LU-13088 ldlm: Fix sleeping function called in atomic

target_recovery_overseer() can sleep while holding a spinlock, which
triggers a BUG warning.

It is easily fixed by dropping the spinlock before waiting.  In the
case where the task waits, no useful information that could be
protected by the spinlock is held, so nothing can be lost by dropping
it.

Lustre-change: https://review.whamcloud.com/#/c/37063/
Lustre-commit: b29b9310dafe17ba78e1db490b79b89d2d6fdcd1

Signed-off-by: Mr NeilBrown <neilb@suse.de>
Change-Id: I8bb3d02523b5dcfadac19f01ccb736d7b7f28239
Reviewed-on: https://review.whamcloud.com/37063
Tested-by: jenkins <devops@whamcloud.com>
Reviewed-by: James Simmons <jsimmons@infradead.org>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Mike Pershin <mpershin@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
Reviewed-on: https://review.whamcloud.com/39283

diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index 6e0ac53..4b13020 100644 (file)
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -2029,11 +2029,13 @@ repeat:
                        if (next_update_transno == 0) {
                                struct l_wait_info lwi = { 0 };
 
+                               spin_unlock(&obd->obd_recovery_task_lock);
                                l_wait_event(tdtd->tdtd_recovery_threads_waitq,
                                       atomic_read(
                                       &tdtd->tdtd_recovery_threads_count) == 0,
                                       &lwi);
 
+                               spin_lock(&obd->obd_recovery_task_lock);
                                next_update_transno =
                                        distribute_txn_get_next_transno(
                                                                lut->lut_tdtd);