iv/linux
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[CIFS] Add new nostrictsync cifs mount option to avoid slow SMB flush

Browse Source 
If this mount option is set, when an application does an
fsync call then the cifs client does not send an SMB Flush
to the server (to force the server to write all dirty data
for this file immediately to disk), although cifs still sends
all dirty (cached) file data to the server and waits for the
server to respond to the write write.  Since SMB Flush can be
very slow, and some servers may be reliable enough (to risk
delaying slightly flushing the data to disk on the server),
turning on this option may be useful to improve performance for
applications that fsync too much, at a small risk of server
crash.  If this mount option is not set, by default cifs will
send an SMB flush request (and wait for a response) on every
fsync call.

Signed-off-by: Steve French <sfrench@us.ibm.com>
This commit is contained in:
Steve French 2009-02-23 15:21:59 +00:00
parent 10e70afa75
commit be652445fd
5 changed files with 31 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
fs/cifs/CHANGES
View File

@ -8,7 +8,9 @@ user's smb session. This fix allows cifs to mount multiple times to the
same server with different userids without risking invalidating earlier
established security contexts. fsync now sends SMB Flush operation
to better ensure that we wait for server to write all of the data to
server disk (not just write it over the network).
server disk (not just write it over the network). Add new mount
parameter to allow user to disable sending the (slow) SMB flush on
fsync if desired (fsync still flushes all cached write data to the server).
Version 1.56
------------

22
fs/cifs/README
View File

@ -472,6 +472,19 @@ A partial list of the supported mount options follows:
even if the cifs server would support posix advisory locks.
"forcemand" is accepted as a shorter form of this mount
option.
nostrictsync If this mount option is set, when an application does an
fsync call then the cifs client does not send an SMB Flush
to the server (to force the server to write all dirty data
for this file immediately to disk), although cifs still sends
all dirty (cached) file data to the server and waits for the
server to respond to the write. Since SMB Flush can be
very slow, and some servers may be reliable enough (to risk
delaying slightly flushing the data to disk on the server),
turning on this option may be useful to improve performance for
applications that fsync too much, at a small risk of server
crash. If this mount option is not set, by default cifs will
send an SMB flush request (and wait for a response) on every
fsync call.
nodfs Disable DFS (global name space support) even if the
server claims to support it. This can help work around
a problem with parsing of DFS paths with Samba server
@ -692,13 +705,14 @@ require this helper. Note that NTLMv2 security (which does not require the
cifs.upcall helper program), instead of using Kerberos, is sufficient for
some use cases.
Enabling DFS support (used to access shares transparently in an MS-DFS
global name space) requires that CONFIG_CIFS_EXPERIMENTAL be enabled. In
addition, DFS support for target shares which are specified as UNC
DFS support allows transparent redirection to shares in an MS-DFS name space.
In addition, DFS support for target shares which are specified as UNC
names which begin with host names (rather than IP addresses) requires
a user space helper (such as cifs.upcall) to be present in order to
translate host names to ip address, and the user space helper must also
be configured in the file /etc/request-key.conf
be configured in the file /etc/request-key.conf. Samba, Windows servers and
many NAS appliances support DFS as a way of constructing a global name
space to ease network configuration and improve reliability.
To use cifs Kerberos and DFS support, the Linux keyutils package should be
installed and something like the following lines should be added to the

1
fs/cifs/cifs_fs_sb.h
View File

@ -32,6 +32,7 @@
#define CIFS_MOUNT_OVERR_GID 0x800 /* override gid returned from server */
#define CIFS_MOUNT_DYNPERM 0x1000 /* allow in-memory only mode setting */
#define CIFS_MOUNT_NOPOSIXBRL 0x2000 /* mandatory not posix byte range lock */
#define CIFS_MOUNT_NO_SSYNC 0x4000 /* don't do slow SMBflush on every sync*/
struct cifs_sb_info {
struct cifsTconInfo *tcon; /* primary mount */

7
fs/cifs/connect.c
View File

@ -95,6 +95,7 @@ struct smb_vol {
bool local_lease:1; /* check leases only on local system, not remote */
bool noblocksnd:1;
bool noautotune:1;
bool nostrictsync:1; /* do not force expensive SMBflush on every sync */
unsigned int rsize;
unsigned int wsize;
unsigned int sockopt;
@ -1274,6 +1275,10 @@ cifs_parse_mount_options(char *options, const char *devname,
vol->intr = 0;
} else if (strnicmp(data, "intr", 4) == 0) {
vol->intr = 1;
} else if (strnicmp(data, "nostrictsync", 12) == 0) {
vol->nostrictsync = 1;
} else if (strnicmp(data, "strictsync", 10) == 0) {
vol->nostrictsync = 0;
} else if (strnicmp(data, "serverino", 7) == 0) {
vol->server_ino = 1;
} else if (strnicmp(data, "noserverino", 9) == 0) {
@ -2160,6 +2165,8 @@ static void setup_cifs_sb(struct smb_vol *pvolume_info,
cifs_sb->mnt_cifs_flags |= CIFS_MOUNT_UNX_EMUL;
if (pvolume_info->nobrl)
cifs_sb->mnt_cifs_flags |= CIFS_MOUNT_NO_BRL;
if (pvolume_info->nostrictsync)
cifs_sb->mnt_cifs_flags |= CIFS_MOUNT_NO_SSYNC;
if (pvolume_info->mand_lock)
cifs_sb->mnt_cifs_flags |= CIFS_MOUNT_NOPOSIXBRL;
if (pvolume_info->cifs_acl)

3
fs/cifs/file.c
View File

@ -1538,7 +1538,8 @@ int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
rc = CIFS_I(inode)->write_behind_rc;
CIFS_I(inode)->write_behind_rc = 0;
tcon = CIFS_SB(inode->i_sb)->tcon;
if (!rc && tcon && smbfile)
if (!rc && tcon && smbfile &&
!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_SSYNC))
rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
}