btrfs-progs: image: Fix data races when reporting progress

Making the code data-race safe requires that reads *and* writes
happen under a mutex lock, if any of the access are writes. See
Dmitri Vyukov, "Benign data races: what could possibly go wrong?"
for more details.

The fix here was to put most of the main loop of restore_worker
under a mutex lock.

This race was detected using fsck-tests/012-leaf-corruption.

==================
WARNING: ThreadSanitizer: data race
  Write of size 4 by main thread:
    #0 add_cluster btrfs-progs/image/main.c:1931
    #1 restore_metadump btrfs-progs/image/main.c:2566
    #2 main btrfs-progs/image/main.c:2859

  Previous read of size 4 by thread T6:
    #0 restore_worker btrfs-progs/image/main.c:1720

  Location is stack of main thread.

  Thread T6 (running) created by main thread at:
    #0 pthread_create <null>
    #1 mdrestore_init btrfs-progs/image/main.c:1868
    #2 restore_metadump btrfs-progs/image/main.c:2534
    #3 main btrfs-progs/image/main.c:2859

SUMMARY: ThreadSanitizer: data race btrfs-progs/image/main.c:1931 in
add_cluster

Signed-off-by: Adam Buchbinder <abuchbinder@google.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This commit is contained in:
Adam Buchbinder 2017-07-12 13:05:10 -07:00 committed by David Sterba
parent 1639cc2a51
commit 76daaec5cc

View file

@ -1720,12 +1720,13 @@ static void *restore_worker(void *data)
}
async = list_entry(mdres->list.next, struct async_work, list);
list_del_init(&async->list);
pthread_mutex_unlock(&mdres->mutex);
if (mdres->compress_method == COMPRESS_ZLIB) {
size = compress_size;
pthread_mutex_unlock(&mdres->mutex);
ret = uncompress(buffer, (unsigned long *)&size,
async->buffer, async->bufsize);
pthread_mutex_lock(&mdres->mutex);
if (ret != Z_OK) {
error("decompressiion failed with %d", ret);
err = -EIO;
@ -1803,7 +1804,6 @@ error:
if (!mdres->multi_devices && async->start == BTRFS_SUPER_INFO_OFFSET)
write_backup_supers(outfd, outbuf);
pthread_mutex_lock(&mdres->mutex);
if (err && !mdres->error)
mdres->error = err;
mdres->num_items--;
@ -1933,7 +1933,9 @@ static int add_cluster(struct meta_cluster *cluster,
u32 i, nritems;
int ret;
pthread_mutex_lock(&mdres->mutex);
mdres->compress_method = header->compress;
pthread_mutex_unlock(&mdres->mutex);
bytenr = le64_to_cpu(header->bytenr) + BLOCK_SIZE;
nritems = le32_to_cpu(header->nritems);