Clear PG_writeback for sync I/O error case

Commit 2b28613 accidentally introduced this issue by only conditionally registering the commit callback in the async case. The error handing code for the dmu_tx_assign() failure case relied on there always being a registered commit callback to clear the PG_writeback bit. Since that is no longer strictly true for the synchronous case we must explicitly invoke the callback. Signed-off-by: Brian Behlendorf <[email protected]> Closes #961
openzfs · Sep 14, 2012 · 8312c6d · 8312c6d
1 parent 8e8e7f3
commit 8312c6d
Showing 1 changed file with 9 additions and 0 deletions.
diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c
@@ -3848,7 +3848,16 @@ zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc)
 		if (err == ERESTART)
 			dmu_tx_wait(tx);
 
+		/* Will call all registered commit callbacks */
 		dmu_tx_abort(tx);
+
+		/*
+		 * For the synchronous case the commit callback must be
+		 * explicitly called because there is no registered callback.
+		 */
+		if (sync)
+			zfs_putpage_commit_cb(pp, ECANCELED);
+
 		return (err);
 	}