1 files changed, 176 insertions, 124 deletions

diff --git a/erts/emulator/drivers/unix/unix_efile.c b/erts/emulator/drivers/unix/unix_efile.c
index 4b3934657c..b250bac4dc 100644
--- a/erts/emulator/drivers/unix/unix_efile.c
+++ b/erts/emulator/drivers/unix/unix_efile.c
@@ -1,7 +1,7 @@
 /*
  * %CopyrightBegin%
  *
- * Copyright Ericsson AB 1997-2010. All Rights Reserved.
+ * Copyright Ericsson AB 1997-2012. All Rights Reserved.
  *
  * The contents of this file are subject to the Erlang Public License,
  * Version 1.1, (the "License"); you may not use this file except in
@@ -33,6 +33,9 @@
 #include <sys/types.h>
 #include <sys/uio.h>
 #endif
+#if defined(HAVE_SENDFILE) && (defined(__linux__) || (defined(__sun) && defined(__SVR4)))
+#include <sys/sendfile.h>
+#endif
 
 #if defined(__APPLE__) && defined(__MACH__) && !defined(__DARWIN__)
 #define DARWIN 1
@@ -104,8 +107,8 @@ static void *ef_safe_alloc(Uint s)
 {
     void *p = EF_ALLOC(s);
     if (!p) erl_exit(1,
-		     "unix efile drv: Can't allocate %d bytes of memory\n",
-		     s);
+		     "unix efile drv: Can't allocate %lu bytes of memory\n",
+		     (unsigned long)s);
     return p;
 }
 
@@ -115,8 +118,8 @@ static void *ef_safe_realloc(void *op, Uint s)
 {
     void *p = EF_REALLOC(op, s);
     if (!p) erl_exit(1,
-		     "unix efile drv: Can't reallocate %d bytes of memory\n",
-		     s);
+		     "unix efile drv: Can't reallocate %lu bytes of memory\n",
+		     (unsigned long)s);
     return p;
 }
 
@@ -813,7 +816,6 @@ efile_fileinfo(Efile_error* errInfo, Efile_info* pInfo,
 	       char* name, int info_for_link)
 {
     struct stat statbuf;	/* Information about the file */
-    struct tm *timep;		/* Broken-apart filetime. */
     int result;
 
 #ifdef VXWORKS
@@ -880,40 +882,17 @@ efile_fileinfo(Efile_error* errInfo, Efile_info* pInfo,
     else
 	pInfo->type = FT_OTHER;
 
-#if defined(HAVE_LOCALTIME_R) || defined(VXWORKS)
-    {
-	/* Use the reentrant version of localtime() */
-	static struct tm local_tm;
-#define localtime(a) (localtime_r((a), &local_tm), &local_tm)
-#endif
-
-
-#define GET_TIME(dst, src) \
-    timep = localtime(&statbuf.src); \
-    (dst).year = timep->tm_year+1900; \
-    (dst).month = timep->tm_mon+1; \
-    (dst).day = timep->tm_mday; \
-    (dst).hour = timep->tm_hour; \
-    (dst).minute = timep->tm_min; \
-    (dst).second = timep->tm_sec
-
-    GET_TIME(pInfo->accessTime, st_atime);
-    GET_TIME(pInfo->modifyTime, st_mtime);
-    GET_TIME(pInfo->cTime, st_ctime);
-
-#undef GET_TIME
+    pInfo->accessTime   = statbuf.st_atime;
+    pInfo->modifyTime   = statbuf.st_mtime;
+    pInfo->cTime        = statbuf.st_ctime;
 
-#if defined(HAVE_LOCALTIME_R) || defined(VXWORKS)
-    }
-#endif
-
-    pInfo->mode = statbuf.st_mode;
-    pInfo->links = statbuf.st_nlink;
+    pInfo->mode         = statbuf.st_mode;
+    pInfo->links        = statbuf.st_nlink;
     pInfo->major_device = statbuf.st_dev;
     pInfo->minor_device = statbuf.st_rdev;
-    pInfo->inode = statbuf.st_ino;
-    pInfo->uid = statbuf.st_uid;
-    pInfo->gid = statbuf.st_gid;
+    pInfo->inode        = statbuf.st_ino;
+    pInfo->uid          = statbuf.st_uid;
+    pInfo->gid          = statbuf.st_gid;
 
     return 1;
 }
@@ -921,6 +900,8 @@ efile_fileinfo(Efile_error* errInfo, Efile_info* pInfo,
 int
 efile_write_info(Efile_error *errInfo, Efile_info *pInfo, char *name)
 {
+    struct utimbuf tval;
+
     CHECK_PATHLEN(name, errInfo);
 
 #ifdef VXWORKS
@@ -973,38 +954,18 @@ efile_write_info(Efile_error *errInfo, Efile_info *pInfo, char *name)
 
 #endif /* !VXWORKS */
 
-    if (pInfo->accessTime.year != -1 && pInfo->modifyTime.year != -1) {
-	struct utimbuf tval;
-	struct tm timebuf;
-
-#define MKTIME(tb, ts) \
-    timebuf.tm_year = ts.year-1900; \
-    timebuf.tm_mon = ts.month-1; \
-    timebuf.tm_mday = ts.day; \
-    timebuf.tm_hour = ts.hour; \
-    timebuf.tm_min = ts.minute; \
-    timebuf.tm_sec = ts.second; \
-    timebuf.tm_isdst = -1; \
-    if ((tb = mktime(&timebuf)) == (time_t) -1) { \
-       errno = EINVAL; \
-       return check_error(-1, errInfo); \
-    }
+    tval.actime  = pInfo->accessTime;
+    tval.modtime = pInfo->modifyTime;
 
-        MKTIME(tval.actime, pInfo->accessTime);
-	MKTIME(tval.modtime, pInfo->modifyTime);
-#undef MKTIME
-	
 #ifdef VXWORKS
-	/* VxWorks' utime doesn't work when the file is a nfs mounted
-	 * one, don't report error if utime fails.
-	 */
-	utime(name, &tval);
-	return 1;
+    /* VxWorks' utime doesn't work when the file is a nfs mounted
+     * one, don't report error if utime fails.
+     */
+    utime(name, &tval);
+    return 1;
 #else
-	return check_error(utime(name, &tval), errInfo);
+    return check_error(utime(name, &tval), errInfo);
 #endif
-    }
-    return 1;
 }
 
 
@@ -1043,13 +1004,11 @@ efile_writev(Efile_error* errInfo,   /* Where to return error codes */
 				      * opened */
 	     int fd,                 /* File descriptor to write to */
 	     SysIOVec* iov,          /* Vector of buffer structs.
-				      * The structs are unchanged 
-				      * after the call */
-	     int iovcnt,             /* Number of structs in vector */
-	     size_t size)            /* Number of bytes to write */
+				      * The structs may be changed i.e.
+				      * due to incomplete writes */
+	     int iovcnt)             /* Number of structs in vector */
 {
     int cnt = 0;                     /* Buffers so far written */
-    int p = 0;                       /* Position in next buffer */
 
     ASSERT(iovcnt >= 0);
     
@@ -1060,66 +1019,47 @@ efile_writev(Efile_error* errInfo,   /* Where to return error codes */
 #endif
 
     while (cnt < iovcnt) {
+	if ((! iov[cnt].iov_base) || (iov[cnt].iov_len <= 0)) {
+	    /* Empty buffer - skip */
+	    cnt++;
+	} else { /* Non-empty buffer */
+	    ssize_t w;                   /* Bytes written in this call */
 #ifdef HAVE_WRITEV
-	int w;                       /* Bytes written in this call */
-	int b = iovcnt - cnt;        /* Buffers to write */
-	if (b > MAXIOV)
-	    b = MAXIOV;
-	if (iov[cnt].iov_base && iov[cnt].iov_len > 0) {
-	    if (b == 1) {
-		/* Degenerated io vector */
-		do {
-		    w = write(fd, iov[cnt].iov_base + p, iov[cnt].iov_len - p);
-		} while (w < 0 && errno == EINTR);
-	    } else {
-		/* Non-empty vector first.
-		 * Adjust pos in first buffer in case of 
-		 * previous incomplete writev */
-		iov[cnt].iov_base += p;
-		iov[cnt].iov_len  -= p;
+	    int b = iovcnt - cnt;        /* Buffers to write */
+	    /* Use as many buffers as MAXIOV allows */
+	    if (b > MAXIOV)
+		b = MAXIOV;
+	    if (b > 1) {
 		do {
 		    w = writev(fd, &iov[cnt], b);
 		} while (w < 0 && errno == EINTR);
-		iov[cnt].iov_base -= p;
-		iov[cnt].iov_len  += p;
-	    }
-	    if (w < 0)
-		return check_error(-1, errInfo);
-	} else {
-	    /* Empty vector first - skip */
-	    cnt++;
-	    continue;
-	}
-	ASSERT(w >= 0);
-	/* Move forward to next vector to write */
-	for (; cnt < iovcnt; cnt++) {
-	    if (iov[cnt].iov_base && iov[cnt].iov_len > 0) {
-		if (w < iov[cnt].iov_len)
-		    break;
-		else
-		    w -= iov[cnt].iov_len;
-	    }
-	}
-	ASSERT(w >= 0);
-	p = w > 0 ? w : 0; /* Skip p bytes next writev */
-#else /* #ifdef HAVE_WRITEV */
-	if (iov[cnt].iov_base && iov[cnt].iov_len > 0) {
-	    /* Non-empty vector */
-	    int w;                   /* Bytes written in this call */
-	    while (p < iov[cnt].iov_len) {
-		do {
-		    w = write(fd, iov[cnt].iov_base + p, iov[cnt].iov_len - p);
-		} while (w < 0 && errno == EINTR);
-		if (w < 0)
-		    return check_error(-1, errInfo);
-		p += w;
+	    } else
+		/* Degenerated io vector - use regular write */
+#endif
+		{
+		    do {
+			w = write(fd, iov[cnt].iov_base, iov[cnt].iov_len);
+		    } while (w < 0 && errno == EINTR);
+		    ASSERT(w <= iov[cnt].iov_len);
+		}
+	    if (w < 0) return check_error(-1, errInfo);
+	    /* Move forward to next buffer to write */
+	    for (; cnt < iovcnt && w > 0; cnt++) {
+		if (iov[cnt].iov_base && iov[cnt].iov_len > 0) {
+		    if (w < iov[cnt].iov_len) {
+			/* Adjust the buffer for next write */
+			iov[cnt].iov_len -= w;
+			iov[cnt].iov_base += w;
+			w = 0;
+			break;
+		    } else {
+			w -= iov[cnt].iov_len;
+		    }
+		}
 	    }
-	}
-	cnt++;
-	p = 0;
-#endif /* #ifdef HAVE_WRITEV */
+	    ASSERT(w == 0);
+	} /* else Non-empty buffer */
     } /* while (cnt< iovcnt) */
-    size = 0; /* Avoid compiler warning */
     return 1;
 }
 
@@ -1464,3 +1404,115 @@ efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset,
     return check_error(0, errInfo);
 #endif
 }
+
+#ifdef HAVE_SENDFILE
+// For some reason the maximum size_t cannot be used as the max size
+// 3GB seems to work on all platforms
+#define SENDFILE_CHUNK_SIZE ((1UL << 30) -1)
+
+/*
+ * sendfile: The implementation of the sendfile system call varies
+ * a lot on different *nix platforms so to make the api similar in all
+ * we have to emulate some things in linux and play with variables on
+ * bsd/darwin.
+ *
+ * All of the calls will split a command which tries to send more than
+ * SENDFILE_CHUNK_SIZE of data at once.
+ *
+ * On platforms where *nbytes of 0 does not mean the entire file, this is
+ * simulated.
+ *
+ * It could be possible to implement header/trailer in sendfile. Though
+ * you would have to emulate it in linux and on BSD/Darwin some complex
+ * calculations have to be made when using a non blocking socket to figure
+ * out how much of the header/file/trailer was sent in each command.
+ *
+ * The semantics of the API is this:
+ * Return value: 1 if all data was sent and the function does not need to
+ * be called again. 0 if an error occures OR if there is more data which
+ * has to be sent (EAGAIN or EINTR will be set appropriately)
+ *
+ * The amount of data written in a call is returned through nbytes.
+ *
+ */
+
+int
+efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
+	       off_t *offset, Uint64 *nbytes, struct t_sendfile_hdtl* hdtl)
+{
+    Uint64 written = 0;
+#if defined(__linux__)
+    ssize_t retval;
+    do {
+      // check if *nbytes is 0 or greater than chunk size
+      if (*nbytes == 0 || *nbytes > SENDFILE_CHUNK_SIZE)
+	retval = sendfile(out_fd, in_fd, offset, SENDFILE_CHUNK_SIZE);
+      else
+	retval = sendfile(out_fd, in_fd, offset, *nbytes);
+      if (retval > 0) {
+	written += retval;
+	*nbytes -= retval;
+      }
+    } while (retval == SENDFILE_CHUNK_SIZE);
+    if (written != 0) {
+      // -1 is not returned by the linux API so we have to simulate it
+      retval = -1;
+      errno = EAGAIN;
+    }
+#elif defined(__sun) && defined(__SVR4) && defined(HAVE_SENDFILEV)
+    ssize_t retval;
+    size_t len;
+    sendfilevec_t fdrec;
+    fdrec.sfv_fd = in_fd;
+    fdrec.sfv_flag = 0;
+    do {
+      fdrec.sfv_off = *offset;
+      len = 0;
+      // check if *nbytes is 0 or greater than chunk size
+      if (*nbytes == 0 || *nbytes > SENDFILE_CHUNK_SIZE)
+	fdrec.sfv_len = SENDFILE_CHUNK_SIZE;
+      else
+	fdrec.sfv_len = *nbytes;
+      retval = sendfilev(out_fd, &fdrec, 1, &len);
+      if (retval != -1 || errno == EAGAIN || errno == EINTR) {
+        *offset += len;
+	*nbytes -= len;
+	written += len;
+      }
+    } while (len == SENDFILE_CHUNK_SIZE);
+#elif defined(DARWIN)
+    int retval;
+    off_t len;
+    do {
+      // check if *nbytes is 0 or greater than chunk size
+      if(*nbytes > SENDFILE_CHUNK_SIZE)
+	len = SENDFILE_CHUNK_SIZE;
+      else
+	len = *nbytes;
+      retval = sendfile(in_fd, out_fd, *offset, &len, NULL, 0);
+      if (retval != -1 || errno == EAGAIN || errno == EINTR) {
+        *offset += len;
+	*nbytes -= len;
+	written += len;
+      }
+    } while (len == SENDFILE_CHUNK_SIZE);
+#elif defined(__FreeBSD__) || defined(__DragonFly__)
+    off_t len;
+    int retval;
+    do {
+      if (*nbytes > SENDFILE_CHUNK_SIZE)
+	retval = sendfile(in_fd, out_fd, *offset, SENDFILE_CHUNK_SIZE,
+			  NULL, &len, 0);
+      else
+	retval = sendfile(in_fd, out_fd, *offset, *nbytes, NULL, &len, 0);
+      if (retval != -1 || errno == EAGAIN || errno == EINTR) {
+	*offset += len;
+	*nbytes -= len;
+	written += len;
+      }
+    } while(len == SENDFILE_CHUNK_SIZE);
+#endif
+    *nbytes = written;
+    return check_error(retval, errInfo);
+}
+#endif /* HAVE_SENDFILE */