On 8/7/21 1:06 am, Jeff Layton wrote:
On Wed, 2021-07-07 at 15:43 +0800, Desmond Cheong Zhi Xi wrote:
Hi,
Sorry for the delay between v1 and v2, there was an unrelated issue with Syzbot testing.
Syzbot reports a possible irq lock inversion dependency:
https://syzkaller.appspot.com/bug?id=923cfc6c6348963f99886a0176ef11dcc429547b
While investigating this error, I discovered that multiple similar lock inversion scenarios can occur. Hence, this series addresses potential deadlocks for two classes of locks, one in each patch:
1. Fix potential deadlocks for &fown_struct.lock
2. Fix potential deadlock for &fasync_struct.fa_lock
v2 -> v3:
- Removed WARN_ON_ONCE, keeping elaboration for why read_lock_irq is safe to use in the commit message. As suggested by Greg KH.
v1 -> v2:
- Added WARN_ON_ONCE(irqs_disabled()) before calls to read_lock_irq, and added elaboration in the commit message. As suggested by Jeff Layton.
Best wishes,
Desmond
Desmond Cheong Zhi Xi (2):
fcntl: fix potential deadlocks for &fown_struct.lock
fcntl: fix potential deadlock for &fasync_struct.fa_lock
fs/fcntl.c | 18 ++++++++++--------
1 file changed, 10 insertions(+), 8 deletions(-)
Looks like these patches are identical to the v1 set, so I'm just going
to leave those in place since linux-next already has them. Let me know
if I've missed something though.
Thanks!
Yep, there's no change outside of the commit message. But I think after
the discussion and with config DEBUG_IRQFLAGS, that is fine.
Thanks again, Jeff!
