_H
+#include <linux/falloc.h>
+#endif
+
+#include <utime.h>
+
+/* Macros for testing file types. */
+#ifdef NO_UMASK
+#define FILE_MODE 0644
+#define DIR_MODE 0755
+#else
+#define FILE_MODE 0666
+#define DIR_MODE 0777
+#endif
+
+/* Old platforms might not have IOV_MAX defined. */
+#if !defined(IOV_MAX) && defined(UIO_MAXIOV)
+#define IOV_MAX UIO_MAXIOV
+#elif !defined(IOV_MAX)
+#define IOV_MAX 16
+#endif
+
+typedef struct {
+ efile_data_t common;
+ int fd;
+} efile_unix_t;
+
+static int has_invalid_null_termination(const ErlNifBinary *path) {
+ const char *null_pos, *end_pos;
+
+ null_pos = memchr(path->data, '\0', path->size);
+ end_pos = (const char*)&path->data[path->size] - 1;
+
+ if(null_pos == NULL) {
+ return 1;
+ }
+
+ /* prim_file:internal_name2native sometimes feeds us data that is "doubly"
+ * NUL-terminated, so we'll accept any number of trailing NULs so long as
+ * they aren't interrupted by anything else. */
+ while(null_pos < end_pos && (*null_pos) == '\0') {
+ null_pos++;
+ }
+
+ return null_pos != end_pos;
+}
+
+posix_errno_t efile_marshal_path(ErlNifEnv *env, ERL_NIF_TERM path, efile_path_t *result) {
+ if(!enif_inspect_binary(env, path, result)) {
+ return EINVAL;
+ }
+
+ if(has_invalid_null_termination(result)) {
+ return EINVAL;
+ }
+
+ return 0;
+}
+
+ERL_NIF_TERM efile_get_handle(ErlNifEnv *env, efile_data_t *d) {
+ efile_unix_t *u = (efile_unix_t*)d;
+
+ ERL_NIF_TERM result;
+ unsigned char *bits;
+
+ bits = enif_make_new_binary(env, sizeof(u->fd), &result);
+ memcpy(bits, &u->fd, sizeof(u->fd));
+
+ return result;
+}
+
+static int open_file_type_check(const efile_path_t *path, int fd) {
+ struct stat file_info;
+ int error;
+
+#ifndef HAVE_FSTAT
+ error = stat((const char*)path->data, &file_info);
+ (void)fd;
+#else
+ error = fstat(fd, &file_info);
+ (void)path;
+#endif
+
+ if(error < 0) {
+ /* If we failed to stat assume success and let the next call handle the
+ * error. The old driver checked whether the file was to be used
+ * immediately in a read within the call, but the new implementation
+ * never does that. */
+ return 1;
+ } else {
+ /* The old driver tolerated opening /dev/null despite the "no devices"
+ * limitation. It provided no explanation for this but we still need
+ * to match the behavior. We're checking through stat(2) instead of
+ * comparing the name to account for links. */
+ struct stat null_device_info;
+ int is_dev_null;
+
+ is_dev_null = (stat("/dev/null", &null_device_info) == 0);
+ is_dev_null &= (file_info.st_ino == null_device_info.st_ino);
+ is_dev_null &= (file_info.st_dev == null_device_info.st_dev);
+
+ if(is_dev_null) {
+ return 1;
+ }
+ }
+
+ if(!S_ISREG(file_info.st_mode)) {
+ return 0;
+ }
+
+ return 1;
+}
+
+posix_errno_t efile_open(const efile_path_t *path, enum efile_modes_t modes,
+ ErlNifResourceType *nif_type, efile_data_t **d) {
+
+ int flags, fd;
+
+ flags = 0;
+
+ if(modes & EFILE_MODE_READ && !(modes & EFILE_MODE_WRITE)) {
+ flags |= O_RDONLY;
+ } else if(modes & EFILE_MODE_WRITE && !(modes & EFILE_MODE_READ)) {
+ if(!(modes & EFILE_MODE_NO_TRUNCATE)) {
+ flags |= O_TRUNC;
+ }
+
+ flags |= O_WRONLY | O_CREAT;
+ } else if(modes & EFILE_MODE_READ_WRITE) {
+ flags |= O_RDWR | O_CREAT;
+ } else {
+ return EINVAL;
+ }
+
+ if(modes & EFILE_MODE_APPEND) {
+ flags &= ~O_TRUNC;
+ flags |= O_APPEND;
+ }
+
+ if(modes & EFILE_MODE_EXCLUSIVE) {
+ flags |= O_EXCL;
+ }
+
+ if(modes & EFILE_MODE_SYNC) {
+#ifndef O_SYNC
+ return ENOTSUP;
+#else
+ flags |= O_SYNC;
+#endif
+ }
+
+ do {
+ fd = open((const char*)path->data, flags, FILE_MODE);
+ } while(fd == -1 && errno == EINTR);
+
+ if(fd != -1) {
+ efile_unix_t *u;
+
+ if(!(modes & EFILE_MODE_SKIP_TYPE_CHECK) && !open_file_type_check(path, fd)) {
+ close(fd);
+
+ /* This is blatantly incorrect, but we're documented as returning
+ * this for everything that isn't a file. */
+ return EISDIR;
+ }
+
+ u = (efile_unix_t*)enif_alloc_resource(nif_type, sizeof(efile_unix_t));
+ u->fd = fd;
+
+ EFILE_INIT_RESOURCE(&u->common, modes);
+ (*d) = &u->common;
+
+ return 0;
+ }
+
+ (*d) = NULL;
+ return errno;
+}
+
+int efile_close(efile_data_t *d) {
+ efile_unix_t *u = (efile_unix_t*)d;
+ int fd;
+
+ ASSERT(erts_atomic32_read_nob(&d->state) == EFILE_STATE_CLOSED);
+ ASSERT(u->fd != -1);
+
+ fd = u->fd;
+ u->fd = -1;
+
+ /* close(2) either always closes (*BSD, Linux) or leaves the fd in an
+ * undefined state (POSIX 2008, Solaris), so we must not retry on EINTR. */
+
+ if(close(fd) < 0) {
+ u->common.posix_errno = errno;
+ return 0;
+ }
+
+ return 1;
+}
+
+static void shift_iov(SysIOVec **iov, int *iovlen, ssize_t shift) {
+ SysIOVec *head_vec = (*iov);
+
+ ASSERT(shift >= 0);
+
+ while(shift > 0) {
+ ASSERT(head_vec < &(*iov)[*iovlen]);
+
+ if(shift < head_vec->iov_len) {
+ head_vec->iov_base = (char*)head_vec->iov_base + shift;
+ head_vec->iov_len -= shift;
+ break;
+ } else {
+ shift -= head_vec->iov_len;
+ head_vec++;
+ }
+ }
+
+ (*iovlen) -= head_vec - (*iov);
+ (*iov) = head_vec;
+}
+
+Sint64 efile_readv(efile_data_t *d, SysIOVec *iov, int iovlen) {
+ efile_unix_t *u = (efile_unix_t*)d;
+
+ Sint64 bytes_read;
+ ssize_t result;
+
+ bytes_read = 0;
+
+ do {
+ int use_fallback = 0;
+
+ if(iovlen < 1) {
+ result = 0;
+ break;
+ }
+
+ /* writev(2) implies readv(2) */
+#ifdef HAVE_WRITEV
+ result = readv(u->fd, iov, MIN(IOV_MAX, iovlen));
+
+ /* Fall back to using read(2) if readv(2) reports that the combined
+ * size of iov is greater than SSIZE_T_MAX. */
+ use_fallback = (result < 0 && errno == EINVAL);
+#else
+ use_fallback = 1;
+#endif
+
+ if(use_fallback) {
+ result = read(u->fd, iov->iov_base, iov->iov_len);
+ }
+
+ if(result > 0) {
+ shift_iov(&iov, &iovlen, result);
+ bytes_read += result;
+ }
+ } while(result > 0 || (result < 0 && errno == EINTR));
+
+ u->common.posix_errno = errno;
+
+ if(result == 0 && bytes_read > 0) {
+ return bytes_read;
+ }
+
+ return result;
+}
+
+Sint64 efile_writev(efile_data_t *d, SysIOVec *iov, int iovlen) {
+ efile_unix_t *u = (efile_unix_t*)d;
+
+ Sint64 bytes_written;
+ ssize_t result;
+
+ bytes_written = 0;
+
+ do {
+ int use_fallback = 0;
+
+ if(iovlen < 1) {
+ result = 0;
+ break;
+ }
+
+#ifdef HAVE_WRITEV
+ result = writev(u->fd, iov, MIN(IOV_MAX, iovlen));
+
+ /* Fall back to using write(2) if writev(2) reports that the combined
+ * size of iov is greater than SSIZE_T_MAX. */
+ use_fallback = (result < 0 && errno == EINVAL);
+#else
+ use_fallback = 1;
+#endif
+
+ if(use_fallback) {
+ result = write(u->fd, iov->iov_base, iov->iov_len);
+ }
+
+ if(result > 0) {
+ shift_iov(&iov, &iovlen, result);
+ bytes_written += result;
+ }
+ } while(result > 0 || (result < 0 && errno == EINTR));
+
+ u->common.posix_errno = errno;
+
+ if(result == 0 && bytes_written > 0) {
+ return bytes_written;
+ }
+
+ return result;
+}
+
+Sint64 efile_preadv(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen) {
+ efile_unix_t *u = (efile_unix_t*)d;
+
+ Uint64 bytes_read;
+ Sint64 result;
+
+#if !defined(HAVE_PREADV) && !defined(HAVE_PREAD)
+ /* This function is documented as leaving the file position undefined, but
+ * the old driver always reset it so there's probably code in the wild that
+ * relies on this behavior. */
+ off_t original_position = lseek(u->fd, 0, SEEK_CUR);
+
+ if(original_position < 0 || lseek(u->fd, offset, SEEK_SET) < 0) {
+ u->common.posix_errno = errno;
+ return -1;
+ }
+#endif
+
+ bytes_read = 0;
+
+ do {
+ if(iovlen < 1) {
+ result = 0;
+ break;
+ }
+
+#if defined(HAVE_PREADV)
+ result = preadv(u->fd, iov, MIN(IOV_MAX, iovlen), offset);
+#elif defined(HAVE_PREAD)
+ result = pread(u->fd, iov->iov_base, iov->iov_len, offset);
+#else
+ result = read(u->fd, iov->iov_base, iov->iov_len);
+#endif
+
+ if(result > 0) {
+ shift_iov(&iov, &iovlen, result);
+ bytes_read += result;
+ offset += result;
+ }
+ } while(result > 0 || (result < 0 && errno == EINTR));
+
+ u->common.posix_errno = errno;
+
+#if !defined(HAVE_PREADV) && !defined(HAVE_PREAD)
+ if(result >= 0) {
+ if(lseek(u->fd, original_position, SEEK_SET) < 0) {
+ u->common.posix_errno = errno;
+ return -1;
+ }
+ }
+#endif
+
+ if(result == 0 && bytes_read > 0) {
+ return bytes_read;
+ }
+
+ return result;
+}
+
+Sint64 efile_pwritev(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen) {
+ efile_unix_t *u = (efile_unix_t*)d;
+
+ Sint64 bytes_written;
+ ssize_t result;
+
+#if !defined(HAVE_PWRITEV) && !defined(HAVE_PWRITE)
+ off_t original_position = lseek(u->fd, 0, SEEK_CUR);
+
+ if(original_position < 0 || lseek(u->fd, offset, SEEK_SET) < 0) {
+ u->common.posix_errno = errno;
+ return -1;
+ }
+#endif
+
+ bytes_written = 0;
+
+ do {
+ if(iovlen < 1) {
+ result = 0;
+ break;
+ }
+
+#if defined(HAVE_PWRITEV)
+ result = pwritev(u->fd, iov, MIN(IOV_MAX, iovlen), offset);
+#elif defined(HAVE_PWRITE)
+ result = pwrite(u->fd, iov->iov_base, iov->iov_len, offset);
+#else
+ result = write(u->fd, iov->iov_base, iov->iov_len);
+#endif
+
+ if(result > 0) {
+ shift_iov(&iov, &iovlen, result);
+ bytes_written += result;
+ offset += result;
+ }
+ } while(result > 0 || (result < 0 && errno == EINTR));
+
+ u->common.posix_errno = errno;
+
+#if !defined(HAVE_PWRITEV) && !defined(HAVE_PWRITE)
+ if(result >= 0) {
+ if(lseek(u->fd, original_position, SEEK_SET) < 0) {
+ u->common.posix_errno = errno;
+ return -1;
+ }
+ }
+#endif
+
+ if(result == 0 && bytes_written > 0) {
+ return bytes_written;
+ }
+
+ return result;
+}
+
+int efile_seek(efile_data_t *d, enum efile_seek_t seek, Sint64 offset, Sint64 *new_position) {
+ efile_unix_t *u = (efile_unix_t*)d;
+ off_t result;
+ int whence;
+
+ switch(seek) {
+ case EFILE_SEEK_BOF: whence = SEEK_SET; break;
+ case EFILE_SEEK_CUR: whence = SEEK_CUR; break;
+ case EFILE_SEEK_EOF: whence = SEEK_END; break;
+ default: ERTS_INTERNAL_ERROR("Invalid seek parameter");
+ }
+
+ result = lseek(u->fd, offset, whence);
+
+ /*
+ * The man page for lseek (on SunOs 5) says:
+ *
+ * "if fildes is a remote file descriptor and offset is negative, lseek()
+ * returns the file pointer even if it is negative."
+ */
+ if(result < 0 && errno == 0) {
+ errno = EINVAL;
+ }
+
+ if(result < 0) {
+ u->common.posix_errno = errno;
+ return 0;
+ }
+
+ (*new_position) = result;
+
+ return 1;
+}
+
+int efile_sync(efile_data_t *d, int data_only) {
+ efile_unix_t *u = (efile_unix_t*)d;
+
+#if defined(HAVE_FDATASYNC) && !defined(__DARWIN__)
+ if(data_only) {
+ if(fdatasync(u->fd) < 0) {
+ u->common.posix_errno = errno;
+ return 0;
+ }
+
+ return 1;
+ }
+#endif
+
+#if defined(__DARWIN__) && defined(F_FULLFSYNC)
+ if(fcntl(u->fd, F_FULLFSYNC) < 0) {
+#else
+ if(fsync(u->fd) < 0) {
+#endif
+ u->common.posix_errno = errno;
+ return 0;
+ }
+
+ return 1;
+}
+
+int efile_advise(efile_data_t *d, Sint64 offset, Sint64 length, enum efile_advise_t advise) {
+ efile_unix_t *u = (efile_unix_t*)d;
+#ifdef HAVE_POSIX_FADVISE
+ int p_advise;
+
+ switch(advise) {
+ case EFILE_ADVISE_NORMAL: p_advise = POSIX_FADV_NORMAL; break;
+ case EFILE_ADVISE_RANDOM: p_advise = POSIX_FADV_RANDOM; break;
+ case EFILE_ADVISE_SEQUENTIAL: p_advise = POSIX_FADV_SEQUENTIAL; break;
+ case EFILE_ADVISE_WILL_NEED: p_advise = POSIX_FADV_WILLNEED; break;
+ case EFILE_ADVISE_DONT_NEED: p_advise = POSIX_FADV_DONTNEED; break;
+ case EFILE_ADVISE_NO_REUSE: p_advise = POSIX_FADV_NOREUSE; break;
+ default:
+ u->common.posix_errno = EINVAL;
+ return 0;
+ }
+
+ if(posix_fadvise(u->fd, offset, length, p_advise) < 0) {
+ u->common.posix_errno = errno;
+ return 0;
+ }
+
+ return 1;
+#else
+ /* We'll pretend to support this syscall as it's only a recommendation even
+ * on systems that do support it. */
+ return 1;
+#endif
+}
+
+int efile_allocate(efile_data_t *d, Sint64 offset, Sint64 length) {
+ efile_unix_t *u = (efile_unix_t*)d;
+ int ret = -1;
+
+ /* We prefer OS-specific methods, but fall back to posix_fallocate on
+ * failure. It's unclear whether this has any practical benefit on
+ * modern systems, but the old driver did it. */
+
+#if defined(HAVE_FALLOCATE)
+ /* Linux-specific */
+ do {
+ ret = fallocate(u->fd, FALLOC_FL_KEEP_SIZE, offset, length);
+ } while(ret < 0 && errno == EINTR);
+#elif defined(F_PREALLOCATE)
+ /* Mac-specific */
+ fstore_t fs = {};
+
+ fs.fst_flags = F_ALLOCATECONTIG;
+ fs.fst_posmode = F_VOLPOSMODE;
+ fs.fst_offset = offset;
+ fs.fst_length = length;
+
+ ret = fcntl(u->fd, F_PREALLOCATE, &fs);
+ if(ret < 0) {
+ fs.fst_flags = F_ALLOCATEALL;
+ ret = fcntl(u->fd, F_PREALLOCATE, &fs);
+ }
+#elif !defined(HAVE_POSIX_FALLOCATE)
+ u->common.posix_errno = ENOTSUP;
+ return 0;
+#endif
+
+#ifdef HAVE_POSIX_FALLOCATE
+ if(ret < 0) {
+ do {
+ ret = posix_fallocate(u->fd, offset, length);
+
+ /* On Linux and Solaris for example, posix_fallocate() returns a
+ * positive error number on error and it does not set errno. On
+ * FreeBSD however (9.0 at least), it returns -1 on error and it
+ * sets errno. */
+ if (ret > 0) {
+ errno = ret;
+ ret = -1;
+ }
+ } while(ret < 0 && errno == EINTR);
+ }
+#endif
+
+ if(ret < 0) {
+ u->common.posix_errno = errno;
+ return 0;
+ }
+
+ return 1;
+}
+
+int efile_truncate(efile_data_t *d) {
+ efile_unix_t *u = (efile_unix_t*)d;
+ off_t offset;
+
+ offset = lseek(u->fd, 0, SEEK_CUR);
+
+ if(offset < 0) {
+ u->common.posix_errno = errno;
+ return 0;
+ }
+
+ if(ftruncate(u->fd, offset) < 0) {
+ u->common.posix_errno = errno;
+ return 0;
+ }
+
+ return 1;
+}
+
+posix_errno_t efile_read_info(const efile_path_t *path, int follow_links, efile_fileinfo_t *result) {
+ struct stat data;
+
+ if(follow_links) {
+ if(stat((const char*)path->data, &data) < 0) {
+ return errno;
+ }
+ } else {
+ if(lstat((const char*)path->data, &data) < 0) {
+ return errno;
+ }
+ }
+
+ if(S_ISCHR(data.st_mode) || S_ISBLK(data.st_mode)) {
+ result->type = EFILE_FILETYPE_DEVICE;
+ } else if(S_ISDIR(data.st_mode)) {
+ result->type = EFILE_FILETYPE_DIRECTORY;
+ } else if(S_ISREG(data.st_mode)) {
+ result->type = EFILE_FILETYPE_REGULAR;
+ } else if(S_ISLNK(data.st_mode)) {
+ result->type = EFILE_FILETYPE_SYMLINK;
+ } else {
+ result->type = EFILE_FILETYPE_OTHER;
+ }
+
+ result->a_time = (Sint64)data.st_atime;
+ result->m_time = (Sint64)data.st_mtime;
+ result->c_time = (Sint64)data.st_ctime;
+ result->size = data.st_size;
+
+ result->major_device = data.st_dev;
+ result->minor_device = data.st_rdev;
+ result->links = data.st_nlink;
+ result->inode = data.st_ino;
+ result->mode = data.st_mode;
+ result->uid = data.st_uid;
+ result->gid = data.st_gid;
+
+#ifndef NO_ACCESS
+ result->access = EFILE_ACCESS_NONE;
+
+ if(access((const char*)path->data, R_OK) == 0) {
+ result->access |= EFILE_ACCESS_READ;
+ }
+ if(access((const char*)path->data, W_OK) == 0) {
+ result->access |= EFILE_ACCESS_WRITE;
+ }
+#else
+ /* Just look at read/write access for owner. */
+ result->access = ((data.st_mode >> 6) & 07) >> 1;
+#endif
+
+ return 0;
+}
+
+posix_errno_t efile_set_permissions(const efile_path_t *path, Uint32 permissions) {
+ const mode_t MUTABLE_MODES = (S_ISUID | S_ISGID | S_IRWXU | S_IRWXG | S_IRWXO);
+ mode_t new_modes = permissions & MUTABLE_MODES;
+
+ if(chmod((const char*)path->data, new_modes) < 0) {
+ new_modes &= ~(S_ISUID | S_ISGID);
+
+ if (chmod((const char*)path->data, new_modes) < 0) {
+ return errno;
+ }
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_set_owner(const efile_path_t *path, Uint32 owner, Uint32 group) {
+ if(chown((const char*)path->data, owner, group) < 0) {
+ return errno;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_set_time(const efile_path_t *path, Sint64 a_time, Sint64 m_time, Sint64 c_time) {
+ struct utimbuf tval;
+
+ tval.actime = (time_t)a_time;
+ tval.modtime = (time_t)m_time;
+
+ (void)c_time;
+
+ if(utime((const char*)path->data, &tval) < 0) {
+ return errno;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_read_link(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result) {
+ ErlNifBinary result_bin;
+
+ if(!enif_alloc_binary(256, &result_bin)) {
+ return ENOMEM;
+ }
+
+ for(;;) {
+ ssize_t bytes_copied;
+
+ bytes_copied = readlink((const char*)path->data, (char*)result_bin.data,
+ result_bin.size);
+
+ if(bytes_copied <= 0) {
+ posix_errno_t saved_errno = errno;
+ enif_release_binary(&result_bin);
+ return saved_errno;
+ } else if(bytes_copied < result_bin.size) {
+ if(!enif_realloc_binary(&result_bin, bytes_copied)) {
+ enif_release_binary(&result_bin);
+ return ENOMEM;
+ }
+
+ (*result) = enif_make_binary(env, &result_bin);
+
+ return 0;
+ }
+
+ /* The result didn't fit into the buffer, so we'll try again with a
+ * larger one. */
+
+ if(!enif_realloc_binary(&result_bin, result_bin.size * 2)) {
+ enif_release_binary(&result_bin);
+ return ENOMEM;
+ }
+ }
+}
+
+static int is_ignored_name(int name_length, const char *name) {
+ if(name_length == 1 && name[0] == '.') {
+ return 1;
+ } else if(name_length == 2 && memcmp(name, "..", 2) == 0) {
+ return 1;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_list_dir(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result) {
+ ERL_NIF_TERM list_head;
+
+ struct dirent *dir_entry;
+ DIR *dir_stream;
+
+ dir_stream = opendir((const char*)path->data);
+ if(dir_stream == NULL) {
+ posix_errno_t saved_errno = errno;
+ *result = enif_make_list(env, 0);
+ return saved_errno;
+ }
+
+ list_head = enif_make_list(env, 0);
+ dir_entry = readdir(dir_stream);
+
+ while(dir_entry != NULL) {
+ int name_length = strlen(dir_entry->d_name);
+
+ if(!is_ignored_name(name_length, dir_entry->d_name)) {
+ unsigned char *name_bytes;
+ ERL_NIF_TERM name_term;
+
+ name_bytes = enif_make_new_binary(env, name_length, &name_term);
+ sys_memcpy(name_bytes, dir_entry->d_name, name_length);
+
+ list_head = enif_make_list_cell(env, name_term, list_head);
+ }
+
+ dir_entry = readdir(dir_stream);
+ }
+
+ (*result) = list_head;
+ closedir(dir_stream);
+
+ return 0;
+}
+
+posix_errno_t efile_rename(const efile_path_t *old_path, const efile_path_t *new_path) {
+ if(rename((const char*)old_path->data, (const char*)new_path->data) < 0) {
+ if(errno == ENOTEMPTY) {
+ return EEXIST;
+ }
+
+ if(strcmp((const char*)old_path->data, "/") == 0) {
+ /* Alpha reports renaming / as EBUSY and Linux reports it as EACCES
+ * instead of EINVAL.*/
+ return EINVAL;
+ }
+
+ return errno;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_make_hard_link(const efile_path_t *existing_path, const efile_path_t *new_path) {
+ if(link((const char*)existing_path->data, (const char*)new_path->data) < 0) {
+ return errno;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_make_soft_link(const efile_path_t *existing_path, const efile_path_t *new_path) {
+ if(symlink((const char*)existing_path->data, (const char*)new_path->data) < 0) {
+ return errno;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_make_dir(const efile_path_t *path) {
+#ifdef NO_MKDIR_MODE
+ if(mkdir((const char*)path->data) < 0) {
+#else
+ if(mkdir((const char*)path->data, DIR_MODE) < 0) {
+#endif
+ return errno;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_del_file(const efile_path_t *path) {
+ if(unlink((const char*)path->data) < 0) {
+ /* Linux sets the wrong error code. */
+ if(errno == EISDIR) {
+ return EPERM;
+ }
+
+ return errno;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_del_dir(const efile_path_t *path) {
+ if(rmdir((const char*)path->data) < 0) {
+ posix_errno_t saved_errno = errno;
+
+ if(saved_errno == ENOTEMPTY) {
+ saved_errno = EEXIST;
+ }
+
+ /* The error code might be wrong if we're trying to delete the current
+ * directory. */
+ if(saved_errno == EEXIST) {
+ struct stat path_stat, cwd_stat;
+ int has_stat;
+
+ has_stat = (stat((const char*)path->data, &path_stat) == 0);
+ has_stat &= (stat(".", &cwd_stat) == 0);
+
+ if(has_stat && path_stat.st_ino == cwd_stat.st_ino) {
+ if(path_stat.st_dev == cwd_stat.st_dev) {
+ return EINVAL;
+ }
+ }
+ }
+
+ return saved_errno;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_set_cwd(const efile_path_t *path) {
+ if(chdir((const char*)path->data) < 0) {
+ return errno;
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_get_device_cwd(ErlNifEnv *env, int device_index, ERL_NIF_TERM *result) {
+ (void)device_index;
+ (void)result;
+ (void)env;
+
+ return ENOTSUP;
+}
+
+posix_errno_t efile_get_cwd(ErlNifEnv *env, ERL_NIF_TERM *result) {
+ ErlNifBinary result_bin;
+ size_t bytes_copied;
+
+ if(!enif_alloc_binary(256, &result_bin)) {
+ return ENOMEM;
+ }
+
+ while(getcwd((char*)result_bin.data, result_bin.size) == NULL) {
+ posix_errno_t saved_errno = errno;
+
+ if(saved_errno != ERANGE) {
+ enif_release_binary(&result_bin);
+ return saved_errno;
+ } else {
+ if(!enif_realloc_binary(&result_bin, result_bin.size * 2)) {
+ enif_release_binary(&result_bin);
+ return ENOMEM;
+ }
+ }
+ }
+
+ /* getcwd(2) guarantees null-termination. */
+ bytes_copied = strlen((const char*)result_bin.data);
+
+ if(!enif_realloc_binary(&result_bin, bytes_copied)) {
+ enif_release_binary(&result_bin);
+ return ENOMEM;
+ }
+
+ (*result) = enif_make_binary(env, &result_bin);
+
+ return 0;
+}
+
+posix_errno_t efile_altname(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result) {
+ (void)path;
+ (void)result;
+
+ return ENOTSUP;
+}
diff --git a/erts/emulator/nifs/win32/win_prim_file.c b/erts/emulator/nifs/win32/win_prim_file.c
new file mode 100644
index 0000000000..9f993f1d24
--- /dev/null
+++ b/erts/emulator/nifs/win32/win_prim_file.c
@@ -0,0 +1,1427 @@
+/*
+ * %CopyrightBegin%
+ *
+ * Copyright Ericsson 2017. All Rights Reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * %CopyrightEnd%
+ */
+
+#include "erl_nif.h"
+#include "config.h"
+#include "sys.h"
+
+#include "prim_file_nif.h"
+
+#include <windows.h>
+#include <strsafe.h>
+#include <wchar.h>
+
+#define IS_SLASH(a) ((a) == L'\\' || (a) == L'/')
+
+#define FILE_SHARE_FLAGS (FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE)
+
+#define LP_PREFIX L"\\\\?\\"
+#define LP_PREFIX_SIZE (sizeof(LP_PREFIX) - sizeof(WCHAR))
+#define LP_PREFIX_LENGTH (LP_PREFIX_SIZE / sizeof(WCHAR))
+
+#define PATH_LENGTH(path) (path->size / sizeof(WCHAR) - 1)
+
+#define ASSERT_PATH_FORMAT(path) \
+ do { \
+ ASSERT(PATH_LENGTH(path) >= 4 && \
+ !memcmp(path->data, LP_PREFIX, LP_PREFIX_SIZE)); \
+ ASSERT(PATH_LENGTH(path) == wcslen((WCHAR*)path->data)); \
+ } while(0)
+
+#define TICKS_PER_SECOND (10000000ULL)
+#define EPOCH_DIFFERENCE (11644473600LL)
+
+#define FILETIME_TO_EPOCH(epoch, ft) \
+ do { \
+ ULARGE_INTEGER ull; \
+ ull.LowPart = (ft).dwLowDateTime; \
+ ull.HighPart = (ft).dwHighDateTime; \
+ (epoch) = ((ull.QuadPart / TICKS_PER_SECOND) - EPOCH_DIFFERENCE); \
+ } while(0)
+
+#define EPOCH_TO_FILETIME(ft, epoch) \
+ do { \
+ ULARGE_INTEGER ull; \
+ ull.QuadPart = (((epoch) + EPOCH_DIFFERENCE) * TICKS_PER_SECOND); \
+ (ft).dwLowDateTime = ull.LowPart; \
+ (ft).dwHighDateTime = ull.HighPart; \
+ } while(0)
+
+typedef struct {
+ efile_data_t common;
+ HANDLE handle;
+} efile_win_t;
+
+static int windows_to_posix_errno(DWORD last_error);
+
+static int has_invalid_null_termination(const ErlNifBinary *path) {
+ const WCHAR *null_pos, *end_pos;
+
+ null_pos = wmemchr((const WCHAR*)path->data, L'\0', path->size);
+ end_pos = (const WCHAR*)&path->data[path->size] - 1;
+
+ if(null_pos == NULL) {
+ return 1;
+ }
+
+ /* prim_file:internal_name2native sometimes feeds us data that is "doubly"
+ * NUL-terminated, so we'll accept any number of trailing NULs so long as
+ * they aren't interrupted by anything else. */
+ while(null_pos < end_pos && (*null_pos) == L'\0') {
+ null_pos++;
+ }
+
+ return null_pos != end_pos;
+}
+
+static posix_errno_t get_full_path(ErlNifEnv *env, WCHAR *input, efile_path_t *result) {
+ DWORD maximum_length, actual_length;
+ int add_long_prefix;
+
+ maximum_length = GetFullPathNameW(input, 0, NULL, NULL);
+ add_long_prefix = 0;
+
+ if(maximum_length == 0) {
+ /* POSIX doesn't have the concept of a "path error" in the same way
+ * Windows does, so we'll return ENOENT since that's what most POSIX
+ * APIs would return if they were fed such garbage. */
+ return ENOENT;
+ }
+
+ maximum_length += LP_PREFIX_LENGTH;
+
+ if(!enif_alloc_binary(maximum_length * sizeof(WCHAR), result)) {
+ return ENOMEM;
+ }
+
+ actual_length = GetFullPathNameW(input, maximum_length, (WCHAR*)result->data, NULL);
+
+ if(actual_length < maximum_length) {
+ int has_long_path_prefix;
+ WCHAR *path_start;
+
+ /* Make sure we have a long-path prefix; GetFullPathNameW only adds one
+ * if the path is relative. */
+ has_long_path_prefix = actual_length >= LP_PREFIX_LENGTH &&
+ !sys_memcmp(result->data, LP_PREFIX, LP_PREFIX_SIZE);
+
+ if(!has_long_path_prefix) {
+ sys_memmove(result->data + LP_PREFIX_SIZE, result->data,
+ (actual_length + 1) * sizeof(WCHAR));
+ sys_memcpy(result->data, LP_PREFIX, LP_PREFIX_SIZE);
+ actual_length += LP_PREFIX_LENGTH;
+ }
+
+ path_start = (WCHAR*)result->data;
+
+ /* We're removing trailing slashes since quite a few APIs refuse to
+ * work with them, and none require them. We only check the last
+ * character since GetFullPathNameW folds slashes together. */
+ if(IS_SLASH(path_start[actual_length - 1])) {
+ if(path_start[actual_length - 2] != L':') {
+ path_start[actual_length - 1] = L'\0';
+ actual_length--;
+ }
+ }
+
+ if(!enif_realloc_binary(result, (actual_length + 1) * sizeof(WCHAR))) {
+ enif_release_binary(result);
+ return ENOMEM;
+ }
+
+ enif_make_binary(env, result);
+ return 0;
+ }
+
+ /* We may end up here if the current directory changes to something longer
+ * between/during GetFullPathName. There's nothing sensible we can do about
+ * this. */
+
+ enif_release_binary(result);
+
+ return EINVAL;
+}
+
+posix_errno_t efile_marshal_path(ErlNifEnv *env, ERL_NIF_TERM path, efile_path_t *result) {
+ ErlNifBinary raw_path;
+
+ if(!enif_inspect_binary(env, path, &raw_path)) {
+ return EINVAL;
+ } else if(raw_path.size % sizeof(WCHAR)) {
+ return EINVAL;
+ }
+
+ if(has_invalid_null_termination(&raw_path)) {
+ return EINVAL;
+ }
+
+ return get_full_path(env, (WCHAR*)raw_path.data, result);
+}
+
+ERL_NIF_TERM efile_get_handle(ErlNifEnv *env, efile_data_t *d) {
+ efile_win_t *w = (efile_win_t*)d;
+
+ ERL_NIF_TERM result;
+ unsigned char *bits;
+
+ bits = enif_make_new_binary(env, sizeof(w->handle), &result);
+ memcpy(bits, &w->handle, sizeof(w->handle));
+
+ return result;
+}
+
+/** @brief Converts a native path to the preferred form in "erlang space,"
+ * without path-prefixes, forward-slashes, or NUL terminators. */
+static int normalize_path_result(ErlNifBinary *path) {
+ WCHAR *path_iterator, *path_start, *path_end;
+ int length;
+
+ path_start = (WCHAR*)path->data;
+ length = wcslen(path_start);
+
+ ASSERT(length < path->size / sizeof(WCHAR));
+
+ /* Get rid of the long-path prefix, if present. */
+ if(length >= LP_PREFIX_LENGTH) {
+ if(!sys_memcmp(path_start, LP_PREFIX, LP_PREFIX_SIZE)) {
+ length -= LP_PREFIX_LENGTH;
+
+ sys_memmove(path_start, &path_start[LP_PREFIX_LENGTH],
+ length * sizeof(WCHAR));
+ }
+ }
+
+ path_end = &path_start[length];
+ path_iterator = path_start;
+
+ /* Convert drive letters to lowercase, if present. */
+ if(length >= 2 && path_start[1] == L':') {
+ WCHAR drive_letter = path_start[0];
+
+ if(drive_letter >= L'A' && drive_letter <= L'Z') {
+ path_start[0] = drive_letter - L'A' + L'a';
+ }
+ }
+
+ while(path_iterator < path_end) {
+ if(*path_iterator == L'\\') {
+ *path_iterator = L'/';
+ }
+
+ path_iterator++;
+ }
+
+ /* Truncate the result to its actual length; we don't want to include the
+ * NUL terminator. */
+ return enif_realloc_binary(path, length * sizeof(WCHAR));
+}
+
+/* @brief Checks whether all the given attributes are set on the object at the
+ * given path. Note that it assumes false on errors. */
+static int has_file_attributes(const efile_path_t *path, DWORD mask) {
+ DWORD attributes = GetFileAttributesW((WCHAR*)path->data);
+
+ if(attributes == INVALID_FILE_ATTRIBUTES) {
+ return 0;
+ }
+
+ return !!((attributes & mask) == mask);
+}
+
+static int is_ignored_name(int name_length, const WCHAR *name) {
+ if(name_length == 1 && name[0] == L'.') {
+ return 1;
+ } else if(name_length == 2 && !sys_memcmp(name, L"..", 2 * sizeof(WCHAR))) {
+ return 1;
+ }
+
+ return 0;
+}
+
+static int get_drive_number(const efile_path_t *path) {
+ const WCHAR *path_start;
+ int length;
+
+ ASSERT_PATH_FORMAT(path);
+
+ path_start = (WCHAR*)path->data + LP_PREFIX_LENGTH;
+ length = PATH_LENGTH(path) - LP_PREFIX_LENGTH;
+
+ if(length >= 2 && path_start[1] == L':') {
+ WCHAR drive_letter = path_start[0];
+
+ if(drive_letter >= L'A' && drive_letter <= L'Z') {
+ return drive_letter - L'A' + 1;
+ } else if(drive_letter >= L'a' && drive_letter <= L'z') {
+ return drive_letter - L'a' + 1;
+ }
+ }
+
+ return -1;
+}
+
+/* @brief Checks whether two *paths* are on the same mount point; they don't
+ * have to refer to existing or accessible files/directories. */
+static int has_same_mount_point(const efile_path_t *path_a, const efile_path_t *path_b) {
+ WCHAR *mount_a, *mount_b;
+ int result = 0;
+
+ mount_a = enif_alloc(path_a->size);
+ mount_b = enif_alloc(path_b->size);
+
+ if(mount_a != NULL && mount_b != NULL) {
+ int length_a, length_b;
+
+ length_a = PATH_LENGTH(path_a);
+ length_b = PATH_LENGTH(path_b);
+
+ if(GetVolumePathNameW((WCHAR*)path_a->data, mount_a, length_a)) {
+ ASSERT(wcslen(mount_a) <= length_a);
+
+ if(GetVolumePathNameW((WCHAR*)path_b->data, mount_b, length_b)) {
+ ASSERT(wcslen(mount_b) <= length_b);
+
+ result = !_wcsicmp(mount_a, mount_b);
+ }
+ }
+ }
+
+ if(mount_b != NULL) {
+ enif_free(mount_b);
+ }
+
+ if(mount_a != NULL) {
+ enif_free(mount_a);
+ }
+
+ return result;
+}
+
+/* Mirrors the PathIsRootW function of the shell API, but doesn't choke on
+ * paths longer than MAX_PATH. */
+static int is_path_root(const efile_path_t *path) {
+ const WCHAR *path_start, *path_end;
+ int length;
+
+ ASSERT_PATH_FORMAT(path);
+
+ path_start = (WCHAR*)path->data + LP_PREFIX_LENGTH;
+ length = PATH_LENGTH(path) - LP_PREFIX_LENGTH;
+
+ path_end = &path_start[length];
+
+ if(length == 1) {
+ /* A single \ refers to the root of the current working directory. */
+ return IS_SLASH(path_start[0]);
+ } else if(length == 3 && iswalpha(path_start[0]) && path_start[1] == L':') {
+ /* Drive letter. */
+ return IS_SLASH(path_start[2]);
+ } else if(length >= 4) {
+ /* Check whether we're a UNC root, eg. \\server, \\server\share */
+ const WCHAR *path_iterator;
+
+ if(!IS_SLASH(path_start[0]) || !IS_SLASH(path_start[1])) {
+ return 0;
+ }
+
+ path_iterator = path_start + 2;
+
+ /* Slide to the slash between the server and share names, if present. */
+ while(path_iterator < path_end && !IS_SLASH(*path_iterator)) {
+ path_iterator++;
+ }
+
+ /* Slide past the end of the string, stopping at the first slash we
+ * encounter. */
+ do {
+ path_iterator++;
+ } while(path_iterator < path_end && !IS_SLASH(*path_iterator));
+
+ /* If we're past the end of the string and it didnt't end with a slash,
+ * then we're a root path. */
+ return path_iterator >= path_end && !IS_SLASH(path_start[length - 1]);
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_open(const efile_path_t *path, enum efile_modes_t modes,
+ ErlNifResourceType *nif_type, efile_data_t **d) {
+
+ DWORD attributes, access_flags, open_mode;
+ HANDLE handle;
+
+ ASSERT_PATH_FORMAT(path);
+
+ access_flags = 0;
+ open_mode = 0;
+
+ if(modes & EFILE_MODE_READ && !(modes & EFILE_MODE_WRITE)) {
+ access_flags = GENERIC_READ;
+ open_mode = OPEN_EXISTING;
+ } else if(modes & EFILE_MODE_WRITE && !(modes & EFILE_MODE_READ)) {
+ access_flags = GENERIC_WRITE;
+ open_mode = CREATE_ALWAYS;
+ } else if(modes & EFILE_MODE_READ_WRITE) {
+ access_flags = GENERIC_READ | GENERIC_WRITE;
+ open_mode = OPEN_ALWAYS;
+ } else {
+ return EINVAL;
+ }
+
+ if(modes & EFILE_MODE_APPEND) {
+ access_flags |= FILE_APPEND_DATA;
+ open_mode = OPEN_ALWAYS;
+ }
+
+ if(modes & EFILE_MODE_EXCLUSIVE) {
+ open_mode = CREATE_NEW;
+ }
+
+ if(modes & EFILE_MODE_SYNC) {
+ attributes = FILE_FLAG_WRITE_THROUGH;
+ } else {
+ attributes = FILE_ATTRIBUTE_NORMAL;
+ }
+
+ handle = CreateFileW((WCHAR*)path->data, access_flags,
+ FILE_SHARE_FLAGS, NULL, open_mode, attributes, NULL);
+
+ if(handle != INVALID_HANDLE_VALUE) {
+ efile_win_t *w;
+
+ w = (efile_win_t*)enif_alloc_resource(nif_type, sizeof(efile_win_t));
+ w->handle = handle;
+
+ EFILE_INIT_RESOURCE(&w->common, modes);
+ (*d) = &w->common;
+
+ return 0;
+ } else {
+ DWORD last_error = GetLastError();
+
+ /* Rewrite all failures on directories to EISDIR to match the old
+ * driver. */
+ if(has_file_attributes(path, FILE_ATTRIBUTE_DIRECTORY)) {
+ return EISDIR;
+ }
+
+ return windows_to_posix_errno(last_error);
+ }
+}
+
+int efile_close(efile_data_t *d) {
+ efile_win_t *w = (efile_win_t*)d;
+ HANDLE handle;
+
+ ASSERT(erts_atomic32_read_nob(&d->state) == EFILE_STATE_CLOSED);
+ ASSERT(w->handle != INVALID_HANDLE_VALUE);
+
+ handle = w->handle;
+ w->handle = INVALID_HANDLE_VALUE;
+
+ if(!CloseHandle(handle)) {
+ w->common.posix_errno = windows_to_posix_errno(GetLastError());
+ return 0;
+ }
+
+ return 1;
+}
+
+static void shift_overlapped(OVERLAPPED *overlapped, DWORD shift) {
+ LARGE_INTEGER offset;
+
+ ASSERT(shift >= 0);
+
+ offset.HighPart = overlapped->OffsetHigh;
+ offset.LowPart = overlapped->Offset;
+
+ /* ~(Uint64)0 is a magic value ("append to end of file") which needs to be
+ * preserved. Other positions resulting in overflow would have errored out
+ * just prior to this point. */
+ if(offset.QuadPart != ERTS_UINT64_MAX) {
+ offset.QuadPart += shift;
+ }
+
+ /* All unused fields must be zeroed for the next call. */
+ sys_memset(overlapped, 0, sizeof(*overlapped));
+ overlapped->OffsetHigh = offset.HighPart;
+ overlapped->Offset = offset.LowPart;
+}
+
+static void shift_iov(SysIOVec **iov, int *iovlen, DWORD shift) {
+ SysIOVec *head_vec = (*iov);
+
+ ASSERT(shift >= 0);
+
+ while(shift > 0) {
+ ASSERT(head_vec < &(*iov)[*iovlen]);
+
+ if(shift < head_vec->iov_len) {
+ head_vec->iov_base = (char*)head_vec->iov_base + shift;
+ head_vec->iov_len -= shift;
+ break;
+ } else {
+ shift -= head_vec->iov_len;
+ head_vec++;
+ }
+ }
+
+ (*iovlen) -= head_vec - (*iov);
+ (*iov) = head_vec;
+}
+
+typedef BOOL (WINAPI *io_op_t)(HANDLE, LPVOID, DWORD, LPDWORD, LPOVERLAPPED);
+
+static Sint64 internal_sync_io(efile_win_t *w, io_op_t operation,
+ SysIOVec *iov, int iovlen, OVERLAPPED *overlapped) {
+
+ Sint64 bytes_processed = 0;
+
+ for(;;) {
+ DWORD block_bytes_processed, last_error;
+ BOOL succeeded;
+
+ if(iovlen < 1) {
+ return bytes_processed;
+ }
+
+ succeeded = operation(w->handle, iov->iov_base, iov->iov_len,
+ &block_bytes_processed, overlapped);
+ last_error = GetLastError();
+
+ if(!succeeded && (last_error != ERROR_HANDLE_EOF)) {
+ w->common.posix_errno = windows_to_posix_errno(last_error);
+ return -1;
+ } else if(block_bytes_processed == 0) {
+ /* EOF */
+ return bytes_processed;
+ }
+
+ if(overlapped != NULL) {
+ shift_overlapped(overlapped, block_bytes_processed);
+ }
+
+ shift_iov(&iov, &iovlen, block_bytes_processed);
+
+ bytes_processed += block_bytes_processed;
+ }
+}
+
+Sint64 efile_readv(efile_data_t *d, SysIOVec *iov, int iovlen) {
+ efile_win_t *w = (efile_win_t*)d;
+
+ return internal_sync_io(w, ReadFile, iov, iovlen, NULL);
+}
+
+Sint64 efile_writev(efile_data_t *d, SysIOVec *iov, int iovlen) {
+ efile_win_t *w = (efile_win_t*)d;
+
+ OVERLAPPED __overlapped, *overlapped;
+ Uint64 bytes_written;
+
+ if(w->common.modes & EFILE_MODE_APPEND) {
+ overlapped = &__overlapped;
+
+ sys_memset(overlapped, 0, sizeof(*overlapped));
+ overlapped->OffsetHigh = 0xFFFFFFFF;
+ overlapped->Offset = 0xFFFFFFFF;
+ } else {
+ overlapped = NULL;
+ }
+
+ return internal_sync_io(w, WriteFile, iov, iovlen, overlapped);
+}
+
+Sint64 efile_preadv(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen) {
+ efile_win_t *w = (efile_win_t*)d;
+
+ OVERLAPPED overlapped;
+
+ sys_memset(&overlapped, 0, sizeof(overlapped));
+ overlapped.OffsetHigh = (offset >> 32) & 0xFFFFFFFF;
+ overlapped.Offset = offset & 0xFFFFFFFF;
+
+ return internal_sync_io(w, ReadFile, iov, iovlen, &overlapped);
+}
+
+Sint64 efile_pwritev(efile_data_t *d, Sint64 offset, SysIOVec *iov, int iovlen) {
+ efile_win_t *w = (efile_win_t*)d;
+
+ OVERLAPPED overlapped;
+
+ sys_memset(&overlapped, 0, sizeof(overlapped));
+ overlapped.OffsetHigh = (offset >> 32) & 0xFFFFFFFF;
+ overlapped.Offset = offset & 0xFFFFFFFF;
+
+ return internal_sync_io(w, WriteFile, iov, iovlen, &overlapped);
+}
+
+int efile_seek(efile_data_t *d, enum efile_seek_t seek, Sint64 offset, Sint64 *new_position) {
+ efile_win_t *w = (efile_win_t*)d;
+
+ LARGE_INTEGER large_offset, large_new_position;
+ DWORD whence;
+
+ switch(seek) {
+ case EFILE_SEEK_BOF: whence = FILE_BEGIN; break;
+ case EFILE_SEEK_CUR: whence = FILE_CURRENT; break;
+ case EFILE_SEEK_EOF: whence = FILE_END; break;
+ default: ERTS_INTERNAL_ERROR("Invalid seek parameter");
+ }
+
+ large_offset.QuadPart = offset;
+
+ if(!SetFilePointerEx(w->handle, large_offset, &large_new_position, whence)) {
+ w->common.posix_errno = windows_to_posix_errno(GetLastError());
+ return 0;
+ }
+
+ (*new_position) = large_new_position.QuadPart;
+
+ return 1;
+}
+
+int efile_sync(efile_data_t *d, int data_only) {
+ efile_win_t *w = (efile_win_t*)d;
+
+ /* Windows doesn't support data-only syncing. */
+ (void)data_only;
+
+ if(!FlushFileBuffers(w->handle)) {
+ w->common.posix_errno = windows_to_posix_errno(GetLastError());
+ return 0;
+ }
+
+ return 1;
+}
+
+int efile_advise(efile_data_t *d, Sint64 offset, Sint64 length, enum efile_advise_t advise) {
+ /* Windows doesn't support this, but we'll pretend it does since the call
+ * is only a recommendation even on systems that do support it. */
+
+ (void)d;
+ (void)offset;
+ (void)length;
+ (void)advise;
+
+ return 1;
+}
+
+int efile_allocate(efile_data_t *d, Sint64 offset, Sint64 length) {
+ efile_win_t *w = (efile_win_t*)d;
+
+ (void)d;
+ (void)offset;
+ (void)length;
+
+ w->common.posix_errno = ENOTSUP;
+
+ return 0;
+}
+
+int efile_truncate(efile_data_t *d) {
+ efile_win_t *w = (efile_win_t*)d;
+
+ if(!SetEndOfFile(w->handle)) {
+ w->common.posix_errno = windows_to_posix_errno(GetLastError());
+ return 0;
+ }
+
+ return 1;
+}
+
+static int is_executable_file(const efile_path_t *path) {
+ /* We're using the file extension in order to be quirks-compliant with the
+ * old driver, which never bothered to check the actual permissions. We
+ * could easily do so now (cf. GetNamedSecurityInfo) but the execute
+ * permission is only relevant for files that are started with the default
+ * loader, and batch files run just fine with read permission alone. */
+
+ int length = PATH_LENGTH(path);
+
+ if(length >= 4) {
+ const WCHAR *last_four = &((WCHAR*)path->data)[length - 4];
+
+ if (!_wcsicmp(last_four, L".exe") ||
+ !_wcsicmp(last_four, L".cmd") ||
+ !_wcsicmp(last_four, L".bat") ||
+ !_wcsicmp(last_four, L".com")) {
+ return 1;
+ }
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_read_info(const efile_path_t *path, int follow_links, efile_fileinfo_t *result) {
+ BY_HANDLE_FILE_INFORMATION native_file_info;
+ DWORD attributes;
+
+ sys_memset(&native_file_info, 0, sizeof(native_file_info));
+
+ attributes = GetFileAttributesW((WCHAR*)path->data);
+
+ if(attributes == INVALID_FILE_ATTRIBUTES) {
+ DWORD last_error = GetLastError();
+
+ /* Querying a network share root fails with ERROR_BAD_NETPATH, so we'll
+ * fake it as a directory just like local roots. */
+ if(!is_path_root(path) || last_error != ERROR_BAD_NETPATH) {
+ return windows_to_posix_errno(last_error);
+ }
+
+ attributes = FILE_ATTRIBUTE_DIRECTORY;
+ } else if(is_path_root(path)) {
+ /* Local (or mounted) roots can be queried with GetFileAttributesW but
+ * lack support for GetFileInformationByHandle, so we'll skip that
+ * part. */
+ } else {
+ HANDLE handle;
+
+ if(follow_links && (attributes & FILE_ATTRIBUTE_REPARSE_POINT)) {
+ posix_errno_t posix_errno;
+ efile_path_t resolved_path;
+
+ posix_errno = internal_read_link(path, &resolved_path);
+
+ if(posix_errno != 0) {
+ return posix_errno;
+ }
+
+ return efile_read_info(&resolved_path, 0, result);
+ }
+
+ handle = CreateFileW((const WCHAR*)path->data, GENERIC_READ,
+ FILE_SHARE_FLAGS, NULL, OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS,
+ NULL);
+
+ /* The old driver never cared whether this succeeded. */
+ if(handle != INVALID_HANDLE_VALUE) {
+ GetFileInformationByHandle(handle, &native_file_info);
+ CloseHandle(handle);
+ }
+
+ FILETIME_TO_EPOCH(result->m_time, native_file_info.ftLastWriteTime);
+ FILETIME_TO_EPOCH(result->a_time, native_file_info.ftLastAccessTime);
+ FILETIME_TO_EPOCH(result->c_time, native_file_info.ftCreationTime);
+
+ if(result->m_time == -EPOCH_DIFFERENCE) {
+ /* Default to 1970 just like the old driver. */
+ result->m_time = 0;
+ }
+
+ if(result->a_time == -EPOCH_DIFFERENCE) {
+ result->a_time = result->m_time;
+ }
+
+ if(result->c_time == -EPOCH_DIFFERENCE) {
+ result->c_time = result->m_time;
+ }
+ }
+
+ if(attributes & FILE_ATTRIBUTE_REPARSE_POINT) {
+ result->type = EFILE_FILETYPE_SYMLINK;
+ /* This should be _S_IFLNK, but the old driver always set
+ * non-directories to _S_IFREG. */
+ result->mode |= _S_IFREG;
+ } else if(attributes & FILE_ATTRIBUTE_DIRECTORY) {
+ result->type = EFILE_FILETYPE_DIRECTORY;
+ result->mode |= _S_IFDIR | _S_IEXEC;
+ } else {
+ if(is_executable_file(path)) {
+ result->mode |= _S_IEXEC;
+ }
+
+ result->type = EFILE_FILETYPE_REGULAR;
+ result->mode |= _S_IFREG;
+ }
+
+ if(!(attributes & FILE_ATTRIBUTE_READONLY)) {
+ result->access = EFILE_ACCESS_READ | EFILE_ACCESS_WRITE;
+ result->mode |= _S_IREAD | _S_IWRITE;
+ } else {
+ result->access = EFILE_ACCESS_READ;
+ result->mode |= _S_IREAD;
+ }
+
+ /* Propagate user mode-bits to group/other fields */
+ result->mode |= (result->mode & 0700) >> 3;
+ result->mode |= (result->mode & 0700) >> 6;
+
+ result->size =
+ ((Uint64)native_file_info.nFileSizeHigh << 32ull) |
+ (Uint64)native_file_info.nFileSizeLow;
+
+ result->links = MAX(1, native_file_info.nNumberOfLinks);
+
+ result->major_device = get_drive_number(path);
+ result->minor_device = 0;
+ result->inode = 0;
+ result->uid = 0;
+ result->gid = 0;
+
+ return 0;
+}
+
+posix_errno_t efile_set_permissions(const efile_path_t *path, Uint32 permissions) {
+ DWORD attributes = GetFileAttributesW((WCHAR*)path->data);
+
+ if(attributes == INVALID_FILE_ATTRIBUTES) {
+ return windows_to_posix_errno(GetLastError());
+ }
+
+ if(permissions & _S_IWRITE) {
+ attributes &= ~FILE_ATTRIBUTE_READONLY;
+ } else {
+ attributes |= FILE_ATTRIBUTE_READONLY;
+ }
+
+ if(SetFileAttributesW((WCHAR*)path->data, attributes)) {
+ return 0;
+ }
+
+ return windows_to_posix_errno(GetLastError());
+}
+
+posix_errno_t efile_set_owner(const efile_path_t *path, Uint32 owner, Uint32 group) {
+ (void)path;
+ (void)owner;
+ (void)group;
+
+ return 0;
+}
+
+posix_errno_t efile_set_time(const efile_path_t *path, Sint64 a_time, Sint64 m_time, Sint64 c_time) {
+ FILETIME accessed, modified, created;
+ DWORD last_error, attributes;
+ HANDLE handle;
+
+ attributes = GetFileAttributesW((WCHAR*)path->data);
+
+ if(attributes == INVALID_FILE_ATTRIBUTES) {
+ return windows_to_posix_errno(GetLastError());
+ }
+
+ /* If the file is read-only, we have to make it temporarily writable while
+ * setting new metadata. */
+ if(attributes & FILE_ATTRIBUTE_READONLY) {
+ DWORD without_readonly = attributes & ~FILE_ATTRIBUTE_READONLY;
+
+ if(!SetFileAttributesW((WCHAR*)path->data, without_readonly)) {
+ return windows_to_posix_errno(GetLastError());
+ }
+ }
+
+ EPOCH_TO_FILETIME(modified, m_time);
+ EPOCH_TO_FILETIME(accessed, a_time);
+ EPOCH_TO_FILETIME(created, c_time);
+
+ handle = CreateFileW((WCHAR*)path->data, GENERIC_READ | GENERIC_WRITE,
+ FILE_SHARE_FLAGS, NULL, OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL);
+ last_error = GetLastError();
+
+ if(handle != INVALID_HANDLE_VALUE) {
+ if(SetFileTime(handle, &created, &accessed, &modified)) {
+ last_error = ERROR_SUCCESS;
+ } else {
+ last_error = GetLastError();
+ }
+
+ CloseHandle(handle);
+ }
+
+ if(attributes & FILE_ATTRIBUTE_READONLY) {
+ SetFileAttributesW((WCHAR*)path->data, attributes);
+ }
+
+ return windows_to_posix_errno(last_error);
+}
+
+static posix_errno_t internal_read_link(const efile_path_t *path, efile_path_t *result) {
+ DWORD required_length, actual_length;
+ HANDLE link_handle;
+ DWORD last_error;
+
+ link_handle = CreateFileW((WCHAR*)path->data, GENERIC_READ,
+ FILE_SHARE_FLAGS, NULL, OPEN_EXISTING, FILE_FLAG_BACKUP_SEMANTICS, NULL);
+ last_error = GetLastError();
+
+ if(link_handle == INVALID_HANDLE_VALUE) {
+ return windows_to_posix_errno(last_error);
+ }
+
+ required_length = GetFinalPathNameByHandleW(link_handle, NULL, 0, 0);
+ last_error = GetLastError();
+
+ if(required_length <= 0) {
+ CloseHandle(link_handle);
+ return windows_to_posix_errno(last_error);
+ }
+
+ /* Unlike many other path functions (eg. GetFullPathNameW), this one
+ * includes the NUL terminator in its required length. */
+ if(!enif_alloc_binary(required_length * sizeof(WCHAR), result)) {
+ CloseHandle(link_handle);
+ return ENOMEM;
+ }
+
+ actual_length = GetFinalPathNameByHandleW(link_handle,
+ (WCHAR*)result->data, required_length, 0);
+ last_error = GetLastError();
+
+ CloseHandle(link_handle);
+
+ if(actual_length == 0 || actual_length >= required_length) {
+ enif_release_binary(result);
+ return windows_to_posix_errno(last_error);
+ }
+
+ /* GetFinalPathNameByHandle always prepends with "\\?\" and NUL-terminates,
+ * so we never have to touch-up the resulting path. */
+
+ ASSERT_PATH_FORMAT(result);
+
+ return 0;
+}
+
+posix_errno_t efile_read_link(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result) {
+ posix_errno_t posix_errno;
+ ErlNifBinary result_bin;
+ DWORD attributes;
+
+ ASSERT_PATH_FORMAT(path);
+
+ attributes = GetFileAttributesW((WCHAR*)path->data);
+
+ if(attributes == INVALID_FILE_ATTRIBUTES) {
+ return windows_to_posix_errno(GetLastError());
+ } else if(!(attributes & FILE_ATTRIBUTE_REPARSE_POINT)) {
+ return EINVAL;
+ }
+
+ posix_errno = internal_read_link(path, &result_bin);
+
+ if(posix_errno == 0) {
+ if(!normalize_path_result(&result_bin)) {
+ enif_release_binary(&result_bin);
+ return ENOMEM;
+ }
+
+ (*result) = enif_make_binary(env, &result_bin);
+ }
+
+ return posix_errno;
+}
+
+posix_errno_t efile_list_dir(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result) {
+ ERL_NIF_TERM list_head;
+ WIN32_FIND_DATAW data;
+ HANDLE search_handle;
+ WCHAR *search_path;
+ DWORD last_error;
+
+ ASSERT_PATH_FORMAT(path);
+
+ search_path = enif_alloc(path->size + 2 * sizeof(WCHAR));
+
+ if(search_path == NULL) {
+ return ENOMEM;
+ }
+
+ sys_memcpy(search_path, path->data, path->size);
+ search_path[PATH_LENGTH(path) + 0] = L'\\';
+ search_path[PATH_LENGTH(path) + 1] = L'*';
+ search_path[PATH_LENGTH(path) + 2] = L'\0';
+
+ search_handle = FindFirstFileW(search_path, &data);
+ last_error = GetLastError();
+
+ enif_free(search_path);
+
+ if(search_handle == INVALID_HANDLE_VALUE) {
+ return windows_to_posix_errno(last_error);
+ }
+
+ list_head = enif_make_list(env, 0);
+
+ do {
+ int name_length = wcslen(data.cFileName);
+
+ if(!is_ignored_name(name_length, data.cFileName)) {
+ unsigned char *name_bytes;
+ ERL_NIF_TERM name_term;
+ size_t name_size;
+
+ name_size = name_length * sizeof(WCHAR);
+
+ name_bytes = enif_make_new_binary(env, name_size, &name_term);
+ sys_memcpy(name_bytes, data.cFileName, name_size);
+
+ list_head = enif_make_list_cell(env, name_term, list_head);
+ }
+ } while(FindNextFileW(search_handle, &data));
+
+ FindClose(search_handle);
+ (*result) = list_head;
+
+ return 0;
+}
+
+posix_errno_t efile_rename(const efile_path_t *old_path, const efile_path_t *new_path) {
+ BOOL old_is_directory, new_is_directory;
+ DWORD move_flags, last_error;
+
+ ASSERT_PATH_FORMAT(old_path);
+ ASSERT_PATH_FORMAT(new_path);
+
+ move_flags = MOVEFILE_COPY_ALLOWED | MOVEFILE_WRITE_THROUGH;
+
+ if(MoveFileExW((WCHAR*)old_path->data, (WCHAR*)new_path->data, move_flags)) {
+ return 0;
+ }
+
+ last_error = GetLastError();
+
+ old_is_directory = has_file_attributes(old_path, FILE_ATTRIBUTE_DIRECTORY);
+ new_is_directory = has_file_attributes(new_path, FILE_ATTRIBUTE_DIRECTORY);
+
+ switch(last_error) {
+ case ERROR_SHARING_VIOLATION:
+ case ERROR_ACCESS_DENIED:
+ if(old_is_directory) {
+ BOOL moved_into_itself;
+
+ moved_into_itself = (old_path->size <= new_path->size) &&
+ !_wcsnicmp((WCHAR*)old_path->data, (WCHAR*)new_path->data,
+ PATH_LENGTH(old_path));
+
+ if(moved_into_itself) {
+ return EINVAL;
+ } else if(is_path_root(old_path)) {
+ return EINVAL;
+ }
+
+ /* Renaming a directory across volumes needs to be rewritten as
+ * EXDEV so that the caller can respond by simulating it with
+ * copy/delete operations.
+ *
+ * Files are handled through MOVEFILE_COPY_ALLOWED. */
+ if(!has_same_mount_point(old_path, new_path)) {
+ return EXDEV;
+ }
+ }
+ break;
+ case ERROR_PATH_NOT_FOUND:
+ case ERROR_FILE_NOT_FOUND:
+ return ENOENT;
+ case ERROR_ALREADY_EXISTS:
+ case ERROR_FILE_EXISTS:
+ if(old_is_directory && !new_is_directory) {
+ return ENOTDIR;
+ } else if(!old_is_directory && new_is_directory) {
+ return EISDIR;
+ } else if(old_is_directory && new_is_directory) {
+ /* This will fail if the destination isn't empty. */
+ if(RemoveDirectoryW((WCHAR*)new_path->data)) {
+ return efile_rename(old_path, new_path);
+ }
+
+ return EEXIST;
+ } else if(!old_is_directory && !new_is_directory) {
+ /* This is pretty iffy; the public documentation says that the
+ * operation may EACCES on some systems when either file is open,
+ * which gives us room to use MOVEFILE_REPLACE_EXISTING and be done
+ * with it, but the old implementation simulated Unix semantics and
+ * there's a lot of code that relies on that.
+ *
+ * The simulation renames the destination to a scratch name to get
+ * around the fact that it's impossible to open (and by extension
+ * rename) a file that's been deleted while open. It has a few
+ * drawbacks though;
+ *
+ * 1) It's not atomic as there's a small window where there's no
+ * file at all on the destination path.
+ * 2) It will confuse applications that subscribe to folder
+ * changes.
+ * 3) It will fail if we lack general permission to write in the
+ * same folder. */
+
+ WCHAR *swap_path = enif_alloc(new_path->size + sizeof(WCHAR) * 64);
+
+ if(swap_path == NULL) {
+ return ENOMEM;
+ } else {
+ static LONGLONG unique_counter = 0;
+ WCHAR *swap_path_end;
+
+ /* We swap in the same folder as the destination to be
+ * reasonably sure that it's on the same volume. Note that
+ * we're avoiding GetTempFileNameW as it will fail on long
+ * paths. */
+
+ sys_memcpy(swap_path, (WCHAR*)new_path->data, new_path->size);
+ swap_path_end = swap_path + PATH_LENGTH(new_path);
+
+ while(!IS_SLASH(*swap_path_end)) {
+ ASSERT(swap_path_end > swap_path);
+ swap_path_end--;
+ }
+
+ StringCchPrintfW(&swap_path_end[1], 64, L"erl-%lx-%llx.tmp",
+ GetCurrentProcessId(), unique_counter);
+ InterlockedIncrement64(&unique_counter);
+ }
+
+ if(MoveFileExW((WCHAR*)new_path->data, swap_path, MOVEFILE_REPLACE_EXISTING)) {
+ if(MoveFileExW((WCHAR*)old_path->data, (WCHAR*)new_path->data, move_flags)) {
+ last_error = ERROR_SUCCESS;
+ DeleteFileW(swap_path);
+ } else {
+ last_error = GetLastError();
+ MoveFileW(swap_path, (WCHAR*)new_path->data);
+ }
+ } else {
+ last_error = GetLastError();
+ DeleteFileW(swap_path);
+ }
+
+ enif_free(swap_path);
+
+ return windows_to_posix_errno(last_error);
+ }
+
+ return EEXIST;
+ }
+
+ return windows_to_posix_errno(last_error);
+}
+
+posix_errno_t efile_make_hard_link(const efile_path_t *existing_path, const efile_path_t *new_path) {
+ ASSERT_PATH_FORMAT(existing_path);
+ ASSERT_PATH_FORMAT(new_path);
+
+ if(!CreateHardLinkW((WCHAR*)new_path->data, (WCHAR*)existing_path->data, NULL)) {
+ return windows_to_posix_errno(GetLastError());
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_make_soft_link(const efile_path_t *existing_path, const efile_path_t *new_path) {
+ DWORD link_flags;
+
+ ASSERT_PATH_FORMAT(existing_path);
+ ASSERT_PATH_FORMAT(new_path);
+
+ if(has_file_attributes(existing_path, FILE_ATTRIBUTE_DIRECTORY)) {
+ link_flags = SYMBOLIC_LINK_FLAG_DIRECTORY;
+ } else {
+ link_flags = 0;
+ }
+
+ if(!CreateSymbolicLinkW((WCHAR*)new_path->data, (WCHAR*)existing_path->data, link_flags)) {
+ return windows_to_posix_errno(GetLastError());
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_make_dir(const efile_path_t *path) {
+ ASSERT_PATH_FORMAT(path);
+
+ if(!CreateDirectoryW((WCHAR*)path->data, NULL)) {
+ return windows_to_posix_errno(GetLastError());
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_del_file(const efile_path_t *path) {
+ ASSERT_PATH_FORMAT(path);
+
+ if(!DeleteFileW((WCHAR*)path->data)) {
+ DWORD last_error = GetLastError();
+
+ switch(last_error) {
+ case ERROR_INVALID_NAME:
+ /* Attempted to delete a device or similar. */
+ return EACCES;
+ case ERROR_ACCESS_DENIED:
+ /* Windows NT reports removing a directory as EACCES instead of
+ * EPERM. */
+ if(has_file_attributes(path, FILE_ATTRIBUTE_DIRECTORY)) {
+ return EPERM;
+ }
+ break;
+ }
+
+ return windows_to_posix_errno(last_error);
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_del_dir(const efile_path_t *path) {
+ ASSERT_PATH_FORMAT(path);
+
+ if(!RemoveDirectoryW((WCHAR*)path->data)) {
+ DWORD last_error = GetLastError();
+
+ if(last_error == ERROR_DIRECTORY) {
+ return ENOTDIR;
+ }
+
+ return windows_to_posix_errno(last_error);
+ }
+
+ return 0;
+}
+
+posix_errno_t efile_set_cwd(const efile_path_t *path) {
+ const WCHAR *path_start;
+
+ ASSERT_PATH_FORMAT(path);
+
+ /* We have to use _wchdir since that's the only function that updates the
+ * per-drive working directory, but it naively assumes that all paths
+ * starting with \\ are UNC paths, so we have to skip the \\?\-prefix. */
+ path_start = (WCHAR*)path->data + LP_PREFIX_LENGTH;
+
+ if(_wchdir(path_start)) {
+ return windows_to_posix_errno(GetLastError());
+ }
+
+ return 0;
+}
+
+static int is_valid_drive(int device_index) {
+ WCHAR drive_path[4] = {L'?', L':', L'\\', L'\0'};
+
+ if(device_index == 0) {
+ /* Default drive; always valid. */
+ return 1;
+ } else if(device_index > (L'Z' - L'A' + 1)) {
+ return 0;
+ }
+
+ drive_path[0] = device_index + L'A' - 1;
+
+ switch(GetDriveTypeW(drive_path)) {
+ case DRIVE_NO_ROOT_DIR:
+ case DRIVE_UNKNOWN:
+ return 0;
+ }
+
+ return 1;
+}
+
+posix_errno_t efile_get_device_cwd(ErlNifEnv *env, int device_index, ERL_NIF_TERM *result) {
+ ErlNifBinary result_bin;
+
+ /* _wgetdcwd might crash the entire emulator on debug builds since the CRT
+ * invalid parameter handler asserts if passed a non-existent drive (Or
+ * simply one that has been unmounted), so we check it ourselves to avoid
+ * that. */
+ if(!is_valid_drive(device_index)) {
+ return EACCES;
+ }
+
+ if(!enif_alloc_binary(MAX_PATH * sizeof(WCHAR), &result_bin)) {
+ return ENOMEM;
+ }
+
+ if(_wgetdcwd(device_index, (WCHAR*)result_bin.data, MAX_PATH) == NULL) {
+ enif_release_binary(&result_bin);
+ return EACCES;
+ }
+
+ if(!normalize_path_result(&result_bin)) {
+ enif_release_binary(&result_bin);
+ return ENOMEM;
+ }
+
+ (*result) = enif_make_binary(env, &result_bin);
+
+ return 0;
+}
+
+posix_errno_t efile_get_cwd(ErlNifEnv *env, ERL_NIF_TERM *result) {
+ return efile_get_device_cwd(env, 0, result);
+}
+
+posix_errno_t efile_altname(ErlNifEnv *env, const efile_path_t *path, ERL_NIF_TERM *result) {
+ ErlNifBinary result_bin;
+
+ ASSERT_PATH_FORMAT(path);
+
+ if(is_path_root(path)) {
+ /* Root paths can't be queried so we'll just return them as they are. */
+ if(!enif_alloc_binary(path->size, &result_bin)) {
+ return ENOMEM;
+ }
+
+ sys_memcpy(result_bin.data, path->data, path->size);
+ } else {
+ WIN32_FIND_DATAW data;
+ HANDLE handle;
+
+ WCHAR *name_buffer;
+ int name_length;
+
+ /* Reject path wildcards. */
+ if(wcspbrk(&((const WCHAR*)path->data)[4], L"?*")) {
+ return ENOENT;
+ }
+
+ handle = FindFirstFileW((const WCHAR*)path->data, &data);
+
+ if(handle == INVALID_HANDLE_VALUE) {
+ return windows_to_posix_errno(GetLastError());
+ }
+
+ FindClose(handle);
+
+ name_length = wcslen(data.cAlternateFileName);
+
+ if(name_length > 0) {
+ name_buffer = data.cAlternateFileName;
+ } else {
+ name_length = wcslen(data.cFileName);
+ name_buffer = data.cFileName;
+ }
+
+ /* Include NUL-terminator; it will be removed after normalization. */
+ name_length += 1;
+
+ if(!enif_alloc_binary(name_length * sizeof(WCHAR), &result_bin)) {
+ return ENOMEM;
+ }
+
+ sys_memcpy(result_bin.data, name_buffer, name_length * sizeof(WCHAR));
+ }
+
+ if(!normalize_path_result(&result_bin)) {
+ enif_release_binary(&result_bin);
+ return ENOMEM;
+ }
+
+ (*result) = enif_make_binary(env, &result_bin);
+
+ return 0;
+}
+
+static int windows_to_posix_errno(DWORD last_error) {
+ switch(last_error) {
+ case ERROR_SUCCESS:
+ return 0;
+ case ERROR_INVALID_FUNCTION:
+ case ERROR_INVALID_DATA:
+ case ERROR_INVALID_PARAMETER:
+ case ERROR_INVALID_TARGET_HANDLE:
+ case ERROR_INVALID_CATEGORY:
+ case ERROR_NEGATIVE_SEEK:
+ return EINVAL;
+ case ERROR_DIR_NOT_EMPTY:
+ return EEXIST;
+ case ERROR_BAD_FORMAT:
+ return ENOEXEC;
+ case ERROR_PATH_NOT_FOUND:
+ case ERROR_FILE_NOT_FOUND:
+ case ERROR_NO_MORE_FILES:
+ return ENOENT;
+ case ERROR_TOO_MANY_OPEN_FILES:
+ return EMFILE;
+ case ERROR_ACCESS_DENIED:
+ case ERROR_INVALID_ACCESS:
+ case ERROR_CURRENT_DIRECTORY:
+ case ERROR_SHARING_VIOLATION:
+ case ERROR_LOCK_VIOLATION:
+ case ERROR_INVALID_PASSWORD:
+ case ERROR_DRIVE_LOCKED:
+ return EACCES;
+ case ERROR_INVALID_HANDLE:
+ return EBADF;
+ case ERROR_NOT_ENOUGH_MEMORY:
+ case ERROR_OUTOFMEMORY:
+ case ERROR_OUT_OF_STRUCTURES:
+ return ENOMEM;
+ case ERROR_INVALID_DRIVE:
+ case ERROR_BAD_UNIT:
+ case ERROR_NOT_READY:
+ case ERROR_REM_NOT_LIST:
+ case ERROR_DUP_NAME:
+ case ERROR_BAD_NETPATH:
+ case ERROR_NETWORK_BUSY:
+ case ERROR_DEV_NOT_EXIST:
+ case ERROR_BAD_NET_NAME:
+ return ENXIO;
+ case ERROR_NOT_SAME_DEVICE:
+ return EXDEV;
+ case ERROR_WRITE_PROTECT:
+ return EROFS;
+ case ERROR_BAD_LENGTH:
+ case ERROR_BUFFER_OVERFLOW:
+ return E2BIG;
+ case ERROR_SEEK:
+ case ERROR_SECTOR_NOT_FOUND:
+ return ESPIPE;
+ case ERROR_NOT_DOS_DISK:
+ return ENODEV;
+ case ERROR_GEN_FAILURE:
+ return ENODEV;
+ case ERROR_SHARING_BUFFER_EXCEEDED:
+ case ERROR_NO_MORE_SEARCH_HANDLES:
+ return EMFILE;
+ case ERROR_HANDLE_EOF:
+ case ERROR_BROKEN_PIPE:
+ return EPIPE;
+ case ERROR_HANDLE_DISK_FULL:
+ case ERROR_DISK_FULL:
+ return ENOSPC;
+ case ERROR_NOT_SUPPORTED:
+ return ENOTSUP;
+ case ERROR_FILE_EXISTS:
+ case ERROR_ALREADY_EXISTS:
+ case ERROR_CANNOT_MAKE:
+ return EEXIST;
+ case ERROR_ALREADY_ASSIGNED:
+ return EBUSY;
+ case ERROR_NO_PROC_SLOTS:
+ return EAGAIN;
+ case ERROR_CANT_RESOLVE_FILENAME:
+ return EMLINK;
+ case ERROR_PRIVILEGE_NOT_HELD:
+ return EPERM;
+ case ERROR_ARENA_TRASHED:
+ case ERROR_INVALID_BLOCK:
+ case ERROR_BAD_ENVIRONMENT:
+ case ERROR_BAD_COMMAND:
+ case ERROR_CRC:
+ case ERROR_OUT_OF_PAPER:
+ case ERROR_READ_FAULT:
+ case ERROR_WRITE_FAULT:
+ case ERROR_WRONG_DISK:
+ case ERROR_NET_WRITE_FAULT:
+ return EIO;
+ default: /* not to do with files I expect. */
+ return EIO;
+ }
+}
diff --git a/erts/preloaded/ebin/erl_prim_loader.beam b/erts/preloaded/ebin/erl_prim_loader.beam
index 198032c18d..2d86dd3843 100644
--- a/erts/preloaded/ebin/erl_prim_loader.beam
+++ b/erts/preloaded/ebin/erl_prim_loader.beam
Binary files differ
diff --git a/erts/preloaded/ebin/init.beam b/erts/preloaded/ebin/init.beam
index fb4fc67148..a2dd41b435 100644
--- a/erts/preloaded/ebin/init.beam
+++ b/erts/preloaded/ebin/init.beam
Binary files differ
diff --git a/erts/preloaded/ebin/prim_file.beam b/erts/preloaded/ebin/prim_file.beam
index eb326ac4b6..902b0945c6 100644
--- a/erts/preloaded/ebin/prim_file.beam
+++ b/erts/preloaded/ebin/prim_file.beam
Binary files differ
diff --git a/erts/preloaded/ebin/prim_inet.beam b/erts/preloaded/ebin/prim_inet.beam
index 32b104ae95..a42f45c180 100644
--- a/erts/preloaded/ebin/prim_inet.beam
+++ b/erts/preloaded/ebin/prim_inet.beam
Binary files differ
diff --git a/erts/preloaded/src/Makefile b/erts/preloaded/src/Makefile
index bb8d272793..a2872082f2 100644
--- a/erts/preloaded/src/Makefile
+++ b/erts/preloaded/src/Makefile
@@ -36,6 +36,7 @@ include $(ERL_TOP)/lib/kernel/vsn.mk
PRE_LOADED_ERL_MODULES = \
erl_prim_loader \
init \
+ prim_buffer \
prim_file \
prim_inet \
zlib \
@@ -46,8 +47,7 @@ PRE_LOADED_ERL_MODULES = \
erts_internal \
erl_tracer \
erts_literal_area_collector \
- erts_dirty_process_code_checker \
- prim_buffer
+ erts_dirty_process_code_checker
PRE_LOADED_BEAM_MODULES = \
prim_eval
@@ -117,7 +117,7 @@ prim_eval.beam: prim_eval.S prim_eval.abstr
# Include dependencies -- list below added by PaN
$(EBIN)/erl_prim_loader.beam: $(KERNEL_SRC)/inet_boot.hrl $(KERNEL_INCLUDE)/file.hrl
-$(EBIN)/prim_file.beam: $(KERNEL_INCLUDE)/file.hrl
+$(EBIN)/prim_file.beam: $(KERNEL_SRC)/file_int.hrl $(KERNEL_INCLUDE)/file.hrl
$(EBIN)/prim_inet.beam: $(KERNEL_SRC)/inet_int.hrl $(KERNEL_INCLUDE)/inet_sctp.hrl
$(EBIN)/prim_zip.beam: zip_internal.hrl $(KERNEL_INCLUDE)/file.hrl $(STDLIB_INCLUDE)/zip.hrl
$(EBIN)/init.erl: $(KERNEL_INCLUDE)/file.hrl
diff --git a/erts/preloaded/src/erl_prim_loader.erl b/erts/preloaded/src/erl_prim_loader.erl
index 1d09aeded9..d16fe5a0bc 100644
--- a/erts/preloaded/src/erl_prim_loader.erl
+++ b/erts/preloaded/src/erl_prim_loader.erl
@@ -151,9 +151,8 @@ start_inet(Parent) ->
loop(State, Parent, []).
start_efile(Parent) ->
- {ok, Port} = prim_file:start(),
%% Check that we started in a valid directory.
- case prim_file:get_cwd(Port) of
+ case prim_file:get_cwd() of
{error, _} ->
%% At this point in the startup, we have no error_logger at all.
Report = "Invalid current directory or invalid filename "
@@ -165,7 +164,7 @@ start_efile(Parent) ->
end,
PS = prim_init(),
State = #state {loader = efile,
- data = Port,
+ data = noport,
timeout = ?EFILE_IDLE_TIMEOUT,
prim_state = PS},
loop(State, Parent, []).
@@ -401,12 +400,12 @@ handle_get_cwd(State = #state{loader = inet}, Drive) ->
?SAFE2(inet_get_cwd(State, Drive), State).
handle_stop(State = #state{loader = efile}) ->
- efile_stop_port(State);
+ State;
handle_stop(State = #state{loader = inet}) ->
inet_stop_port(State).
-handle_exit(State = #state{loader = efile}, Who, Reason) ->
- efile_exit_port(State, Who, Reason);
+handle_exit(State = #state{loader = efile}, _Who, _Reason) ->
+ State;
handle_exit(State = #state{loader = inet}, Who, Reason) ->
inet_exit_port(State, Who, Reason).
@@ -475,15 +474,6 @@ efile_get_cwd(#state{prim_state = PS} = State, Drive) ->
{Res, PS2} = prim_get_cwd(PS, Drive),
{Res, State#state{prim_state = PS2}}.
-efile_stop_port(#state{data=Port}=State) ->
- prim_file:close(Port),
- State#state{data=noport}.
-
-efile_exit_port(State, Port, Reason) when State#state.data =:= Port ->
- exit({port_died,Reason});
-efile_exit_port(State, _Port, _Reason) ->
- State.
-
efile_timeout_handler(State, _Parent) ->
prim_purge_cache(),
State.
diff --git a/erts/preloaded/src/init.erl b/erts/preloaded/src/init.erl
index 34a9f6b8b9..679a2241d2 100644
--- a/erts/preloaded/src/init.erl
+++ b/erts/preloaded/src/init.erl
@@ -200,10 +200,11 @@ boot(BootArgs) ->
register(init, self()),
process_flag(trap_exit, true),
- %% Load the zlib nif
+ %% Load the static nifs
zlib:on_load(),
- %% Load the tracer nif
erl_tracer:on_load(),
+ prim_buffer:on_load(),
+ prim_file:on_load(),
{Start0,Flags,Args} = parse_boot_args(BootArgs),
%% We don't get to profile parsing of BootArgs
diff --git a/erts/preloaded/src/prim_file.erl b/erts/preloaded/src/prim_file.erl
index ab5359ebbc..35042a7c72 100644
--- a/erts/preloaded/src/prim_file.erl
+++ b/erts/preloaded/src/prim_file.erl
@@ -1,7 +1,7 @@
%%
%% %CopyrightBegin%
%%
-%% Copyright Ericsson AB 2000-2016. All Rights Reserved.
+%% Copyright Ericsson AB 2000-2017. All Rights Reserved.
%%
%% Licensed under the Apache License, Version 2.0 (the "License");
%% you may not use this file except in compliance with the License.
@@ -19,147 +19,45 @@
%%
-module(prim_file).
-%% Interface module to the file driver.
+-export([on_load/0]).
+-export([open/2, close/1,
+ sync/1, datasync/1, truncate/1, advise/4, allocate/3,
+ read_line/1, read/2, write/2, position/2,
+ pread/2, pread/3, pwrite/2, pwrite/3]).
+%% OTP internal.
+-export([ipread_s32bu_p32bu/3, sendfile/8, altname/1, get_handle/1]).
-%%% Interface towards a single file's contents. Uses ?FD_DRV.
+-export([read_file/1, write_file/2]).
-%% Generic file contents operations
--export([open/2, close/1, datasync/1, sync/1, advise/4, position/2, truncate/1,
- write/2, pwrite/2, pwrite/3, read/2, read_line/1, pread/2, pread/3,
- copy/3, sendfile/8, allocate/3]).
+-export([read_link/1, read_link_all/1,
+ read_link_info/1, read_link_info/2,
+ read_file_info/1, read_file_info/2,
+ write_file_info/2, write_file_info/3]).
-%% Specialized file operations
--export([open/1, open/3]).
--export([read_file/1, read_file/2, write_file/2]).
--export([ipread_s32bu_p32bu/3]).
+-export([list_dir/1, list_dir_all/1]).
+-export([get_cwd/0, get_cwd/1, set_cwd/1,
+ delete/1, rename/2,
+ make_dir/1, del_dir/1,
+ make_link/2, make_symlink/2]).
-
-%%% Interface towards file system and metadata. Uses ?DRV.
-
-%% Takes an optional port (opens a ?DRV port per default) as first argument.
--export([get_cwd/0, get_cwd/1, get_cwd/2,
- set_cwd/1, set_cwd/2,
- delete/1, delete/2,
- rename/2, rename/3,
- make_dir/1, make_dir/2,
- del_dir/1, del_dir/2,
- read_file_info/1, read_file_info/2, read_file_info/3,
- altname/1, altname/2,
- write_file_info/2, write_file_info/3, write_file_info/4,
- make_link/2, make_link/3,
- make_symlink/2, make_symlink/3,
- read_link/1, read_link/2, read_link_all/1, read_link_all/2,
- read_link_info/1, read_link_info/2, read_link_info/3,
- list_dir/1, list_dir/2, list_dir_all/1, list_dir_all/2]).
-%% How to start and stop the ?DRV port.
--export([start/0, stop/1]).
-
-%% Debug exports
--export([open_int/4, open_mode/1, open_mode/4]).
-
-%%%-----------------------------------------------------------------
-%%% Includes and defines
-
--include("file.hrl").
-
--define(DRV, "efile").
--define(FD_DRV, "efile").
-
+-define(MIN_READLINE_SIZE, 256).
-define(LARGEFILESIZE, (1 bsl 63)).
-%% Driver commands
--define(FILE_OPEN, 1).
--define(FILE_READ, 2).
--define(FILE_LSEEK, 3).
--define(FILE_WRITE, 4).
--define(FILE_FSTAT, 5).
--define(FILE_PWD, 6).
--define(FILE_READDIR, 7).
--define(FILE_CHDIR, 8).
--define(FILE_FSYNC, 9).
--define(FILE_MKDIR, 10).
--define(FILE_DELETE, 11).
--define(FILE_RENAME, 12).
--define(FILE_RMDIR, 13).
--define(FILE_TRUNCATE, 14).
--define(FILE_READ_FILE, 15).
--define(FILE_WRITE_INFO, 16).
--define(FILE_LSTAT, 19).
--define(FILE_READLINK, 20).
--define(FILE_LINK, 21).
--define(FILE_SYMLINK, 22).
--define(FILE_CLOSE, 23).
--define(FILE_PWRITEV, 24).
--define(FILE_PREADV, 25).
--define(FILE_SETOPT, 26).
--define(FILE_IPREAD, 27).
--define(FILE_ALTNAME, 28).
--define(FILE_READ_LINE, 29).
--define(FILE_FDATASYNC, 30).
--define(FILE_ADVISE, 31).
--define(FILE_SENDFILE, 32).
--define(FILE_ALLOCATE, 33).
-
-%% Driver responses
--define(FILE_RESP_OK, 0).
--define(FILE_RESP_ERROR, 1).
--define(FILE_RESP_DATA, 2).
--define(FILE_RESP_NUMBER, 3).
--define(FILE_RESP_INFO, 4).
--define(FILE_RESP_NUMERR, 5).
--define(FILE_RESP_LDATA, 6).
--define(FILE_RESP_N2DATA, 7).
--define(FILE_RESP_EOF, 8).
--define(FILE_RESP_FNAME, 9).
--define(FILE_RESP_ALL_DATA, 10).
--define(FILE_RESP_LFNAME, 11).
-
-%% Open modes for the driver's open function.
--define(EFILE_MODE_READ, 1).
--define(EFILE_MODE_WRITE, 2).
--define(EFILE_MODE_READ_WRITE, 3).
--define(EFILE_MODE_APPEND, 4).
--define(EFILE_COMPRESSED, 8).
--define(EFILE_MODE_EXCL, 16).
-%% Note: bit 5 (32) is used internally for VxWorks
--define(EFILE_MODE_SYNC, 64).
-
-%% Use this mask to get just the mode bits to be passed to the driver.
--define(EFILE_MODE_MASK, 127).
-
-%% Seek modes for the driver's seek function.
--define(EFILE_SEEK_SET, 0).
--define(EFILE_SEEK_CUR, 1).
--define(EFILE_SEEK_END, 2).
-
-%% Options
--define(FILE_OPT_DELAYED_WRITE, 0).
--define(FILE_OPT_READ_AHEAD, 1).
-
-%% IPREAD variants
--define(IPREAD_S32BU_P32BU, 0).
-
-%% POSIX file advises
--define(POSIX_FADV_NORMAL, 0).
--define(POSIX_FADV_RANDOM, 1).
--define(POSIX_FADV_SEQUENTIAL, 2).
--define(POSIX_FADV_WILLNEED, 3).
--define(POSIX_FADV_DONTNEED, 4).
--define(POSIX_FADV_NOREUSE, 5).
-
-%% Sendfile flags
--define(EFILE_SENDFILE_USE_THREADS, 1).
+-export([copy/3]).
+-include("file_int.hrl").
+
+-type prim_file_ref() :: term().
%%% BIFs
-export([internal_name2native/1,
internal_native2name/1,
internal_normalize_utf8/1,
- is_translatable/1]).
+ is_translatable/1]).
-type prim_file_name() :: string() | unicode:unicode_binary().
-type prim_file_name_error() :: 'error' | 'ignore' | 'warning'.
@@ -185,1316 +83,733 @@ internal_normalize_utf8(_) ->
is_translatable(_) ->
erlang:nif_error(undefined).
-%%% End of BIFs
-
-%%%-----------------------------------------------------------------
-%%% Functions operating on a file through a handle. ?FD_DRV.
-%%%
-%%% Generic file contents operations.
-%%%
-%%% Supposed to be called by applications through module file.
-
-
-%% Opens a file using the driver port Port. Returns {error, Reason}
-%% | {ok, FileDescriptor}
-open(Port, File, ModeList) when is_port(Port),
- (is_list(File) orelse is_binary(File)),
- is_list(ModeList) ->
- case open_mode(ModeList) of
- {Mode, _Portopts, _Setopts} ->
- open_int(Port, File, Mode, []);
- Reason ->
- {error, Reason}
- end;
-open(_,_,_) ->
- {error, badarg}.
-
-%% Opens a file. Returns {error, Reason} | {ok, FileDescriptor}.
-open(File, ModeList) when (is_list(File) orelse is_binary(File)),
- is_list(ModeList) ->
- case open_mode(ModeList) of
- {Mode, Portopts, Setopts} ->
- open_int({?FD_DRV, Portopts},File, Mode, Setopts);
- Reason ->
- {error, Reason}
- end;
-open(_, _) ->
- {error, badarg}.
+%%
-%% Opens a port that can be used for open/3 or read_file/2.
-%% Returns {ok, Port} | {error, Reason}.
-open(Portopts) when is_list(Portopts) ->
- drv_open(?FD_DRV, [binary|Portopts]);
-open(_) ->
- {error, badarg}.
+%% Returns {error, Reason} | {ok, BytesCopied}
+copy(#file_descriptor{module = ?MODULE} = Source,
+ #file_descriptor{module = ?MODULE} = Dest,
+ Length)
+ when is_integer(Length), Length >= 0;
+ is_atom(Length) ->
+ %% XXX Should be moved down to the driver for optimization.
+ file:copy_opened(Source, Dest, Length).
-open_int({Driver, Portopts}, File, Mode, Setopts) ->
- case drv_open(Driver, Portopts) of
- {ok, Port} ->
- open_int(Port, File, Mode, Setopts);
- {error, _} = Error ->
- Error
- end;
-open_int(Port, File, Mode, Setopts) ->
- M = Mode band ?EFILE_MODE_MASK,
- case drv_command(Port, [<<?FILE_OPEN, M:32>>, pathname(File)]) of
- {ok, Number} ->
- open_int_setopts(Port, Number, Setopts);
- Error ->
- drv_close(Port),
- Error
- end.
+on_load() ->
+ ok = erlang:load_nif(atom_to_list(?MODULE), 0).
-open_int_setopts(Port, Number, []) ->
- {ok, #file_descriptor{module = ?MODULE, data = {Port, Number}}};
-open_int_setopts(Port, Number, [Cmd | Tail]) ->
- case drv_command(Port, Cmd) of
- ok ->
- open_int_setopts(Port, Number, Tail);
- Error ->
- drv_close(Port),
- Error
+open(Name, Modes) ->
+ %% The try/catch pattern seen here is used throughout the file to adhere to
+ %% the public file interface, which has leaked through for ages because of
+ %% "raw files."
+ try open_nif(encode_path(Name), Modes) of
+ {ok, Ref} -> {ok, make_fd(Ref, Modes)};
+ {error, Reason} -> {error, Reason}
+ catch
+ error:badarg -> {error, badarg}
end.
+make_fd(FRef, Modes) ->
+ #file_descriptor{module = ?MODULE, data = build_fd_data(FRef, Modes) }.
-
-%% Returns ok.
-
-close(#file_descriptor{module = ?MODULE, data = {Port, _}}) ->
- case drv_command(Port, <<?FILE_CLOSE>>) of
- ok ->
- drv_close(Port);
- Error ->
- Error
- end;
-%% Closes a port opened with open/1.
-close(Port) when is_port(Port) ->
- drv_close(Port).
-
--define(ADVISE(Offs, Len, Adv),
- <<?FILE_ADVISE, Offs:64/signed, Len:64/signed,
- Adv:32/signed>>).
-
-%% Returns {error, Reason} | ok.
-advise(#file_descriptor{module = ?MODULE, data = {Port, _}},
- Offset, Length, Advise) ->
- case Advise of
- normal ->
- Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_NORMAL),
- drv_command(Port, Cmd);
- random ->
- Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_RANDOM),
- drv_command(Port, Cmd);
- sequential ->
- Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_SEQUENTIAL),
- drv_command(Port, Cmd);
- will_need ->
- Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_WILLNEED),
- drv_command(Port, Cmd);
- dont_need ->
- Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_DONTNEED),
- drv_command(Port, Cmd);
- no_reuse ->
- Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_NOREUSE),
- drv_command(Port, Cmd);
- _ ->
- {error, einval}
+close(Fd) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ close_nif(FRef)
+ catch
+ error:badarg -> {error, badarg}
end.
-%% Returns {error, Reason} | ok.
-allocate(#file_descriptor{module = ?MODULE, data = {Port, _}}, Offset, Length) ->
- Cmd = <<?FILE_ALLOCATE, Offset:64/signed, Length:64/signed>>,
- drv_command(Port, Cmd).
-
-%% Returns {error, Reason} | ok.
-write(#file_descriptor{module = ?MODULE, data = {Port, _}}, Bytes) ->
- case drv_command_nt(Port, [?FILE_WRITE,erlang:dt_prepend_vm_tag_data(Bytes)],undefined) of
- {ok, _Size} ->
- ok;
- Error ->
- Error
+read(Fd, Size) ->
+ try
+ #{ handle := FRef,
+ r_ahead_size := RASz,
+ r_buffer := RBuf } = get_fd_data(Fd),
+ read_1(FRef, RBuf, prim_buffer:size(RBuf), RASz, Size)
+ catch
+ error:badarg -> {error, badarg}
end.
-%% Returns ok | {error, {WrittenCount, Reason}}
-pwrite(#file_descriptor{module = ?MODULE, data = {Port, _}}, L)
- when is_list(L) ->
- pwrite_int(Port, L, 0, [], []).
-
-pwrite_int(_, [], 0, [], []) ->
- ok;
-pwrite_int(Port, [], N, Spec, Data) ->
- Header = list_to_binary([?FILE_PWRITEV, erlang:dt_prepend_vm_tag_data(<<N:32>>) | reverse(Spec)]),
- case drv_command_nt(Port, [Header | reverse(Data)], undefined) of
- {ok, _Size} ->
- ok;
- Error ->
- Error
- end;
-pwrite_int(Port, [{Offs, Bytes} | T], N, Spec, Data)
- when is_integer(Offs) ->
- if
- -(?LARGEFILESIZE) =< Offs, Offs < ?LARGEFILESIZE ->
- pwrite_int(Port, T, N, Spec, Data, Offs, Bytes);
- true ->
- {error, einval}
+-spec read_1(FRef, RBuf, RBufSz, RASz, RSz) -> Result when
+ FRef :: prim_file_ref(),
+ RBuf :: term(),
+ RBufSz :: non_neg_integer(),
+ RASz :: non_neg_integer(),
+ RSz :: non_neg_integer(),
+ Result :: eof | {ok, binary()} | {error, Reason :: atom()}.
+read_1(_FRef, RBuf, RBufSz, _RASz, RSz) when RBufSz >= RSz ->
+ {ok, prim_buffer:read(RBuf, RSz)};
+read_1(FRef, RBuf, RBufSz, RASz, RSz) when RBufSz > 0 ->
+ Buffered = prim_buffer:read(RBuf, RBufSz),
+ case read_1(FRef, RBuf, 0, RASz, RSz - RBufSz) of
+ {ok, Data} ->
+ {ok, <<Buffered/binary, Data/binary>>};
+ eof ->
+ {ok, Buffered};
+ {error, Reason} ->
+ {error, Reason}
end;
-pwrite_int(_, [_|_], _N, _Spec, _Data) ->
- {error, badarg}.
+read_1(FRef, RBuf, RBufSz, RASz, RSz) when RBufSz =:= 0 ->
+ case read_nif(FRef, RASz + RSz) of
+ {ok, Data} when byte_size(Data) > RSz ->
+ {First, Rest} = split_binary(Data, RSz),
+ prim_buffer:write(RBuf, [Rest]),
+ {ok, First};
+ {ok, Data} when byte_size(Data) =< RSz ->
+ {ok, Data};
+ eof ->
+ eof;
+ {error, Reason} ->
+ {error, Reason}
+ end.
-pwrite_int(Port, T, N, Spec, Data, Offs, Bin)
- when is_binary(Bin) ->
- Size = byte_size(Bin),
- pwrite_int(Port, T, N+1,
- [<<Offs:64/signed, Size:64>> | Spec],
- [Bin | Data]);
-pwrite_int(Port, T, N, Spec, Data, Offs, Bytes) ->
- try list_to_binary(Bytes) of
- Bin ->
- pwrite_int(Port, T, N, Spec, Data, Offs, Bin)
+read_line(Fd) ->
+ try
+ #{ handle := FRef,
+ r_ahead_size := RASz,
+ r_buffer := RBuf } = get_fd_data(Fd),
+ SearchResult = prim_buffer:find_byte_index(RBuf, $\n),
+ LineSize = max(?MIN_READLINE_SIZE, RASz),
+ read_line_1(FRef, RBuf, SearchResult, LineSize)
catch
- error:Reason ->
- {error, Reason}
+ error:badarg -> {error, badarg}
end.
-
-
-%% Returns {error, Reason} | ok.
-pwrite(#file_descriptor{module = ?MODULE, data = {Port, _}}, Offs, Bytes)
- when is_integer(Offs) ->
- if
- -(?LARGEFILESIZE) =< Offs, Offs < ?LARGEFILESIZE ->
- case pwrite_int(Port, [], 0, [], [], Offs, Bytes) of
- {error, {_, Reason}} ->
- {error, Reason};
- Result ->
- Result
- end;
- true ->
- {error, einval}
+-spec read_line_1(FRef, RBuf, SearchResult, LineSize) -> Result when
+ FRef :: prim_file_ref(),
+ RBuf :: term(),
+ SearchResult :: not_found | {ok, non_neg_integer()},
+ LineSize :: non_neg_integer(),
+ Result :: eof | {ok, binary()} | {error, Reason :: atom()}.
+read_line_1(FRef, RBuf, not_found, LineSize) ->
+ case read_nif(FRef, LineSize) of
+ {ok, Data} ->
+ prim_buffer:write(RBuf, [Data]),
+ SearchResult = prim_buffer:find_byte_index(RBuf, $\n),
+ read_line_1(FRef, RBuf, SearchResult, LineSize);
+ eof ->
+ case prim_buffer:size(RBuf) of
+ Size when Size > 0 -> {ok, prim_buffer:read(RBuf, Size)};
+ Size when Size =:= 0 -> eof
+ end;
+ {error, Reason} ->
+ {error, Reason}
end;
-pwrite(#file_descriptor{module = ?MODULE}, _, _) ->
- {error, badarg}.
-
+read_line_1(_FRef, RBuf, {ok, LFIndex}, _LineSize) ->
+ %% Translate CRLF into just LF, completely ignoring which encoding is used.
+ CRIndex = (LFIndex - 1),
+ case prim_buffer:read(RBuf, LFIndex + 1) of
+ <<Line:CRIndex/binary, "\r\n">> -> {ok, <<Line/binary, "\n">>};
+ Line -> {ok, Line}
+ end.
-%% Returns {error, Reason} | ok.
-datasync(#file_descriptor{module = ?MODULE, data = {Port, _}}) ->
- drv_command(Port, [?FILE_FDATASYNC]).
-
-%% Returns {error, Reason} | ok.
-sync(#file_descriptor{module = ?MODULE, data = {Port, _}}) ->
- drv_command(Port, [?FILE_FSYNC]).
-
-%% Returns {ok, Data} | eof | {error, Reason}.
-read_line(#file_descriptor{module = ?MODULE, data = {Port, _}}) ->
- case drv_command(Port, <<?FILE_READ_LINE>>) of
- {ok, {0, _Data}} ->
- eof;
- {ok, {_Size, Data}} ->
- {ok, Data};
- {error, enomem} ->
- erlang:garbage_collect(),
- case drv_command(Port, <<?FILE_READ_LINE>>) of
- {ok, {0, _Data}} ->
- eof;
- {ok, {_Size, Data}} ->
- {ok, Data};
- Other ->
- Other
- end;
- Error ->
- Error
+write(Fd, IOData) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ reset_write_position(Fd),
+ write_1(FRef, erlang:iolist_to_iovec(IOData))
+ catch
+ error:badarg -> {error, badarg}
end.
-
-%% Returns {ok, Data} | eof | {error, Reason}.
-read(#file_descriptor{module = ?MODULE, data = {Port, _}}, Size)
- when is_integer(Size), 0 =< Size ->
- if
- Size < ?LARGEFILESIZE ->
- case drv_command(Port, <<?FILE_READ, Size:64>>) of
- {ok, {0, _Data}} when Size =/= 0 ->
- eof;
- {ok, {_Size, Data}} ->
- {ok, Data};
- {error, enomem} ->
- %% Garbage collecting here might help if
- %% the current processes have some old binaries left.
- erlang:garbage_collect(),
- case drv_command(Port, <<?FILE_READ, Size:64>>) of
- {ok, {0, _Data}} when Size =/= 0 ->
- eof;
- {ok, {_Size, Data}} ->
- {ok, Data};
- Other ->
- Other
- end;
- Error ->
- Error
- end;
- true ->
- {error, einval}
+write_1(FRef, IOVec) ->
+ case write_nif(FRef, IOVec) of
+ {continue, Remainder} ->
+ write_1(FRef, Remainder);
+ ok ->
+ ok;
+ {error, Reason} ->
+ {error, Reason}
end.
-%% Returns {ok, [Data|eof, ...]} | {error, Reason}
-pread(#file_descriptor{module = ?MODULE, data = {Port, _}}, L)
- when is_list(L) ->
- pread_int(Port, L, 0, []).
-
-pread_int(_, [], 0, []) ->
- {ok, []};
-pread_int(Port, [], N, Spec) ->
- drv_command_nt(Port, [?FILE_PREADV, erlang:dt_prepend_vm_tag_data(<<0:32, N:32>>) | reverse(Spec)],undefined);
-pread_int(Port, [{Offs, Size} | T], N, Spec)
- when is_integer(Offs), is_integer(Size), 0 =< Size ->
- if
- -(?LARGEFILESIZE) =< Offs, Offs < ?LARGEFILESIZE,
- Size < ?LARGEFILESIZE ->
- pread_int(Port, T, N+1, [<<Offs:64/signed, Size:64>> | Spec]);
- true ->
- {error, einval}
- end;
-pread_int(_, [_|_], _N, _Spec) ->
- {error, badarg}.
-
-
-
-%% Returns {ok, Data} | eof | {error, Reason}.
-pread(#file_descriptor{module = ?MODULE, data = {Port, _}}, Offs, Size)
- when is_integer(Offs), is_integer(Size), 0 =< Size ->
- if
- -(?LARGEFILESIZE) =< Offs, Offs < ?LARGEFILESIZE,
- Size < ?LARGEFILESIZE ->
- case drv_command_nt(Port,
- [?FILE_PREADV, erlang:dt_prepend_vm_tag_data(<<0:32, 1:32,
- Offs:64/signed, Size:64>>)], undefined) of
- {ok, [eof]} ->
- eof;
- {ok, [Data]} ->
- {ok, Data};
- Error ->
- Error
- end;
- true ->
- {error, einval}
- end;
-pread(#file_descriptor{module = ?MODULE, data = {_, _}}, _, _) ->
- {error, badarg}.
-
-
-
-%% Returns {ok, Position} | {error, Reason}.
-position(#file_descriptor{module = ?MODULE, data = {Port, _}}, At) ->
- case lseek_position(At) of
- {Offs, Whence}
- when -(?LARGEFILESIZE) =< Offs, Offs < ?LARGEFILESIZE ->
- drv_command(Port, <<?FILE_LSEEK, Offs:64/signed, Whence:32>>);
- {_, _} ->
- {error, einval};
- Reason ->
- {error, Reason}
+truncate(Fd) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ reset_write_position(Fd),
+ truncate_nif(FRef)
+ catch
+ error:badarg -> {error, badarg}
end.
-%% Returns {error, Reason} | ok.
-truncate(#file_descriptor{module = ?MODULE, data = {Port, _}}) ->
- drv_command(Port, <<?FILE_TRUNCATE>>).
-
-
-
-%% Returns {error, Reason} | {ok, BytesCopied}
-copy(#file_descriptor{module = ?MODULE} = Source,
- #file_descriptor{module = ?MODULE} = Dest,
- Length)
- when is_integer(Length), Length >= 0;
- is_atom(Length) ->
- %% XXX Should be moved down to the driver for optimization.
- file:copy_opened(Source, Dest, Length).
-
-
-
-ipread_s32bu_p32bu(#file_descriptor{module = ?MODULE,
- data = {_, _}} = Handle,
- Offs,
- Infinity) when is_atom(Infinity) ->
- ipread_s32bu_p32bu(Handle, Offs, (1 bsl 31)-1);
-ipread_s32bu_p32bu(#file_descriptor{module = ?MODULE, data = {Port, _}},
- Offs,
- MaxSize)
- when is_integer(Offs), is_integer(MaxSize) ->
- if
- -(?LARGEFILESIZE) =< Offs, Offs < ?LARGEFILESIZE,
- 0 =< MaxSize, MaxSize < (1 bsl 31) ->
- drv_command(Port, <<?FILE_IPREAD, ?IPREAD_S32BU_P32BU,
- Offs:64, MaxSize:32>>);
- true ->
- {error, einval}
- end;
-ipread_s32bu_p32bu(#file_descriptor{module = ?MODULE, data = {_, _}},
- _Offs,
- _MaxSize) ->
- {error, badarg}.
+advise(Fd, Offset, Length, Advise) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ advise_nif(FRef, Offset, Length, Advise)
+ catch
+ error:badarg -> {error, badarg}
+ end.
+allocate(Fd, Offset, Length) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ allocate_nif(FRef, Offset, Length)
+ catch
+ error:badarg -> {error, badarg}
+ end.
+sync(Fd) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ sync_nif(FRef, 0)
+ catch
+ error:badarg -> {error, badarg}
+ end.
-%% Returns {ok, Contents} | {error, Reason}
-read_file(File) when (is_list(File) orelse is_binary(File)) ->
- case drv_open(?FD_DRV, [binary]) of
- {ok, Port} ->
- Result = read_file(Port, File),
- close(Port),
- Result;
- {error, _} = Error ->
- Error
- end;
-read_file(_) ->
- {error, badarg}.
+datasync(Fd) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ sync_nif(FRef, 1)
+ catch
+ error:badarg -> {error, badarg}
+ end.
-%% Takes a Port opened with open/1.
-read_file(Port, File) when is_port(Port),
- (is_list(File) orelse is_binary(File)) ->
- Cmd = [?FILE_READ_FILE | pathname(File)],
- case drv_command(Port, Cmd) of
- {error, enomem} ->
- %% It could possibly help to do a
- %% garbage collection here,
- %% if the file server has some references
- %% to binaries read earlier.
- erlang:garbage_collect(),
- drv_command(Port, Cmd);
- Result ->
- Result
+position(Fd, {cur, Offset}) ->
+ try
+ %% Adjust our current position according to how much we've read ahead.
+ #{ r_buffer := RBuf } = get_fd_data(Fd),
+ position_1(Fd, cur, Offset - prim_buffer:size(RBuf))
+ catch
+ error:badarg -> {error, badarg}
end;
-read_file(_,_) ->
- {error, badarg}.
-
-
-
-%% Returns {error, Reason} | ok.
-write_file(File, Bin) when (is_list(File) orelse is_binary(File)) ->
- case open(File, [binary, write]) of
- {ok, Handle} ->
- Result = write(Handle, Bin),
- close(Handle),
- Result;
- Error ->
- Error
+position(Fd, {Mark, Offset}) ->
+ try
+ position_1(Fd, Mark, Offset)
+ catch
+ error:badarg -> {error, badarg}
end;
-write_file(_, _) ->
- {error, badarg}.
+position(Fd, cur) -> position(Fd, {cur, 0});
+position(Fd, bof) -> position(Fd, {bof, 0});
+position(Fd, eof) -> position(Fd, {eof, 0});
+position(Fd, Offset) -> position(Fd, {bof, Offset}).
+position_1(Fd, Mark, Offset) ->
+ #{ handle := FRef, r_buffer := RBuf } = get_fd_data(Fd),
+ prim_buffer:wipe(RBuf),
+ seek_nif(FRef, Mark, Offset).
-%% Returns {error, Reason} | {ok, BytesCopied}
-%sendfile(_,_,_,_,_,_,_,_,_,_) ->
-% {error, enotsup};
-sendfile(#file_descriptor{module = ?MODULE, data = {Port, _}},
- Dest, Offset, Bytes, _ChunkSize, Headers, Trailers,
- Flags) ->
- case erlang:port_get_data(Dest) of
- Data when Data == inet_tcp; Data == inet6_tcp ->
- ok = inet:lock_socket(Dest,true),
- {ok, DestFD} = prim_inet:getfd(Dest),
- IntFlags = translate_sendfile_flags(Flags),
- try drv_command(Port, [<<?FILE_SENDFILE, DestFD:32,
- IntFlags:8,
- Offset:64/unsigned,
- Bytes:64/unsigned,
- (iolist_size(Headers)):32/unsigned,
- (iolist_size(Trailers)):32/unsigned>>,
- Headers,Trailers])
- after
- ok = inet:lock_socket(Dest,false)
- end;
- _Else ->
- {error,badarg}
+pread(Fd, Offset, Size) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ pread_nif(FRef, Offset, Size)
+ catch
+ error:badarg -> {error, badarg}
end.
-translate_sendfile_flags([{use_threads,true}|T]) ->
- ?EFILE_SENDFILE_USE_THREADS bor translate_sendfile_flags(T);
-translate_sendfile_flags([_|T]) ->
- translate_sendfile_flags(T);
-translate_sendfile_flags([]) ->
- 0.
-
-
-%%%-----------------------------------------------------------------
-%%% Functions operating on files without handle to the file. ?DRV.
-%%%
-%%% Supposed to be called by applications through module file.
-
-
-
-%% Returns {ok, Port}, the Port should be used as first argument in all
-%% the following functions. Returns {error, Reason} upon failure.
-start() ->
- drv_open(?DRV, [binary]).
-
-stop(Port) when is_port(Port) ->
- try erlang:port_close(Port) of
- _ ->
- ok
+pread(Fd, LocNums) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ pread_list(FRef, LocNums, [])
catch
- _:_ ->
- ok
+ error:badarg -> {error, badarg}
end.
+-spec pread_list(FRef, LocNums, ResultList) -> Result when
+ FRef :: prim_file_ref(),
+ LocNums :: list({Offset :: non_neg_integer(),
+ Size :: non_neg_integer()}),
+ ResultList :: list(eof | binary()),
+ Result :: {ok, ResultList} | {error, Reason :: atom()}.
+pread_list(_FRef, [], ResultList) ->
+ {ok, reverse_list(ResultList)};
+pread_list(FRef, [{Offset, Size} | Rest], ResultList) ->
+ case pread_nif(FRef, Offset, Size) of
+ {ok, Data} ->
+ pread_list(FRef, Rest, [Data | ResultList]);
+ eof ->
+ pread_list(FRef, Rest, [eof | ResultList]);
+ {error, Reason} ->
+ {error, Reason}
+ end.
+pwrite(Fd, Offset, IOData) ->
+ try
+ #{ handle := FRef, r_buffer := RBuf } = get_fd_data(Fd),
+ prim_buffer:wipe(RBuf),
+ pwrite_plain(FRef, Offset, erlang:iolist_to_iovec(IOData))
+ catch
+ error:badarg -> {error, badarg}
+ end.
+pwrite_plain(FRef, Offset, IOVec) ->
+ case pwrite_nif(FRef, Offset, IOVec) of
+ {continue, BytesWritten, Remainder} ->
+ pwrite_plain(FRef, Offset + BytesWritten, Remainder);
+ ok ->
+ ok;
+ {error, Reason} ->
+ {error, Reason}
+ end.
-%%% The following functions take an optional Port as first argument.
-%%% If the port is not supplied, a temporary one is opened and then
-%%% closed after the request has been performed.
-
-
-
-%% get_cwd/{0,1,2}
-
-get_cwd() ->
- get_cwd_int(0).
-
-get_cwd(Port) when is_port(Port) ->
- get_cwd_int(Port, 0);
-get_cwd([]) ->
- get_cwd_int(0);
-get_cwd([Letter, $: | _]) when $a =< Letter, Letter =< $z ->
- get_cwd_int(Letter - $a + 1);
-get_cwd([Letter, $: | _]) when $A =< Letter, Letter =< $Z ->
- get_cwd_int(Letter - $A + 1);
-get_cwd([_|_]) ->
- {error, einval};
-get_cwd(_) ->
- {error, badarg}.
-
-get_cwd(Port, []) when is_port(Port) ->
- get_cwd_int(Port, 0);
-get_cwd(Port, [Letter, $: | _])
- when is_port(Port), $a =< Letter, Letter =< $z ->
- get_cwd_int(Port, Letter - $a + 1);
-get_cwd(Port, [Letter, $: | _])
- when is_port(Port), $A =< Letter, Letter =< $Z ->
- get_cwd_int(Port, Letter - $A + 1);
-get_cwd(Port, [_|_]) when is_port(Port) ->
- {error, einval};
-get_cwd(_, _) ->
- {error, badarg}.
-
-get_cwd_int(Drive) ->
- get_cwd_int({?DRV, [binary]}, Drive).
-
-get_cwd_int(Port, Drive) ->
- drv_command(Port, <<?FILE_PWD, Drive>>,
- fun handle_fname_response/1).
-
-
-
-%% set_cwd/{1,2}
+pwrite(Fd, LocBytes) ->
+ try
+ #{ handle := FRef, r_buffer := RBuf } = get_fd_data(Fd),
+ prim_buffer:wipe(RBuf),
+ pwrite_list(FRef, LocBytes, 0)
+ catch
+ error:badarg -> {error, badarg}
+ end.
-set_cwd(Dir) ->
- set_cwd_int({?DRV, [binary]}, Dir).
+-spec pwrite_list(FRef, LocBytes, Successes) -> Result when
+ FRef :: prim_file_ref(),
+ LocBytes :: list({Offset :: non_neg_integer(),
+ IOData :: iodata()}),
+ Successes :: non_neg_integer(),
+ Result :: ok | {error, {Successes, Reason :: atom()}}.
+pwrite_list(_FRef, [], _Successes) ->
+ ok;
+pwrite_list(FRef, [{Offset, IOData} | Rest], Successes) ->
+ case pwrite_plain(FRef, Offset, erlang:iolist_to_iovec(IOData)) of
+ {error, Reason} -> {error, {Successes, Reason}};
+ ok -> pwrite_list(FRef, Rest, Successes + 1)
+ end.
-set_cwd(Port, Dir) when is_port(Port) ->
- set_cwd_int(Port, Dir).
+sendfile(Fd, Socket, Offset, Bytes, _ChunkSize, [], [], _Flags) ->
+ %% There's a very nasty race in here; if we die just prior to duplicating
+ %% the handle down in the sendfile call, it might get reused by something
+ %% entirely different and we'll leak unknown data to the socket until it
+ %% dies soon after.
+ %%
+ %% This bug was inherited from the old driver, except it was vulnerable to
+ %% the bug at any point and not just during setup.
+ %%
+ %% We'll have to live with this until we have a way to unambiguously
+ %% transfer things between drivers or NIFs. Current ideas all fall afoul
+ %% of the Two Generals problem.
+ try
+ advise(Fd, Offset, Bytes, sequential),
+ prim_inet:sendfile(Socket, get_handle(Fd), Offset, Bytes)
+ catch
+ error:badarg -> {error, badarg}
+ end;
+sendfile(_Fd, _Socket, _Offset, _Bytes, _ChunkSize, _Headers, _Trailers, _Flags) ->
+ {error, enotsup}.
-set_cwd_int(Port, Dir) when is_binary(Dir) ->
- case prim_file:is_translatable(Dir) of
- false ->
- {error, no_translation};
- true ->
- drv_command(Port, [?FILE_CHDIR, pathname(Dir)])
+%% Undocumented internal function that reads a data block with indirection.
+%%
+%% This is only used once in DETS and can easily be emulated with pread/2, but
+%% it's pretty performance-sensitive so we've implemented it down in the NIF to
+%% avoid excessive rescheduling.
+-spec ipread_s32bu_p32bu(Fd, Offset, MaxSize) -> Result when
+ Fd :: #file_descriptor{},
+ Offset :: non_neg_integer(),
+ MaxSize :: non_neg_integer() | infinity,
+ Result :: {ok, Size :: non_neg_integer(),
+ Pointer :: non_neg_integer(),
+ Data :: iodata() | eof} |
+ eof |
+ {error, Reason :: atom()}.
+ipread_s32bu_p32bu(Fd, Offset, Infinity) when is_atom(Infinity) ->
+ ipread_s32bu_p32bu(Fd, Offset, (1 bsl 31) - 1);
+ipread_s32bu_p32bu(Fd, Offset, MaxSize)
+ when is_integer(Offset), is_integer(MaxSize) ->
+ try
+ #{ handle := FRef } = get_fd_data(Fd),
+ ipread_s32bu_p32bu_nif(FRef, Offset, MaxSize)
+ catch
+ error:badarg -> {error, badarg}
end;
-set_cwd_int(Port, Dir) when is_list(Dir) ->
- drv_command(Port, [?FILE_CHDIR, pathname(Dir)]);
-set_cwd_int(_, _) ->
+ipread_s32bu_p32bu(_Fd, _Offset, _MaxSize) ->
{error, badarg}.
-
-
-
-%% delete/{1,2}
-
-delete(File) ->
- delete_int({?DRV, [binary]}, File).
-
-delete(Port, File) when is_port(Port) ->
- delete_int(Port, File).
-
-delete_int(Port, File) ->
- drv_command(Port, [?FILE_DELETE, pathname(File)]).
-
-
-
-%% rename/{2,3}
-
-rename(From, To) ->
- rename_int({?DRV, [binary]}, From, To).
-
-rename(Port, From, To) when is_port(Port) ->
- rename_int(Port, From, To).
-
-rename_int(Port, From, To) ->
- drv_command(Port, [?FILE_RENAME, pathname(From), pathname(To)]).
-
-
-
-%% make_dir/{1,2}
-
-make_dir(Dir) ->
- make_dir_int({?DRV, [binary]}, Dir).
-
-make_dir(Port, Dir) when is_port(Port) ->
- make_dir_int(Port, Dir).
-
-make_dir_int(Port, Dir) ->
- drv_command(Port, [?FILE_MKDIR, pathname(Dir)]).
-
-
-
-%% del_dir/{1,2}
-
-del_dir(Dir) ->
- del_dir_int({?DRV, [binary]}, Dir).
-
-del_dir(Port, Dir) when is_port(Port) ->
- del_dir_int(Port, Dir).
-
-del_dir_int(Port, Dir) ->
- drv_command(Port, [?FILE_RMDIR, pathname(Dir)]).
-
-
-
-%% read_file_info/{1,2,3}
-
-read_file_info(File) ->
- read_file_info_int({?DRV, [binary]}, File, local).
-
-read_file_info(Port, File) when is_port(Port) ->
- read_file_info_int(Port, File, local);
-read_file_info(File, Opts) ->
- read_file_info_int({?DRV, [binary]}, File, plgv(time, Opts, local)).
-
-read_file_info(Port, File, Opts) when is_port(Port) ->
- read_file_info_int(Port, File, plgv(time, Opts, local)).
-
-read_file_info_int(Port, File, TimeType) ->
+ipread_s32bu_p32bu_nif(_FRef, _Offset, _MaxSize) ->
+ erlang:nif_error(undef).
+
+%% Returns the binary representation of the underlying handle, for use in
+%% tricky operations like sendfile/8.
+-spec get_handle(Fd) -> Result when
+ Fd :: #file_descriptor{},
+ Result :: binary() | {error, Reason :: atom()}.
+get_handle(Fd) ->
try
- case drv_command(Port, [?FILE_FSTAT, pathname(File)]) of
- {ok, FI} -> {ok, FI#file_info{
- ctime = from_seconds(FI#file_info.ctime, TimeType),
- mtime = from_seconds(FI#file_info.mtime, TimeType),
- atime = from_seconds(FI#file_info.atime, TimeType)
- }};
- Error -> Error
- end
+ #{ handle := FRef } = get_fd_data(Fd),
+ get_handle_nif(FRef)
catch
- error:_ -> {error, badarg}
+ error:badarg -> {error, badarg}
end.
+%% Resets the write head to the position the user believes we're at, which may
+%% not be the same as the real one when read caching is in effect.
+reset_write_position(Fd) ->
+ #{ r_buffer := RBuf } = Fd#file_descriptor.data,
+ case prim_buffer:size(RBuf) of
+ Size when Size > 0 -> position(Fd, cur);
+ Size when Size =:= 0 -> ok
+ end.
-%% altname/{1,2}
-
-altname(File) ->
- altname_int({?DRV, [binary]}, File).
-
-altname(Port, File) when is_port(Port) ->
- altname_int(Port, File).
-
-altname_int(Port, File) ->
- drv_command(Port, [?FILE_ALTNAME, pathname(File)],
- fun handle_fname_response/1).
-
-%% write_file_info/{2,3,4}
-
-write_file_info(File, Info) ->
- write_file_info_int({?DRV, [binary]}, File, Info, local).
-
-write_file_info(Port, File, Info) when is_port(Port) ->
- write_file_info_int(Port, File, Info, local);
-write_file_info(File, Info, Opts) ->
- write_file_info_int({?DRV, [binary]}, File, Info, plgv(time, Opts, local)).
-
-write_file_info(Port, File, Info, Opts) when is_port(Port) ->
- write_file_info_int(Port, File, Info, plgv(time, Opts, local)).
+get_fd_data(#file_descriptor{ data = Data }) ->
+ #{ owner := Owner } = Data,
+ case self() of
+ Owner -> Data;
+ _ -> error(not_on_controlling_process)
+ end.
-write_file_info_int(Port, File,
- #file_info{mode=Mode,
- uid=Uid,
- gid=Gid,
- atime=Atime0,
- mtime=Mtime0,
- ctime=Ctime0},
- TimeType) ->
+build_fd_data(FRef, Modes) ->
+ Defaults =
+ #{ owner => self(),
+ handle => FRef,
+ r_ahead_size => 0,
+ r_buffer => prim_buffer:new() },
+ fill_fd_option_map(Modes, Defaults).
+
+fill_fd_option_map([], Map) ->
+ Map;
+
+fill_fd_option_map([read_ahead | Modes], Map) ->
+ fill_fd_option_map([{read_ahead, 64 bsl 10} | Modes], Map);
+fill_fd_option_map([{read_ahead, Size} | Modes], Map) ->
+ fill_fd_option_map(Modes, Map#{ r_ahead_size => Size });
+
+fill_fd_option_map([_Ignored | Modes], Map) ->
+ fill_fd_option_map(Modes, Map).
+
+open_nif(_Name, _Modes) ->
+ erlang:nif_error(undef).
+close_nif(_FileRef) ->
+ erlang:nif_error(undef).
+read_nif(_FileRef, _Size) ->
+ erlang:nif_error(undef).
+write_nif(_FileRef, _IOVec) ->
+ erlang:nif_error(undef).
+pread_nif(_FileRef, _Offset, _Size) ->
+ erlang:nif_error(undef).
+pwrite_nif(_FileRef, _Offset, _IOVec) ->
+ erlang:nif_error(undef).
+seek_nif(_FileRef, _Mark, _Offset) ->
+ erlang:nif_error(undef).
+sync_nif(_FileRef, _DataOnly) ->
+ erlang:nif_error(undef).
+advise_nif(_FileRef, _Offset, _Length, _Advise) ->
+ erlang:nif_error(undef).
+allocate_nif(_FileRef, _Offset, _Length) ->
+ erlang:nif_error(undef).
+truncate_nif(_FileRef) ->
+ erlang:nif_error(undef).
+get_handle_nif(_FileRef) ->
+ erlang:nif_error(undef).
- % Atime and/or Mtime might be undefined
- % - use localtime() for atime, if atime is undefined
- % - use atime as mtime if mtime is undefined
- % - use mtime as ctime if ctime is undefined
+%%
+%% Quality-of-life helpers
+%%
+read_file(Filename) ->
+ %% We're doing this operation in the NIF to avoid excessive rescheduling.
try
- Atime = file_info_validate_atime(Atime0, TimeType),
- Mtime = file_info_validate_mtime(Mtime0, Atime),
- Ctime = file_info_validate_ctime(Ctime0, Mtime),
-
- drv_command(Port, [?FILE_WRITE_INFO,
- int_to_int32bytes(Mode),
- int_to_int32bytes(Uid),
- int_to_int32bytes(Gid),
- int_to_int64bytes(to_seconds(Atime, TimeType)),
- int_to_int64bytes(to_seconds(Mtime, TimeType)),
- int_to_int64bytes(to_seconds(Ctime, TimeType)),
- pathname(File)])
+ read_file_nif(encode_path(Filename))
catch
- error:_ -> {error, badarg}
+ error:badarg -> {error, badarg}
+ end.
+read_file_nif(_Filename) ->
+ erlang:nif_error(undef).
+
+write_file(Filename, Bytes) ->
+ write_file(Filename, Bytes, []).
+write_file(Filename, Bytes, Modes) ->
+ case open(Filename, [write, binary | Modes]) of
+ {ok, Fd} ->
+ Result = write(Fd, Bytes),
+ close(Fd),
+ Result;
+ {error, Reason} ->
+ {error, Reason}
end.
+%%
+%% Filesystem operations
+%%
-file_info_validate_atime(Atime, _) when Atime =/= undefined -> Atime;
-file_info_validate_atime(undefined, local) -> erlang:localtime();
-file_info_validate_atime(undefined, universal) -> erlang:universaltime();
-file_info_validate_atime(undefined, posix) -> erlang:universaltime_to_posixtime(erlang:universaltime()).
-
-file_info_validate_mtime(undefined, Atime) -> Atime;
-file_info_validate_mtime(Mtime, _) -> Mtime.
-
-file_info_validate_ctime(undefined, Mtime) -> Mtime;
-file_info_validate_ctime(Ctime, _) -> Ctime.
-
-%% make_link/{2,3}
-
-make_link(Old, New) ->
- make_link_int({?DRV, [binary]}, Old, New).
-
-make_link(Port, Old, New) when is_port(Port) ->
- make_link_int(Port, Old, New).
-
-make_link_int(Port, Old, New) ->
- drv_command(Port, [?FILE_LINK, pathname(Old), pathname(New)]).
-
-
-
-%% make_symlink/{2,3}
-
-make_symlink(Old, New) ->
- make_symlink_int({?DRV, [binary]}, Old, New).
-
-make_symlink(Port, Old, New) when is_port(Port) ->
- make_symlink_int(Port, Old, New).
-
-make_symlink_int(Port, Old, New) ->
- drv_command(Port, [?FILE_SYMLINK, pathname(Old), pathname(New)]).
-
-
-
-%% read_link/{2,3}
-
-read_link(Link) ->
- read_link_int({?DRV, [binary]}, Link).
-
-read_link(Port, Link) when is_port(Port) ->
- read_link_int(Port, Link).
-
-read_link_int(Port, Link) ->
- drv_command(Port, [?FILE_READLINK, pathname(Link)],
- fun handle_fname_response/1).
-
-%% read_link_all/{2,3}
-
-read_link_all(Link) ->
- read_link_all_int({?DRV, [binary]}, Link).
-
-read_link_all(Port, Link) when is_port(Port) ->
- read_link_all_int(Port, Link).
-
-read_link_all_int(Port, Link) ->
- drv_command(Port, [?FILE_READLINK, pathname(Link)],
- fun handle_fname_response_all/1).
-
-
-
-%% read_link_info/{2,3}
-
-read_link_info(Link) ->
- read_link_info_int({?DRV, [binary]}, Link, local).
-
-read_link_info(Port, Link) when is_port(Port) ->
- read_link_info_int(Port, Link, local);
+read_link(Name) -> read_link_1(Name, false).
+read_link_all(Name) -> read_link_1(Name, true).
-read_link_info(Link, Opts) ->
- read_link_info_int({?DRV, [binary]}, Link, plgv(time, Opts, local)).
+read_link_1(Name, AcceptRawNames) ->
+ try read_link_nif(encode_path(Name)) of
+ {ok, RawName} -> translate_raw_name(RawName, AcceptRawNames);
+ {error, Reason} -> {error, Reason}
+ catch
+ error:badarg -> {error, badarg}
+ end.
-read_link_info(Port, Link, Opts) when is_port(Port) ->
- read_link_info_int(Port, Link, plgv(time, Opts, local)).
+translate_raw_name(RawName, SilentFailure) ->
+ case decode_path(RawName) of
+ Converted when is_list(Converted) -> {ok, Converted};
+ {error, _Reason} when SilentFailure =:= false -> {error, einval};
+ {error, _Reason} when SilentFailure =:= true -> {ok, RawName}
+ end.
+list_dir(Name) -> list_dir_1(Name, true).
+list_dir_all(Name) -> list_dir_1(Name, false).
-read_link_info_int(Port, Link, TimeType) ->
- try
- case drv_command(Port, [?FILE_LSTAT, pathname(Link)]) of
- {ok, FI} -> {ok, FI#file_info{
- ctime = from_seconds(FI#file_info.ctime, TimeType),
- mtime = from_seconds(FI#file_info.mtime, TimeType),
- atime = from_seconds(FI#file_info.atime, TimeType)
- }};
- Error -> Error
- end
+list_dir_1(Name, SkipInvalid) ->
+ try list_dir_nif(encode_path(Name)) of
+ {ok, RawNames} -> list_dir_convert(RawNames, SkipInvalid, []);
+ {error, Reason} -> {error, Reason}
catch
- error:_ -> {error, badarg}
+ error:badarg -> {error, badarg}
end.
-%% list_dir/{1,2}
-
-list_dir(Dir) ->
- list_dir_int({?DRV, [binary]}, Dir).
-
-list_dir(Port, Dir) when is_port(Port) ->
- list_dir_int(Port, Dir).
-
-list_dir_int(Port, Dir) ->
- drv_command(Port, [?FILE_READDIR, pathname(Dir)],
- fun(P) ->
- case list_dir_response(P, []) of
- {ok, RawNames} ->
- try
- {ok, list_dir_convert(RawNames)}
- catch
- throw:Reason ->
- Reason
- end;
- Error ->
- Error
- end
- end).
-
-list_dir_all(Dir) ->
- list_dir_all_int({?DRV, [binary]}, Dir).
-
-list_dir_all(Port, Dir) when is_port(Port) ->
- list_dir_all_int(Port, Dir).
-
-list_dir_all_int(Port, Dir) ->
- drv_command(Port, [?FILE_READDIR, pathname(Dir)],
- fun(P) ->
- case list_dir_response(P, []) of
- {ok, RawNames} ->
- {ok, list_dir_convert_all(RawNames)};
- Error ->
- Error
- end
- end).
-
-list_dir_response(Port, Acc0) ->
- case drv_get_response(Port) of
- {lfname, []} ->
- {ok, Acc0};
- {lfname, Names} ->
- Acc = [Name || <<L:16,Name:L/binary>> <= Names] ++ Acc0,
- list_dir_response(Port, Acc);
- Error ->
- Error
+list_dir_convert([], _SkipInvalid, Result) ->
+ {ok, Result};
+list_dir_convert([RawName | Rest], SkipInvalid, Result) ->
+ case decode_path(RawName) of
+ Converted when is_list(Converted) ->
+ list_dir_convert(Rest, SkipInvalid, [Converted | Result]);
+ {error, _} when SkipInvalid =:= false ->
+ list_dir_convert(Rest, SkipInvalid, [RawName | Result]);
+
+ %% If the filename cannot be converted, return error or ignore with
+ %% optional error logger warning depending on +fn{u|a}{i|e|w} emulator
+ %% switches.
+ {error, ignore} ->
+ list_dir_convert(Rest, SkipInvalid, Result);
+ {error, warning} ->
+ error_logger:warning_msg(
+ "Non-unicode filename ~p ignored\n", [RawName]),
+ list_dir_convert(Rest, SkipInvalid, Result);
+ {error, _} ->
+ {error, {no_translation, RawName}}
end.
-list_dir_convert([Name|Names]) ->
- %% If the filename cannot be converted, return error or ignore
- %% with optional error logger warning, depending on +fn{u|a}{i|e|w}
- %% emulator switches.
- case prim_file:internal_native2name(Name) of
- {error, warning} ->
- error_logger:warning_msg("Non-unicode filename ~p ignored\n",
- [Name]),
- list_dir_convert(Names);
- {error, ignore} ->
- list_dir_convert(Names);
- {error, error} ->
- throw({error, {no_translation, Name}});
- Converted when is_list(Converted) ->
- [Converted|list_dir_convert(Names)]
- end;
-list_dir_convert([]) -> [].
-
-list_dir_convert_all([Name|Names]) ->
- %% If the filename cannot be converted, retain the filename as
- %% a binary.
- case prim_file:internal_native2name(Name) of
- {error, _} ->
- [Name|list_dir_convert_all(Names)];
- Converted when is_list(Converted) ->
- [Converted|list_dir_convert_all(Names)]
- end;
-list_dir_convert_all([]) -> [].
-
-%%%-----------------------------------------------------------------
-%%% Functions to communicate with the driver
-
-handle_fname_response(Port) ->
- case drv_get_response(Port) of
- {fname, Name} ->
- case prim_file:internal_native2name(Name) of
- {error, warning} ->
- error_logger:warning_msg("Non-unicode filename ~p "
- "ignored when reading link\n",
- [Name]),
- {error, einval};
- {error, _} ->
- {error, einval};
- Converted when is_list(Converted) ->
- {ok, Converted}
- end;
- Error ->
- Error
+read_file_info(Filename) ->
+ read_info_1(Filename, 1, local).
+read_file_info(Filename, Opts) ->
+ read_info_1(Filename, 1, proplist_get_value(time, Opts, local)).
+
+read_link_info(Name) ->
+ read_info_1(Name, 0, local).
+read_link_info(Name, Opts) ->
+ read_info_1(Name, 0, proplist_get_value(time, Opts, local)).
+
+read_info_1(Name, FollowLinks, TimeType) ->
+ try read_info_nif(encode_path(Name), FollowLinks) of
+ {error, Reason} -> {error, Reason};
+ FileInfo ->
+ CTime = from_posix_seconds(FileInfo#file_info.ctime, TimeType),
+ MTime = from_posix_seconds(FileInfo#file_info.mtime, TimeType),
+ ATime = from_posix_seconds(FileInfo#file_info.atime, TimeType),
+ {ok, FileInfo#file_info{ ctime = CTime, mtime = MTime, atime = ATime }}
+ catch
+ error:badarg -> {error, badarg}
end.
-handle_fname_response_all(Port) ->
- case drv_get_response(Port) of
- {fname, Name} ->
- case prim_file:internal_native2name(Name) of
- {error, _} ->
- {ok, Name};
- Converted when is_list(Converted) ->
- {ok, Converted}
- end;
- Error ->
- Error
+write_file_info(Filename, Info) ->
+ write_file_info_1(Filename, Info, local).
+write_file_info(Filename, Info, Opts) ->
+ write_file_info_1(Filename, Info, proplist_get_value(time, Opts, local)).
+
+write_file_info_1(Filename, Info, TimeType) ->
+ #file_info{ mode = Modes,
+ uid = Uid,
+ gid = Gid,
+ atime = ATime0,
+ mtime = MTime0,
+ ctime = CTime0} = Info,
+ try
+ % ATime and/or MTime might be undefined
+ % - use localtime() for atime, if atime is undefined
+ % - use atime as mtime if mtime is undefined
+ % - use mtime as ctime if ctime is undefined
+ ATime = file_info_convert_atime(ATime0, TimeType),
+ MTime = file_info_convert_mtime(MTime0, ATime, TimeType),
+ CTime = file_info_convert_ctime(CTime0, MTime, TimeType),
+ EncodedName = encode_path(Filename),
+
+ %% This is a bit ugly but we need to handle partial failures the same
+ %% way the old driver did.
+ throw_on_error(set_owner(EncodedName, Uid, Gid)),
+ throw_on_error(set_permissions(EncodedName, Modes)),
+ throw_on_error(set_time(EncodedName, ATime, MTime, CTime))
+ catch
+ throw:Reason -> {error, Reason};
+ error:_ -> {error, badarg}
end.
-%% Opens a driver port and converts any problems into {error, emfile}.
-%% Returns {ok, Port} when successful.
+set_owner(_EncodedName, undefined, undefined) ->
+ ok;
+set_owner(EncodedName, Uid, Gid) ->
+ set_owner_nif(EncodedName, Uid, Gid).
+set_owner_nif(_Path, _Uid, _Gid) ->
+ erlang:nif_error(undef).
-drv_open(Driver, Portopts) ->
- try erlang:open_port({spawn_driver, Driver}, Portopts) of
- Port ->
- {ok, Port}
+set_permissions(_EncodedName, undefined) ->
+ ok;
+set_permissions(EncodedName, Permissions) ->
+ set_permissions_nif(EncodedName, Permissions).
+set_permissions_nif(_Path, _Permissions) ->
+ erlang:nif_error(undef).
+
+set_time(EncodedName, ATime, MTime, CTime) ->
+ set_time_nif(EncodedName, ATime, MTime, CTime).
+set_time_nif(_Path, _ATime, _MTime, _CTime) ->
+ erlang:nif_error(undef).
+
+throw_on_error(ok) -> ok;
+throw_on_error({error, enotsup}) -> ok;
+throw_on_error({error, Reason}) -> throw(Reason).
+
+file_info_convert_atime(ATime, TimeType) when ATime =/= undefined ->
+ to_posix_seconds(ATime, TimeType);
+file_info_convert_atime(undefined, local) ->
+ to_posix_seconds(erlang:localtime(), local);
+file_info_convert_atime(undefined, universal) ->
+ to_posix_seconds(erlang:universaltime(), universal);
+file_info_convert_atime(undefined, posix) ->
+ erlang:universaltime_to_posixtime(erlang:universaltime()).
+
+file_info_convert_mtime(undefined, ATime, _TimeType) ->
+ ATime;
+file_info_convert_mtime(MTime, _ATime, TimeType) ->
+ to_posix_seconds(MTime, TimeType).
+
+file_info_convert_ctime(undefined, MTime, _TimeType) ->
+ MTime;
+file_info_convert_ctime(CTime, _MTime, TimeType) ->
+ to_posix_seconds(CTime, TimeType).
+
+%% This is only relevant on Windows, so we assume that format to simplify the
+%% internals.
+get_cwd([Letter, $:]) when Letter >= $A, Letter =< $Z ->
+ get_dcwd(Letter - $A + 1);
+get_cwd([Letter, $:]) when Letter >= $a, Letter =< $z ->
+ get_dcwd(Letter - $a + 1);
+get_cwd([_|_]) ->
+ {error, einval};
+get_cwd(_) ->
+ {error, badarg}.
+get_dcwd(Index) ->
+ try get_device_cwd_nif(Index) of
+ {ok, RawPath} -> {ok, decode_path(RawPath)};
+ {error, Reason} -> {error, Reason}
catch
- error:Reason ->
- {error, Reason}
+ error:badarg -> {error, badarg}
end.
-
-
-%% Closes a port in a safe way. Returns ok.
-
-drv_close(Port) ->
- Save = erlang:dt_spread_tag(false),
+get_cwd() ->
+ try get_cwd_nif() of
+ {ok, RawPath} -> {ok, decode_path(RawPath)};
+ {error, Reason} -> {error, Reason}
+ catch
+ error:badarg -> {error, badarg}
+ end.
+set_cwd(Path) ->
try
- try erlang:port_close(Port) catch error:_ -> ok end,
- receive %% Ugly workaround in case the caller==owner traps exits
- {'EXIT', Port, _Reason} ->
- ok
- after 0 ->
- ok
- end
- after
- erlang:dt_restore_tag(Save)
+ case is_path_translatable(Path) of
+ true -> set_cwd_nif(encode_path(Path));
+ false -> {error, no_translation}
+ end
+ catch
+ error:badarg -> {error, badarg}
end.
-
-
-%% Issues a command to a port and gets the response.
-%% If Port is {Driver, Portopts} a port is first opened and
-%% then closed after the result has been received.
-%% Returns {ok, Result} or {error, Reason}.
-
-drv_command(Port, Command) ->
- drv_command(Port, Command, undefined).
-
-drv_command(Port, Command, R) when is_binary(Command) ->
- drv_command(Port, Command, true, R);
-drv_command(Port, Command, R) ->
- try erlang:iolist_size(Command) of
- _ ->
- drv_command(Port, Command, true, R)
+delete(Path) ->
+ try
+ del_file_nif(encode_path(Path))
catch
- error:Reason ->
- {error, Reason}
+ error:badarg -> {error, badarg}
end.
-drv_command(Port, Command, Validated, R) when is_port(Port) ->
- Save = erlang:dt_spread_tag(false),
- try erlang:port_command(Port, erlang:dt_append_vm_tag_data(Command)) of
- true ->
- drv_get_response(Port, R)
+rename(Source, Destination) ->
+ try
+ rename_nif(encode_path(Source), encode_path(Destination))
catch
- %% If the Command is valid, knowing that the port is a port,
- %% a badarg error must mean it is a dead port, that is:
- %% a currently invalid filehandle, -> einval, not badarg.
- error:badarg when Validated ->
- {error, einval};
- error:badarg ->
- try erlang:iolist_size(Command) of
- _ -> % Valid
- {error, einval}
- catch
- error:_ ->
- {error, badarg}
- end;
- error:Reason ->
- {error, Reason}
- after
- erlang:dt_restore_tag(Save)
- end;
-drv_command({Driver, Portopts}, Command, Validated, R) ->
- case drv_open(Driver, Portopts) of
- {ok, Port} ->
- Result = drv_command(Port, Command, Validated, R),
- drv_close(Port),
- Result;
- Error ->
- Error
+ error:badarg -> {error, badarg}
end.
-drv_command_nt(Port, Command, R) when is_port(Port) ->
- Save = erlang:dt_spread_tag(false),
- try erlang:port_command(Port, Command) of
- true ->
- drv_get_response(Port, R)
+make_dir(Path) ->
+ try
+ make_dir_nif(encode_path(Path))
catch
- error:badarg ->
- try erlang:iolist_size(Command) of
- _ -> % Valid
- {error, einval}
- catch
- error:_ ->
- {error, badarg}
- end;
- error:Reason ->
- {error, Reason}
- after
- erlang:dt_restore_tag(Save)
+ error:badarg -> {error, badarg}
end.
-
-
-
-%% Receives the response from a driver port.
-%% Returns: {ok, ListOrBinary}|{error, Reason}
-
-drv_get_response(Port, undefined) ->
- drv_get_response(Port);
-drv_get_response(Port, Fun) when is_function(Fun, 1) ->
- Fun(Port).
-
-drv_get_response(Port) ->
- erlang:bump_reductions(100),
- receive
- {Port, {data, [Response|Rest] = Data}} ->
- try translate_response(Response, Rest)
- catch
- error:Reason ->
- {error, {bad_response_from_port, Data,
- {Reason, erlang:get_stacktrace()}}}
- end;
- {'EXIT', Port, Reason} ->
- {error, {port_died, Reason}}
+del_dir(Path) ->
+ try
+ del_dir_nif(encode_path(Path))
+ catch
+ error:badarg -> {error, badarg}
end.
-
-
-%%%-----------------------------------------------------------------
-%%% Utility functions.
-
-%% Converts a list of mode atoms into a mode word for the driver.
-%% Returns {Mode, Portopts, Setopts} where Portopts is a list of
-%% options for erlang:open_port/2 and Setopts is a list of
-%% setopt commands to send to the port, or error Reason upon failure.
-
-open_mode(List) when is_list(List) ->
- case open_mode(List, 0, [], []) of
- {Mode, Portopts, Setopts} when Mode band
- (?EFILE_MODE_READ bor ?EFILE_MODE_WRITE)
- =:= 0 ->
- {Mode bor ?EFILE_MODE_READ, Portopts, Setopts};
- Other ->
- Other
+make_link(Existing, New) ->
+ try
+ make_hard_link_nif(encode_path(Existing), encode_path(New))
+ catch
+ error:badarg -> {error, badarg}
end.
-
-open_mode([raw|Rest], Mode, Portopts, Setopts) ->
- open_mode(Rest, Mode, Portopts, Setopts);
-open_mode([read|Rest], Mode, Portopts, Setopts) ->
- open_mode(Rest, Mode bor ?EFILE_MODE_READ, Portopts, Setopts);
-open_mode([write|Rest], Mode, Portopts, Setopts) ->
- open_mode(Rest, Mode bor ?EFILE_MODE_WRITE, Portopts, Setopts);
-open_mode([binary|Rest], Mode, Portopts, Setopts) ->
- open_mode(Rest, Mode, [binary | Portopts], Setopts);
-open_mode([compressed|Rest], Mode, Portopts, Setopts) ->
- open_mode(Rest, Mode bor ?EFILE_COMPRESSED, Portopts, Setopts);
-open_mode([append|Rest], Mode, Portopts, Setopts) ->
- open_mode(Rest, Mode bor ?EFILE_MODE_APPEND bor ?EFILE_MODE_WRITE,
- Portopts, Setopts);
-open_mode([exclusive|Rest], Mode, Portopts, Setopts) ->
- open_mode(Rest, Mode bor ?EFILE_MODE_EXCL, Portopts, Setopts);
-open_mode([sync|Rest], Mode, Portopts, Setopts) ->
- open_mode(Rest, Mode bor ?EFILE_MODE_SYNC, Portopts, Setopts);
-open_mode([delayed_write|Rest], Mode, Portopts, Setopts) ->
- open_mode([{delayed_write, 64*1024, 2000}|Rest], Mode,
- Portopts, Setopts);
-open_mode([{delayed_write, Size, Delay}|Rest], Mode, Portopts, Setopts)
- when is_integer(Size), 0 =< Size, is_integer(Delay), 0 =< Delay ->
- if
- Size < ?LARGEFILESIZE, Delay < 1 bsl 64 ->
- open_mode(Rest, Mode, Portopts,
- [<<?FILE_SETOPT, ?FILE_OPT_DELAYED_WRITE,
- Size:64, Delay:64>>
- | Setopts]);
- true ->
- einval
- end;
-open_mode([read_ahead|Rest], Mode, Portopts, Setopts) ->
- open_mode([{read_ahead, 64*1024}|Rest], Mode, Portopts, Setopts);
-open_mode([{read_ahead, Size}|Rest], Mode, Portopts, Setopts)
- when is_integer(Size), 0 =< Size ->
- if
- Size < ?LARGEFILESIZE ->
- open_mode(Rest, Mode, Portopts,
- [<<?FILE_SETOPT, ?FILE_OPT_READ_AHEAD,
- Size:64>> | Setopts]);
- true ->
- einval
- end;
-open_mode([], Mode, Portopts, Setopts) ->
- {Mode, reverse(Portopts), reverse(Setopts)};
-open_mode(_, _Mode, _Portopts, _Setopts) ->
- badarg.
-
-
-
-%% Converts a position tuple {bof, X} | {cur, X} | {eof, X} into
-%% {Offset, OriginCode} for the driver.
-%% Returns badarg upon failure.
-
-lseek_position(Pos)
- when is_integer(Pos) ->
- lseek_position({bof, Pos});
-lseek_position(bof) ->
- lseek_position({bof, 0});
-lseek_position(cur) ->
- lseek_position({cur, 0});
-lseek_position(eof) ->
- lseek_position({eof, 0});
-lseek_position({bof, Offset})
- when is_integer(Offset) ->
- {Offset, ?EFILE_SEEK_SET};
-lseek_position({cur, Offset})
- when is_integer(Offset) ->
- {Offset, ?EFILE_SEEK_CUR};
-lseek_position({eof, Offset})
- when is_integer(Offset) ->
- {Offset, ?EFILE_SEEK_END};
-lseek_position(_) ->
- badarg.
-
-
-
-%% Translates the response from the driver into
-%% {ok, Result} or {error, Reason}.
-
--dialyzer({no_improper_lists, translate_response/2}).
-translate_response(?FILE_RESP_OK, []) ->
- ok;
-translate_response(?FILE_RESP_ERROR, List) when is_list(List) ->
- {error, list_to_atom(List)};
-translate_response(?FILE_RESP_NUMBER, List) ->
- {N, []} = get_uint64(List),
- {ok, N};
-translate_response(?FILE_RESP_DATA, List) ->
- {_N, _Data} = ND = get_uint64(List),
- {ok, ND};
-translate_response(?FILE_RESP_INFO, List) when is_list(List) ->
- {ok, transform_info(List)};
-translate_response(?FILE_RESP_NUMERR, L0) ->
- {N, L1} = get_uint64(L0),
- {error, {N, list_to_atom(L1)}};
-translate_response(?FILE_RESP_LDATA, List) ->
- {ok, transform_ldata(List)};
-translate_response(?FILE_RESP_N2DATA,
- <<Offset:64, 0:64, Size:64>>) ->
- {ok, {Size, Offset, eof}};
-translate_response(?FILE_RESP_N2DATA,
- [<<Offset:64, 0:64, Size:64>> | <<>>]) ->
- {ok, {Size, Offset, eof}};
-translate_response(?FILE_RESP_N2DATA = X,
- [<<_:64, 0:64, _:64>> | _] = Data) ->
- {error, {bad_response_from_port, [X | Data]}};
-translate_response(?FILE_RESP_N2DATA = X,
- [<<_:64, _:64, _:64>> | <<>>] = Data) ->
- {error, {bad_response_from_port, [X | Data]}};
-translate_response(?FILE_RESP_N2DATA,
- [<<Offset:64, _ReadSize:64, Size:64>> | D]) ->
- {ok, {Size, Offset, D}};
-translate_response(?FILE_RESP_N2DATA = X, L0) when is_list(L0) ->
- {Offset, L1} = get_uint64(L0),
- {ReadSize, L2} = get_uint64(L1),
- {Size, L3} = get_uint64(L2),
- case {ReadSize, L3} of
- {0, []} ->
- {ok, {Size, Offset, eof}};
- {0, _} ->
- {error, {bad_response_from_port, [X | L0]}};
- {_, []} ->
- {error, {bad_response_from_port, [X | L0]}};
- _ ->
- {ok, {Size, Offset, L3}}
- end;
-translate_response(?FILE_RESP_EOF, []) ->
- eof;
-translate_response(?FILE_RESP_FNAME, Data) ->
- {fname, Data};
-translate_response(?FILE_RESP_LFNAME, Data) ->
- {lfname, Data};
-translate_response(?FILE_RESP_ALL_DATA, Data) ->
- {ok, Data};
-translate_response(X, Data) ->
- {error, {bad_response_from_port, [X | Data]}}.
-
-transform_info([
- Hsize1, Hsize2, Hsize3, Hsize4,
- Lsize1, Lsize2, Lsize3, Lsize4,
- Type1, Type2, Type3, Type4,
- Atime1, Atime2, Atime3, Atime4, Atime5, Atime6, Atime7, Atime8,
- Mtime1, Mtime2, Mtime3, Mtime4, Mtime5, Mtime6, Mtime7, Mtime8,
- Ctime1, Ctime2, Ctime3, Ctime4, Ctime5, Ctime6, Ctime7, Ctime8,
- Mode1, Mode2, Mode3, Mode4,
- Links1, Links2, Links3, Links4,
- Major1, Major2, Major3, Major4,
- Minor1, Minor2, Minor3, Minor4,
- Inode1, Inode2, Inode3, Inode4,
- Uid1, Uid2, Uid3, Uid4,
- Gid1, Gid2, Gid3, Gid4,
- Access1,Access2,Access3,Access4]) ->
- #file_info {
- size = uint32(Hsize1,Hsize2,Hsize3,Hsize4)*16#100000000 + uint32(Lsize1,Lsize2,Lsize3,Lsize4),
- type = file_type(uint32(Type1,Type2,Type3,Type4)),
- access = file_access(uint32(Access1,Access2,Access3,Access4)),
- atime = sint64(Atime1, Atime2, Atime3, Atime4, Atime5, Atime6, Atime7, Atime8),
- mtime = sint64(Mtime1, Mtime2, Mtime3, Mtime4, Mtime5, Mtime6, Mtime7, Mtime8),
- ctime = sint64(Ctime1, Ctime2, Ctime3, Ctime4, Ctime5, Ctime6, Ctime7, Ctime8),
- mode = uint32(Mode1,Mode2,Mode3,Mode4),
- links = uint32(Links1,Links2,Links3,Links4),
- major_device = uint32(Major1,Major2,Major3,Major4),
- minor_device = uint32(Minor1,Minor2,Minor3,Minor4),
- inode = uint32(Inode1,Inode2,Inode3,Inode4),
- uid = uint32(Uid1,Uid2,Uid3,Uid4),
- gid = uint32(Gid1,Gid2,Gid3,Gid4)
- }.
-
-
-file_type(1) -> device;
-file_type(2) -> directory;
-file_type(3) -> regular;
-file_type(4) -> symlink;
-file_type(_) -> other.
-
-file_access(0) -> none;
-file_access(1) -> write;
-file_access(2) -> read;
-file_access(3) -> read_write.
-
-int_to_int32bytes(Int) when is_integer(Int) ->
- <<Int:32>>;
-int_to_int32bytes(undefined) ->
- <<-1:32>>.
-
-int_to_int64bytes(Int) when is_integer(Int) ->
- <<Int:64/signed>>.
-
-
-sint64(I1,I2,I3,I4,I5,I6,I7,I8) when I1 > 127 ->
- ((I1 bsl 56) bor (I2 bsl 48) bor (I3 bsl 40) bor (I4 bsl 32) bor
- (I5 bsl 24) bor (I6 bsl 16) bor (I7 bsl 8) bor I8) - (1 bsl 64);
-sint64(I1,I2,I3,I4,I5,I6,I7,I8) ->
- ((I1 bsl 56) bor (I2 bsl 48) bor (I3 bsl 40) bor (I4 bsl 32) bor
- (I5 bsl 24) bor (I6 bsl 16) bor (I7 bsl 8) bor I8).
-
-
-uint32(X1,X2,X3,X4) ->
- (X1 bsl 24) bor (X2 bsl 16) bor (X3 bsl 8) bor X4.
-
-get_uint64(L0) ->
- {X1, L1} = get_uint32(L0),
- {X2, L2} = get_uint32(L1),
- {(X1 bsl 32) bor X2, L2}.
-
-get_uint32([X1,X2,X3,X4|List]) ->
- {(((((X1 bsl 8) bor X2) bsl 8) bor X3) bsl 8) bor X4, List}.
-
-
-%% Binary mode
-transform_ldata(<<0:32, 0:32>>) ->
- [];
-transform_ldata([<<0:32, N:32, Sizes/binary>> | Datas]) ->
- transform_ldata(N, Sizes, Datas, []);
-%% List mode
-transform_ldata([_,_,_,_,_,_,_,_|_] = L0) ->
- {0, L1} = get_uint32(L0),
- {N, L2} = get_uint32(L1),
- transform_ldata(N, L2, []).
-
-%% List mode
-transform_ldata(0, List, Sizes) ->
- transform_ldata(0, List, reverse(Sizes), []);
-transform_ldata(N, L0, Sizes) ->
- {Size, L1} = get_uint64(L0),
- transform_ldata(N-1, L1, [Size | Sizes]).
-
-%% Binary mode
-transform_ldata(1, <<0:64>>, <<>>, R) ->
- reverse(R, [eof]);
-transform_ldata(1, <<Size:64>>, Data, R)
- when byte_size(Data) =:= Size ->
- reverse(R, [Data]);
-transform_ldata(N, <<0:64, Sizes/binary>>, [<<>> | Datas], R) ->
- transform_ldata(N-1, Sizes, Datas, [eof | R]);
-transform_ldata(N, <<Size:64, Sizes/binary>>, [Data | Datas], R)
- when byte_size(Data) =:= Size ->
- transform_ldata(N-1, Sizes, Datas, [Data | R]);
-%% List mode
-transform_ldata(0, [], [], R) ->
- reverse(R);
-transform_ldata(0, List, [0 | Sizes], R) ->
- transform_ldata(0, List, Sizes, [eof | R]);
-transform_ldata(0, List, [Size | Sizes], R) ->
- {Front, Rear} = lists_split(List, Size),
- transform_ldata(0, Rear, Sizes, [Front | R]).
-
-lists_split(List, 0) when is_list(List) ->
- {[], List};
-lists_split(List, N) when is_list(List), is_integer(N), N < 0 ->
- erlang:error(badarg, [List, N]);
-lists_split(List, N) when is_list(List), is_integer(N) ->
- case lists_split(List, N, []) of
- premature_end_of_list ->
- erlang:error(badarg, [List, N]);
- Result ->
- Result
+make_symlink(Existing, New) ->
+ try
+ make_soft_link_nif(encode_path(Existing), encode_path(New))
+ catch
+ error:badarg -> {error, badarg}
end.
-lists_split(List, 0, Rev) ->
- {reverse(Rev), List};
-lists_split([], _, _) ->
- premature_end_of_list;
-lists_split([Hd | Tl], N, Rev) ->
- lists_split(Tl, N-1, [Hd | Rev]).
-
-%% We KNOW that lists:reverse/2 is a BIF.
-
-reverse(X) -> lists:reverse(X, []).
-reverse(L, T) -> lists:reverse(L, T).
+altname(Path) ->
+ try altname_nif(encode_path(Path)) of
+ {ok, RawPath} -> {ok, decode_path(RawPath)};
+ Other -> Other
+ catch
+ error:badarg -> {error, badarg}
+ end.
-% Will add zero termination too
-% The 'EXIT' tuple from a bad argument will eventually generate an error
-% in list_to_binary, which is caught and generates the {error,badarg} return
-pathname(File) ->
- (catch prim_file:internal_name2native(File)).
+list_dir_nif(_Path) ->
+ erlang:nif_error(undef).
+read_link_nif(_Path) ->
+ erlang:nif_error(undef).
+read_info_nif(_Path, _FollowLinks) ->
+ erlang:nif_error(undef).
+make_hard_link_nif(_Existing, _New) ->
+ erlang:nif_error(undef).
+make_soft_link_nif(_Existing, _New) ->
+ erlang:nif_error(undef).
+rename_nif(_Source, _Destination) ->
+ erlang:nif_error(undef).
+make_dir_nif(_Path) ->
+ erlang:nif_error(undef).
+del_file_nif(_Path) ->
+ erlang:nif_error(undef).
+del_dir_nif(_Path) ->
+ erlang:nif_error(undef).
+get_device_cwd_nif(_DevicePath) ->
+ erlang:nif_error(undef).
+set_cwd_nif(_Path) ->
+ erlang:nif_error(undef).
+get_cwd_nif() ->
+ erlang:nif_error(undef).
+altname_nif(_Path) ->
+ erlang:nif_error(undef).
+%%
+%% General helper functions.
+%%
-%% proplist:get_value/3
-plgv(K, [{K, V}|_], _) -> V;
-plgv(K, [_|KVs], D) -> plgv(K, KVs, D);
-plgv(_, [], D) -> D.
+%% We know for certain that lists:reverse/2 is a BIF, so it's safe to use it
+%% even though this module is preloaded.
+reverse_list(List) -> lists:reverse(List).
+
+proplist_get_value(_Key, [], Default) ->
+ Default;
+proplist_get_value(Key, [{Key, Value} | _Rest], _Default) ->
+ Value;
+proplist_get_value(Key, [Key | _Rest], _Default) ->
+ true;
+proplist_get_value(Key, [_Other | Rest], Default) ->
+ proplist_get_value(Key, Rest, Default).
+
+encode_path(Path) ->
+ prim_file:internal_name2native(Path).
+decode_path(NativePath) when is_binary(NativePath) ->
+ prim_file:internal_native2name(NativePath).
+
+is_path_translatable(Path) when is_list(Path) ->
+ true;
+is_path_translatable(Path) ->
+ prim_file:is_translatable(Path).
-%%
%% We don't actually want this here
+%%
%% We want to use posix time in all prim but erl_prim_loader makes that tricky
%% It is probably needed to redo the whole erl_prim_loader
-from_seconds(Seconds, posix) when is_integer(Seconds) ->
+from_posix_seconds(Seconds, posix) when is_integer(Seconds) ->
Seconds;
-from_seconds(Seconds, universal) when is_integer(Seconds) ->
+from_posix_seconds(Seconds, universal) when is_integer(Seconds) ->
erlang:posixtime_to_universaltime(Seconds);
-from_seconds(Seconds, local) when is_integer(Seconds) ->
+from_posix_seconds(Seconds, local) when is_integer(Seconds) ->
erlang:universaltime_to_localtime(erlang:posixtime_to_universaltime(Seconds)).
-to_seconds(Seconds, posix) when is_integer(Seconds) ->
+to_posix_seconds(Seconds, posix) when is_integer(Seconds) ->
Seconds;
-to_seconds({_,_} = Datetime, universal) ->
+to_posix_seconds({_,_} = Datetime, universal) ->
erlang:universaltime_to_posixtime(Datetime);
-to_seconds({_,_} = Datetime, local) ->
+to_posix_seconds({_,_} = Datetime, local) ->
erlang:universaltime_to_posixtime(erlang:localtime_to_universaltime(Datetime)).
diff --git a/erts/preloaded/src/prim_inet.erl b/erts/preloaded/src/prim_inet.erl
index 017a706a8b..2a3605260d 100644
--- a/erts/preloaded/src/prim_inet.erl
+++ b/erts/preloaded/src/prim_inet.erl
@@ -31,7 +31,7 @@
-export([connect/3, connect/4, async_connect/4]).
-export([accept/1, accept/2, async_accept/2]).
-export([shutdown/2]).
--export([send/2, send/3, sendto/4, sendmsg/3]).
+-export([send/2, send/3, sendto/4, sendmsg/3, sendfile/4]).
-export([recv/2, recv/3, async_recv/3]).
-export([unrecv/2]).
-export([recvfrom/2, recvfrom/3]).
@@ -500,6 +500,55 @@ sendmsg(S, #sctp_sndrcvinfo{}=SRI, Data) when is_port(S) ->
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%
+%% SENDFILE(outsock(), Fd, Offset, Length) -> {ok,BytesSent} | {error, Reason}
+%%
+%% send Length data bytes from a file handle, to a socket, starting at Offset
+%%
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%% "sendfile" is for TCP:
+
+sendfile(S, FileHandle, Offset, Length)
+ when not is_port(S);
+ not is_binary(FileHandle);
+ not is_integer(Offset);
+ not is_integer(Length) ->
+ {error, badarg};
+sendfile(S, FileHandle, Offset, Length) ->
+ case erlang:port_info(S, connected) of
+ {connected, Pid} when Pid =:= self() ->
+ sendfile_1(S, FileHandle, Offset, Length);
+ {connected, Pid} when Pid =/= self() ->
+ {error, not_owner};
+ _Other ->
+ {error, einval}
+ end.
+
+sendfile_1(S, FileHandle, Offset, 0) ->
+ sendfile_1(S, FileHandle, Offset, (1 bsl 63) - 1);
+sendfile_1(_S, _FileHandle, Offset, Length) when
+ Offset < 0; Offset > ((1 bsl 63) - 1);
+ Length < 0; Length > ((1 bsl 63) - 1) ->
+ {error, einval};
+sendfile_1(S, FileHandle, Offset, Length) ->
+ Args = [FileHandle,
+ ?int64(Offset),
+ ?int64(Length)],
+ case ctl_cmd(S, ?TCP_REQ_SENDFILE, Args) of
+ {ok, []} ->
+ receive
+ {sendfile, S, {ok, SentLow, SentHigh}} ->
+ {ok, SentLow bor (SentHigh bsl 32)};
+ {sendfile, S, {error, Reason}} ->
+ {error, Reason};
+ {'EXIT', S, _Reason} ->
+ {error, closed}
+ end;
+ {error, Reason} ->
+ {error, Reason}
+ end.
+
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%%
%% RECV(insock(), Length, [Timeout]) -> {ok,Data} | {error, Reason}
%%
%% receive Length data bytes from a socket
diff --git a/lib/kernel/src/Makefile b/lib/kernel/src/Makefile
index 5946620f0f..4a713b2a99 100644
--- a/lib/kernel/src/Makefile
+++ b/lib/kernel/src/Makefile
@@ -120,6 +120,13 @@ MODULES = \
user \
user_drv \
user_sup \
+ raw_file_io \
+ raw_file_io_compressed \
+ raw_file_io_inflate \
+ raw_file_io_deflate \
+ raw_file_io_delayed \
+ raw_file_io_list \
+ raw_file_io_raw \
wrap_log_reader
HRL_FILES= ../include/file.hrl ../include/inet.hrl ../include/inet_sctp.hrl \
@@ -226,7 +233,8 @@ $(EBIN)/disk_log_server.beam: disk_log.hrl
$(EBIN)/dist_util.beam: ../include/dist_util.hrl ../include/dist.hrl
$(EBIN)/erl_boot_server.beam: inet_boot.hrl
$(EBIN)/erl_epmd.beam: inet_int.hrl erl_epmd.hrl
-$(EBIN)/file.beam: ../include/file.hrl
+$(EBIN)/file.beam: ../include/file.hrl file_int.hrl
+$(EBIN)/file_io_server.beam: ../include/file.hrl file_int.hrl
$(EBIN)/gen_tcp.beam: inet_int.hrl
$(EBIN)/gen_udp.beam: inet_int.hrl
$(EBIN)/gen_sctp.beam: ../include/inet_sctp.hrl
@@ -254,3 +262,10 @@ $(EBIN)/net_kernel.beam: ../include/net_address.hrl
$(EBIN)/os.beam: ../include/file.hrl
$(EBIN)/ram_file.beam: ../include/file.hrl
$(EBIN)/wrap_log_reader.beam: disk_log.hrl ../include/file.hrl
+$(EBIN)/raw_file_io.beam: ../include/file.hrl file_int.hrl
+$(EBIN)/raw_file_io_compressed.beam: ../include/file.hrl file_int.hrl
+$(EBIN)/raw_file_io_inflate.beam: ../include/file.hrl file_int.hrl
+$(EBIN)/raw_file_io_deflate.beam: ../include/file.hrl file_int.hrl
+$(EBIN)/raw_file_io_delayed.beam: ../include/file.hrl file_int.hrl
+$(EBIN)/raw_file_io_list.beam: ../include/file.hrl file_int.hrl
+$(EBIN)/raw_file_io_raw.beam: ../include/file.hrl file_int.hrl
diff --git a/lib/kernel/src/file.erl b/lib/kernel/src/file.erl
index 933f2d5f65..b5a51c3410 100644
--- a/lib/kernel/src/file.erl
+++ b/lib/kernel/src/file.erl
@@ -72,7 +72,7 @@
io_device/0, name/0, name_all/0, posix/0]).
%%% Includes and defines
--include("file.hrl").
+-include("file_int.hrl").
-define(FILE_IO_SERVER_TABLE, file_io_servers).
@@ -454,41 +454,23 @@ raw_write_file_info(Name, #file_info{} = Info) ->
Reason :: posix() | badarg | system_limit.
open(Item, ModeList) when is_list(ModeList) ->
- case lists:member(raw, ModeList) of
- %% Raw file, use ?PRIM_FILE to handle this file
- true ->
+ case {lists:member(raw, ModeList), lists:member(ram, ModeList)} of
+ {false, false} ->
+ %% File server file
Args = [file_name(Item) | ModeList],
case check_args(Args) of
ok ->
[FileName | _] = Args,
- %% We rely on the returned Handle (in {ok, Handle})
- %% being a pid() or a #file_descriptor{}
- ?PRIM_FILE:open(FileName, ModeList);
+ call(open, [FileName, ModeList]);
Error ->
Error
- end;
- false ->
- case lists:member(ram, ModeList) of
- %% RAM file, use ?RAM_FILE to handle this file
- true ->
- case check_args(ModeList) of
- ok ->
- ?RAM_FILE:open(Item, ModeList);
- Error ->
- Error
- end;
- %% File server file
- false ->
- Args = [file_name(Item) | ModeList],
- case check_args(Args) of
- ok ->
- [FileName | _] = Args,
- call(open, [FileName, ModeList]);
- Error ->
- Error
- end
- end
+ end;
+ {true, _Either} ->
+ raw_file_io:open(file_name(Item), ModeList);
+ {false, true} ->
+ ram_file:open(Item, ModeList)
end;
+
%% Old obsolete mode specification in atom or 2-tuple format
open(Item, Mode) ->
open(Item, mode_list(Mode)).
diff --git a/lib/kernel/src/file_int.hrl b/lib/kernel/src/file_int.hrl
new file mode 100644
index 0000000000..bafc330c04
--- /dev/null
+++ b/lib/kernel/src/file_int.hrl
@@ -0,0 +1,33 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 2017. All Rights Reserved.
+%%
+%% Licensed under the Apache License, Version 2.0 (the "License");
+%% you may not use this file except in compliance with the License.
+%% You may obtain a copy of the License at
+%%
+%% http://www.apache.org/licenses/LICENSE-2.0
+%%
+%% Unless required by applicable law or agreed to in writing, software
+%% distributed under the License is distributed on an "AS IS" BASIS,
+%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+%% See the License for the specific language governing permissions and
+%% limitations under the License.
+%%
+%% %CopyrightEnd%
+%%
+
+%%
+%% Internal definitions for the 'file' module and friends.
+%%
+
+-ifndef(FILE_INTERNAL_HRL_).
+-define(FILE_INTERNAL_HRL_, 1).
+
+-include("file.hrl").
+
+-define(CALL_FD(Fd, Method, Args),
+ apply(Fd#file_descriptor.module, Method, [Fd | Args])).
+
+-endif.
diff --git a/lib/kernel/src/file_io_server.erl b/lib/kernel/src/file_io_server.erl
index deb7b315b1..2b35d2acfb 100644
--- a/lib/kernel/src/file_io_server.erl
+++ b/lib/kernel/src/file_io_server.erl
@@ -1,7 +1,7 @@
%%
%% %CopyrightBegin%
%%
-%% Copyright Ericsson AB 2000-2015. All Rights Reserved.
+%% Copyright Ericsson AB 2000-2017. All Rights Reserved.
%%
%% Licensed under the Apache License, Version 2.0 (the "License");
%% you may not use this file except in compliance with the License.
@@ -28,7 +28,8 @@
-record(state, {handle,owner,mref,buf,read_mode,unic}).
--define(PRIM_FILE, prim_file).
+-include("file_int.hrl").
+
-define(READ_SIZE_LIST, 128).
-define(READ_SIZE_BINARY, (8*1024)).
@@ -68,7 +69,7 @@ do_start(Spawn, Owner, FileName, ModeList) ->
%% process_flag(trap_exit, true),
case parse_options(ModeList) of
{ReadMode, UnicodeMode, Opts} ->
- case ?PRIM_FILE:open(FileName, Opts) of
+ case raw_file_io:open(FileName, [raw | Opts]) of
{error, Reason} = Error ->
Self ! {Ref, Error},
exit(Reason);
@@ -205,7 +206,7 @@ io_reply(From, ReplyAs, Reply) ->
file_request({advise,Offset,Length,Advise},
#state{handle=Handle}=State) ->
- case ?PRIM_FILE:advise(Handle, Offset, Length, Advise) of
+ case ?CALL_FD(Handle, advise, [Offset, Length, Advise]) of
{error,Reason}=Reply ->
{stop,Reason,Reply,State};
Reply ->
@@ -213,7 +214,7 @@ file_request({advise,Offset,Length,Advise},
end;
file_request({allocate, Offset, Length},
#state{handle = Handle} = State) ->
- Reply = ?PRIM_FILE:allocate(Handle, Offset, Length),
+ Reply = ?CALL_FD(Handle, allocate, [Offset, Length]),
{reply, Reply, State};
file_request({pread,At,Sz}, State)
when At =:= cur;
@@ -256,7 +257,7 @@ file_request({pwrite,At,Data},
end;
file_request(datasync,
#state{handle=Handle}=State) ->
- case ?PRIM_FILE:datasync(Handle) of
+ case ?CALL_FD(Handle, datasync, []) of
{error,Reason}=Reply ->
{stop,Reason,Reply,State};
Reply ->
@@ -264,7 +265,7 @@ file_request(datasync,
end;
file_request(sync,
#state{handle=Handle}=State) ->
- case ?PRIM_FILE:sync(Handle) of
+ case ?CALL_FD(Handle, sync, []) of
{error,Reason}=Reply ->
{stop,Reason,Reply,State};
Reply ->
@@ -272,7 +273,7 @@ file_request(sync,
end;
file_request(close,
#state{handle=Handle}=State) ->
- case ?PRIM_FILE:close(Handle) of
+ case ?CALL_FD(Handle, close, []) of
{error,Reason}=Reply ->
{stop,Reason,Reply,State#state{buf= <<>>}};
Reply ->
@@ -288,7 +289,7 @@ file_request({position,At},
end;
file_request(truncate,
#state{handle=Handle}=State) ->
- case ?PRIM_FILE:truncate(Handle) of
+ case ?CALL_FD(Handle, truncate, []) of
{error,Reason}=Reply ->
{stop,Reason,Reply,State#state{buf= <<>>}};
Reply ->
@@ -398,7 +399,7 @@ io_request_loop([Request|Tail],
%%
put_chars(Chars, latin1, #state{handle=Handle, unic=latin1}=State) ->
NewState = State#state{buf = <<>>},
- case ?PRIM_FILE:write(Handle, Chars) of
+ case ?CALL_FD(Handle, write, [Chars]) of
{error,Reason}=Reply ->
{stop,Reason,Reply,NewState};
Reply ->
@@ -408,7 +409,7 @@ put_chars(Chars, InEncoding, #state{handle=Handle, unic=OutEncoding}=State) ->
NewState = State#state{buf = <<>>},
case unicode:characters_to_binary(Chars,InEncoding,OutEncoding) of
Bin when is_binary(Bin) ->
- case ?PRIM_FILE:write(Handle, Bin) of
+ case ?CALL_FD(Handle, write, [Bin]) of
{error,Reason}=Reply ->
{stop,Reason,Reply,NewState};
Reply ->
@@ -422,7 +423,7 @@ put_chars(Chars, InEncoding, #state{handle=Handle, unic=OutEncoding}=State) ->
get_line(S, {<<>>, Cont}, OutEnc,
#state{handle=Handle, read_mode=Mode, unic=InEnc}=State) ->
- case ?PRIM_FILE:read(Handle, read_size(Mode)) of
+ case ?CALL_FD(Handle, read, [read_size(Mode)]) of
{ok,Bin} ->
get_line(S, convert_enc([Cont, Bin], InEnc, OutEnc), OutEnc, State);
eof ->
@@ -472,7 +473,7 @@ get_chars(N, OutEnc,#state{handle=Handle,buf=Buf,read_mode=ReadMode,unic=latin1}
BufSize = byte_size(Buf),
NeedSize = N-BufSize,
Size = erlang:max(NeedSize, ?READ_SIZE_BINARY),
- case ?PRIM_FILE:read(Handle, Size) of
+ case ?CALL_FD(Handle, read, [Size]) of
{ok, B} ->
if BufSize+byte_size(B) < N ->
std_reply(cat(Buf, B, ReadMode,latin1,OutEnc), State);
@@ -504,7 +505,7 @@ get_chars(N, OutEnc,#state{handle=Handle,buf=Buf,read_mode=ReadMode,unic=InEncod
%% Need more, Try to read 4*needed in bytes...
NeedSize = (N - BufCount) * 4,
Size = erlang:max(NeedSize, ?READ_SIZE_BINARY),
- case ?PRIM_FILE:read(Handle, Size) of
+ case ?CALL_FD(Handle, read, [Size]) of
{ok, B} ->
NewBuf = list_to_binary([Buf,B]),
{NewCount,NewSplit} = count_and_find(NewBuf,N,InEncoding),
@@ -544,7 +545,7 @@ get_chars(Mod, Func, XtraArg, OutEnc, #state{buf=Buf}=State) ->
get_chars_empty(Mod, Func, XtraArg, S, latin1,
#state{handle=Handle,read_mode=ReadMode, unic=latin1}=State) ->
- case ?PRIM_FILE:read(Handle, read_size(ReadMode)) of
+ case ?CALL_FD(Handle, read, [read_size(ReadMode)]) of
{ok,Bin} ->
get_chars_apply(Mod, Func, XtraArg, S, latin1, State, Bin);
eof ->
@@ -554,7 +555,7 @@ get_chars_empty(Mod, Func, XtraArg, S, latin1,
end;
get_chars_empty(Mod, Func, XtraArg, S, OutEnc,
#state{handle=Handle,read_mode=ReadMode}=State) ->
- case ?PRIM_FILE:read(Handle, read_size(ReadMode)) of
+ case ?CALL_FD(Handle, read, [read_size(ReadMode)]) of
{ok,Bin} ->
get_chars_apply(Mod, Func, XtraArg, S, OutEnc, State, Bin);
eof ->
@@ -564,7 +565,7 @@ get_chars_empty(Mod, Func, XtraArg, S, OutEnc,
end.
get_chars_notempty(Mod, Func, XtraArg, S, OutEnc,
#state{handle=Handle,read_mode=ReadMode,buf = B}=State) ->
- case ?PRIM_FILE:read(Handle, read_size(ReadMode)) of
+ case ?CALL_FD(Handle, read, [read_size(ReadMode)]) of
{ok,Bin} ->
get_chars_apply(Mod, Func, XtraArg, S, OutEnc, State, list_to_binary([B,Bin]));
eof ->
@@ -918,13 +919,10 @@ cbv({utf32,little},_) ->
%% Compensates ?PRIM_FILE:position/2 for the number of bytes
%% we have buffered
position(Handle, At, Buf) ->
- ?PRIM_FILE:position(
- Handle,
- case At of
- cur ->
- {cur, -byte_size(Buf)};
- {cur, Offs} ->
- {cur, Offs-byte_size(Buf)};
- _ ->
- At
- end).
+ SeekTo =
+ case At of
+ {cur, Offs} -> {cur, Offs-byte_size(Buf)};
+ cur -> {cur, -byte_size(Buf)};
+ _ -> At
+ end,
+ ?CALL_FD(Handle, position, [SeekTo]).
diff --git a/lib/kernel/src/file_server.erl b/lib/kernel/src/file_server.erl
index 6e8f64d932..ecc1ffbdd6 100644
--- a/lib/kernel/src/file_server.erl
+++ b/lib/kernel/src/file_server.erl
@@ -63,7 +63,7 @@ stop() ->
%%% Callback functions from gen_server
%%%----------------------------------------------------------------------
--type state() :: port(). % Internal type
+-type state() :: term(). % Internal type
%%----------------------------------------------------------------------
%% Func: init/1
@@ -77,14 +77,8 @@ stop() ->
init([]) ->
process_flag(trap_exit, true),
- case ?PRIM_FILE:start() of
- {ok, Handle} ->
- ?FILE_IO_SERVER_TABLE =
- ets:new(?FILE_IO_SERVER_TABLE, [named_table]),
- {ok, Handle};
- {error, Reason} ->
- {stop, Reason}
- end.
+ ?FILE_IO_SERVER_TABLE = ets:new(?FILE_IO_SERVER_TABLE, [named_table]),
+ {ok, undefined}.
%%----------------------------------------------------------------------
%% Func: handle_call/3
@@ -101,7 +95,7 @@ init([]) ->
{'reply', 'eof' | 'ok' | {'error', term()} | {'ok', term()}, state()} |
{'stop', 'normal', 'stopped', state()}.
-handle_call({open, Name, ModeList}, {Pid, _Tag} = _From, Handle)
+handle_call({open, Name, ModeList}, {Pid, _Tag} = _From, State)
when is_list(ModeList) ->
Child = ?FILE_IO_SERVER:start_link(Pid, Name, ModeList),
case Child of
@@ -110,78 +104,78 @@ handle_call({open, Name, ModeList}, {Pid, _Tag} = _From, Handle)
_ ->
ok
end,
- {reply, Child, Handle};
+ {reply, Child, State};
-handle_call({open, _Name, _Mode}, _From, Handle) ->
- {reply, {error, einval}, Handle};
+handle_call({open, _Name, _Mode}, _From, State) ->
+ {reply, {error, einval}, State};
-handle_call({read_file, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:read_file(Name), Handle};
+handle_call({read_file, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:read_file(Name), State};
-handle_call({write_file, Name, Bin}, _From, Handle) ->
- {reply, ?PRIM_FILE:write_file(Name, Bin), Handle};
+handle_call({write_file, Name, Bin}, _From, State) ->
+ {reply, ?PRIM_FILE:write_file(Name, Bin), State};
-handle_call({set_cwd, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:set_cwd(Handle, Name), Handle};
+handle_call({set_cwd, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:set_cwd(Name), State};
-handle_call({delete, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:delete(Handle, Name), Handle};
+handle_call({delete, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:delete(Name), State};
-handle_call({rename, Fr, To}, _From, Handle) ->
- {reply, ?PRIM_FILE:rename(Handle, Fr, To), Handle};
+handle_call({rename, Fr, To}, _From, State) ->
+ {reply, ?PRIM_FILE:rename(Fr, To), State};
-handle_call({make_dir, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:make_dir(Handle, Name), Handle};
+handle_call({make_dir, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:make_dir(Name), State};
-handle_call({del_dir, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:del_dir(Handle, Name), Handle};
+handle_call({del_dir, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:del_dir(Name), State};
-handle_call({list_dir, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:list_dir(Handle, Name), Handle};
-handle_call({list_dir_all, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:list_dir_all(Handle, Name), Handle};
+handle_call({list_dir, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:list_dir(Name), State};
+handle_call({list_dir_all, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:list_dir_all(Name), State};
-handle_call(get_cwd, _From, Handle) ->
- {reply, ?PRIM_FILE:get_cwd(Handle), Handle};
-handle_call({get_cwd}, _From, Handle) ->
- {reply, ?PRIM_FILE:get_cwd(Handle), Handle};
-handle_call({get_cwd, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:get_cwd(Handle, Name), Handle};
+handle_call(get_cwd, _From, State) ->
+ {reply, ?PRIM_FILE:get_cwd(), State};
+handle_call({get_cwd}, _From, State) ->
+ {reply, ?PRIM_FILE:get_cwd(), State};
+handle_call({get_cwd, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:get_cwd(Name), State};
-handle_call({read_file_info, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:read_file_info(Handle, Name), Handle};
+handle_call({read_file_info, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:read_file_info(Name), State};
-handle_call({read_file_info, Name, Opts}, _From, Handle) ->
- {reply, ?PRIM_FILE:read_file_info(Handle, Name, Opts), Handle};
+handle_call({read_file_info, Name, Opts}, _From, State) ->
+ {reply, ?PRIM_FILE:read_file_info(Name, Opts), State};
-handle_call({altname, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:altname(Handle, Name), Handle};
+handle_call({altname, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:altname(Name), State};
-handle_call({write_file_info, Name, Info}, _From, Handle) ->
- {reply, ?PRIM_FILE:write_file_info(Handle, Name, Info), Handle};
+handle_call({write_file_info, Name, Info}, _From, State) ->
+ {reply, ?PRIM_FILE:write_file_info(Name, Info), State};
-handle_call({write_file_info, Name, Info, Opts}, _From, Handle) ->
- {reply, ?PRIM_FILE:write_file_info(Handle, Name, Info, Opts), Handle};
+handle_call({write_file_info, Name, Info, Opts}, _From, State) ->
+ {reply, ?PRIM_FILE:write_file_info(Name, Info, Opts), State};
-handle_call({read_link_info, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:read_link_info(Handle, Name), Handle};
+handle_call({read_link_info, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:read_link_info(Name), State};
-handle_call({read_link_info, Name, Opts}, _From, Handle) ->
- {reply, ?PRIM_FILE:read_link_info(Handle, Name, Opts), Handle};
+handle_call({read_link_info, Name, Opts}, _From, State) ->
+ {reply, ?PRIM_FILE:read_link_info(Name, Opts), State};
-handle_call({read_link, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:read_link(Handle, Name), Handle};
-handle_call({read_link_all, Name}, _From, Handle) ->
- {reply, ?PRIM_FILE:read_link_all(Handle, Name), Handle};
+handle_call({read_link, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:read_link(Name), State};
+handle_call({read_link_all, Name}, _From, State) ->
+ {reply, ?PRIM_FILE:read_link_all(Name), State};
-handle_call({make_link, Old, New}, _From, Handle) ->
- {reply, ?PRIM_FILE:make_link(Handle, Old, New), Handle};
+handle_call({make_link, Old, New}, _From, State) ->
+ {reply, ?PRIM_FILE:make_link(Old, New), State};
-handle_call({make_symlink, Old, New}, _From, Handle) ->
- {reply, ?PRIM_FILE:make_symlink(Handle, Old, New), Handle};
+handle_call({make_symlink, Old, New}, _From, State) ->
+ {reply, ?PRIM_FILE:make_symlink(Old, New), State};
handle_call({copy, SourceName, SourceOpts, DestName, DestOpts, Length},
- _From, Handle) ->
+ _From, State) ->
Reply =
case ?PRIM_FILE:open(SourceName, [read, binary | SourceOpts]) of
{ok, Source} ->
@@ -201,14 +195,14 @@ handle_call({copy, SourceName, SourceOpts, DestName, DestOpts, Length},
{error, _} = Error ->
Error
end,
- {reply, Reply, Handle};
+ {reply, Reply, State};
-handle_call(stop, _From, Handle) ->
- {stop, normal, stopped, Handle};
+handle_call(stop, _From, State) ->
+ {stop, normal, stopped, State};
-handle_call(Request, From, Handle) ->
+handle_call(Request, From, State) ->
error_logger:error_msg("handle_call(~tp, ~tp, _)", [Request, From]),
- {noreply, Handle}.
+ {noreply, State}.
%%----------------------------------------------------------------------
%% Func: handle_cast/2
@@ -233,14 +227,9 @@ handle_cast(Msg, State) ->
-spec handle_info(term(), state()) ->
{'noreply', state()} | {'stop', 'normal', state()}.
-handle_info({'EXIT', Pid, _Reason}, Handle) when is_pid(Pid) ->
+handle_info({'EXIT', Pid, _Reason}, State) when is_pid(Pid) ->
ets:delete(?FILE_IO_SERVER_TABLE, Pid),
- {noreply, Handle};
-
-handle_info({'EXIT', Handle, _Reason}, Handle) ->
- error_logger:error_msg("Port controlling ~w terminated in ~w",
- [?FILE_SERVER, ?MODULE]),
- {stop, normal, Handle};
+ {noreply, State};
handle_info(Info, State) ->
error_logger:error_msg("handle_Info(~tp, _)", [Info]),
@@ -254,8 +243,8 @@ handle_info(Info, State) ->
-spec terminate(term(), state()) -> 'ok'.
-terminate(_Reason, Handle) ->
- ?PRIM_FILE:stop(Handle).
+terminate(_Reason, _State) ->
+ ok.
%%----------------------------------------------------------------------
%% Func: code_change/3
diff --git a/lib/kernel/src/inet_int.hrl b/lib/kernel/src/inet_int.hrl
index bc5b67f7bf..357e27826c 100644
--- a/lib/kernel/src/inet_int.hrl
+++ b/lib/kernel/src/inet_int.hrl
@@ -100,6 +100,8 @@
-define(TCP_REQ_RECV, 42).
-define(TCP_REQ_UNRECV, 43).
-define(TCP_REQ_SHUTDOWN, 44).
+-define(TCP_REQ_SENDFILE, 45).
+
%% UDP and SCTP requests
-define(PACKET_REQ_RECV, 60).
%%-define(SCTP_REQ_LISTEN, 61). MERGED
@@ -319,6 +321,12 @@
[((X) bsr 24) band 16#ff, ((X) bsr 16) band 16#ff,
((X) bsr 8) band 16#ff, (X) band 16#ff]).
+-define(int64(X),
+ [((X) bsr 56) band 16#ff, ((X) bsr 48) band 16#ff,
+ ((X) bsr 40) band 16#ff, ((X) bsr 32) band 16#ff,
+ ((X) bsr 24) band 16#ff, ((X) bsr 16) band 16#ff,
+ ((X) bsr 8) band 16#ff, (X) band 16#ff]).
+
-define(intAID(X), % For SCTP AssocID
?int32(X)).
diff --git a/lib/kernel/src/kernel.app.src b/lib/kernel/src/kernel.app.src
index 080b11fc4d..e4852a6e75 100644
--- a/lib/kernel/src/kernel.app.src
+++ b/lib/kernel/src/kernel.app.src
@@ -88,6 +88,13 @@
inet_udp,
inet_sctp,
pg2,
+ raw_file_io,
+ raw_file_io_compressed,
+ raw_file_io_deflate,
+ raw_file_io_delayed,
+ raw_file_io_inflate,
+ raw_file_io_list,
+ raw_file_io_raw,
seq_trace,
standard_error,
wrap_log_reader]},
diff --git a/lib/kernel/src/raw_file_io.erl b/lib/kernel/src/raw_file_io.erl
new file mode 100644
index 0000000000..e3c07c8f78
--- /dev/null
+++ b/lib/kernel/src/raw_file_io.erl
@@ -0,0 +1,75 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 2017. All Rights Reserved.
+%%
+%% Licensed under the Apache License, Version 2.0 (the "License");
+%% you may not use this file except in compliance with the License.
+%% You may obtain a copy of the License at
+%%
+%% http://www.apache.org/licenses/LICENSE-2.0
+%%
+%% Unless required by applicable law or agreed to in writing, software
+%% distributed under the License is distributed on an "AS IS" BASIS,
+%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+%% See the License for the specific language governing permissions and
+%% limitations under the License.
+%%
+%% %CopyrightEnd%
+%%
+-module(raw_file_io).
+
+-export([open/2]).
+
+open(Filename, Modes) ->
+ %% Layers are applied in this order, and the listed modules will call this
+ %% function again as necessary. eg. a raw compressed delayed file in list
+ %% mode will walk through [_list -> _compressed -> _delayed -> _raw].
+ ModuleOrder = [{raw_file_io_list, fun match_list/1},
+ {raw_file_io_compressed, fun match_compressed/1},
+ {raw_file_io_delayed, fun match_delayed/1},
+ {raw_file_io_raw, fun match_raw/1}],
+ open_1(ModuleOrder, Filename, add_implicit_modes(Modes)).
+open_1([], _Filename, _Modes) ->
+ error(badarg);
+open_1([{Module, Match} | Rest], Filename, Modes) ->
+ case lists:any(Match, Modes) of
+ true ->
+ {Options, ChildModes} =
+ lists:partition(fun(Mode) -> Match(Mode) end, Modes),
+ Module:open_layer(Filename, ChildModes, Options);
+ false ->
+ open_1(Rest, Filename, Modes)
+ end.
+
+%% 'read' and 'list' mode are enabled unless disabled by another option, so
+%% we'll explicitly add them to avoid duplicating this logic in child layers.
+add_implicit_modes(Modes0) ->
+ Modes1 = add_unless_matched(Modes0, fun match_writable/1, read),
+ add_unless_matched(Modes1, fun match_binary/1, list).
+add_unless_matched(Modes, Match, Default) ->
+ case lists:any(Match, Modes) of
+ false -> [Default | Modes];
+ true -> Modes
+ end.
+
+match_list(list) -> true;
+match_list(_Other) -> false.
+
+match_compressed(compressed) -> true;
+match_compressed(_Other) -> false.
+
+match_delayed({delayed_write, _Size, _Timeout}) -> true;
+match_delayed(delayed_write) -> true;
+match_delayed(_Other) -> false.
+
+match_raw(raw) -> true;
+match_raw(_Other) -> false.
+
+match_writable(write) -> true;
+match_writable(append) -> true;
+match_writable(exclusive) -> true;
+match_writable(_Other) -> false.
+
+match_binary(binary) -> true;
+match_binary(_Other) -> false.
diff --git a/lib/kernel/src/raw_file_io_compressed.erl b/lib/kernel/src/raw_file_io_compressed.erl
new file mode 100644
index 0000000000..d5ab042d25
--- /dev/null
+++ b/lib/kernel/src/raw_file_io_compressed.erl
@@ -0,0 +1,134 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 2017. All Rights Reserved.
+%%
+%% Licensed under the Apache License, Version 2.0 (the "License");
+%% you may not use this file except in compliance with the License.
+%% You may obtain a copy of the License at
+%%
+%% http://www.apache.org/licenses/LICENSE-2.0
+%%
+%% Unless required by applicable law or agreed to in writing, software
+%% distributed under the License is distributed on an "AS IS" BASIS,
+%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+%% See the License for the specific language governing permissions and
+%% limitations under the License.
+%%
+%% %CopyrightEnd%
+%%
+-module(raw_file_io_compressed).
+
+-export([close/1, sync/1, datasync/1, truncate/1, advise/4, allocate/3,
+ position/2, write/2, pwrite/2, pwrite/3,
+ read_line/1, read/2, pread/2, pread/3]).
+
+%% OTP internal.
+-export([ipread_s32bu_p32bu/3, sendfile/8]).
+
+-export([open_layer/3]).
+
+-include("file_int.hrl").
+
+open_layer(Filename, Modes, Options) ->
+ IsAppend = lists:member(append, Modes),
+ IsDeflate = lists:member(write, Modes),
+ IsInflate = lists:member(read, Modes),
+ if
+ IsDeflate, IsInflate; IsAppend ->
+ {error, einval};
+ IsDeflate, not IsInflate ->
+ start_server_module(raw_file_io_deflate, Filename, Modes, Options);
+ IsInflate ->
+ start_server_module(raw_file_io_inflate, Filename, Modes, Options)
+ end.
+
+start_server_module(Module, Filename, Modes, Options) ->
+ Secret = make_ref(),
+ case gen_statem:start(Module, {self(), Secret, Options}, []) of
+ {ok, Pid} -> open_next_layer(Pid, Secret, Filename, Modes);
+ Other -> Other
+ end.
+
+open_next_layer(Pid, Secret, Filename, Modes) ->
+ case gen_statem:call(Pid, {'$open', Secret, Filename, Modes}, infinity) of
+ ok ->
+ PublicFd = #file_descriptor{
+ module = raw_file_io_compressed, data = {self(), Pid} },
+ {ok, PublicFd};
+ Other -> Other
+ end.
+
+close(Fd) ->
+ wrap_call(Fd, [close]).
+
+sync(Fd) ->
+ wrap_call(Fd, [sync]).
+datasync(Fd) ->
+ wrap_call(Fd, [datasync]).
+
+truncate(Fd) ->
+ wrap_call(Fd, [truncate]).
+
+advise(Fd, Offset, Length, Advise) ->
+ wrap_call(Fd, [advise, Offset, Length, Advise]).
+allocate(Fd, Offset, Length) ->
+ wrap_call(Fd, [allocate, Offset, Length]).
+
+position(Fd, Mark) ->
+ wrap_call(Fd, [position, Mark]).
+
+write(Fd, IOData) ->
+ try
+ CompactedData = erlang:iolist_to_iovec(IOData),
+ wrap_call(Fd, [write, CompactedData])
+ catch
+ error:badarg -> {error, badarg}
+ end.
+
+pwrite(Fd, Offset, IOData) ->
+ try
+ CompactedData = erlang:iolist_to_iovec(IOData),
+ wrap_call(Fd, [pwrite, Offset, CompactedData])
+ catch
+ error:badarg -> {error, badarg}
+ end.
+pwrite(Fd, LocBytes) ->
+ try
+ CompactedLocBytes =
+ [ {Offset, erlang:iolist_to_iovec(IOData)} ||
+ {Offset, IOData} <- LocBytes ],
+ wrap_call(Fd, [pwrite, CompactedLocBytes])
+ catch
+ error:badarg -> {error, badarg}
+ end.
+
+read_line(Fd) ->
+ wrap_call(Fd, [read_line]).
+read(Fd, Size) ->
+ wrap_call(Fd, [read, Size]).
+pread(Fd, Offset, Size) ->
+ wrap_call(Fd, [pread, Offset, Size]).
+pread(Fd, LocNums) ->
+ wrap_call(Fd, [pread, LocNums]).
+
+ipread_s32bu_p32bu(Fd, Offset, MaxSize) ->
+ wrap_call(Fd, [ipread_s32bu_p32bu, Offset, MaxSize]).
+
+sendfile(_,_,_,_,_,_,_,_) ->
+ {error, enotsup}.
+
+wrap_call(Fd, Command) ->
+ {_Owner, Pid} = get_fd_data(Fd),
+ try gen_statem:call(Pid, Command, infinity) of
+ Result -> Result
+ catch
+ exit:{noproc, _StackTrace} -> {error, einval}
+ end.
+
+get_fd_data(#file_descriptor{ data = Data }) ->
+ {Owner, _ServerPid} = Data,
+ case self() of
+ Owner -> Data;
+ _ -> error(not_on_controlling_process)
+ end.
diff --git a/lib/kernel/src/raw_file_io_deflate.erl b/lib/kernel/src/raw_file_io_deflate.erl
new file mode 100644
index 0000000000..acfc546743
--- /dev/null
+++ b/lib/kernel/src/raw_file_io_deflate.erl
@@ -0,0 +1,159 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 2017. All Rights Reserved.
+%%
+%% Licensed under the Apache License, Version 2.0 (the "License");
+%% you may not use this file except in compliance with the License.
+%% You may obtain a copy of the License at
+%%
+%% http://www.apache.org/licenses/LICENSE-2.0
+%%
+%% Unless required by applicable law or agreed to in writing, software
+%% distributed under the License is distributed on an "AS IS" BASIS,
+%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+%% See the License for the specific language governing permissions and
+%% limitations under the License.
+%%
+%% %CopyrightEnd%
+%%
+-module(raw_file_io_deflate).
+
+-behavior(gen_statem).
+
+-export([init/1, callback_mode/0, terminate/3]).
+-export([opening/3, opened/3]).
+
+-include("file_int.hrl").
+
+-define(GZIP_WBITS, 16 + 15).
+
+callback_mode() -> state_functions.
+
+init({Owner, Secret, [compressed]}) ->
+ Monitor = monitor(process, Owner),
+ Z = zlib:open(),
+ ok = zlib:deflateInit(Z, default, deflated, ?GZIP_WBITS, 8, default),
+ Data =
+ #{ owner => Owner,
+ monitor => Monitor,
+ secret => Secret,
+ position => 0,
+ zlib => Z },
+ {ok, opening, Data}.
+
+opening({call, From}, {'$open', Secret, Filename, Modes}, #{ secret := Secret } = Data) ->
+ case raw_file_io:open(Filename, Modes) of
+ {ok, PrivateFd} ->
+ NewData = Data#{ handle => PrivateFd },
+ {next_state, opened, NewData, [{reply, From, ok}]};
+ Other ->
+ {stop_and_reply, normal, [{reply, From, Other}]}
+ end;
+opening(_Event, _Contents, _Data) ->
+ {keep_state_and_data, [postpone]}.
+
+%%
+
+opened(info, {'DOWN', Monitor, process, _Owner, Reason}, #{ monitor := Monitor } = Data) ->
+ if
+ Reason =/= kill -> flush_deflate_state(Data);
+ Reason =:= kill -> ignored
+ end,
+ {stop, shutdown};
+
+opened(info, _Message, _Data) ->
+ keep_state_and_data;
+
+opened({call, {Owner, _Tag} = From}, [close], #{ owner := Owner } = Data) ->
+ #{ handle := PrivateFd } = Data,
+ Response =
+ case flush_deflate_state(Data) of
+ ok -> ?CALL_FD(PrivateFd, close, []);
+ Other -> Other
+ end,
+ {stop_and_reply, normal, [{reply, From, Response}]};
+
+opened({call, {Owner, _Tag} = From}, [position, Mark], #{ owner := Owner } = Data) ->
+ case position(Data, Mark) of
+ {ok, NewData, Result} ->
+ Response = {ok, Result},
+ {keep_state, NewData, [{reply, From, Response}]};
+ Other ->
+ {keep_state_and_data, [{reply, From, Other}]}
+ end;
+
+opened({call, {Owner, _Tag} = From}, [write, IOVec], #{ owner := Owner } = Data) ->
+ case write(Data, IOVec) of
+ {ok, NewData} -> {keep_state, NewData, [{reply, From, ok}]};
+ Other -> {keep_state_and_data, [{reply, From, Other}]}
+ end;
+
+opened({call, {Owner, _Tag} = From}, [read, _Size], #{ owner := Owner }) ->
+ Response = {error, ebadf},
+ {keep_state_and_data, [{reply, From, Response}]};
+
+opened({call, {Owner, _Tag} = From}, [read_line], #{ owner := Owner }) ->
+ Response = {error, ebadf},
+ {keep_state_and_data, [{reply, From, Response}]};
+
+opened({call, {Owner, _Tag} = From}, _Command, #{ owner := Owner }) ->
+ Response = {error, enotsup},
+ {keep_state_and_data, [{reply, From, Response}]};
+
+opened({call, _From}, _Command, _Data) ->
+ %% The client functions filter this out, so we'll crash if the user does
+ %% anything stupid on purpose.
+ {shutdown, protocol_violation};
+
+opened(_Event, _Request, _Data) ->
+ keep_state_and_data.
+
+write(Data, IOVec) ->
+ #{ handle := PrivateFd, position := Position, zlib := Z } = Data,
+ UncompressedSize = iolist_size(IOVec),
+ case ?CALL_FD(PrivateFd, write, [zlib:deflate(Z, IOVec)]) of
+ ok -> {ok, Data#{ position := (Position + UncompressedSize) }};
+ Other -> Other
+ end.
+
+%%
+%% We support "seeking" forward as long as it isn't relative to EOF.
+%%
+%% Seeking is a bit of a misnomer as it's really just compressing zeroes until
+%% we reach the desired point, but it has always behaved like this.
+%%
+
+position(Data, Mark) when is_atom(Mark) ->
+ position(Data, {Mark, 0});
+position(Data, Offset) when is_integer(Offset) ->
+ position(Data, {bof, Offset});
+position(Data, {bof, Offset}) when is_integer(Offset) ->
+ position_1(Data, Offset);
+position(Data, {cur, Offset}) when is_integer(Offset) ->
+ #{ position := Position } = Data,
+ position_1(Data, Position + Offset);
+position(_Data, {eof, Offset}) when is_integer(Offset) ->
+ {error, einval};
+position(_Data, _Any) ->
+ {error, badarg}.
+
+position_1(#{ position := Desired } = Data, Desired) ->
+ {ok, Data, Desired};
+position_1(#{ position := Current } = Data, Desired) when Current < Desired ->
+ BytesToWrite = min(Desired - Current, 4 bsl 20),
+ case write(Data, <<0:(BytesToWrite)/unit:8>>) of
+ {ok, NewData} -> position_1(NewData, Desired);
+ Other -> Other
+ end;
+position_1(#{ position := Current }, Desired) when Current > Desired ->
+ {error, einval}.
+
+flush_deflate_state(#{ handle := PrivateFd, zlib := Z }) ->
+ case ?CALL_FD(PrivateFd, write, [zlib:deflate(Z, [], finish)]) of
+ ok -> ok;
+ Other -> Other
+ end.
+
+terminate(_Reason, _State, _Data) ->
+ ok.
diff --git a/lib/kernel/src/raw_file_io_delayed.erl b/lib/kernel/src/raw_file_io_delayed.erl
new file mode 100644
index 0000000000..d2ad7550a1
--- /dev/null
+++ b/lib/kernel/src/raw_file_io_delayed.erl
@@ -0,0 +1,320 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 2017. All Rights Reserved.
+%%
+%% Licensed under the Apache License, Version 2.0 (the "License");
+%% you may not use this file except in compliance with the License.
+%% You may obtain a copy of the License at
+%%
+%% http://www.apache.org/licenses/LICENSE-2.0
+%%
+%% Unless required by applicable law or agreed to in writing, software
+%% distributed under the License is distributed on an "AS IS" BASIS,
+%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+%% See the License for the specific language governing permissions and
+%% limitations under the License.
+%%
+%% %CopyrightEnd%
+%%
+-module(raw_file_io_delayed).
+
+-behavior(gen_statem).
+
+-export([close/1, sync/1, datasync/1, truncate/1, advise/4, allocate/3,
+ position/2, write/2, pwrite/2, pwrite/3,
+ read_line/1, read/2, pread/2, pread/3]).
+
+%% OTP internal.
+-export([ipread_s32bu_p32bu/3, sendfile/8]).
+
+-export([open_layer/3]).
+
+-export([init/1, callback_mode/0, terminate/3]).
+-export([opening/3, opened/3]).
+
+-include("file_int.hrl").
+
+open_layer(Filename, Modes, Options) ->
+ Secret = make_ref(),
+ case gen_statem:start(?MODULE, {self(), Secret, Options}, []) of
+ {ok, Pid} ->
+ gen_statem:call(Pid, {'$open', Secret, Filename, Modes}, infinity);
+ Other ->
+ Other
+ end.
+
+callback_mode() -> state_functions.
+
+init({Owner, Secret, Options}) ->
+ Monitor = monitor(process, Owner),
+ Defaults =
+ #{ owner => Owner,
+ monitor => Monitor,
+ secret => Secret,
+ timer => none,
+ pid => self(),
+ buffer => prim_buffer:new(),
+ delay_size => 64 bsl 10,
+ delay_time => 2000 },
+ Data = fill_delay_values(Defaults, Options),
+ {ok, opening, Data}.
+
+fill_delay_values(Data, []) ->
+ Data;
+fill_delay_values(Data, [{delayed_write, Size, Time} | Options]) ->
+ fill_delay_values(Data#{ delay_size => Size, delay_time => Time }, Options);
+fill_delay_values(Data, [_ | Options]) ->
+ fill_delay_values(Data, Options).
+
+opening({call, From}, {'$open', Secret, Filename, Modes}, #{ secret := Secret } = Data) ->
+ case raw_file_io:open(Filename, Modes) of
+ {ok, PrivateFd} ->
+ PublicData = maps:with([owner, buffer, delay_size, pid], Data),
+ PublicFd = #file_descriptor{ module = ?MODULE, data = PublicData },
+
+ NewData = Data#{ handle => PrivateFd },
+ Response = {ok, PublicFd},
+ {next_state, opened, NewData, [{reply, From, Response}]};
+ Other ->
+ {stop_and_reply, normal, [{reply, From, Other}]}
+ end;
+opening(_Event, _Contents, _Data) ->
+ {keep_state_and_data, [postpone]}.
+
+%%
+
+opened(info, {'$timed_out', Secret}, #{ secret := Secret } = Data) ->
+ %% If the user writes something at this exact moment, the flush will fail
+ %% and the timer won't reset on the next write since the buffer won't be
+ %% empty (Unless we collided on a flush). We therefore reset the timeout to
+ %% ensure that data won't sit idle for extended periods of time.
+ case try_flush_write_buffer(Data) of
+ busy -> gen_statem:cast(self(), '$reset_timeout');
+ ok -> ok
+ end,
+ {keep_state, Data#{ timer => none }, []};
+
+opened(info, {'DOWN', Monitor, process, _Owner, Reason}, #{ monitor := Monitor } = Data) ->
+ if
+ Reason =/= kill -> try_flush_write_buffer(Data);
+ Reason =:= kill -> ignored
+ end,
+ {stop, shutdown};
+
+opened(info, _Message, _Data) ->
+ keep_state_and_data;
+
+opened({call, {Owner, _Tag} = From}, [close], #{ owner := Owner } = Data) ->
+ case flush_write_buffer(Data) of
+ ok ->
+ #{ handle := PrivateFd } = Data,
+ Response = ?CALL_FD(PrivateFd, close, []),
+ {stop_and_reply, normal, [{reply, From, Response}]};
+ Other ->
+ {stop_and_reply, normal, [{reply, From, Other}]}
+ end;
+
+opened({call, {Owner, _Tag} = From}, '$wait', #{ owner := Owner }) ->
+ %% Used in write/2 to synchronize writes on lock conflicts.
+ {keep_state_and_data, [{reply, From, ok}]};
+
+opened({call, {Owner, _Tag} = From}, '$synchronous_flush', #{ owner := Owner } = Data) ->
+ cancel_flush_timeout(Data),
+ Response = flush_write_buffer(Data),
+ {keep_state_and_data, [{reply, From, Response}]};
+
+opened({call, {Owner, _Tag} = From}, Command, #{ owner := Owner } = Data) ->
+ Response =
+ case flush_write_buffer(Data) of
+ ok -> dispatch_command(Data, Command);
+ Other -> Other
+ end,
+ {keep_state_and_data, [{reply, From, Response}]};
+
+opened({call, _From}, _Command, _Data) ->
+ %% The client functions filter this out, so we'll crash if the user does
+ %% anything stupid on purpose.
+ {shutdown, protocol_violation};
+
+opened(cast, '$reset_timeout', #{ delay_time := Timeout, secret := Secret } = Data) ->
+ cancel_flush_timeout(Data),
+ Timer = erlang:send_after(Timeout, self(), {'$timed_out', Secret}),
+ {keep_state, Data#{ timer => Timer }, []};
+
+opened(cast, _Message, _Data) ->
+ {keep_state_and_data, []}.
+
+dispatch_command(Data, [Function | Args]) ->
+ #{ handle := Handle } = Data,
+ Module = Handle#file_descriptor.module,
+ apply(Module, Function, [Handle | Args]).
+
+cancel_flush_timeout(#{ timer := none }) ->
+ ok;
+cancel_flush_timeout(#{ timer := Timer }) ->
+ _ = erlang:cancel_timer(Timer, [{async, true}]),
+ ok.
+
+try_flush_write_buffer(#{ buffer := Buffer, handle := PrivateFd }) ->
+ case prim_buffer:try_lock(Buffer) of
+ acquired ->
+ flush_write_buffer_1(Buffer, PrivateFd),
+ prim_buffer:unlock(Buffer),
+ ok;
+ busy ->
+ busy
+ end.
+
+%% This is only safe to use when there is no chance of conflict with the owner
+%% process, or in other words, "during synchronous calls outside of the locked
+%% section of write/2"
+flush_write_buffer(#{ buffer := Buffer, handle := PrivateFd }) ->
+ acquired = prim_buffer:try_lock(Buffer),
+ Result = flush_write_buffer_1(Buffer, PrivateFd),
+ prim_buffer:unlock(Buffer),
+ Result.
+
+flush_write_buffer_1(Buffer, PrivateFd) ->
+ case prim_buffer:size(Buffer) of
+ Size when Size > 0 ->
+ ?CALL_FD(PrivateFd, write, [prim_buffer:read_iovec(Buffer, Size)]);
+ 0 ->
+ ok
+ end.
+
+terminate(_Reason, _State, _Data) ->
+ ok.
+
+%% Client functions
+
+write(Fd, IOData) ->
+ try
+ enqueue_write(Fd, erlang:iolist_to_iovec(IOData))
+ catch
+ error:badarg -> {error, badarg}
+ end.
+enqueue_write(_Fd, []) ->
+ ok;
+enqueue_write(Fd, IOVec) ->
+ %% get_fd_data will reject everyone except the process that opened the Fd,
+ %% so we can't race with anyone except the wrapper process.
+ #{ delay_size := DelaySize,
+ buffer := Buffer,
+ pid := Pid } = get_fd_data(Fd),
+ case prim_buffer:try_lock(Buffer) of
+ acquired ->
+ %% (The wrapper process will exit without flushing if we're killed
+ %% while holding the lock).
+ enqueue_write_locked(Pid, Buffer, DelaySize, IOVec);
+ busy ->
+ %% This can only happen while we're processing a timeout in the
+ %% wrapper process, so we perform a bogus call to get a completion
+ %% notification before trying again.
+ gen_statem:call(Pid, '$wait'),
+ enqueue_write(Fd, IOVec)
+ end.
+enqueue_write_locked(Pid, Buffer, DelaySize, IOVec) ->
+ %% The synchronous operations (write, forced flush) are safe since we're
+ %% running on the only process that can fill the buffer; a timeout being
+ %% processed just before $synchronous_flush will cause the flush to nop,
+ %% and a timeout sneaking in just before a synchronous write won't do
+ %% anything since the buffer is guaranteed to be empty at that point.
+ BufSize = prim_buffer:size(Buffer),
+ case is_iovec_smaller_than(IOVec, DelaySize - BufSize) of
+ true when BufSize > 0 ->
+ prim_buffer:write(Buffer, IOVec),
+ prim_buffer:unlock(Buffer);
+ true ->
+ prim_buffer:write(Buffer, IOVec),
+ prim_buffer:unlock(Buffer),
+ gen_statem:cast(Pid, '$reset_timeout');
+ false when BufSize > 0 ->
+ prim_buffer:write(Buffer, IOVec),
+ prim_buffer:unlock(Buffer),
+ gen_statem:call(Pid, '$synchronous_flush');
+ false ->
+ prim_buffer:unlock(Buffer),
+ gen_statem:call(Pid, [write, IOVec])
+ end.
+
+%% iolist_size/1 will always look through the entire list to get a precise
+%% amount, which is pretty inefficient since we only need to know whether we've
+%% hit the buffer threshold or not.
+%%
+%% We only handle the binary case since write/2 forcibly translates input to
+%% erlang:iovec().
+is_iovec_smaller_than(IOVec, Max) ->
+ is_iovec_smaller_than_1(IOVec, Max, 0).
+is_iovec_smaller_than_1(_IOVec, Max, Acc) when Acc >= Max ->
+ false;
+is_iovec_smaller_than_1([], _Max, _Acc) ->
+ true;
+is_iovec_smaller_than_1([Binary | Rest], Max, Acc) when is_binary(Binary) ->
+ is_iovec_smaller_than_1(Rest, Max, Acc + byte_size(Binary)).
+
+close(Fd) ->
+ wrap_call(Fd, [close]).
+
+sync(Fd) ->
+ wrap_call(Fd, [sync]).
+datasync(Fd) ->
+ wrap_call(Fd, [datasync]).
+
+truncate(Fd) ->
+ wrap_call(Fd, [truncate]).
+
+advise(Fd, Offset, Length, Advise) ->
+ wrap_call(Fd, [advise, Offset, Length, Advise]).
+allocate(Fd, Offset, Length) ->
+ wrap_call(Fd, [allocate, Offset, Length]).
+
+position(Fd, Mark) ->
+ wrap_call(Fd, [position, Mark]).
+
+pwrite(Fd, Offset, IOData) ->
+ try
+ CompactedData = erlang:iolist_to_iovec(IOData),
+ wrap_call(Fd, [pwrite, Offset, CompactedData])
+ catch
+ error:badarg -> {error, badarg}
+ end.
+pwrite(Fd, LocBytes) ->
+ try
+ CompactedLocBytes =
+ [ {Offset, erlang:iolist_to_iovec(IOData)} ||
+ {Offset, IOData} <- LocBytes ],
+ wrap_call(Fd, [pwrite, CompactedLocBytes])
+ catch
+ error:badarg -> {error, badarg}
+ end.
+
+read_line(Fd) ->
+ wrap_call(Fd, [read_line]).
+read(Fd, Size) ->
+ wrap_call(Fd, [read, Size]).
+pread(Fd, Offset, Size) ->
+ wrap_call(Fd, [pread, Offset, Size]).
+pread(Fd, LocNums) ->
+ wrap_call(Fd, [pread, LocNums]).
+
+ipread_s32bu_p32bu(Fd, Offset, MaxSize) ->
+ wrap_call(Fd, [ipread_s32bu_p32bu, Offset, MaxSize]).
+
+sendfile(_,_,_,_,_,_,_,_) ->
+ {error, enotsup}.
+
+wrap_call(Fd, Command) ->
+ #{ pid := Pid } = get_fd_data(Fd),
+ try gen_statem:call(Pid, Command, infinity) of
+ Result -> Result
+ catch
+ exit:{noproc, _StackTrace} -> {error, einval}
+ end.
+
+get_fd_data(#file_descriptor{ data = Data }) ->
+ #{ owner := Owner } = Data,
+ case self() of
+ Owner -> Data;
+ _ -> error(not_on_controlling_process)
+ end.
diff --git a/lib/kernel/src/raw_file_io_inflate.erl b/lib/kernel/src/raw_file_io_inflate.erl
new file mode 100644
index 0000000000..7e9780310c
--- /dev/null
+++ b/lib/kernel/src/raw_file_io_inflate.erl
@@ -0,0 +1,261 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 2017. All Rights Reserved.
+%%
+%% Licensed under the Apache License, Version 2.0 (the "License");
+%% you may not use this file except in compliance with the License.
+%% You may obtain a copy of the License at
+%%
+%% http://www.apache.org/licenses/LICENSE-2.0
+%%
+%% Unless required by applicable law or agreed to in writing, software
+%% distributed under the License is distributed on an "AS IS" BASIS,
+%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+%% See the License for the specific language governing permissions and
+%% limitations under the License.
+%%
+%% %CopyrightEnd%
+%%
+-module(raw_file_io_inflate).
+
+-behavior(gen_statem).
+
+-export([init/1, callback_mode/0, terminate/3]).
+-export([opening/3, opened_gzip/3, opened_passthrough/3]).
+
+-include("file_int.hrl").
+
+-define(INFLATE_CHUNK_SIZE, (1 bsl 10)).
+-define(GZIP_WBITS, (16 + 15)).
+
+callback_mode() -> state_functions.
+
+init({Owner, Secret, [compressed]}) ->
+ Monitor = monitor(process, Owner),
+ %% We're using the undocumented inflateInit/3 to open the stream in
+ %% 'reset mode', which resets the inflate state at the end of every stream,
+ %% allowing us to read concatenated gzip files.
+ Z = zlib:open(),
+ ok = zlib:inflateInit(Z, ?GZIP_WBITS, reset),
+ Data =
+ #{ owner => Owner,
+ monitor => Monitor,
+ secret => Secret,
+ position => 0,
+ buffer => prim_buffer:new(),
+ zlib => Z },
+ {ok, opening, Data}.
+
+%% The old driver fell back to plain reads if the file didn't start with the
+%% magic gzip bytes.
+choose_decompression_state(PrivateFd) ->
+ State =
+ case ?CALL_FD(PrivateFd, read, [2]) of
+ {ok, <<16#1F, 16#8B>>} -> opened_gzip;
+ _Other -> opened_passthrough
+ end,
+ {ok, 0} = ?CALL_FD(PrivateFd, position, [0]),
+ State.
+
+opening({call, From}, {'$open', Secret, Filename, Modes}, #{ secret := Secret } = Data) ->
+ case raw_file_io:open(Filename, Modes) of
+ {ok, PrivateFd} ->
+ NextState = choose_decompression_state(PrivateFd),
+ NewData = Data#{ handle => PrivateFd },
+ {next_state, NextState, NewData, [{reply, From, ok}]};
+ Other ->
+ {stop_and_reply, normal, [{reply, From, Other}]}
+ end;
+opening(_Event, _Contents, _Data) ->
+ {keep_state_and_data, [postpone]}.
+
+internal_close(From, Data) ->
+ #{ handle := PrivateFd } = Data,
+ Response = ?CALL_FD(PrivateFd, close, []),
+ {stop_and_reply, normal, [{reply, From, Response}]}.
+
+opened_passthrough(info, {'DOWN', Monitor, process, _Owner, _Reason}, #{ monitor := Monitor }) ->
+ {stop, shutdown};
+
+opened_passthrough(info, _Message, _Data) ->
+ keep_state_and_data;
+
+opened_passthrough({call, {Owner, _Tag} = From}, [close], #{ owner := Owner } = Data) ->
+ internal_close(From, Data);
+
+opened_passthrough({call, {Owner, _Tag} = From}, [Method | Args], #{ owner := Owner } = Data) ->
+ #{ handle := PrivateFd } = Data,
+ Response = ?CALL_FD(PrivateFd, Method, Args),
+ {keep_state_and_data, [{reply, From, Response}]};
+
+opened_passthrough({call, _From}, _Command, _Data) ->
+ %% The client functions filter this out, so we'll crash if the user does
+ %% anything stupid on purpose.
+ {shutdown, protocol_violation};
+
+opened_passthrough(_Event, _Request, _Data) ->
+ keep_state_and_data.
+
+%%
+
+opened_gzip(info, {'DOWN', Monitor, process, _Owner, _Reason}, #{ monitor := Monitor }) ->
+ {stop, shutdown};
+
+opened_gzip(info, _Message, _Data) ->
+ keep_state_and_data;
+
+opened_gzip({call, {Owner, _Tag} = From}, [close], #{ owner := Owner } = Data) ->
+ internal_close(From, Data);
+
+opened_gzip({call, {Owner, _Tag} = From}, [position, Mark], #{ owner := Owner } = Data) ->
+ case position(Data, Mark) of
+ {ok, NewData, Result} ->
+ Response = {ok, Result},
+ {keep_state, NewData, [{reply, From, Response}]};
+ Other ->
+ {keep_state_and_data, [{reply, From, Other}]}
+ end;
+
+opened_gzip({call, {Owner, _Tag} = From}, [read, Size], #{ owner := Owner } = Data) ->
+ case read(Data, Size) of
+ {ok, NewData, Result} ->
+ Response = {ok, Result},
+ {keep_state, NewData, [{reply, From, Response}]};
+ Other ->
+ {keep_state_and_data, [{reply, From, Other}]}
+ end;
+
+opened_gzip({call, {Owner, _Tag} = From}, [read_line], #{ owner := Owner } = Data) ->
+ case read_line(Data) of
+ {ok, NewData, Result} ->
+ Response = {ok, Result},
+ {keep_state, NewData, [{reply, From, Response}]};
+ Other ->
+ {keep_state_and_data, [{reply, From, Other}]}
+ end;
+
+opened_gzip({call, {Owner, _Tag} = From}, [write, _IOData], #{ owner := Owner }) ->
+ Response = {error, ebadf},
+ {keep_state_and_data, [{reply, From, Response}]};
+
+opened_gzip({call, {Owner, _Tag} = From}, _Request, #{ owner := Owner }) ->
+ Response = {error, enotsup},
+ {keep_state_and_data, [{reply, From, Response}]};
+
+opened_gzip({call, _From}, _Request, _Data) ->
+ %% The client functions filter this out, so we'll crash if the user does
+ %% anything stupid on purpose.
+ {shutdown, protocol_violation};
+
+opened_gzip(_Event, _Request, _Data) ->
+ keep_state_and_data.
+
+%%
+
+read(#{ buffer := Buffer } = Data, Size) ->
+ try read_1(Data, Buffer, prim_buffer:size(Buffer), Size) of
+ Result -> Result
+ catch
+ error:badarg -> {error, badarg};
+ error:_ -> {error, eio}
+ end.
+read_1(Data, Buffer, BufferSize, ReadSize) when BufferSize >= ReadSize ->
+ #{ position := Position } = Data,
+ Decompressed = prim_buffer:read(Buffer, ReadSize),
+ {ok, Data#{ position => (Position + ReadSize) }, Decompressed};
+read_1(Data, Buffer, BufferSize, ReadSize) when BufferSize < ReadSize ->
+ #{ handle := PrivateFd } = Data,
+ case ?CALL_FD(PrivateFd, read, [?INFLATE_CHUNK_SIZE]) of
+ {ok, Compressed} ->
+ #{ zlib := Z } = Data,
+ Uncompressed = erlang:iolist_to_iovec(zlib:inflate(Z, Compressed)),
+ prim_buffer:write(Buffer, Uncompressed),
+ read_1(Data, Buffer, prim_buffer:size(Buffer), ReadSize);
+ eof when BufferSize > 0 ->
+ read_1(Data, Buffer, BufferSize, BufferSize);
+ Other ->
+ Other
+ end.
+
+read_line(#{ buffer := Buffer } = Data) ->
+ try read_line_1(Data, Buffer, prim_buffer:find_byte_index(Buffer, $\n)) of
+ {ok, NewData, Decompressed} -> {ok, NewData, Decompressed};
+ Other -> Other
+ catch
+ error:badarg -> {error, badarg};
+ error:_ -> {error, eio}
+ end.
+
+read_line_1(Data, Buffer, not_found) ->
+ #{ handle := PrivateFd, zlib := Z } = Data,
+ case ?CALL_FD(PrivateFd, read, [?INFLATE_CHUNK_SIZE]) of
+ {ok, Compressed} ->
+ Uncompressed = erlang:iolist_to_iovec(zlib:inflate(Z, Compressed)),
+ prim_buffer:write(Buffer, Uncompressed),
+ read_line_1(Data, Buffer, prim_buffer:find_byte_index(Buffer, $\n));
+ eof ->
+ case prim_buffer:size(Buffer) of
+ Size when Size > 0 -> {ok, prim_buffer:read(Buffer, Size)};
+ Size when Size =:= 0 -> eof
+ end;
+ Error ->
+ Error
+ end;
+read_line_1(Data, Buffer, {ok, LFIndex}) ->
+ %% Translate CRLF into just LF, completely ignoring which encoding is used,
+ %% but treat the file position as including CR.
+ #{ position := Position } = Data,
+ NewData = Data#{ position => (Position + LFIndex + 1) },
+ CRIndex = (LFIndex - 1),
+ TranslatedLine =
+ case prim_buffer:read(Buffer, LFIndex + 1) of
+ <<Line:CRIndex/binary, "\r\n">> -> <<Line/binary, "\n">>;
+ Line -> Line
+ end,
+ {ok, NewData, TranslatedLine}.
+
+%%
+%% We support seeking in both directions as long as it isn't relative to EOF.
+%%
+%% Seeking backwards is extremely inefficient since we have to seek to the very
+%% beginning and then decompress up to the desired point.
+%%
+
+position(Data, Mark) when is_atom(Mark) ->
+ position(Data, {Mark, 0});
+position(Data, Offset) when is_integer(Offset) ->
+ position(Data, {bof, Offset});
+position(Data, {bof, Offset}) when is_integer(Offset) ->
+ position_1(Data, Offset);
+position(Data, {cur, Offset}) when is_integer(Offset) ->
+ #{ position := Position } = Data,
+ position_1(Data, Position + Offset);
+position(_Data, {eof, Offset}) when is_integer(Offset) ->
+ {error, einval};
+position(_Data, _Other) ->
+ {error, badarg}.
+
+position_1(_Data, Desired) when Desired < 0 ->
+ {error, einval};
+position_1(#{ position := Desired } = Data, Desired) ->
+ {ok, Data, Desired};
+position_1(#{ position := Current } = Data, Desired) when Current < Desired ->
+ case read(Data, min(Desired - Current, ?INFLATE_CHUNK_SIZE)) of
+ {ok, NewData, _Data} -> position_1(NewData, Desired);
+ eof -> {ok, Data, Current};
+ Other -> Other
+ end;
+position_1(#{ position := Current } = Data, Desired) when Current > Desired ->
+ #{ handle := PrivateFd, buffer := Buffer, zlib := Z } = Data,
+ case ?CALL_FD(PrivateFd, position, [bof]) of
+ {ok, 0} ->
+ ok = zlib:inflateReset(Z),
+ prim_buffer:wipe(Buffer),
+ position_1(Data#{ position => 0 }, Desired);
+ Other ->
+ Other
+ end.
+
+terminate(_Reason, _State, _Data) ->
+ ok.
diff --git a/lib/kernel/src/raw_file_io_list.erl b/lib/kernel/src/raw_file_io_list.erl
new file mode 100644
index 0000000000..2e16e63f0e
--- /dev/null
+++ b/lib/kernel/src/raw_file_io_list.erl
@@ -0,0 +1,128 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 2017. All Rights Reserved.
+%%
+%% Licensed under the Apache License, Version 2.0 (the "License");
+%% you may not use this file except in compliance with the License.
+%% You may obtain a copy of the License at
+%%
+%% http://www.apache.org/licenses/LICENSE-2.0
+%%
+%% Unless required by applicable law or agreed to in writing, software
+%% distributed under the License is distributed on an "AS IS" BASIS,
+%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+%% See the License for the specific language governing permissions and
+%% limitations under the License.
+%%
+%% %CopyrightEnd%
+%%
+-module(raw_file_io_list).
+
+-export([close/1, sync/1, datasync/1, truncate/1, advise/4, allocate/3,
+ position/2, write/2, pwrite/2, pwrite/3,
+ read_line/1, read/2, pread/2, pread/3]).
+
+%% OTP internal.
+-export([ipread_s32bu_p32bu/3, sendfile/8]).
+
+-export([open_layer/3]).
+
+-include("file_int.hrl").
+
+open_layer(Filename, Modes, [list]) ->
+ case raw_file_io:open(Filename, [binary | Modes]) of
+ {ok, PrivateFd} -> {ok, make_public_fd(PrivateFd, Modes)};
+ Other -> Other
+ end.
+
+%% We can skip wrapping the file if it's write-only since only read operations
+%% are affected by list mode. Since raw_file_io fills in all implicit options
+%% for us, all we need to do is check whether 'read' is among them.
+make_public_fd(PrivateFd, Modes) ->
+ case lists:member(read, Modes) of
+ true -> #file_descriptor{ module = ?MODULE, data = PrivateFd };
+ false -> PrivateFd
+ end.
+
+close(Fd) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, close, []).
+
+sync(Fd) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, sync, []).
+datasync(Fd) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, datasync, []).
+
+truncate(Fd) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, truncate, []).
+
+advise(Fd, Offset, Length, Advise) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, advise, [Offset, Length, Advise]).
+allocate(Fd, Offset, Length) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, allocate, [Offset, Length]).
+
+position(Fd, Mark) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, position, [Mark]).
+
+write(Fd, IOData) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, write, [IOData]).
+
+pwrite(Fd, Offset, IOData) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, pwrite, [Offset, IOData]).
+pwrite(Fd, LocBytes) ->
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, pwrite, [LocBytes]).
+
+read_line(Fd) ->
+ PrivateFd = Fd#file_descriptor.data,
+ case ?CALL_FD(PrivateFd, read_line, []) of
+ {ok, Binary} -> {ok, binary_to_list(Binary)};
+ Other -> Other
+ end.
+read(Fd, Size) ->
+ PrivateFd = Fd#file_descriptor.data,
+ case ?CALL_FD(PrivateFd, read, [Size]) of
+ {ok, Binary} -> {ok, binary_to_list(Binary)};
+ Other -> Other
+ end.
+pread(Fd, Offset, Size) ->
+ PrivateFd = Fd#file_descriptor.data,
+ case ?CALL_FD(PrivateFd, pread, [Offset, Size]) of
+ {ok, Binary} -> {ok, binary_to_list(Binary)};
+ Other -> Other
+ end.
+pread(Fd, LocNums) ->
+ PrivateFd = Fd#file_descriptor.data,
+ case ?CALL_FD(PrivateFd, pread, [LocNums]) of
+ {ok, LocResults} ->
+ TranslatedResults =
+ [ case Result of
+ Result when is_binary(Result) -> binary_to_list(Result);
+ eof -> eof
+ end || Result <- LocResults ],
+ {ok, TranslatedResults};
+ Other -> Other
+ end.
+
+ipread_s32bu_p32bu(Fd, Offset, MaxSize) ->
+ PrivateFd = Fd#file_descriptor.data,
+ case ?CALL_FD(PrivateFd, ipread_s32bu_p32bu, [Offset, MaxSize]) of
+ {ok, {Size, Pointer, Binary}} when is_binary(Binary) ->
+ {ok, {Size, Pointer, binary_to_list(Binary)}};
+ Other ->
+ Other
+ end.
+
+sendfile(Fd, Dest, Offset, Bytes, ChunkSize, Headers, Trailers, Flags) ->
+ Args = [Dest, Offset, Bytes, ChunkSize, Headers, Trailers, Flags],
+ PrivateFd = Fd#file_descriptor.data,
+ ?CALL_FD(PrivateFd, sendfile, Args).
diff --git a/lib/kernel/src/raw_file_io_raw.erl b/lib/kernel/src/raw_file_io_raw.erl
new file mode 100644
index 0000000000..9a9fe78eb1
--- /dev/null
+++ b/lib/kernel/src/raw_file_io_raw.erl
@@ -0,0 +1,25 @@
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 2017. All Rights Reserved.
+%%
+%% Licensed under the Apache License, Version 2.0 (the "License");
+%% you may not use this file except in compliance with the License.
+%% You may obtain a copy of the License at
+%%
+%% http://www.apache.org/licenses/LICENSE-2.0
+%%
+%% Unless required by applicable law or agreed to in writing, software
+%% distributed under the License is distributed on an "AS IS" BASIS,
+%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+%% See the License for the specific language governing permissions and
+%% limitations under the License.
+%%
+%% %CopyrightEnd%
+%%
+-module(raw_file_io_raw).
+
+-export([open_layer/3]).
+
+open_layer(Filename, Modes, [raw]) ->
+ prim_file:open(Filename, [raw | Modes]).