LU-15994 llite: use fatal_signal_pending in range_lock

FIO io_uring failed with one file shared by two FIO processes
under Unubtu 2204 kernel.
After analyzed, we found that range_lock() function returns
-ERESTARTSYS when there is pending signal on current process in
Lustre I/O. This causes -EINTR returned to the application.

The reason that we have pending signal is because that io_uring
is using signal based task_work running in new kernel. Thus when
I/O process tries to acquire range lock and checks whether
signal_pending(), it may always find a pending signal and return
-ERESTARTSYS.

we solve this bug by replacing @signal_pending(current) with
@fatal_signal_pending(current) in range_lock(). The range_lock()
function only returns -ERESTARTSYS when the current process has
fatal pending signal such as SIGKILL.

Lustre-change: https://review.whamcloud.com/48106
Lustre-commit: TBD(91244566e9d2762c2f64a67e7d4fad8f301b556c)

Signed-off-by: Qian Yingjin <qian@ddn.com>
Change-Id: I0a0be8fa3b4ba5c89f7866286b2bdc6595f18026
Reviewed-on: https://review.whamcloud.com/48111
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>

diff --git a/lustre/obdclass/range_lock.c b/lustre/obdclass/range_lock.c
index 140337f..9ebb623 100644 (file)
--- a/lustre/obdclass/range_lock.c
+++ b/lustre/obdclass/range_lock.c
@@ -238,7 +238,7 @@ int range_lock(struct range_lock_tree *tree, struct range_lock *lock)
                spin_unlock(&tree->rlt_lock);
                schedule();
 
-               if (signal_pending(current)) {
+               if (fatal_signal_pending(current)) {
                        range_unlock(tree, lock);
                        GOTO(out, rc = -ERESTARTSYS);
                }