cifs: Reduce number of socket retries in large write path

Backport of upstream commit edf1ae403896cb7750800508b14996ba6be39a53
for -stable.

[CIFS] Reduce number of socket retries in large write path

CIFS in some heavy stress conditions cifs could get EAGAIN
repeatedly in smb_send2 which led to repeated retries and eventually
failure of large writes which could lead to data corruption.

There are three changes that were suggested by various network
developers:

1) convert cifs from non-blocking to blocking tcp sendmsg
(we left in the retry on failure)
2) change cifs to not set sendbuf and rcvbuf size for the socket
(let tcp autotune the buffer sizes since that works much better
in the TCP stack now)
3) if we have a partial frame sent in smb_send2, mark the tcp
session as invalid (close the socket and reconnect) so we do
not corrupt the remaining part of the SMB with the beginning
of the next SMB.

This does not appear to hurt performance measurably and has
been run in various scenarios, but it definately removes
a corruption that we were seeing in some high stress
test cases.

Acked-by: Shirish Pargaonkar <shirishp@us.ibm.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

1 files changed, 43 insertions, 15 deletions

diff --git a/fs/cifs/connect.c b/fs/cifs/connect.c
index 4c13bcdb92a..63d3cbc8605 100644
--- a/fs/cifs/connect.c
+++ b/fs/cifs/connect.c
@@ -90,6 +90,8 @@ struct smb_vol {
 	bool nocase:1;     /* request case insensitive filenames */
 	bool nobrl:1;      /* disable sending byte range locks to srv */
 	bool seal:1;       /* request transport encryption on share */
+	bool noblocksnd:1;
+	bool noautotune:1;
 	unsigned int rsize;
 	unsigned int wsize;
 	unsigned int sockopt;
@@ -100,9 +102,11 @@ struct smb_vol {
 static int ipv4_connect(struct sockaddr_in *psin_server,
 			struct socket **csocket,
 			char *netb_name,
-			char *server_netb_name);
+			char *server_netb_name,
+			bool noblocksnd,
+			bool nosndbuf); /* ipv6 never set sndbuf size */
 static int ipv6_connect(struct sockaddr_in6 *psin_server,
-			struct socket **csocket);
+			struct socket **csocket, bool noblocksnd);
 
 
 	/*
@@ -188,12 +192,13 @@ cifs_reconnect(struct TCP_Server_Info *server)
 		try_to_freeze();
 		if (server->protocolType == IPV6) {
 			rc = ipv6_connect(&server->addr.sockAddr6,
-					  &server->ssocket);
+					  &server->ssocket, server->noautotune);
 		} else {
 			rc = ipv4_connect(&server->addr.sockAddr,
 					&server->ssocket,
 					server->workstation_RFC1001_name,
-					server->server_RFC1001_name);
+					server->server_RFC1001_name,
+					server->noblocksnd, server->noautotune);
 		}
 		if (rc) {
 			cFYI(1, ("reconnect error %d", rc));
@@ -409,8 +414,14 @@ incomplete_rcv:
 			msleep(1); /* minimum sleep to prevent looping
 				allowing socket to clear and app threads to set
 				tcpStatus CifsNeedReconnect if server hung */
-			if (pdu_length < 4)
+			if (pdu_length < 4) {
+				iov.iov_base = (4 - pdu_length) +
+						(char *)smb_buffer;
+				iov.iov_len = pdu_length;
+				smb_msg.msg_control = NULL;
+				smb_msg.msg_controllen = 0;
 				goto incomplete_rcv;
+			}
 			else
 				continue;
 		} else if (length <= 0) {
@@ -1186,6 +1197,10 @@ cifs_parse_mount_options(char *options, const char *devname,
 			/* ignore */
 		} else if (strnicmp(data, "rw", 2) == 0) {
 			vol->rw = true;
+		} else if (strnicmp(data, "noblocksnd", 11) == 0) {
+			vol->noblocksnd = true;
+		} else if (strnicmp(data, "noautotune", 10) == 0) {
+			vol->noautotune = true;
 		} else if ((strnicmp(data, "suid", 4) == 0) ||
 				   (strnicmp(data, "nosuid", 6) == 0) ||
 				   (strnicmp(data, "exec", 4) == 0) ||
@@ -1506,7 +1521,8 @@ static void rfc1002mangle(char *target, char *source, unsigned int length)
 
 static int
 ipv4_connect(struct sockaddr_in *psin_server, struct socket **csocket,
-	     char *netbios_name, char *target_name)
+		char *netbios_name, char *target_name,
+		bool noblocksnd, bool noautotune)
 {
 	int rc = 0;
 	int connected = 0;
@@ -1578,11 +1594,15 @@ ipv4_connect(struct sockaddr_in *psin_server, struct socket **csocket,
 		 (*csocket)->sk->sk_sndbuf,
 		 (*csocket)->sk->sk_rcvbuf, (*csocket)->sk->sk_rcvtimeo));
 	(*csocket)->sk->sk_rcvtimeo = 7 * HZ;
+	if (!noblocksnd)
+		(*csocket)->sk->sk_sndtimeo = 3 * HZ;
 	/* make the bufsizes depend on wsize/rsize and max requests */
-	if ((*csocket)->sk->sk_sndbuf < (200 * 1024))
-		(*csocket)->sk->sk_sndbuf = 200 * 1024;
-	if ((*csocket)->sk->sk_rcvbuf < (140 * 1024))
-		(*csocket)->sk->sk_rcvbuf = 140 * 1024;
+	if (noautotune) {
+		if ((*csocket)->sk->sk_sndbuf < (200 * 1024))
+			(*csocket)->sk->sk_sndbuf = 200 * 1024;
+		if ((*csocket)->sk->sk_rcvbuf < (140 * 1024))
+			(*csocket)->sk->sk_rcvbuf = 140 * 1024;
+	}
 
 	/* send RFC1001 sessinit */
 	if (psin_server->sin_port == htons(RFC1001_PORT)) {
@@ -1619,7 +1639,7 @@ ipv4_connect(struct sockaddr_in *psin_server, struct socket **csocket,
 			/* sizeof RFC1002_SESSION_REQUEST with no scope */
 			smb_buf->smb_buf_length = 0x81000044;
 			rc = smb_send(*csocket, smb_buf, 0x44,
-				(struct sockaddr *)psin_server);
+				(struct sockaddr *)psin_server, noblocksnd);
 			kfree(ses_init_buf);
 			msleep(1); /* RFC1001 layer in at least one server
 				      requires very short break before negprot
@@ -1639,7 +1659,8 @@ ipv4_connect(struct sockaddr_in *psin_server, struct socket **csocket,
 }
 
 static int
-ipv6_connect(struct sockaddr_in6 *psin_server, struct socket **csocket)
+ipv6_connect(struct sockaddr_in6 *psin_server, struct socket **csocket,
+		bool noblocksnd)
 {
 	int rc = 0;
 	int connected = 0;
@@ -1708,6 +1729,8 @@ ipv6_connect(struct sockaddr_in6 *psin_server, struct socket **csocket)
 		the default. sock_setsockopt not used because it expects
 		user space buffer */
 	(*csocket)->sk->sk_rcvtimeo = 7 * HZ;
+	if (!noblocksnd)
+		(*csocket)->sk->sk_sndtimeo = 3 * HZ;
 
 	return rc;
 }
@@ -1961,11 +1984,14 @@ cifs_mount(struct super_block *sb, struct cifs_sb_info *cifs_sb,
 			cFYI(1, ("attempting ipv6 connect"));
 			/* BB should we allow ipv6 on port 139? */
 			/* other OS never observed in Wild doing 139 with v6 */
-			rc = ipv6_connect(&sin_server6, &csocket);
+			rc = ipv6_connect(&sin_server6, &csocket,
+					volume_info.noblocksnd);
 		} else
 			rc = ipv4_connect(&sin_server, &csocket,
-				  volume_info.source_rfc1001_name,
-				  volume_info.target_rfc1001_name);
+					volume_info.source_rfc1001_name,
+					volume_info.target_rfc1001_name,
+					volume_info.noblocksnd,
+					volume_info.noautotune);
 		if (rc < 0) {
 			cERROR(1, ("Error connecting to IPv4 socket. "
 				   "Aborting operation"));
@@ -1980,6 +2006,8 @@ cifs_mount(struct super_block *sb, struct cifs_sb_info *cifs_sb,
 			sock_release(csocket);
 			goto out;
 		} else {
+			srvTcp->noblocksnd = volume_info.noblocksnd;
+			srvTcp->noautotune = volume_info.noautotune;
 			memcpy(&srvTcp->addr.sockAddr, &sin_server,
 				sizeof(struct sockaddr_in));
 			atomic_set(&srvTcp->inFlight, 0);