[PATCH] ext4: fix a data race at inode->i_blocks

From: Qian Cai <cai@lca.pw>
To: tytso@mit.edu
Cc: adilger.kernel@dilger.ca, elver@google.com,
	linux-ext4@vger.kernel.org, linux-kernel@vger.kernel.org,
	Qian Cai <cai@lca.pw>
Subject: [PATCH] ext4: fix a data race at inode->i_blocks
Date: Fri, 21 Feb 2020 23:32:58 -0500	[thread overview]
Message-ID: <20200222043258.2279-1-cai@lca.pw> (raw)

inode->i_blocks could be accessed concurrently as noticed by KCSAN,

 BUG: KCSAN: data-race in ext4_do_update_inode [ext4] / inode_add_bytes

 write to 0xffff9a00d4b982d0 of 8 bytes by task 22100 on cpu 118:
  inode_add_bytes+0x65/0xf0
  __inode_add_bytes at fs/stat.c:689
  (inlined by) inode_add_bytes at fs/stat.c:702
  ext4_mb_new_blocks+0x418/0xca0 [ext4]
  ext4_ext_map_blocks+0x1a6b/0x27b0 [ext4]
  ext4_map_blocks+0x1a9/0x950 [ext4]
  _ext4_get_block+0xfc/0x270 [ext4]
  ext4_get_block_unwritten+0x33/0x50 [ext4]
  __block_write_begin_int+0x22e/0xae0
  __block_write_begin+0x39/0x50
  ext4_write_begin+0x388/0xb50 [ext4]
  ext4_da_write_begin+0x35f/0x8f0 [ext4]
  generic_perform_write+0x15d/0x290
  ext4_buffered_write_iter+0x11f/0x210 [ext4]
  ext4_file_write_iter+0xce/0x9e0 [ext4]
  new_sync_write+0x29c/0x3b0
  __vfs_write+0x92/0xa0
  vfs_write+0x103/0x260
  ksys_write+0x9d/0x130
  __x64_sys_write+0x4c/0x60
  do_syscall_64+0x91/0xb05
  entry_SYSCALL_64_after_hwframe+0x49/0xbe

 read to 0xffff9a00d4b982d0 of 8 bytes by task 8 on cpu 65:
  ext4_do_update_inode+0x4a0/0xf60 [ext4]
  ext4_inode_blocks_set at fs/ext4/inode.c:4815
  ext4_mark_iloc_dirty+0xaf/0x160 [ext4]
  ext4_mark_inode_dirty+0x129/0x3e0 [ext4]
  ext4_convert_unwritten_extents+0x253/0x2d0 [ext4]
  ext4_convert_unwritten_io_end_vec+0xc5/0x150 [ext4]
  ext4_end_io_rsv_work+0x22c/0x350 [ext4]
  process_one_work+0x54f/0xb90
  worker_thread+0x80/0x5f0
  kthread+0x1cd/0x1f0
  ret_from_fork+0x27/0x50

 4 locks held by kworker/u256:0/8:
  #0: ffff9a025abc4328 ((wq_completion)ext4-rsv-conversion){+.+.}, at: process_one_work+0x443/0xb90
  #1: ffffab5a862dbe20 ((work_completion)(&ei->i_rsv_conversion_work)){+.+.}, at: process_one_work+0x443/0xb90
  #2: ffff9a025a9d0f58 (jbd2_handle){++++}, at: start_this_handle+0x1c1/0x9d0 [jbd2]
  #3: ffff9a00d4b985d8 (&(&ei->i_raw_lock)->rlock){+.+.}, at: ext4_do_update_inode+0xaa/0xf60 [ext4]
 irq event stamp: 3009267
 hardirqs last  enabled at (3009267): [<ffffffff980da9b7>] __find_get_block+0x107/0x790
 hardirqs last disabled at (3009266): [<ffffffff980da8f9>] __find_get_block+0x49/0x790
 softirqs last  enabled at (3009230): [<ffffffff98a0034c>] __do_softirq+0x34c/0x57c
 softirqs last disabled at (3009223): [<ffffffff97cc67a2>] irq_exit+0xa2/0xc0

 Reported by Kernel Concurrency Sanitizer on:
 CPU: 65 PID: 8 Comm: kworker/u256:0 Tainted: G L 5.6.0-rc2-next-20200221+ #7
 Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 07/10/2019
 Workqueue: ext4-rsv-conversion ext4_end_io_rsv_work [ext4]

The plain read is outside of inode->i_lock critical section which
results in a data race. Fix it by adding READ_ONCE() there.

Signed-off-by: Qian Cai <cai@lca.pw>
---
 fs/ext4/inode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index e60aca791d3f..98cadd111942 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -4812,7 +4812,7 @@ static int ext4_inode_blocks_set(handle_t *handle,
 				struct ext4_inode_info *ei)
 {
 	struct inode *inode = &(ei->vfs_inode);
-	u64 i_blocks = inode->i_blocks;
+	u64 i_blocks = READ_ONCE(inode->i_blocks);
 	struct super_block *sb = inode->i_sb;
 
 	if (i_blocks <= ~0U) {
-- 
2.21.0 (Apple Git-122.2)

