Add file:allocate/3 operation

This operation allows pre-allocation of space for files.
It succeeds only on systems that support such operation.

The POSIX standard defines the optional system call
posix_fallocate() to implement this feature. However,
some systems implement more specific functions to
accomplish the same operation.

On Linux, if the more specific function fallocate() is
implemented, it is used instead of posix_fallocate(),
falling back to posix_fallocate() if the fallocate()
call failed (it's only supported for the ext4, ocfs2,
xfs and btrfs file systems at the moment).

On Mac OS X it uses the specific fcntl() operation
F_PREALLOCATE, falling back to posix_fallocate() if
it's available (at the moment Mac OS X doesn't provide
posix_fallocate()).

On any other UNIX system, it uses posix_fallocate() if it's
available. Any other system not providing this system call
or any function to pre-allocate space for files, this operation
always fails with the ENOTSUP POSIX error.

7 files changed, 183 insertions, 4 deletions

diff --git a/erts/configure.in b/erts/configure.in
index 3256b0cb59..be9a753f7f 100644
--- a/erts/configure.in
+++ b/erts/configure.in
@@ -1075,8 +1075,43 @@ fi
 
 AC_SUBST(ERTS_BUILD_SMP_EMU)
 
-AC_CHECK_FUNCS([posix_fadvise])
+AC_CHECK_FUNCS([posix_fadvise, fallocate])
+AC_CHECK_HEADERS([linux/falloc.h])
 
+dnl * Old glibcs have broken posix_fallocate(). Make sure not to use it.
+dnl * It may also be broken in AIX.
+AC_CACHE_CHECK([whether posix_fallocate() works],i_cv_posix_fallocate_works,[
+    AC_TRY_RUN([
+        #if !defined(__sun) && !defined(__sun__)
+        #define _XOPEN_SOURCE 600
+        #endif
+        #include <stdio.h>
+        #include <stdlib.h>
+        #include <fcntl.h>
+        #include <unistd.h>
+        #if defined(__GLIBC__) && (__GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ < 7))
+            possibly broken posix_fallocate
+        #endif
+        int main() {
+            int fd = creat("conftest.temp", 0600);
+            int ret;
+            if (-1 == fd) {
+               perror("creat()");
+               return 2;
+            }
+            ret = posix_fallocate(fd, 1024, 1024) < 0 ? 1 : 0;
+            unlink("conftest.temp");
+            return ret;
+        }
+    ], [
+        i_cv_posix_fallocate_works=yes
+    ], [
+        i_cv_posix_fallocate_works=no
+    ])
+])
+if test $i_cv_posix_fallocate_works = yes; then
+   AC_DEFINE(HAVE_POSIX_FALLOCATE,, Define if you have a working posix_fallocate())
+fi
 
 #
 # Figure out if the emulator should use threads. The default is set above
diff --git a/erts/emulator/drivers/common/efile_drv.c b/erts/emulator/drivers/common/efile_drv.c
index 912f5d3d8b..2c38e7256a 100644
--- a/erts/emulator/drivers/common/efile_drv.c
+++ b/erts/emulator/drivers/common/efile_drv.c
@@ -56,6 +56,7 @@
 #define FILE_FDATASYNC          30
 #define FILE_FADVISE            31
 #define FILE_SENDFILE           32
+#define FILE_FALLOCATE          33
 
 /* Return codes */
 
@@ -503,6 +504,10 @@ struct t_data
 	    Uint64 written;
 	} sendfile;
 #endif /* HAVE_SENDFILE */
+	struct {
+	    Sint64 offset;
+	    Sint64 length;
+	} fallocate;
     } c;
     char b[1];
 };
@@ -1953,6 +1958,17 @@ static int flush_sendfile(file_descriptor *desc,void *_) {
 #endif /* HAVE_SENDFILE */
 
 
+static void invoke_fallocate(void *data)
+{
+    struct t_data *d = (struct t_data *) data;
+    int fd = (int) d->fd;
+    Sint64 offset = d->c.fallocate.offset;
+    Sint64 length = d->c.fallocate.length;
+
+    d->again = 0;
+    d->result_ok = efile_fallocate(&d->errInfo, fd, offset, length);
+}
+
 static void free_readdir(void *data)
 {
     struct t_data *d = (struct t_data *) data;
@@ -2348,6 +2364,7 @@ file_async_ready(ErlDrvData e, ErlDrvThreadData data)
       case FILE_RENAME:
       case FILE_WRITE_INFO:
       case FILE_FADVISE:
+      case FILE_FALLOCATE:
 	reply(desc, d->result_ok, &d->errInfo);
 	free_data(data);
 	break;
@@ -2958,6 +2975,20 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
         goto done;
     }
 
+    case FILE_FALLOCATE:
+    {
+        d = EF_SAFE_ALLOC(sizeof(struct t_data));
+
+        d->fd = fd;
+        d->command = command;
+        d->invoke = invoke_fallocate;
+        d->free = free_data;
+        d->level = 2;
+        d->c.fallocate.offset = get_int64((uchar*) buf);
+        d->c.fallocate.length = get_int64(((uchar*) buf) + sizeof(Sint64));
+        goto done;
+    }
+
     }
 
     /*
diff --git a/erts/emulator/drivers/common/erl_efile.h b/erts/emulator/drivers/common/erl_efile.h
index 69ad02633c..b29b4f971c 100644
--- a/erts/emulator/drivers/common/erl_efile.h
+++ b/erts/emulator/drivers/common/erl_efile.h
@@ -185,3 +185,4 @@ int efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length,
 int efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
 		      off_t *offset, Uint64 *nbytes, struct t_sendfile_hdtl *hdtl);
 #endif /* HAVE_SENDFILE */
+int efile_fallocate(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length);
diff --git a/erts/emulator/drivers/common/ram_file_drv.c b/erts/emulator/drivers/common/ram_file_drv.c
index a109e40333..7f7cd7cd91 100644
--- a/erts/emulator/drivers/common/ram_file_drv.c
+++ b/erts/emulator/drivers/common/ram_file_drv.c
@@ -48,6 +48,7 @@
 #define RAM_FILE_SIZE          37  /* get file size */
 #define RAM_FILE_ADVISE        38  /* predeclare the access
                                     * pattern for file data */
+#define RAM_FILE_ALLOCATE      39  /* allocate space for a file */
 /* possible new operations include:
    DES_ENCRYPT
    DES_DECRYPT
@@ -720,6 +721,13 @@ static void rfile_command(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	else
 	    reply(f, 1, 0);
 	break;
+
+    case RAM_FILE_ALLOCATE:
+	if (f->flags == 0)
+	    error_reply(f, EBADF);
+	else
+	    reply(f, 1, 0);
+	break;
     }
     /*
      * Ignore anything else -- let the caller hang.
diff --git a/erts/emulator/drivers/unix/unix_efile.c b/erts/emulator/drivers/unix/unix_efile.c
index cf7af71b92..558651fff9 100644
--- a/erts/emulator/drivers/unix/unix_efile.c
+++ b/erts/emulator/drivers/unix/unix_efile.c
@@ -22,6 +22,12 @@
 #ifdef HAVE_CONFIG_H
 #  include "config.h"
 #endif
+#if defined(HAVE_POSIX_FALLOCATE) && !defined(__sun) && !defined(__sun__)
+#define _XOPEN_SOURCE 600
+#endif
+#if !defined(_GNU_SOURCE) && defined(HAVE_LINUX_FALLOC_H)
+#define _GNU_SOURCE
+#endif
 #include "sys.h"
 #include "erl_driver.h"
 #include "erl_efile.h"
@@ -41,9 +47,13 @@
 #define DARWIN 1
 #endif
 
-#ifdef DARWIN
+#if defined(DARWIN) || defined(HAVE_LINUX_FALLOC_H) || defined(HAVE_POSIX_FALLOCATE)
 #include <fcntl.h>
-#endif /* DARWIN */
+#endif
+
+#ifdef HAVE_LINUX_FALLOC_H
+#include <linux/falloc.h>
+#endif
 
 #ifdef SUNOS4
 #  define getcwd(buf, size) getwd(buf)
@@ -967,3 +977,81 @@ efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
     return check_error(retval, errInfo);
 }
 #endif /* HAVE_SENDFILE */
+
+#ifdef HAVE_POSIX_FALLOCATE
+static int
+call_posix_fallocate(int fd, Sint64 offset, Sint64 length)
+{
+    int ret;
+
+    /*
+     * On Linux and Solaris for example, posix_fallocate() returns
+     * a positive error number on error and it does not set errno.
+     * On FreeBSD however (9.0 at least), it returns -1 on error
+     * and it sets errno.
+     */
+    do {
+        ret = posix_fallocate(fd, (off_t) offset, (off_t) length);
+        if (ret > 0) {
+            errno = ret;
+            ret = -1;
+        }
+    } while (ret != 0 && errno == EINTR);
+
+    return ret;
+}
+#endif /* HAVE_POSIX_FALLOCATE */
+
+int
+efile_fallocate(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length)
+{
+#if defined HAVE_FALLOCATE
+    /* Linux specific, more efficient than posix_fallocate. */
+    int ret;
+
+    do {
+        ret = fallocate(fd, FALLOC_FL_KEEP_SIZE, (off_t) offset, (off_t) length);
+    } while (ret != 0 && errno == EINTR);
+
+#if defined HAVE_POSIX_FALLOCATE
+    /* Fallback to posix_fallocate if available. */
+    if (ret != 0) {
+        ret = call_posix_fallocate(fd, offset, length);
+    }
+#endif
+
+    return check_error(ret, errInfo);
+#elif defined F_PREALLOCATE
+    /* Mac OS X specific, equivalent to posix_fallocate. */
+    int ret;
+    fstore_t fs;
+
+    memset(&fs, 0, sizeof(fs));
+    fs.fst_flags = F_ALLOCATECONTIG;
+    fs.fst_posmode = F_VOLPOSMODE;
+    fs.fst_offset = (off_t) offset;
+    fs.fst_length = (off_t) length;
+
+    ret = fcntl(fd, F_PREALLOCATE, &fs);
+
+    if (-1 == ret) {
+        fs.fst_flags = F_ALLOCATEALL;
+        ret = fcntl(fd, F_PREALLOCATE, &fs);
+
+#if defined HAVE_POSIX_FALLOCATE
+        /* Fallback to posix_fallocate if available. */
+        if (-1 == ret) {
+            ret = call_posix_fallocate(fd, offset, length);
+        }
+#endif
+    }
+
+    return check_error(ret, errInfo);
+#elif defined HAVE_POSIX_FALLOCATE
+    /* Other Unixes, use posix_fallocate if available. */
+    return check_error(call_posix_fallocate(fd, offset, length), errInfo);
+#else
+    errno = ENOTSUP;
+    return check_error(-1, errInfo);
+#endif
+}
diff --git a/erts/emulator/drivers/win32/win_efile.c b/erts/emulator/drivers/win32/win_efile.c
index dc7add01f7..f5011d11a5 100644
--- a/erts/emulator/drivers/win32/win_efile.c
+++ b/erts/emulator/drivers/win32/win_efile.c
@@ -1558,3 +1558,13 @@ efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset,
     errno = ERROR_SUCCESS;
     return check_error(0, errInfo);
 }
+
+int
+efile_fallocate(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length)
+{
+    /* No file preallocation method available in Windows. */
+    errno = errno_map(ERROR_NOT_SUPPORTED);
+    SetLastError(ERROR_NOT_SUPPORTED);
+
+    return check_error(-1, errInfo);
+}
diff --git a/erts/preloaded/src/prim_file.erl b/erts/preloaded/src/prim_file.erl
index eafab1bae4..c412b7faf2 100644
--- a/erts/preloaded/src/prim_file.erl
+++ b/erts/preloaded/src/prim_file.erl
@@ -27,7 +27,7 @@
 %% Generic file contents operations
 -export([open/2, close/1, datasync/1, sync/1, advise/4, position/2, truncate/1,
 	 write/2, pwrite/2, pwrite/3, read/2, read_line/1, pread/2, pread/3,
-	 copy/3, sendfile/10]).
+	 copy/3, sendfile/10, allocate/3]).
 
 %% Specialized file operations
 -export([open/1, open/3]).
@@ -100,6 +100,7 @@
 -define(FILE_FDATASYNC,        30).
 -define(FILE_ADVISE,           31).
 -define(FILE_SENDFILE,         32).
+-define(FILE_ALLOCATE,         33).
 
 %% Driver responses
 -define(FILE_RESP_OK,          0).
@@ -293,6 +294,11 @@ advise(#file_descriptor{module = ?MODULE, data = {Port, _}},
     end.
 
 %% Returns {error, Reason} | ok.
+allocate(#file_descriptor{module = ?MODULE, data = {Port, _}}, Offset, Length) ->
+    Cmd = <<?FILE_ALLOCATE, Offset:64/signed, Length:64/signed>>,
+    drv_command(Port, Cmd).
+
+%% Returns {error, Reason} | ok.
 write(#file_descriptor{module = ?MODULE, data = {Port, _}}, Bytes) ->
     case drv_command_nt(Port, [?FILE_WRITE,erlang:dt_prepend_vm_tag_data(Bytes)],undefined) of
 	{ok, _Size} ->