FS-Cache: Retain the netfs context in the retrieval op earlier
authorDavid Howells <dhowells@redhat.com>
Tue, 24 Feb 2015 10:05:29 +0000 (10:05 +0000)
committerDavid Howells <dhowells@redhat.com>
Thu, 2 Apr 2015 13:28:53 +0000 (14:28 +0100)
commit4a47132ff472a0c2c5441baeb50cf97f2580bc43
treed6ef9c9d598ed89c3b503728f9ce2a8b3ce21e3d
parentd3b97ca4a99e4e6c78f5a21c968eadf5c8ba9971
FS-Cache: Retain the netfs context in the retrieval op earlier

Now that the retrieval operation may be disposed of by fscache_put_operation()
before we actually set the context, the retrieval-specific cleanup operation
can produce a NULL-pointer dereference when it tries to unconditionally clean
up the netfs context.

Given that it is expected that we'll get at least as far as the place where we
currently set the context pointer and it is unlikely we'll go through the
error handling paths prior to that point, retain the context right from the
point that the retrieval op is allocated.

Concomitant to this, we need to retain the cookie pointer in the retrieval op
also so that we can call the netfs to release its context in the release
method.

In addition, we might now get into fscache_release_retrieval_op() with the op
only initialised.  To this end, set the operation to DEAD only after the
release method has been called and skip the n_pages test upon cleanup if the
op is still in the INITIALISED state.

Without these changes, the following oops might be seen:

BUG: unable to handle kernel NULL pointer dereference at 00000000000000b8
...
RIP: 0010:[<ffffffffa0089c98>] fscache_release_retrieval_op+0xae/0x100
...
Call Trace:
 [<ffffffffa0088560>] fscache_put_operation+0x117/0x2e0
 [<ffffffffa008b8f5>] __fscache_read_or_alloc_pages+0x351/0x3ac
 [<ffffffffa00b761f>] __nfs_readpages_from_fscache+0x59/0xbf [nfs]
 [<ffffffffa00b06c5>] nfs_readpages+0x10c/0x185 [nfs]
 [<ffffffff81124925>] ? alloc_pages_current+0x119/0x13e
 [<ffffffff810ee5fd>] ? __page_cache_alloc+0xfb/0x10a
 [<ffffffff810f87f8>] __do_page_cache_readahead+0x188/0x22c
 [<ffffffff810f8b3a>] ondemand_readahead+0x29e/0x2af
 [<ffffffff810f8c92>] page_cache_sync_readahead+0x38/0x3a
 [<ffffffff810ef337>] generic_file_read_iter+0x1a2/0x55a
 [<ffffffffa00a9dff>] ? nfs_revalidate_mapping+0xd6/0x288 [nfs]
 [<ffffffffa00a6a23>] nfs_file_read+0x49/0x70 [nfs]
 [<ffffffff811363be>] new_sync_read+0x78/0x9c
 [<ffffffff81137164>] __vfs_read+0x13/0x38
 [<ffffffff8113721e>] vfs_read+0x95/0x121
 [<ffffffff811372f6>] SyS_read+0x4c/0x8a
 [<ffffffff81557a52>] system_call_fastpath+0x12/0x17

Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Steve Dickson <steved@redhat.com>
Acked-by: Jeff Layton <jeff.layton@primarydata.com>
fs/fscache/operation.c
fs/fscache/page.c
include/linux/fscache-cache.h