aboutsummaryrefslogtreecommitdiffstats
path: root/erts/emulator/nifs
diff options
context:
space:
mode:
authorJohn Högberg <[email protected]>2017-11-03 11:49:27 +0100
committerJohn Högberg <[email protected]>2017-11-30 15:44:33 +0100
commitebbd26eeea4115c946d1254d94acd50f150b4455 (patch)
treeab2a8ba1023aa0684695dc36ff5567ae9e3f59c4 /erts/emulator/nifs
parent6cb62e44eba1db8d1917ebb0db84298e91582c4e (diff)
downloadotp-ebbd26eeea4115c946d1254d94acd50f150b4455.tar.gz
otp-ebbd26eeea4115c946d1254d94acd50f150b4455.tar.bz2
otp-ebbd26eeea4115c946d1254d94acd50f150b4455.zip
Reimplement efile_drv as a dirty NIF
This improves the latency of file operations as dirty schedulers are a bit more eager to run jobs than async threads, and use a single global queue rather than per-thread queues, eliminating the risk of a job stalling behind a long-running job on the same thread while other async threads sit idle. There's no such thing as a free lunch though; the lowered latency comes at the cost of increased busy-waiting which may have an adverse effect on some applications. This behavior can be tweaked with the +sbwt flag, but unfortunately it affects all types of schedulers and not just dirty ones. We plan to add type-specific flags at a later stage. sendfile has been moved to inet_drv to lessen the effect of a nasty race; the cooperation between inet_drv and efile has never been airtight and the socket dying at the wrong time (Regardless of reason) could result in fd aliasing. Moving it to the inet driver makes it impossible to trigger this by closing the socket in the middle of a sendfile operation, while still allowing it to be aborted -- something that can't be done if it stays in the file driver. The race still occurs if the controlling process dies in the short window between dispatching the sendfile operation and the dup(2) call in the driver, but it's much less likely to happen now. A proper fix is in the works. -- Notable functional differences: * The use_threads option for file:sendfile/5 no longer has any effect. * The file-specific DTrace probes have been removed. The same effect can be achieved with normal tracing together with the nif__entry/nif__return probes to track scheduling. -- OTP-14256
Diffstat (limited to 'erts/emulator/nifs')
-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
4 files changed, 3861 insertions, 0 deletions
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;
+ }
+}