[CHECKER] deadlock in 2.5.62/fs/lockd/svc*.c?

Dawson Engler (engler@csl.stanford.edu)
Sat, 22 Mar 2003 02:10:28 -0800 (PST)


Hi All,

here's a more interesting potential deadlock in the
2.5.62/fs/lockd/svc*.c
code that requires the miscordination of three threads. It might be
a false because of
1. infeasible callchain path or

2. the two struct nlm_file.f_sema pointers cannot actually point
to the same object.

It requires three threads:
thread 1: acquires nlm_file_sema then tries to get file->f_sema
thread 2: acquires file->f_sema and tries to get nlm_host_sema
thread 3: acquires nlm_host_sema and tries to get nlm_file_sema

Any feedback on this one would be great.

Dawson

BUG: ERROR: 3 thread global-global deadlock.
<&nlm_file_sema>-><&nlm_host_sema> occurred 4 times
<&nlm_host_sema>-><&nlm_file_sema> occurred 5 times

Plausible chain
thread 1: <&nlm_file_sema,struct nlm_file.f_sema>
depth = 3:
2.5.62/fs/lockd/svcsubs.c:nlm_traverse_files:218

down(&nlm_file_sema);

->2.5.62/fs/lockd/svcsubs.c:nlm_traverse_files:218
->nlm_traverse_files:224
->nlm_inspect_file:202
->end=fs/lockd/svclock.c:nlmsvc_traverse_blocks:274:down

down(&file->f_sema);

Seems like a plausible chain.
thread 2: <struct nlm_file.f_sema,&nlm_host_sema>
depth = 4:
2.5.62/fs/lockd/svclock.c:nlmsvc_lock:309

/* Lock file against concurrent access */
down(&file->f_sema);

->2.5.62/fs/lockd/svclock.c:nlmsvc_lock:309
->nlmsvc_lock:351
->nlmsvc_create_block:179
->2.5.62/fs/lockd/host.c:nlmclnt_lookup_host:44
->end=nlm_lookup_host:74:down

/* Lock hash table */
down(&nlm_host_sema);

Seems reasonable.
<&nlm_host_sema>-><&nlm_file_sema> =
depth = 4:
2.5.62/fs/lockd/host.c:nlm_lookup_host:74
/* Lock hash table */
down(&nlm_host_sema);

->2.5.62/fs/lockd/host.c:nlm_lookup_host:74
->nlm_lookup_host:77
->nlm_gc_hosts:319
->/u2/engler/mc/oses/linux/linux-2.5.62/fs/lockd/svcsubs.c:nlmsvc_mark_resources:279
->end=nlm_traverse_files:218:down
->2.5.62/fs/lockd/svcsubs.c:nlm_traverse_files:218

down(&nlm_file_sema);

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/