2.6.0-test10 Invalidate cached inode attributes after rename]

I initially mailed this to the NFS mailing list on sourceforge.net but it doesn't seem to have made it through.

------------------------------------------------------------------------------
This message is intended only for the personal and confidential use of the designated recipient(s) named above. If you are not the intended recipient of this message you are hereby notified that any review, dissemination, distribution or copying of this message is strictly prohibited. This communication is for information purposes only and should not be regarded as an offer to sell or as a solicitation of an offer to buy any financial product, an official confirmation of any transaction, or as an official statement of Lehman Brothers. Email transmission cannot be guaranteed to be secure or error-free. Therefore, we do not represent that this information is complete or accurate and it should not be relied upon as such. All information is subject to change without notice.

Hi Trond et al,

The following one line patch invalidates the attributes of the underlying inode when a file is renamed. Some filesystems update ctime upon rename(). One such filesystem is ext3, and a comment in the relevant code there indicates this is true with other Unix filesystems as well. The following problem was observed on Fedora Core 1 running a stock kernel.org 2.6.0-test10.
The server is running Solaris 2.8 but I have verified the same issue exists with a server running Linix 2.4.22. These operations a done via a Perl script.

1. Check out a CVS repository into an NFS mounted directory.
2. Move files from CVS working directory into another directory in the same filesystem.
3. Tar up the resultant directory.
4. Tar prints lots of "file changed after we read it" messages.

Tar obtains ctime via stat() before reading the file and compares it to the ctime obtained via fstat() after having read the file. The two differ because the intervening open() forces an attribute refresh due to CTO consistency at which time ctime is updated. Forcing a cache invalidation during rename() eliminates this particular scenario.

Thanks,
Shantanu

@@ -1257,6 +1257,7 @@
 
    nfs_zap_caches(new_dir);
    nfs_zap_caches(old_dir);
+    NFS_CACHEINV(old_inode);
    error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
                       new_dir, &new_dentry->d_name);

out:

> I initially mailed this to the NFS mailing list on > sourceforge.net but it doesn't seem to have made it through.

It's not a critical bugfix. It will make it in, but not while the current code-freeze is in place.

Cheers,
Trond

BTW, this also affects 2.4 kernels (e.g. RedHat's AS3 kernel) which include your CTO patches that aren't in the upstream kernels yet.

Thanks,
Shantanu

------------------------------------------------------------------------------
This message is intended only for the personal and confidential use of the designated recipient(s) named above. If you are not the intended recipient of this message you are hereby notified that any review, dissemination, distribution or copying of this message is strictly prohibited. This communication is for information purposes only and should not be regarded as an offer to sell or as a solicitation of an offer to buy any financial product, an official confirmation of any transaction, or as an official statement of Lehman Brothers. Email transmission cannot be guaranteed to be secure or error-free. Therefore, we do not represent that this information is complete or accurate and it should not be relied upon as such. All information is subject to change without notice.

> BTW, this also affects 2.4 kernels (e.g. RedHat's AS3 kernel) > which include your CTO patches that aren't in the upstream > kernels yet.

AFAIK Steve Dickson is already aware of the need for this particular patch for the AS3 kernel. He was one of the people that brought it to my attention a couple of weeks ago.

Cheers,
Trond