nfsd: allow fh_want_write to be called twice
authorJ. Bruce Fields <bfields@redhat.com>
Fri, 12 Apr 2019 20:37:30 +0000 (16:37 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 22 Jun 2019 06:17:15 +0000 (08:17 +0200)
commit101e808f7e75da2e0e9fb5aefc2bc3c92209eacd
treeb3fd7b48a8c752dfa06ac4fdac8f1ca978481387
parent4edf907dcd42d1d5790b54db472d08ee67c348a2
nfsd: allow fh_want_write to be called twice

[ Upstream commit 0b8f62625dc309651d0efcb6a6247c933acd8b45 ]

A fuzzer recently triggered lockdep warnings about potential sb_writers
deadlocks caused by fh_want_write().

Looks like we aren't careful to pair each fh_want_write() with an
fh_drop_write().

It's not normally a problem since fh_put() will call fh_drop_write() for
us.  And was OK for NFSv3 where we'd do one operation that might call
fh_want_write(), and then put the filehandle.

But an NFSv4 protocol fuzzer can do weird things like call unlink twice
in a compound, and then we get into trouble.

I'm a little worried about this approach of just leaving everything to
fh_put().  But I think there are probably a lot of
fh_want_write()/fh_drop_write() imbalances so for now I think we need it
to be more forgiving.

Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
fs/nfsd/vfs.h