[CIFS] Fix byte range locking to Windows when Windows server returns
illegal RFC1001 length (which had caused the lock to block forever
until killed).
diff --git a/fs/cifs/CHANGES b/fs/cifs/CHANGES
index 5351772..1c249a2 100644
--- a/fs/cifs/CHANGES
+++ b/fs/cifs/CHANGES
@@ -1,3 +1,15 @@
+Version 1.38
+------------
+Fix tcp socket retransmission timeouts (e.g. on ENOSPACE from the socket)
+to be smaller at first (but increasing) so large write performance performance
+over GigE is better. Do not hang thread on illegal byte range lock response
+from Windows (Windows can send an RFC1001 size which does not match smb size) by
+allowing an SMBs TCP length to be up to a few bytes longer than it should be.
+wsize and rsize can now be larger than negotiated buffer size if server
+supports large readx/writex, even when directio mount flag not specified.
+Write size will in many cases now be 16K instead of 4K which greatly helps
+file copy performance on lightly loaded networks.
+
Version 1.37
------------
Fix readdir caching when unlink removes file in current search buffer,
diff --git a/fs/cifs/README b/fs/cifs/README
index 3b610d0..5d9a953 100644
--- a/fs/cifs/README
+++ b/fs/cifs/README
@@ -294,8 +294,8 @@
during the local client kernel build will be used.
If server does not support Unicode, this parameter is
unused.
- rsize default read size
- wsize default write size
+ rsize default read size (usually 16K)
+ wsize default write size (usually 16K, 32K is often better over GigE)
rw mount the network share read-write (note that the
server may still consider the share read-only)
ro mount network share read-only
diff --git a/fs/cifs/TODO b/fs/cifs/TODO
index 0593f64..c909298 100644
--- a/fs/cifs/TODO
+++ b/fs/cifs/TODO
@@ -1,4 +1,4 @@
-version 1.36 September 6, 2005
+version 1.37 October 9, 2005
A Partial List of Missing Features
==================================
@@ -38,8 +38,8 @@
i) quota support (needs minor kernel change since quota calls
to make it to network filesystems or deviceless filesystems)
-j) finish writepages support (multi-page write behind for improved
-performance) and syncpage. Started by Shaggy.
+j) investigate sync behavior (including syncpage) and check
+for proper behavior of intr/nointr
k) hook lower into the sockets api (as NFS/SunRPC does) to avoid the
extra copy in/out of the socket buffers in some cases.
diff --git a/fs/cifs/misc.c b/fs/cifs/misc.c
index 8a0edd6..eba1de9 100644
--- a/fs/cifs/misc.c
+++ b/fs/cifs/misc.c
@@ -420,6 +420,7 @@
checkSMB(struct smb_hdr *smb, __u16 mid, int length)
{
__u32 len = smb->smb_buf_length;
+ __u32 clc_len; /* calculated length */
cFYI(0,
("Entering checkSMB with Length: %x, smb_buf_length: %x ",
length, len));
@@ -440,20 +441,27 @@
cERROR(1,
("smb_buf_length greater than MaxBufSize"));
cERROR(1,
- ("bad smb detected. Illegal length. The mid=%d",
+ ("bad smb detected. Illegal length. mid=%d",
smb->Mid));
return 1;
}
if (checkSMBhdr(smb, mid))
return 1;
-
- if ((4 + len != smbCalcSize_LE(smb))
+ clc_len = smbCalcSize_LE(smb);
+ if ((4 + len != clc_len)
|| (4 + len != (unsigned int)length)) {
- cERROR(1, ("smbCalcSize %x ", smbCalcSize_LE(smb)));
- cERROR(1,
- ("bad smb size detected. The Mid=%d", smb->Mid));
- return 1;
+ cERROR(1, ("Calculated size 0x%x vs actual length 0x%x",
+ clc_len, 4 + len));
+ cERROR(1, ("bad smb size detected for Mid=%d", smb->Mid));
+ /* Windows XP can return a few bytes too much, presumably
+ an illegal pad, at the end of byte range lock responses
+ so we allow for up to eight byte pad, as long as actual
+ received length is as long or longer than calculated length */
+ if((4+len > clc_len) && (len <= clc_len + 3))
+ return 0;
+ else
+ return 1;
}
return 0;
}