summaryrefslogtreecommitdiffstats
path: root/fs/cifs/transport.c
diff options
context:
space:
mode:
authorSteve French <sfrench@us.ibm.com>2008-10-29 00:47:57 +0000
committerSteve French <sfrench@us.ibm.com>2008-10-29 00:47:57 +0000
commitedf1ae403896cb7750800508b14996ba6be39a53 (patch)
treeff792ea77e558d473a9f6515397728d31e73fd09 /fs/cifs/transport.c
parent49fdf6785fd660e18a1eb4588928f47e9fa29a9a (diff)
downloadop-kernel-dev-edf1ae403896cb7750800508b14996ba6be39a53.zip
op-kernel-dev-edf1ae403896cb7750800508b14996ba6be39a53.tar.gz
[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>
Diffstat (limited to 'fs/cifs/transport.c')
-rw-r--r--fs/cifs/transport.c41
1 files changed, 31 insertions, 10 deletions
diff --git a/fs/cifs/transport.c b/fs/cifs/transport.c
index bf0e6d8..ba4d666 100644
--- a/fs/cifs/transport.c
+++ b/fs/cifs/transport.c
@@ -161,7 +161,7 @@ void DeleteTconOplockQEntries(struct cifsTconInfo *tcon)
int
smb_send(struct socket *ssocket, struct smb_hdr *smb_buffer,
- unsigned int smb_buf_length, struct sockaddr *sin)
+ unsigned int smb_buf_length, struct sockaddr *sin, bool noblocksnd)
{
int rc = 0;
int i = 0;
@@ -178,7 +178,10 @@ smb_send(struct socket *ssocket, struct smb_hdr *smb_buffer,
smb_msg.msg_namelen = sizeof(struct sockaddr);
smb_msg.msg_control = NULL;
smb_msg.msg_controllen = 0;
- smb_msg.msg_flags = MSG_DONTWAIT + MSG_NOSIGNAL; /* BB add more flags?*/
+ if (noblocksnd)
+ smb_msg.msg_flags = MSG_DONTWAIT + MSG_NOSIGNAL;
+ else
+ smb_msg.msg_flags = MSG_NOSIGNAL;
/* smb header is converted in header_assemble. bcc and rest of SMB word
area, and byte area if necessary, is converted to littleendian in
@@ -229,8 +232,8 @@ smb_send(struct socket *ssocket, struct smb_hdr *smb_buffer,
}
static int
-smb_send2(struct socket *ssocket, struct kvec *iov, int n_vec,
- struct sockaddr *sin)
+smb_send2(struct TCP_Server_Info *server, struct kvec *iov, int n_vec,
+ struct sockaddr *sin, bool noblocksnd)
{
int rc = 0;
int i = 0;
@@ -240,6 +243,7 @@ smb_send2(struct socket *ssocket, struct kvec *iov, int n_vec,
unsigned int total_len;
int first_vec = 0;
unsigned int smb_buf_length = smb_buffer->smb_buf_length;
+ struct socket *ssocket = server->ssocket;
if (ssocket == NULL)
return -ENOTSOCK; /* BB eventually add reconnect code here */
@@ -248,7 +252,10 @@ smb_send2(struct socket *ssocket, struct kvec *iov, int n_vec,
smb_msg.msg_namelen = sizeof(struct sockaddr);
smb_msg.msg_control = NULL;
smb_msg.msg_controllen = 0;
- smb_msg.msg_flags = MSG_DONTWAIT + MSG_NOSIGNAL; /* BB add more flags?*/
+ if (noblocksnd)
+ smb_msg.msg_flags = MSG_DONTWAIT + MSG_NOSIGNAL;
+ else
+ smb_msg.msg_flags = MSG_NOSIGNAL;
/* smb header is converted in header_assemble. bcc and rest of SMB word
area, and byte area if necessary, is converted to littleendian in
@@ -312,6 +319,16 @@ smb_send2(struct socket *ssocket, struct kvec *iov, int n_vec,
i = 0; /* in case we get ENOSPC on the next send */
}
+ if ((total_len > 0) && (total_len != smb_buf_length + 4)) {
+ cFYI(1, ("partial send (%d remaining), terminating session",
+ total_len));
+ /* If we have only sent part of an SMB then the next SMB
+ could be taken as the remainder of this one. We need
+ to kill the socket so the server throws away the partial
+ SMB */
+ server->tcpStatus = CifsNeedReconnect;
+ }
+
if (rc < 0) {
cERROR(1, ("Error %d sending data on socket to server", rc));
} else
@@ -518,8 +535,9 @@ SendReceive2(const unsigned int xid, struct cifsSesInfo *ses,
#ifdef CONFIG_CIFS_STATS2
atomic_inc(&ses->server->inSend);
#endif
- rc = smb_send2(ses->server->ssocket, iov, n_vec,
- (struct sockaddr *) &(ses->server->addr.sockAddr));
+ rc = smb_send2(ses->server, iov, n_vec,
+ (struct sockaddr *) &(ses->server->addr.sockAddr),
+ ses->server->noblocksnd);
#ifdef CONFIG_CIFS_STATS2
atomic_dec(&ses->server->inSend);
midQ->when_sent = jiffies;
@@ -711,7 +729,8 @@ SendReceive(const unsigned int xid, struct cifsSesInfo *ses,
atomic_inc(&ses->server->inSend);
#endif
rc = smb_send(ses->server->ssocket, in_buf, in_buf->smb_buf_length,
- (struct sockaddr *) &(ses->server->addr.sockAddr));
+ (struct sockaddr *) &(ses->server->addr.sockAddr),
+ ses->server->noblocksnd);
#ifdef CONFIG_CIFS_STATS2
atomic_dec(&ses->server->inSend);
midQ->when_sent = jiffies;
@@ -851,7 +870,8 @@ send_nt_cancel(struct cifsTconInfo *tcon, struct smb_hdr *in_buf,
return rc;
}
rc = smb_send(ses->server->ssocket, in_buf, in_buf->smb_buf_length,
- (struct sockaddr *) &(ses->server->addr.sockAddr));
+ (struct sockaddr *) &(ses->server->addr.sockAddr),
+ ses->server->noblocksnd);
up(&ses->server->tcpSem);
return rc;
}
@@ -941,7 +961,8 @@ SendReceiveBlockingLock(const unsigned int xid, struct cifsTconInfo *tcon,
atomic_inc(&ses->server->inSend);
#endif
rc = smb_send(ses->server->ssocket, in_buf, in_buf->smb_buf_length,
- (struct sockaddr *) &(ses->server->addr.sockAddr));
+ (struct sockaddr *) &(ses->server->addr.sockAddr),
+ ses->server->noblocksnd);
#ifdef CONFIG_CIFS_STATS2
atomic_dec(&ses->server->inSend);
midQ->when_sent = jiffies;
OpenPOWER on IntegriCloud