Merge branch 'ta/sendfile/OTP-9240'

* ta/sendfile/OTP-9240: (31 commits)
  Add sendfile server printouts
  Skip recv/send during tests for fallback platforms
  Remove header/trailer support
  Remove windows implementation
  Expand sendfile documentation
  Only allow tcp sockets as target for sendfile
  Move sendfile api to file module
  Preliminary work on header/trailer
  Use free_sendfile explicitly for non-async
  Remove debug printouts
  Add tests for send/recv/sendfile interactions
  Remove tests for file_server sendfile
  sendfile caller now has to be the controlling_process
  Remove support for file_server, sendfile has to be raw
  Set chunk size to 3 GB
  Change type of fd to be ErlDrvEvent
  Add ifdef's for HAVE_SENDFILE
  Fix freebsd support for sendfile
  Change nbytes to 64 bit
  Implement ignorefd for TCP
  ...

7 files changed, 411 insertions, 16 deletions

diff --git a/erts/configure.in b/erts/configure.in
index 378d0ab220..10d303e241 100644
--- a/erts/configure.in
+++ b/erts/configure.in
@@ -1742,6 +1742,22 @@ dnl       fdatasync requires linking against -lrt on SunOS <= 5.10.
 dnl       OpenSolaris 2009.06 is SunOS 5.11 and does not require -lrt.
 AC_SEARCH_LIBS(fdatasync, [rt])
 
+
+dnl sendfile syscall
+case $host_os in
+    linux*|freebsd*|dragonfly*|darwin*)
+		AC_CHECK_FUNCS([sendfile])
+		;;
+    solaris*)
+		AC_SEARCH_LIBS(sendfile, sendfile, AC_DEFINE(HAVE_SENDFILE, 1))
+		;;
+    win32)
+		LIBS="$LIBS -lmswsock"
+		;;
+    *)
+		;;
+esac
+
 dnl ----------------------------------------------------------------------
 dnl Checks for library functions.
 dnl ----------------------------------------------------------------------
diff --git a/erts/emulator/drivers/common/efile_drv.c b/erts/emulator/drivers/common/efile_drv.c
index 901d98c09d..5c52b99348 100644
--- a/erts/emulator/drivers/common/efile_drv.c
+++ b/erts/emulator/drivers/common/efile_drv.c
@@ -55,6 +55,7 @@
 #define FILE_READ_LINE          29
 #define FILE_FDATASYNC          30
 #define FILE_FADVISE            31
+#define FILE_SENDFILE           32
 
 /* Return codes */
 
@@ -98,7 +99,13 @@
 #  include "config.h"
 #endif
 #include <stdlib.h>
+
+// Need (NON)BLOCKING macros for sendfile
+#ifndef WANT_NONBLOCKING
+#define WANT_NONBLOCKING
+#endif
 #include "sys.h"
+
 #include "erl_driver.h"
 #include "erl_efile.h"
 #include "erl_threads.h"
@@ -225,9 +232,16 @@ static void file_outputv(ErlDrvData, ErlIOVec*);
 static void file_async_ready(ErlDrvData, ErlDrvThreadData);
 static void file_flush(ErlDrvData);
 
+#ifdef HAVE_SENDFILE
+static void file_ready_output(ErlDrvData data, ErlDrvEvent event);
+static void file_stop_select(ErlDrvEvent event, void* _);
+#endif /* HAVE_SENDFILE */
 
 
 enum e_timer {timer_idle, timer_again, timer_write};
+#ifdef HAVE_SENDFILE
+enum e_sendfile {sending, not_sending};
+#endif /* HAVE_SENDFILE */
 
 struct t_data;
 
@@ -242,6 +256,9 @@ typedef struct {
     struct t_data  *cq_head;  /* Queue of incoming commands */
     struct t_data  *cq_tail;  /* -""- */
     enum e_timer    timer_state;
+#ifdef HAVE_SENDFILE
+    enum e_sendfile sendfile_state;
+#endif /* HAVE_SENDFILE */
     size_t          read_bufsize;
     ErlDrvBinary   *read_binp;
     size_t          read_offset;
@@ -264,7 +281,11 @@ struct erl_drv_entry efile_driver_entry = {
     file_stop,
     file_output,
     NULL,
+#ifdef HAVE_SENDFILE
+    file_ready_output,
+#else
     NULL,
+#endif /* HAVE_SENDFILE */
     "efile",
     NULL,
     NULL,
@@ -279,7 +300,13 @@ struct erl_drv_entry efile_driver_entry = {
     ERL_DRV_EXTENDED_MAJOR_VERSION,
     ERL_DRV_EXTENDED_MINOR_VERSION,
     ERL_DRV_FLAG_USE_PORT_LOCKING,
+    NULL,
+    NULL,
+#ifdef HAVE_SENDFILE
+    file_stop_select
+#else
     NULL
+#endif /* HAVE_SENDFILE */
 };
 
 
@@ -398,6 +425,14 @@ struct t_data
 	    Sint64 length;
 	    int advise;
 	} fadvise;
+#ifdef HAVE_SENDFILE
+	struct {
+	    int out_fd;
+	    off_t offset;
+	    Uint64 nbytes;
+	    Uint64 written;
+	} sendfile;
+#endif /* HAVE_SENDFILE */
     } c;
     char b[1];
 };
@@ -485,7 +520,6 @@ static void *ef_safe_realloc(void *op, Uint s)
      : 0)
 
 
-
 #if 0
 
 static void ev_clear(ErlIOVec *ev) {
@@ -613,7 +647,6 @@ static struct t_data *cq_deq(file_descriptor *desc) {
 }
 
 
-
 /*********************************************************************
  * Driver entry point -> init
  */
@@ -628,6 +661,7 @@ file_init(void)
 			    ? atoi(buf)
 			    : 0);
     driver_system_info(&sys_info, sizeof(ErlDrvSysInfo));
+
     return 0;
 }
 
@@ -655,6 +689,9 @@ file_start(ErlDrvPort port, char* command)
     desc->cq_head = NULL;
     desc->cq_tail = NULL;
     desc->timer_state = timer_idle;
+#ifdef HAVE_SENDFILE
+    desc->sendfile_state = not_sending;
+#endif
     desc->read_bufsize = 0;
     desc->read_binp = NULL;
     desc->read_offset = 0;
@@ -893,8 +930,6 @@ static int reply_eof(file_descriptor *desc) {
     driver_output2(desc->port, &c, 1, NULL, 0);
     return 0;
 }
-
-
  
 static void invoke_name(void *data, int (*f)(Efile_error *, char *))
 {
@@ -1694,6 +1729,66 @@ static void invoke_fadvise(void *data)
     d->result_ok = efile_fadvise(&d->errInfo, fd, offset, length, advise);
 }
 
+#ifdef HAVE_SENDFILE
+static void invoke_sendfile(void *data)
+{
+    struct t_data *d = (struct t_data *)data;
+    int fd = d->fd;
+    int out_fd = (int)d->c.sendfile.out_fd;
+    Uint64 nbytes = d->c.sendfile.nbytes;
+    int result = 0;
+    d->again = 0;
+
+    result = efile_sendfile(&d->errInfo, fd, out_fd, &d->c.sendfile.offset, &nbytes, NULL);
+
+    d->c.sendfile.written += nbytes;
+
+    if (result == 1) {
+	if (sys_info.async_threads != 0) {
+	    d->result_ok = 0;
+	} else if (d->c.sendfile.nbytes == 0 && nbytes != 0) {
+	    d->result_ok = 1;
+	} else if ((d->c.sendfile.nbytes - nbytes) != 0) {
+	    d->result_ok = 1;
+	    d->c.sendfile.nbytes -= nbytes;
+	} else {
+	    d->result_ok = 0;
+	}
+    } else if (result == 0 && (d->errInfo.posix_errno == EAGAIN
+				 || d->errInfo.posix_errno == EINTR)) {
+	d->result_ok = 1;
+    } else {
+	d->result_ok = -1;
+    }
+}
+
+static void free_sendfile(void *data) {
+    EF_FREE(data);
+}
+
+static void file_ready_output(ErlDrvData data, ErlDrvEvent event)
+{
+    file_descriptor* fd = (file_descriptor*) data;
+
+    switch (fd->d->command) {
+    case FILE_SENDFILE:
+	driver_select(fd->port, event,
+		      (int)ERL_DRV_WRITE,(int) 0);
+	invoke_sendfile((void *)fd->d);
+	file_async_ready(data, (ErlDrvThreadData)fd->d);
+	break;
+    default:
+	break;
+    }
+}
+
+static void file_stop_select(ErlDrvEvent event, void* _)
+{
+
+}
+#endif /* HAVE_SENDFILE */
+
+
 static void free_readdir(void *data)
 {
     struct t_data *d = (struct t_data *) data;
@@ -1755,6 +1850,10 @@ static void cq_execute(file_descriptor *desc) {
     register void *void_ptr; /* Soft cast variable */
     if (desc->timer_state == timer_again)
 	return;
+#ifdef HAVE_SENDFILE
+    if (desc->sendfile_state == sending)
+	return;
+#endif
     if (! (d = cq_deq(desc)))
 	return;
     TRACE_F(("x%i", (int) d->command));
@@ -2105,6 +2204,37 @@ file_async_ready(ErlDrvData e, ErlDrvThreadData data)
 	  }
 	  free_preadv(data);
 	  break;
+#ifdef HAVE_SENDFILE
+      case FILE_SENDFILE:
+	  if (d->result_ok == -1) {
+	      desc->sendfile_state = not_sending;
+	      reply_error(desc, &d->errInfo);
+	      if (sys_info.async_threads != 0) {
+		  SET_NONBLOCKING(d->c.sendfile.out_fd);
+		  free_sendfile(data);
+	      } else {
+		driver_select(desc->port, (ErlDrvEvent)(long)d->c.sendfile.out_fd,
+			ERL_DRV_USE, 0);
+		free_sendfile(data);
+	      }
+	  } else if (d->result_ok == 0) {
+	      desc->sendfile_state = not_sending;
+	      reply_Sint64(desc, d->c.sendfile.written);
+	      if (sys_info.async_threads != 0) {
+		SET_NONBLOCKING(d->c.sendfile.out_fd);
+		free_sendfile(data);
+	      } else {
+		driver_select(desc->port, (ErlDrvEvent)(long)d->c.sendfile.out_fd, ERL_DRV_USE, 0);
+		free_sendfile(data);
+	      }
+	  } else if (d->result_ok == 1) { // If we are using select to send the rest of the data
+	      desc->sendfile_state = sending;
+	      desc->d = d;
+	      driver_select(desc->port, (ErlDrvEvent)(long)d->c.sendfile.out_fd,
+			    ERL_DRV_USE|ERL_DRV_WRITE, 1);
+	  }
+	  break;
+#endif
       default:
 	abort();
     }
@@ -3245,9 +3375,69 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	    goto done;
 	} /* case FILE_OPT_DELAYED_WRITE: */
     } ASSERT(0); goto done; /* case FILE_SETOPT: */
-    
+
+    case FILE_SENDFILE: {
+
+#ifdef HAVE_SENDFILE
+        struct t_data *d;
+	Uint32 out_fd, offsetH, offsetL, hd_len, tl_len;
+	Uint64 nbytes;
+	char flags;
+
+	if (ev->size < 1 + 7 * sizeof(Uint32) + sizeof(char)
+		|| !EV_GET_UINT32(ev, &out_fd, &p, &q)
+		|| !EV_GET_CHAR(ev, &flags, &p, &q)
+		|| !EV_GET_UINT32(ev, &offsetH, &p, &q)
+		|| !EV_GET_UINT32(ev, &offsetL, &p, &q)
+		|| !EV_GET_UINT64(ev, &nbytes, &p, &q)
+		|| !EV_GET_UINT32(ev, &hd_len, &p, &q)
+		|| !EV_GET_UINT32(ev, &tl_len, &p, &q)) {
+	    /* Buffer has wrong length to contain all the needed values */
+	    reply_posix_error(desc, EINVAL);
+	    goto done;
+	}
+
+	if (hd_len != 0 || tl_len != 0 || flags != 0) {
+	    // We do not allow header, trailers and/or flags right now
+	    reply_posix_error(desc, EINVAL);
+	    goto done;
+	}
+
+	d = EF_SAFE_ALLOC(sizeof(struct t_data));
+	d->fd = desc->fd;
+	d->command = command;
+	d->invoke = invoke_sendfile;
+	d->free = NULL;
+	d->level = 2;
+
+	d->c.sendfile.out_fd = (int) out_fd;
+	d->c.sendfile.written = 0;
+
+    #if SIZEOF_OFF_T == 4
+	if (offsetH != 0) {
+	    reply_posix_error(desc, EINVAL);
+	    goto done;
+	}
+	d->c.sendfile.offset = (off_t) offsetL;
+    #else
+	d->c.sendfile.offset = ((off_t) offsetH << 32) | offsetL;
+    #endif
+
+	d->c.sendfile.nbytes = nbytes;
+
+	if (sys_info.async_threads != 0) {
+	    SET_BLOCKING(d->c.sendfile.out_fd);
+	}
+
+	cq_enq(desc, d);
+#else
+	reply_posix_error(desc, ENOTSUP);
+#endif
+	goto done;
+        } /* case FILE_SENDFILE: */
+
     } /* switch(command) */
-    
+
     if (lseek_flush_read(desc, &err) < 0) {
 	reply_posix_error(desc, err);
 	goto done;
diff --git a/erts/emulator/drivers/common/erl_efile.h b/erts/emulator/drivers/common/erl_efile.h
index 3097ded3f1..349ab0e17b 100644
--- a/erts/emulator/drivers/common/erl_efile.h
+++ b/erts/emulator/drivers/common/erl_efile.h
@@ -118,6 +118,19 @@ typedef struct _Efile_info {
 				 */
 } Efile_info;
 
+
+#ifdef HAVE_SENDFILE
+/*
+ * Described the structure of header/trailers for sendfile
+ */
+struct t_sendfile_hdtl {
+    SysIOVec *headers;
+    int hdr_cnt;
+    SysIOVec *trailers;
+    int trl_cnt;
+};
+#endif /* HAVE_SENDFILE */
+
 /*
  * Functions.
  */
@@ -162,3 +175,7 @@ int efile_symlink(Efile_error* errInfo, char* old, char* new);
 int efile_may_openfile(Efile_error* errInfo, char *name);
 int efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length,
 		  int advise);
+#ifdef HAVE_SENDFILE
+int efile_sendfile(Efile_error* errInfo, int in_fd, int out_fd,
+		      off_t *offset, Uint64 *nbytes, struct t_sendfile_hdtl *hdtl);
+#endif /* HAVE_SENDFILE */
diff --git a/erts/emulator/drivers/common/inet_drv.c b/erts/emulator/drivers/common/inet_drv.c
index db052523a8..7d952b0c71 100644
--- a/erts/emulator/drivers/common/inet_drv.c
+++ b/erts/emulator/drivers/common/inet_drv.c
@@ -445,6 +445,7 @@ static int my_strncasecmp(const char *s1, const char *s2, size_t n)
                                     driver_select(port, e, mode | (on?ERL_DRV_USE:0), on)
 
 #define sock_select(d, flags, onoff) do { \
+        ASSERT(!onoff || !(d)->is_ignored); \
         (d)->event_mask = (onoff) ? \
                  ((d)->event_mask | (flags)) : \
                  ((d)->event_mask & ~(flags)); \
@@ -538,6 +539,8 @@ static int my_strncasecmp(const char *s1, const char *s2, size_t n)
 #define INET_REQ_GETIFADDRS    25
 #define INET_REQ_ACCEPT        26
 #define INET_REQ_LISTEN        27
+#define INET_REQ_IGNOREFD      28
+
 /* TCP requests */
 /* #define TCP_REQ_ACCEPT         40 MOVED */
 /* #define TCP_REQ_LISTEN         41 MERGED */
@@ -725,6 +728,11 @@ static int my_strncasecmp(const char *s1, const char *s2, size_t n)
 /* Max interface name */
 #define INET_IFNAMSIZ          16
 
+/* INET Ignore states */
+#define INET_IGNORE_NONE 0
+#define INET_IGNORE_READ 1
+#define INET_IGNORE_WRITE 1 << 1
+
 /* Max length of Erlang Term Buffer (for outputting structured terms):  */
 #ifdef  HAVE_SCTP
 #define PACKET_ERL_DRV_TERM_DATA_LEN  512
@@ -864,6 +872,9 @@ typedef struct {
     double send_avg;            /* average packet size sent */
 
     subs_list empty_out_q_subs; /* Empty out queue subscribers */
+    int is_ignored;             /* if a fd is ignored by from the inet_drv,
+				   this should be set to true when the fd is used
+				   outside of inet_drv. */
 } inet_descriptor;
 
 
@@ -7344,6 +7355,8 @@ static ErlDrvData inet_start(ErlDrvPort port, int size, int protocol)
 
     sys_memzero((char *)&desc->remote,sizeof(desc->remote));
 
+    desc->is_ignored = 0;
+
     return (ErlDrvData)desc;
 }
 
@@ -7626,6 +7639,33 @@ static int inet_ctl(inet_descriptor* desc, int cmd, char* buf, int len,
 	return ctl_reply(INET_REP_OK, tbuf, 2, rbuf, rsize);
     }
 
+    case INET_REQ_IGNOREFD: {
+      DEBUGF(("inet_ctl(%ld): IGNOREFD, IGNORED = %d\r\n",
+	      (long)desc->port,(int)*buf));
+
+      /*
+       * FD can only be ignored for connected TCP connections for now,
+       * possible to add UDP and SCTP support if needed.
+       */
+      if (!IS_CONNECTED(desc))
+	  return ctl_error(ENOTCONN, rbuf, rsize);
+
+      if (!desc->stype == SOCK_STREAM)
+	  return ctl_error(EINVAL, rbuf, rsize);
+
+      if (*buf == 1 && !desc->is_ignored) {
+	  desc->is_ignored = INET_IGNORE_READ;
+	  sock_select(desc, (FD_READ|FD_WRITE|FD_CLOSE|ERL_DRV_USE_NO_CALLBACK), 0);
+      } else if (*buf == 0 && desc->is_ignored) {
+	  int flags = (FD_READ|FD_CLOSE|((desc->is_ignored & INET_IGNORE_WRITE)?FD_WRITE:0));
+	  desc->is_ignored = INET_IGNORE_NONE;
+	  sock_select(desc, flags, 1);
+      } else
+	  return ctl_error(EINVAL, rbuf, rsize);
+
+      return ctl_reply(INET_REP_OK, NULL, 0, rbuf, rsize);
+    }
+
 #ifndef VXWORKS
 
     case INET_REQ_GETSERVBYNAME: { /* L1 Name-String L2 Proto-String */
@@ -8001,6 +8041,7 @@ static int tcp_inet_ctl(ErlDrvData e, unsigned int cmd, char* buf, int len,
 			char** rbuf, int rsize)
 {
     tcp_descriptor* desc = (tcp_descriptor*)e;
+
     switch(cmd) {
     case INET_REQ_OPEN: { /* open socket and return internal index */
 	int domain;
@@ -8266,13 +8307,14 @@ static int tcp_inet_ctl(ErlDrvData e, unsigned int cmd, char* buf, int len,
 	if (enq_async(INETP(desc), tbuf, TCP_REQ_RECV) < 0)
 	    return ctl_error(EALREADY, rbuf, rsize);
 
-	if (tcp_recv(desc, n) == 0) {
+	if (INETP(desc)->is_ignored || tcp_recv(desc, n) == 0) {
 	    if (timeout == 0)
 		async_error_am(INETP(desc), am_timeout);
 	    else {
 		if (timeout != INET_INFINITY)
-		    driver_set_timer(desc->inet.port, timeout); 
-		sock_select(INETP(desc),(FD_READ|FD_CLOSE),1);
+		    driver_set_timer(desc->inet.port, timeout);
+		if (!INETP(desc)->is_ignored)
+		    sock_select(INETP(desc),(FD_READ|FD_CLOSE),1);
 	    }
 	}
 	return ctl_reply(INET_REP_OK, tbuf, 2, rbuf, rsize);
@@ -9012,6 +9054,7 @@ static int tcp_inet_input(tcp_descriptor* desc, HANDLE event)
 #ifdef DEBUG
     long port = (long) desc->inet.port;  /* Used after driver_exit() */
 #endif
+    ASSERT(!INETP(desc)->is_ignored);
     DEBUGF(("tcp_inet_input(%ld) {s=%d\r\n", port, desc->inet.s));
     if (desc->inet.state == INET_STATE_ACCEPTING) {
 	SOCKET s;
@@ -9273,7 +9316,11 @@ static int tcp_sendv(tcp_descriptor* desc, ErlIOVec* ev)
 	
 	DEBUGF(("tcp_sendv(%ld): s=%d, about to send %d,%d bytes\r\n",
 		(long)desc->inet.port, desc->inet.s, h_len, len));
-	if (desc->tcp_add_flags & TCP_ADDF_DELAY_SEND) {
+
+	if (INETP(desc)->is_ignored) {
+	    INETP(desc)->is_ignored |= INET_IGNORE_WRITE;
+	    n = 0;
+	} else if (desc->tcp_add_flags & TCP_ADDF_DELAY_SEND) {
 	    n = 0;
 	} else if (IS_SOCKET_ERROR(sock_sendv(desc->inet.s, ev->iov,
 					      vsize, &n, 0))) {
@@ -9301,7 +9348,8 @@ static int tcp_sendv(tcp_descriptor* desc, ErlIOVec* ev)
 	DEBUGF(("tcp_sendv(%ld): s=%d, Send failed, queuing\r\n", 
 		(long)desc->inet.port, desc->inet.s));
 	driver_enqv(ix, ev, n); 
-	sock_select(INETP(desc),(FD_WRITE|FD_CLOSE), 1);
+	if (!INETP(desc)->is_ignored)
+	    sock_select(INETP(desc),(FD_WRITE|FD_CLOSE), 1);
     }
     return 0;
 }
@@ -9366,7 +9414,10 @@ static int tcp_send(tcp_descriptor* desc, char* ptr, int len)
 
 	DEBUGF(("tcp_send(%ld): s=%d, about to send %d,%d bytes\r\n",
 		(long)desc->inet.port, desc->inet.s, h_len, len));
-	if (desc->tcp_add_flags & TCP_ADDF_DELAY_SEND) {
+	if (INETP(desc)->is_ignored) {
+	    INETP(desc)->is_ignored |= INET_IGNORE_WRITE;
+	    n = 0;
+	} else if (desc->tcp_add_flags & TCP_ADDF_DELAY_SEND) {
 	    sock_send(desc->inet.s, buf, 0, 0);
 	    n = 0;
 	} else 	if (IS_SOCKET_ERROR(sock_sendv(desc->inet.s,iov,2,&n,0))) {
@@ -9397,7 +9448,8 @@ static int tcp_send(tcp_descriptor* desc, char* ptr, int len)
 	    n -= h_len;
 	    driver_enq(ix, ptr+n, len-n);
 	}
-	sock_select(INETP(desc),(FD_WRITE|FD_CLOSE), 1);
+	if (!INETP(desc)->is_ignored)
+	    sock_select(INETP(desc),(FD_WRITE|FD_CLOSE), 1);
     }
     return 0;
 }
@@ -9421,6 +9473,7 @@ static int tcp_inet_output(tcp_descriptor* desc, HANDLE event)
     int ret = 0;
     ErlDrvPort ix = desc->inet.port;
 
+    ASSERT(!INETP(desc)->is_ignored);
     DEBUGF(("tcp_inet_output(%ld) {s=%d\r\n", 
 	    (long)desc->inet.port, desc->inet.s));
     if (desc->inet.state == INET_STATE_CONNECTING) {
diff --git a/erts/emulator/drivers/unix/unix_efile.c b/erts/emulator/drivers/unix/unix_efile.c
index 4b3934657c..72911641d3 100644
--- a/erts/emulator/drivers/unix/unix_efile.c
+++ b/erts/emulator/drivers/unix/unix_efile.c
@@ -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
@@ -1464,3 +1467,77 @@ efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset,
     return check_error(0, errInfo);
 #endif
 }
+
+#ifdef HAVE_SENDFILE
+#define SENDFILE_CHUNK_SIZE ((1 << 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.
+ *
+ * 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.
+ */
+
+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__) || (defined(__sun) && defined(__SVR4))
+    ssize_t retval;
+    do {
+      // check if *nbytes is 0 or greater than the largest size_t
+      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 != -1 && retval == SENDFILE_CHUNK_SIZE);
+    *nbytes = written;
+    return check_error(retval == -1 ? -1 : 0, errInfo);
+#elif defined(DARWIN)
+    int retval;
+    off_t len;
+    do {
+      // check if *nbytes is 0 or greater than the largest off_t
+      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);
+    *nbytes = written;
+    return check_error(retval, errInfo);
+#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);
+    *nbytes = written;
+    return check_error(retval, errInfo);
+#endif
+}
+#endif /* HAVE_SENDFILE */
diff --git a/erts/preloaded/src/prim_file.erl b/erts/preloaded/src/prim_file.erl
index 30b7a5246a..7316e0be99 100644
--- a/erts/preloaded/src/prim_file.erl
+++ b/erts/preloaded/src/prim_file.erl
@@ -26,7 +26,8 @@
 
 %% 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]).
+	 write/2, pwrite/2, pwrite/3, read/2, read_line/1, pread/2, pread/3,
+	 copy/3, sendfile/10]).
 
 %% Specialized file operations
 -export([open/1, open/3]).
@@ -98,6 +99,7 @@
 -define(FILE_READ_LINE,        29).
 -define(FILE_FDATASYNC,        30).
 -define(FILE_ADVISE,           31).
+-define(FILE_SENDFILE,         32).
 
 %% Driver responses
 -define(FILE_RESP_OK,          0).
@@ -537,7 +539,31 @@ write_file(File, Bin) when (is_list(File) orelse is_binary(File)) ->
     end;
 write_file(_, _) -> 
     {error, badarg}.
-    
+
+
+%% Returns {error, Reason} | {ok, BytesCopied}
+%sendfile(_,_,_,_,_,_,_,_,_,_) ->
+%    {error, enotsup};
+sendfile(#file_descriptor{module = ?MODULE, data = {Port, _}},
+	 Dest, Offset, Bytes, _ChunkSize, Headers, Trailers,
+	 _Nodiskio, _MNowait, _Sync) ->
+    case erlang:port_get_data(Dest) of
+	Data when Data == inet_tcp; Data == inet6_tcp ->
+	    ok = inet:lock_socket(Dest,true),
+	    {ok, DestFD} = prim_inet:getfd(Dest),
+	    try drv_command(Port, [<<?FILE_SENDFILE, DestFD:32,
+				     0:8,
+				     Offset:64/unsigned,
+				     Bytes:64/unsigned,
+				     (iolist_size(Headers)):32/unsigned,
+				     (iolist_size(Trailers)):32/unsigned>>,
+				   Headers,Trailers])
+	    after
+		ok = inet:lock_socket(Dest,false)
+	    end;
+	_Else ->
+	    {error,badarg}
+    end.
 
 
 %%%-----------------------------------------------------------------
diff --git a/erts/preloaded/src/prim_inet.erl b/erts/preloaded/src/prim_inet.erl
index f144f73d68..0cedd284db 100644
--- a/erts/preloaded/src/prim_inet.erl
+++ b/erts/preloaded/src/prim_inet.erl
@@ -36,7 +36,8 @@
 -export([recvfrom/2, recvfrom/3]).
 -export([setopt/3, setopts/2, getopt/2, getopts/2, is_sockopt_val/2]).
 -export([chgopt/3, chgopts/2]).
--export([getstat/2, getfd/1, getindex/1, getstatus/1, gettype/1, 
+-export([getstat/2, getfd/1, ignorefd/2,
+	 getindex/1, getstatus/1, gettype/1,
 	 getifaddrs/1, getiflist/1, ifget/3, ifset/3,
 	 gethostname/1]).
 -export([getservbyname/3, getservbyport/3]).
@@ -842,6 +843,21 @@ getfd(S) when is_port(S) ->
 
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 %%
+%% IGNOREFD(insock(),boolean()) -> {ok,integer()} | {error, Reason}
+%%
+%% steal internal file descriptor
+%%
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+
+ignorefd(S,Bool) when is_port(S) ->
+    Val = if Bool -> 1; true -> 0 end,
+    case ctl_cmd(S, ?INET_REQ_IGNOREFD, [Val]) of
+	{ok, _} -> ok;
+	Error -> Error
+    end.
+
+%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+%%
 %% GETIX(insock()) -> {ok,integer()} | {error, Reason}
 %%
 %% get internal socket index