[PATCH] fs: fix a data race in i_size_write/i_size_read

From: Qian Cai <cai@lca.pw>
To: viro@zeniv.linux.org.uk
Cc: hch@infradead.org, darrick.wong@oracle.com, elver@google.com,
	linux-xfs@vger.kernel.org, linux-fsdevel@vger.kernel.org,
	linux-kernel@vger.kernel.org, Qian Cai <cai@lca.pw>
Subject: [PATCH] fs: fix a data race in i_size_write/i_size_read
Date: Tue, 18 Feb 2020 23:04:26 -0500	[thread overview]
Message-ID: <20200219040426.1140-1-cai@lca.pw> (raw)

inode::i_size could be accessed concurently as noticed by KCSAN,

 BUG: KCSAN: data-race in iomap_do_writepage / iomap_write_end

 write to 0xffff8bf68fc0cac0 of 8 bytes by task 7484 on cpu 71:
  iomap_write_end+0xea/0x530
  i_size_write at include/linux/fs.h:888
  (inlined by) iomap_write_end at fs/iomap/buffered-io.c:782
  iomap_write_actor+0x132/0x200
  iomap_apply+0x245/0x8a5
  iomap_file_buffered_write+0xbd/0xf0
  xfs_file_buffered_aio_write+0x1c2/0x790 [xfs]
  xfs_file_write_iter+0x232/0x2d0 [xfs]
  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 0xffff8bf68fc0cac0 of 8 bytes by task 5901 on cpu 70:
  iomap_do_writepage+0xf4/0x450
  i_size_read at include/linux/fs.h:866
  (inlined by) iomap_do_writepage at fs/iomap/buffered-io.c:1558
  write_cache_pages+0x523/0xb20
  iomap_writepages+0x47/0x80
  xfs_vm_writepages+0xc7/0x100 [xfs]
  do_writepages+0x5e/0x130
  __writeback_single_inode+0xd5/0xb20
  writeback_sb_inodes+0x429/0x910
  __writeback_inodes_wb+0xc4/0x150
  wb_writeback+0x47b/0x830
  wb_workfn+0x688/0x930
  process_one_work+0x54f/0xb90
  worker_thread+0x80/0x5f0
  kthread+0x1cd/0x1f0
  ret_from_fork+0x27/0x50

 Reported by Kernel Concurrency Sanitizer on:
 CPU: 70 PID: 5901 Comm: kworker/u257:2 Tainted: G             L    5.6.0-rc2-next-20200218+ #2
 Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 07/10/2019
 Workqueue: writeback wb_workfn (flush-254:0)

The write is protected by exclusive inode::i_rwsem (in
xfs_file_buffered_aio_write()) but the read is not. A shattered value
could introduce a logic bug. Fixed it using a pair of WRITE/READ_ONCE().

Signed-off-by: Qian Cai <cai@lca.pw>
---
 include/linux/fs.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 3cd4fe6b845e..25f98da90cf3 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -863,7 +863,7 @@ static inline loff_t i_size_read(const struct inode *inode)
 	preempt_enable();
 	return i_size;
 #else
-	return inode->i_size;
+	return READ_ONCE(inode->i_size);
 #endif
 }
 
@@ -885,7 +885,7 @@ static inline void i_size_write(struct inode *inode, loff_t i_size)
 	inode->i_size = i_size;
 	preempt_enable();
 #else
-	inode->i_size = i_size;
+	WRITE_ONCE(inode->i_size, i_size);
 #endif
 }
 
-- 
2.21.0 (Apple Git-122.2)

