[SCM] Samba Shared Repository - branch master updated

Jeremy Allison jra at samba.org
Tue Dec 14 18:25:02 MST 2010


The branch, master has been updated
       via  a8eed18 Implement "use sendfile = yes" for SMB2. (cherry picked from commit 95cb7adcd03a1abbd0af395b6c96dd8e0eebd3d1)
       via  e071fd7 Fix read/write calls over sockets to cope with EAGAIN/EWOULDBLOCK for non-blocking sockets.
       via  14c7034 Change interface of schedule_smb2_aio_read() to allocate the return DATA_BLOB. Change smb2_read code to allocate return DATA_BLOB just before the read.
      from  3bc1cf9 s4-test/repl_schema: remove unused and commented out code

http://gitweb.samba.org/?p=samba.git;a=shortlog;h=master


- Log -----------------------------------------------------------------
commit a8eed184a0f9e5fdeec9a40d8ffbc2f3d56beb74
Author: Jeremy Allison <jra at samba.org>
Date:   Tue Dec 14 13:36:08 2010 -0800

    Implement "use sendfile = yes" for SMB2. (cherry picked from commit 95cb7adcd03a1abbd0af395b6c96dd8e0eebd3d1)
    
    Autobuild-User: Jeremy Allison <jra at samba.org>
    Autobuild-Date: Wed Dec 15 02:24:08 CET 2010 on sn-devel-104

commit e071fd711dee0a5b03df53979b355f2e5a2e58b0
Author: Jeremy Allison <jra at samba.org>
Date:   Tue Dec 14 15:30:06 2010 -0800

    Fix read/write calls over sockets to cope with EAGAIN/EWOULDBLOCK for non-blocking sockets.

commit 14c70346fc88d05dfabc14b0a932632470fee6fe
Author: Jeremy Allison <jra at samba.org>
Date:   Tue Dec 14 16:32:10 2010 -0800

    Change interface of schedule_smb2_aio_read() to allocate the return DATA_BLOB.
    Change smb2_read code to allocate return DATA_BLOB just before the read.
    
    Preparing for SMB2 sendfile change which will not need to allocate
    return buffer.
    
    Jeremy

-----------------------------------------------------------------------

Summary of changes:
 source3/include/proto.h    |    8 ++-
 source3/lib/sendfile.c     |   36 ++++++++--
 source3/lib/system.c       |   48 ++++++++++---
 source3/smbd/aio.c         |   14 +++-
 source3/smbd/reply.c       |    5 +-
 source3/smbd/smb2_read.c   |  168 ++++++++++++++++++++++++++++++++++++++++++--
 source3/smbd/smb2_server.c |    9 +++
 7 files changed, 259 insertions(+), 29 deletions(-)


Changeset truncated at 500 lines:

diff --git a/source3/include/proto.h b/source3/include/proto.h
index 084d97f..0f02dfb 100644
--- a/source3/include/proto.h
+++ b/source3/include/proto.h
@@ -4438,7 +4438,8 @@ NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
 NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
 				struct smb_request *smbreq,
 				files_struct *fsp,
-				char *inbuf,
+				TALLOC_CTX *ctx,
+				DATA_BLOB *preadbuf,
 				SMB_OFF_T startpos,
 				size_t smb_maxcnt);
 NTSTATUS schedule_aio_smb2_write(connection_struct *conn,
@@ -5271,6 +5272,11 @@ NTSTATUS unlink_internals(connection_struct *conn, struct smb_request *req,
 			  uint32 dirtype, struct smb_filename *smb_fname,
 			  bool has_wild);
 void reply_unlink(struct smb_request *req);
+ssize_t fake_sendfile(files_struct *fsp, SMB_OFF_T startpos, size_t nread);
+void sendfile_short_send(files_struct *fsp,
+				ssize_t nread,
+				size_t headersize,
+				size_t smb_maxcnt);
 void reply_readbraw(struct smb_request *req);
 void reply_lockread(struct smb_request *req);
 void reply_read(struct smb_request *req);
diff --git a/source3/lib/sendfile.c b/source3/lib/sendfile.c
index 3003246..c2099bc 100644
--- a/source3/lib/sendfile.c
+++ b/source3/lib/sendfile.c
@@ -63,7 +63,11 @@ ssize_t sys_sendfile(int tofd, int fromfd, const DATA_BLOB *header, SMB_OFF_T of
 #else
 			nwritten = sendfile(tofd, fromfd, &offset, total);
 #endif
-		} while (nwritten == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+		} while (nwritten == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+		} while (nwritten == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 		if (nwritten == -1) {
 			if (errno == ENOSYS || errno == EINVAL) {
 				/* Ok - we're in a world of pain here. We just sent
@@ -145,7 +149,11 @@ ssize_t sys_sendfile(int tofd, int fromfd, const DATA_BLOB *header, SMB_OFF_T of
 		int32 nwritten;
 		do {
 			nwritten = sendfile(tofd, fromfd, &small_offset, small_total);
-		} while (nwritten == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+		} while (nwritten == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+		} while (nwritten == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 		if (nwritten == -1) {
 			if (errno == ENOSYS || errno == EINVAL) {
 				/* Ok - we're in a world of pain here. We just sent
@@ -226,7 +234,11 @@ ssize_t sys_sendfile(int tofd, int fromfd, const DATA_BLOB *header, SMB_OFF_T of
 #else
 			nwritten = sendfilev(tofd, vec, sfvcnt, &xferred);
 #endif
-		if (nwritten == -1 && errno == EINTR) {
+#if defined(EWOULDBLOCK)
+		if  (nwritten == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK)) {
+#else
+		if (nwritten == -1 && (errno == EINTR || errno == EAGAIN)) {
+#endif
 			if (xferred == 0)
 				continue; /* Nothing written yet. */
 			else
@@ -300,7 +312,11 @@ ssize_t sys_sendfile(int tofd, int fromfd, const DATA_BLOB *header, SMB_OFF_T of
 #else
 			nwritten = sendfile(tofd, fromfd, offset, total, &hdtrl[0], 0);
 #endif
-		} while (nwritten == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+		} while (nwritten == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+		} while (nwritten == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 		if (nwritten == -1)
 			return -1;
 		if (nwritten == 0)
@@ -371,7 +387,11 @@ ssize_t sys_sendfile(int tofd, int fromfd, const DATA_BLOB *header, SMB_OFF_T of
 
 		do {
 			ret = sendfile(fromfd, tofd, offset, total, &hdr, &nwritten, 0);
-		} while (ret == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+		} while (ret == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+		} while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 		if (ret == -1)
 			return -1;
 
@@ -449,7 +469,11 @@ ssize_t sys_sendfile(int tofd, int fromfd, const DATA_BLOB *header, SMB_OFF_T of
 		*/
 		do {
 			ret = send_file(&tofd, &hdtrl, 0);
-		} while ( (ret == 1) || (ret == -1 && errno == EINTR) );
+#if defined(EWOULDBLOCK)
+		} while ((ret == 1) || (ret == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK)));
+#else
+		} while ((ret == 1) || (ret == -1 && (errno == EINTR || errno == EAGAIN)));
+#endif
 		if ( ret == -1 )
 			return -1;
 	}
diff --git a/source3/lib/system.c b/source3/lib/system.c
index ee8efe8..1c00ad8 100644
--- a/source3/lib/system.c
+++ b/source3/lib/system.c
@@ -123,7 +123,11 @@ ssize_t sys_read(int fd, void *buf, size_t count)
 
 	do {
 		ret = read(fd, buf, count);
-	} while (ret == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 	return ret;
 }
 
@@ -137,7 +141,11 @@ ssize_t sys_write(int fd, const void *buf, size_t count)
 
 	do {
 		ret = write(fd, buf, count);
-	} while (ret == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 	return ret;
 }
 
@@ -162,7 +170,11 @@ ssize_t sys_writev(int fd, const struct iovec *iov, int iovcnt)
 
 	do {
 		ret = writev(fd, iov, iovcnt);
-	} while (ret == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 	return ret;
 }
 
@@ -207,7 +219,7 @@ ssize_t sys_pwrite(int fd, const void *buf, size_t count, SMB_OFF_T off)
 #endif
 
 /*******************************************************************
-A send wrapper that will deal with EINTR.
+A send wrapper that will deal with EINTR or EAGAIN or EWOULDBLOCK.
 ********************************************************************/
 
 ssize_t sys_send(int s, const void *msg, size_t len, int flags)
@@ -216,12 +228,16 @@ ssize_t sys_send(int s, const void *msg, size_t len, int flags)
 
 	do {
 		ret = send(s, msg, len, flags);
-	} while (ret == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 	return ret;
 }
 
 /*******************************************************************
-A sendto wrapper that will deal with EINTR.
+A sendto wrapper that will deal with EINTR or EAGAIN or EWOULDBLOCK.
 ********************************************************************/
 
 ssize_t sys_sendto(int s,  const void *msg, size_t len, int flags, const struct sockaddr *to, socklen_t tolen)
@@ -230,12 +246,16 @@ ssize_t sys_sendto(int s,  const void *msg, size_t len, int flags, const struct
 
 	do {
 		ret = sendto(s, msg, len, flags, to, tolen);
-	} while (ret == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 	return ret;
 }
 
 /*******************************************************************
-A recv wrapper that will deal with EINTR.
+A recv wrapper that will deal with EINTR or EAGAIN or EWOULDBLOCK.
 ********************************************************************/
 
 ssize_t sys_recv(int fd, void *buf, size_t count, int flags)
@@ -244,7 +264,11 @@ ssize_t sys_recv(int fd, void *buf, size_t count, int flags)
 
 	do {
 		ret = recv(fd, buf, count, flags);
-	} while (ret == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 	return ret;
 }
 
@@ -258,7 +282,11 @@ ssize_t sys_recvfrom(int s, void *buf, size_t len, int flags, struct sockaddr *f
 
 	do {
 		ret = recvfrom(s, buf, len, flags, from, fromlen);
-	} while (ret == -1 && errno == EINTR);
+#if defined(EWOULDBLOCK)
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN || errno == EWOULDBLOCK));
+#else
+	} while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+#endif
 	return ret;
 }
 
diff --git a/source3/smbd/aio.c b/source3/smbd/aio.c
index 7a23d37..eb8ed67 100644
--- a/source3/smbd/aio.c
+++ b/source3/smbd/aio.c
@@ -385,7 +385,8 @@ NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
 NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
 				struct smb_request *smbreq,
 				files_struct *fsp,
-				char *inbuf,
+				TALLOC_CTX *ctx,
+				DATA_BLOB *preadbuf,
 				SMB_OFF_T startpos,
 				size_t smb_maxcnt)
 {
@@ -427,6 +428,12 @@ NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
 		return NT_STATUS_RETRY;
 	}
 
+	/* Create the out buffer. */
+	*preadbuf = data_blob_talloc(ctx, NULL, smb_maxcnt);
+	if (preadbuf->data == NULL) {
+		return NT_STATUS_NO_MEMORY;
+	}
+
 	if (!(aio_ex = create_aio_extra(smbreq->smb2req, fsp, 0))) {
 		return NT_STATUS_NO_MEMORY;
 	}
@@ -447,7 +454,7 @@ NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
 	/* Now set up the aio record for the read call. */
 
 	a->aio_fildes = fsp->fh->fd;
-	a->aio_buf = inbuf;
+	a->aio_buf = preadbuf->data;
 	a->aio_nbytes = smb_maxcnt;
 	a->aio_offset = startpos;
 	a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
@@ -1031,7 +1038,8 @@ NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
 NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
                                 struct smb_request *smbreq,
                                 files_struct *fsp,
-                                char *inbuf,
+				TALLOC_CTX *ctx,
+				DATA_BLOB *preadbuf,
                                 SMB_OFF_T startpos,
                                 size_t smb_maxcnt)
 {
diff --git a/source3/smbd/reply.c b/source3/smbd/reply.c
index 9601f53..26badc4 100644
--- a/source3/smbd/reply.c
+++ b/source3/smbd/reply.c
@@ -2804,8 +2804,7 @@ static void fail_readraw(void)
  Fake (read/write) sendfile. Returns -1 on read or write fail.
 ****************************************************************************/
 
-static ssize_t fake_sendfile(files_struct *fsp, SMB_OFF_T startpos,
-			     size_t nread)
+ssize_t fake_sendfile(files_struct *fsp, SMB_OFF_T startpos, size_t nread)
 {
 	size_t bufsize;
 	size_t tosend = nread;
@@ -2869,7 +2868,7 @@ static ssize_t fake_sendfile(files_struct *fsp, SMB_OFF_T startpos,
  requested. Fill with zeros (all we can do).
 ****************************************************************************/
 
-static void sendfile_short_send(files_struct *fsp,
+void sendfile_short_send(files_struct *fsp,
 				ssize_t nread,
 				size_t headersize,
 				size_t smb_maxcnt)
diff --git a/source3/smbd/smb2_read.c b/source3/smbd/smb2_read.c
index b1866d1..d5f6896 100644
--- a/source3/smbd/smb2_read.c
+++ b/source3/smbd/smb2_read.c
@@ -172,6 +172,7 @@ static void smbd_smb2_request_read_done(struct tevent_req *subreq)
 struct smbd_smb2_read_state {
 	struct smbd_smb2_request *smb2req;
 	files_struct *fsp;
+	uint64_t in_file_id_volatile;
 	uint32_t in_length;
 	uint64_t in_offset;
 	uint32_t in_minimum;
@@ -179,6 +180,139 @@ struct smbd_smb2_read_state {
 	uint32_t out_remaining;
 };
 
+/* struct smbd_smb2_read_state destructor. Send the SMB2_READ data. */
+static int smb2_sendfile_send_data(struct smbd_smb2_read_state *state)
+{
+	struct lock_struct lock;
+	uint32_t in_length = state->in_length;
+	uint64_t in_offset = state->in_offset;
+	files_struct *fsp = state->fsp;
+	ssize_t nread;
+
+	nread = SMB_VFS_SENDFILE(fsp->conn->sconn->sock,
+					fsp,
+					NULL,
+					in_offset,
+					in_length);
+	DEBUG(10,("smb2_sendfile_send_data: SMB_VFS_SENDFILE returned %d on file %s\n",
+		(int)nread,
+		fsp_str_dbg(fsp) ));
+
+	if (nread == -1) {
+		if (errno == ENOSYS || errno == EINTR) {
+			/*
+			 * Special hack for broken systems with no working
+			 * sendfile. Fake this up by doing read/write calls.
+			*/
+			set_use_sendfile(SNUM(fsp->conn), false);
+			nread = fake_sendfile(fsp, in_offset, in_length);
+			if (nread == -1) {
+				DEBUG(0,("smb2_sendfile_send_data: "
+					"fake_sendfile failed for "
+					"file %s (%s).\n",
+					fsp_str_dbg(fsp),
+					strerror(errno)));
+				exit_server_cleanly("smb2_sendfile_send_data: "
+					"fake_sendfile failed");
+			}
+			goto out;
+		}
+
+		DEBUG(0,("smb2_sendfile_send_data: sendfile failed for file "
+			"%s (%s). Terminating\n",
+			fsp_str_dbg(fsp),
+			strerror(errno)));
+		exit_server_cleanly("smb2_sendfile_send_data: sendfile failed");
+	} else if (nread == 0) {
+		/*
+		 * Some sendfile implementations return 0 to indicate
+		 * that there was a short read, but nothing was
+		 * actually written to the socket.  In this case,
+		 * fallback to the normal read path so the header gets
+		 * the correct byte count.
+		 */
+		DEBUG(3, ("send_file_readX: sendfile sent zero bytes "
+			"falling back to the normal read: %s\n",
+			fsp_str_dbg(fsp)));
+
+		nread = fake_sendfile(fsp, in_offset, in_length);
+		if (nread == -1) {
+			DEBUG(0,("smb2_sendfile_send_data: "
+				"fake_sendfile failed for file "
+				"%s (%s). Terminating\n",
+				fsp_str_dbg(fsp),
+				strerror(errno)));
+			exit_server_cleanly("smb2_sendfile_send_data: "
+				"fake_sendfile failed");
+		}
+	}
+
+  out:
+
+	if (nread < in_length) {
+		sendfile_short_send(fsp, nread, 0, in_length);
+	}
+
+	init_strict_lock_struct(fsp,
+				state->in_file_id_volatile,
+				in_offset,
+				in_length,
+				READ_LOCK,
+				&lock);
+
+	SMB_VFS_STRICT_UNLOCK(fsp->conn, fsp, &lock);
+	return 0;
+}
+
+static NTSTATUS schedule_smb2_sendfile_read(struct smbd_smb2_request *smb2req,
+					struct smbd_smb2_read_state *state)
+{
+	struct smbd_smb2_read_state *state_copy = NULL;
+	files_struct *fsp = state->fsp;
+
+	/*
+	 * We cannot use sendfile if...
+	 * We were not configured to do so OR
+	 * Signing is active OR
+	 * This is a compound SMB2 operation OR
+	 * fsp is a STREAM file OR
+	 * We're using a write cache OR
+	 * It's not a regular file OR
+	 * Requested offset is greater than file size OR
+	 * there's not enough data in the file.
+	 * Phew :-). Luckily this means most
+	 * reads on most normal files. JRA.
+	*/
+
+	if (!_lp_use_sendfile(SNUM(fsp->conn)) ||
+			smb2req->do_signing ||
+			smb2req->in.vector_count != 4 ||
+			(fsp->base_fsp != NULL) ||
+			(fsp->wcp != NULL) ||
+			(!S_ISREG(fsp->fsp_name->st.st_ex_mode)) ||
+			(state->in_offset >= fsp->fsp_name->st.st_ex_size) ||
+			(fsp->fsp_name->st.st_ex_size < state->in_offset +
+				state->in_length)) {
+		return NT_STATUS_RETRY;
+	}
+
+	/* We've already checked there's this amount of data
+	   to read. */
+	state->out_data.length = state->in_length;
+	state->out_remaining = 0;
+
+	/* Make a copy of state attached to the smb2req. Attach
+	   the destructor here as this will trigger the sendfile
+	   call when the request is destroyed. */
+	state_copy = TALLOC_P(smb2req, struct smbd_smb2_read_state);
+	if (!state_copy) {
+		return NT_STATUS_NO_MEMORY;
+	}
+	*state_copy = *state;
+	talloc_set_destructor(state_copy, smb2_sendfile_send_data);
+	return NT_STATUS_OK;
+}
+
 static void smbd_smb2_read_pipe_done(struct tevent_req *subreq);
 
 /*******************************************************************
@@ -290,16 +424,17 @@ static struct tevent_req *smbd_smb2_read_send(TALLOC_CTX *mem_ctx,
 		return tevent_req_post(req, ev);
 	}
 
-	state->out_data = data_blob_talloc(state, NULL, in_length);
-	if (in_length > 0 && tevent_req_nomem(state->out_data.data, req)) {
-		return tevent_req_post(req, ev);
-	}
-
 	state->fsp = fsp;
+	state->in_file_id_volatile = in_file_id_volatile;
 
 	if (IS_IPC(smbreq->conn)) {
 		struct tevent_req *subreq = NULL;
 
+		state->out_data = data_blob_talloc(state, NULL, in_length);
+		if (in_length > 0 && tevent_req_nomem(state->out_data.data, req)) {
+			return tevent_req_post(req, ev);
+		}
+
 		if (!fsp_is_np(fsp)) {
 			tevent_req_nterror(req, NT_STATUS_FILE_CLOSED);
 			return tevent_req_post(req, ev);
@@ -326,7 +461,8 @@ static struct tevent_req *smbd_smb2_read_send(TALLOC_CTX *mem_ctx,
 	status = schedule_smb2_aio_read(fsp->conn,
 				smbreq,
 				fsp,
-				(char *)state->out_data.data,
+				state,
+				&state->out_data,
 				(SMB_OFF_T)in_offset,
 				(size_t)in_length);
 
@@ -363,6 +499,26 @@ static struct tevent_req *smbd_smb2_read_send(TALLOC_CTX *mem_ctx,
 		return tevent_req_post(req, ev);
 	}
 
+	/* Try sendfile in preference. */
+	status = schedule_smb2_sendfile_read(smb2req, state);
+	if (NT_STATUS_IS_OK(status)) {
+		tevent_req_done(req);
+		return tevent_req_post(req, ev);
+	} else {
+		if (!NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
+			SMB_VFS_STRICT_UNLOCK(conn, fsp, &lock);
+			tevent_req_nterror(req, status);
+			return tevent_req_post(req, ev);
+		}
+	}
+
+	/* Ok, read into memory. Allocate the out buffer. */
+	state->out_data = data_blob_talloc(state, NULL, in_length);
+	if (in_length > 0 && tevent_req_nomem(state->out_data.data, req)) {
+		SMB_VFS_STRICT_UNLOCK(conn, fsp, &lock);
+		return tevent_req_post(req, ev);


-- 
Samba Shared Repository


More information about the samba-cvs mailing list