On 2025/5/29 18:46, Yangtao Li wrote:
Hi,
在 2025/5/29 18:34, wangjianjian (C) 写道:
[You don't often get email from wangjianjian3@xxxxxxxxxx. Learn why
this is important at https://aka.ms/LearnAboutSenderIdentification ]
On 2025/5/29 14:18, Yangtao Li wrote:
Syzbot reported an issue in hfsplus filesystem:
------------[ cut here ]------------
WARNING: CPU: 0 PID: 4400 at fs/hfsplus/extents.c:346
hfsplus_free_extents+0x700/0xad0
Call Trace:
<TASK>
hfsplus_file_truncate+0x768/0xbb0 fs/hfsplus/extents.c:606
hfsplus_write_begin+0xc2/0xd0 fs/hfsplus/inode.c:56
cont_expand_zero fs/buffer.c:2383 [inline]
cont_write_begin+0x2cf/0x860 fs/buffer.c:2446
hfsplus_write_begin+0x86/0xd0 fs/hfsplus/inode.c:52
generic_cont_expand_simple+0x151/0x250 fs/buffer.c:2347
hfsplus_setattr+0x168/0x280 fs/hfsplus/inode.c:263
notify_change+0xe38/0x10f0 fs/attr.c:420
do_truncate+0x1fb/0x2e0 fs/open.c:65
do_sys_ftruncate+0x2eb/0x380 fs/open.c:193
do_syscall_x64 arch/x86/entry/common.c:50 [inline]
do_syscall_64+0x3d/0xb0 arch/x86/entry/common.c:80
entry_SYSCALL_64_after_hwframe+0x63/0xcd
To avoid deadlock, Commit 31651c607151 ("hfsplus: avoid deadlock
on file truncation") unlock extree before hfsplus_free_extents(),
and add check wheather extree is locked in hfsplus_free_extents().
However, when operations such as hfsplus_file_release,
hfsplus_setattr, hfsplus_unlink, and hfsplus_get_block are executed
concurrently in different files, it is very likely to trigger the
WARN_ON, which will lead syzbot and xfstest to consider it as an
abnormality.
The comment above this warning also describes one of the easy
triggering situations, which can easily trigger and cause
xfstest&syzbot to report errors.
[task A] [task B]
->hfsplus_file_release
->hfsplus_file_truncate
->hfs_find_init
->mutex_lock
->mutex_unlock
->hfsplus_write_begin
->hfsplus_get_block
->hfsplus_file_extend
->hfsplus_ext_read_extent
->hfs_find_init
->mutex_lock
->hfsplus_free_extents
WARN_ON(mutex_is_locked) !!!
I am not familiar with hfsplus, but hfsplus_file_release calls
hfsplus_file_truncate with inode lock, and hfsplus_write_begin can be
called from hfsplus_file_truncate and buffer write, which should also
grab inode lock, so that I think task B should be writeback process,
which call hfsplus_get_block.
Did you read the commit log carefully? I mentioned different files.
sorry, didn't not notice that.
And ->opencnt seems serves as something like link count of other fs, may
be we can move hfsplus_file_truncate to hfsplus_evict_inode, which can
only be called when all users of this inode disappear and writeback
process should also finished for this inode.
Several threads could try to lock the shared extents tree.
And warning can be triggered in one thread when another thread
has locked the tree. This is the wrong behavior of the code and
we need to remove the warning.
Fixes: 31651c607151f ("hfsplus: avoid deadlock on file truncation")
Reported-by: syzbot+8c0bc9f818702ff75b76@xxxxxxxxxxxxxxxxxxxxxxxxx
Closes: https://apc01.safelinks.protection.outlook.com/?
url=https%3A%2F%2Flore.kernel.org%2Fall%2F00000000000057fa4605ef101c4c%40google.com%2F&data=05%7C02%7Cfrank.li%40vivo.com%7C54c2b4030d4c4a98c6c908dd9e9c71b1%7C923e42dc48d54cbeb5821a797a6412ed%7C0%7C0%7C638841116945048579%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=4snXozMFZN%2FsmsL9CP5VJb4mf2p0ReNhQIEuA%2B3tD3A%3D&reserved=0
Signed-off-by: Yangtao Li <frank.li@xxxxxxxx>
---
fs/hfsplus/extents.c | 3 ---
1 file changed, 3 deletions(-)
diff --git a/fs/hfsplus/extents.c b/fs/hfsplus/extents.c
index a6d61685ae79..b1699b3c246a 100644
--- a/fs/hfsplus/extents.c
+++ b/fs/hfsplus/extents.c
@@ -342,9 +342,6 @@ static int hfsplus_free_extents(struct
super_block *sb,
int i;
int err = 0;
- /* Mapping the allocation file may lock the extent tree */
- WARN_ON(mutex_is_locked(&HFSPLUS_SB(sb)->ext_tree->tree_lock));
-
hfsplus_dump_extent(extent);
for (i = 0; i < 8; extent++, i++) {
count = be32_to_cpu(extent->block_count);
--
Regards
Thx,
Yangtao
--
Regards
