Fix i_mutex vs. readdir handling in nfsd
commit2f9092e1020246168b1309b35e085ecd7ff9ff72
authorDavid Woodhouse <dwmw2@infradead.org>
Mon, 20 Apr 2009 22:18:37 +0000 (20 23:18 +0100)
committerAl Viro <viro@zeniv.linux.org.uk>
Tue, 21 Apr 2009 03:01:16 +0000 (20 23:01 -0400)
treef8318c1e62e789718ae7637869f6c075b815bcb2
parent1ba0c7dbbbc24230394100c5f0d0df38cb400cff
Fix i_mutex vs. readdir handling in nfsd

Commit 14f7dd63 ("Copy XFS readdir hack into nfsd code") introduced a
bug to generic code which had been extant for a long time in the XFS
version -- it started to call through into lookup_one_len() and hence
into the file systems' ->lookup() methods without i_mutex held on the
directory.

This patch fixes it by locking the directory's i_mutex again before
calling the filldir functions. The original deadlocks which commit
14f7dd63 was designed to avoid are still avoided, because they were due
to fs-internal locking, not i_mutex.

While we're at it, fix the return type of nfsd_buffered_readdir() which
should be a __be32 not an int -- it's an NFS errno, not a Linux errno.
And return nfserrno(-ENOMEM) when allocation fails, not just -ENOMEM.
Sparse would have caught that, if it wasn't so busy bitching about
__cold__.

Commit 05f4f678 ("nfsd4: don't do lookup within readdir in recovery
code") introduced a similar problem with calling lookup_one_len()
without i_mutex, which this patch also addresses. To fix that, it was
necessary to fix the called functions so that they expect i_mutex to be
held; that part was done by J. Bruce Fields.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Umm-I-can-live-with-that-by: Al Viro <viro@zeniv.linux.org.uk>
Reported-by: J. R. Okajima <hooanon05@yahoo.co.jp>
Tested-by: J. Bruce Fields <bfields@citi.umich.edu>
LKML-Reference: <8036.1237474444@jrobl>
Cc: stable@kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/namei.c
fs/nfsd/nfs4recover.c
fs/nfsd/vfs.c