Messages in this thread Patch in this message | | | From | Jia-Ju Bai <> | Subject | [PATCH 2/4] fs: btrfs: fix data races in extent_write_cache_pages() | Date | Sat, 9 May 2020 13:27:01 +0800 |
| |
The function extent_write_cache_pages is concurrently executed with itself at runtime in the following call contexts:
Thread 1: btrfs_sync_file() start_ordered_ops() btrfs_fdatawrite_range() btrfs_writepages() [via function pointer] extent_writepages() extent_write_cache_pages()
Thread 2: btrfs_writepages() extent_writepages() extent_write_cache_pages()
In extent_write_cache_pages(): index = mapping->writeback_index; ... mapping->writeback_index = done_index;
The accesses to mapping->writeback_index are not synchronized, and thus data races for this value can occur. These data races were found and actually reproduced by our concurrency fuzzer.
To fix these races, the spinlock mapping->private_lock is used to protect the accesses to mapping->writeback_index.
Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com> --- fs/btrfs/extent_io.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 39e45b8a5031..8c33a60bde1d 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -4160,7 +4160,9 @@ static int extent_write_cache_pages(struct address_space *mapping, pagevec_init(&pvec); if (wbc->range_cyclic) { + spin_lock(&mapping->private_lock); index = mapping->writeback_index; /* Start from prev offset */ + spin_unlock(&mapping->private_lock); end = -1; /* * Start from the beginning does not need to cycle over the @@ -4271,8 +4273,11 @@ static int extent_write_cache_pages(struct address_space *mapping, goto retry; } - if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) + if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) { + spin_lock(&mapping->private_lock); mapping->writeback_index = done_index; + spin_unlock(&mapping->private_lock); + } btrfs_add_delayed_iput(inode); return ret; -- 2.17.1
| |