dlm_load: handle EBUSY after cancel

Nothing we can do to be certain we won't get EBUSY after cancel sometimes,
so don't make it an error.

Signed-off-by: David Teigland <teigland@redhat.com>

1 files changed, 40 insertions, 0 deletions

diff --git a/dlm/dlm_load.c b/dlm/dlm_load.c
index d54f414..30cf161 100644
--- a/dlm/dlm_load.c
+++ b/dlm/dlm_load.c
@@ -80,6 +80,7 @@ struct lk {
 	int lastop;
 	int last_status;
 	int bast;
+	int cancel_busy;
 	time_t wait_start;
 	struct dlm_lksb lksb;
 };
@@ -428,11 +429,25 @@ static int do_convert(struct lk *lk, int mode, int noqueue)
 	rv = dlm_ls_lockx(dh, mode, &lk->lksb, flags, name, strlen(name), 0,
 			  astfn, (void *)lk, bastfn, &our_xid, timeout_arg);
 
+	/* Unfortunately we don't have a way to know when the dlm will be
+	   done dealing with a cancel and won't return -EBUSY for ops.
+	   (We don't get a callback for the cancel when got a callback for
+	   the original op and the cancel did nothing.) */
+
+	if (rv == -1 && errno == EBUSY && lk->lastop == Op_cancel) {
+		lk->cancel_busy++;
+		log_debug("lk %03u convert cancel_busy %d", lk->cancel_busy);
+		goto out;
+	} else {
+		lk->cancel_busy = 0;
+	}
+
 	if (rv) {
 		log_error("lk %03u convert errno %d wait r%d c%d u%d c%d lastop %s",
 			  i, errno, lk->wait_request, lk->wait_convert,
 			  lk->wait_unlock, lk->wait_cancel, op_str(lk->lastop));
 	}
+ out:
 	return rv;
 }
 
@@ -447,14 +462,39 @@ static int do_unlock(struct lk *lk)
 
 	rv = dlm_ls_unlock(dh, lkid, flags, &lk->lksb, lk);
 
+	/* See comment above. */
+
+	if (rv == -1 && errno == EBUSY && lk->lastop == Op_cancel) {
+		lk->cancel_busy++;
+		log_debug("lk %03u unlock cancel_busy %d", lk->cancel_busy);
+		goto out;
+	} else {
+		lk->cancel_busy = 0;
+	}
+
 	if (rv) {
 		log_error("lk %03u unlock errno %d wait r%d c%d u%d c%d lastop %s",
 			  i, errno, lk->wait_request, lk->wait_convert,
 			  lk->wait_unlock, lk->wait_cancel, op_str(lk->lastop));
 	}
+ out:
 	return rv;
 }
 
+/* The op following a cancel can return -EBUSY.  I don't think there is any
+   way to know in this case when we can do the next op without getting
+   EBUSY back.  See lk->cancel_busy.  Without the cancel_busy check we get
+   this error:
+   lk 025 request mode 4 noqueue 0
+   lk 025 busy locked 0 unlocked 0 wait r1 c0 u0 c0
+   lk 025 cancel
+   lk 025 busy locked 0 unlocked 0 wait r1 c0 u0 c1
+   lk 025 cast 0706 0       01060001 wait r1 c0 u0 c1 gr -1 rq  4 last cancel
+   lk 025 bast 0707
+   lk 025 convert mode 0 noqueue 0
+   ERROR lk 025 convert errno 16 wait r0 c0 u0 c0 lastop cancel
+*/
+
 static int do_cancel(struct lk *lk)
 {
 	uint32_t lkid = lk->lksb.sb_lkid;