Skip to contentSkip to conversation

btrfs-progs: Fix data races in btrfs-image.

PATCH Applied Adam Buchbinder 2 participants 2 messages Jul 12, 2017 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 btrfs-progs: Fix data races in btrfs-image.

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

---
 image/main.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)
image/main.c +9 -7
index 1eca414..a5d01d8 100644
@@ -1715,14 +1715,15 @@ 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);
+				error("decompression failed with %d", ret);
 				err = -EIO;
 			}
 			outbuf = buffer;
@@ -1798,7 +1799,6 @@ 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--;
@@ -1899,7 +1899,7 @@ static int fill_mdres_info(struct mdrestore_struct *mdres,
 		ret = uncompress(buffer, (unsigned long *)&size,
 				 async->buffer, async->bufsize);
 		if (ret != Z_OK) {
-			error("decompressiion failed with %d", ret);
+			error("decompression failed with %d", ret);
 			free(buffer);
 			return -EIO;
 		}
@@ -1928,7 +1928,9 @@ 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);
@@ -2171,7 +2173,7 @@ static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
 				continue;
 			}
 			error(
-	"unknown state after reading cluster at %llu, probably crrupted data",
+	"unknown state after reading cluster at %llu, probably corrupted data",
 					cluster_bytenr);
 			ret = -EIO;
 			break;
@@ -2220,7 +2222,7 @@ static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
 						 (unsigned long *)&size, tmp,
 						 bufsize);
 				if (ret != Z_OK) {
-					error("decompressiion failed with %d",
+					error("decompression failed with %d",
 							ret);
 					ret = -EIO;
 					break;
@@ -2340,7 +2342,7 @@ static int build_chunk_tree(struct mdrestore_struct *mdres,
 		ret = uncompress(tmp, (unsigned long *)&size,
 				 buffer, le32_to_cpu(item->size));
 		if (ret != Z_OK) {
-			error("decompressiion failed with %d", ret);
+			error("decompression failed with %d", ret);
 			free(buffer);
 			free(tmp);
 			return -EIO;
Signature · 1 line
2.13.2.932.g7449e964c-goog
Applied
#1 lore raw
Quoted 28 lines · Adam Buchbinder

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

Quoted 4 lines

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

Applied, thanks.

Quoted 13 lines
---
 image/main.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)
image/main.c +0 -1
index 1eca414..a5d01d8 100644
@@ -1715,14 +1715,15 @@ 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);

image/main.c static void *restore_worker(void *data) · line 1726

Quoted 10 lines
image/main.c +3 -1
 
 		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);
+				error("decompression failed with %d", ret);

The typo fixes belong to a separate patch, as they fix a different problem. We stick to the same style as in kernel "one patch per logical change", so I'll split them to another patch as it's fairly trivial.

Quoted 64 lines
image/main.c +6 -5
 				err = -EIO;
 			}
 			outbuf = buffer;
@@ -1798,7 +1799,6 @@ 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--;
@@ -1899,7 +1899,7 @@ static int fill_mdres_info(struct mdrestore_struct *mdres,
 		ret = uncompress(buffer, (unsigned long *)&size,
 				 async->buffer, async->bufsize);
 		if (ret != Z_OK) {
-			error("decompressiion failed with %d", ret);
+			error("decompression failed with %d", ret);
 			free(buffer);
 			return -EIO;
 		}
@@ -1928,7 +1928,9 @@ 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);
@@ -2171,7 +2173,7 @@ static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
 				continue;
 			}
 			error(
-	"unknown state after reading cluster at %llu, probably crrupted data",
+	"unknown state after reading cluster at %llu, probably corrupted data",
 					cluster_bytenr);
 			ret = -EIO;
 			break;
@@ -2220,7 +2222,7 @@ static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
 						 (unsigned long *)&size, tmp,
 						 bufsize);
 				if (ret != Z_OK) {
-					error("decompressiion failed with %d",
+					error("decompression failed with %d",
 							ret);
 					ret = -EIO;
 					break;
@@ -2340,7 +2342,7 @@ static int build_chunk_tree(struct mdrestore_struct *mdres,
 		ret = uncompress(tmp, (unsigned long *)&size,
 				 buffer, le32_to_cpu(item->size));
 		if (ret != Z_OK) {
-			error("decompressiion failed with %d", ret);
+			error("decompression failed with %d", ret);
 			free(buffer);
 			free(tmp);
 			return -EIO;
Signature · 6 lines
2.13.2.932.g7449e964c-goog

--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html