1 files changed, 103 insertions, 9 deletions

diff --git a/erts/emulator/drivers/unix/unix_efile.c b/erts/emulator/drivers/unix/unix_efile.c
index 2aa373aa7d..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)
@@ -851,8 +861,8 @@ efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset,
 }
 
 #ifdef HAVE_SENDFILE
-// For some reason the maximum size_t cannot be used as the max size
-// 3GB seems to work on all platforms
+/* For some reason the maximum size_t cannot be used as the max size
+   3GB seems to work on all platforms */
 #define SENDFILE_CHUNK_SIZE ((1UL << 30) -1)
 
 /*
@@ -889,7 +899,7 @@ efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
 #if defined(__linux__)
     ssize_t retval;
     do {
-      // check if *nbytes is 0 or greater than chunk size
+      /* check if *nbytes is 0 or greater than chunk size */
       if (*nbytes == 0 || *nbytes > SENDFILE_CHUNK_SIZE)
 	retval = sendfile(out_fd, in_fd, offset, SENDFILE_CHUNK_SIZE);
       else
@@ -900,7 +910,7 @@ efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
       }
     } while (retval == SENDFILE_CHUNK_SIZE);
     if (written != 0) {
-      // -1 is not returned by the linux API so we have to simulate it
+      /* -1 is not returned by the linux API so we have to simulate it */
       retval = -1;
       errno = EAGAIN;
     }
@@ -913,23 +923,29 @@ efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
     do {
       fdrec.sfv_off = *offset;
       len = 0;
-      // check if *nbytes is 0 or greater than chunk size
+      /* check if *nbytes is 0 or greater than chunk size */
       if (*nbytes == 0 || *nbytes > SENDFILE_CHUNK_SIZE)
 	fdrec.sfv_len = SENDFILE_CHUNK_SIZE;
       else
 	fdrec.sfv_len = *nbytes;
       retval = sendfilev(out_fd, &fdrec, 1, &len);
-      if (retval != -1 || errno == EAGAIN || errno == EINTR) {
+
+      /* Sometimes sendfilev can return -1 and still send data. 
+         When that happens we just pretend that no error happend. */
+      if (retval != -1 || errno == EAGAIN || errno == EINTR ||
+	  len != 0) {
         *offset += len;
 	*nbytes -= len;
 	written += len;
+	if (errno != EAGAIN && errno != EINTR && len != 0)
+	  retval = len;
       }
     } while (len == SENDFILE_CHUNK_SIZE);
 #elif defined(DARWIN)
     int retval;
     off_t len;
     do {
-      // check if *nbytes is 0 or greater than chunk size
+      /* check if *nbytes is 0 or greater than chunk size */
       if(*nbytes > SENDFILE_CHUNK_SIZE)
 	len = SENDFILE_CHUNK_SIZE;
       else
@@ -961,3 +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
+}