When HUNG_TASK_PANIC is _not_ set, this patch doesn't take any action hence it's still possible to observe stack dumps ( as before ) whenever a FUSE server stop responding. This issue is seen in the 5.4 / 5.10 and 6.6 stable branch and IMO we need a solution that can easily be backported. Following Bernd's comment on the previous implementation [3], this patch now leverage the fuse_check_timeout() logic from Joanne [1] with minor modifications: - Runs from a work queue instead of a timer to address the deadlock reported by Sergey [2] - Use time_after() for jiffies arithmetic - Add chained work to kick back the timer logic - Set the timeout to be half the HUNG_TASK_TIMEOUT period in order to stay under the radar I've been running some tests on rc1 and so far things looks good. [1] https://lore.kernel.org/linux-fsdevel/20241114191332.669127-3-joannelkoong@xxxxxxxxx/ [2] https://lore.kernel.org/linux-fsdevel/20241203110147.GD886051@xxxxxxxxxx/ [3] https://lore.kernel.org/linux-fsdevel/20241210171621.64645-1-etmartin4313@xxxxxxxxx/T/#m8f6f370d39087adcd4a7b325fc0562eebab7b998 thanks, Etienne
