Re: 5.2.0-rc1 UBIFS bad unlock balance

[Richard Weinberger <richard@xxxxxx>]

----- Ursprüngliche Mail -----
>> Huh? The bad commit was dropped. Are you still seeing this on Linus' tree?
>> Maybe something else is bad.
>>
> 
> Should be related to this commit. Seems working fine after reverting the patch.
> 
> commit 8009ce956c3d28022af6b122e50213ad830fc902
> Author: Richard Weinberger <richard@xxxxxx>
> Date:   Wed May 15 21:52:34 2019 +0200
> 
>    ubifs: Don't leak orphans on memory during commit

You are perfectly right, there is still a problem!
Does the following patch make the problem go away?
After orphan_delete() c->orphan_lock is being unlocked,
so a double unlock can happen.

diff --git a/fs/ubifs/orphan.c b/fs/ubifs/orphan.c
index b52624e28fa1..3b4b4114f208 100644
--- a/fs/ubifs/orphan.c
+++ b/fs/ubifs/orphan.c
@@ -129,7 +129,6 @@ static void __orphan_drop(struct ubifs_info *c, struct ubifs_orphan *o)
 static void orphan_delete(struct ubifs_info *c, struct ubifs_orphan *orph)
 {
 	if (orph->del) {
-		spin_unlock(&c->orphan_lock);
 		dbg_gen("deleted twice ino %lu", orph->inum);
 		return;
 	}
@@ -138,7 +137,6 @@ static void orphan_delete(struct ubifs_info *c, struct ubifs_orphan *orph)
 		orph->del = 1;
 		orph->dnext = c->orph_dnext;
 		c->orph_dnext = orph;
-		spin_unlock(&c->orphan_lock);
 		dbg_gen("delete later ino %lu", orph->inum);
 		return;
 	}

Thanks,
//richard

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/