On Tue, May 27, 2014 at 02:40:54AM +0100, Al Viro wrote:
> It looks plausible, but I doubt that serializing check_submounts_and_drop()
> will suffice - shrink_dcache_parent() is just as unpleasant and it *is*
> triggered in the same situations. Moreover, the lack of loop in memory
> shrinkers doesn't help - we might get shrink_dentry_list() from one of
> those and loops that keep calling d_walk() from check_submounts_and_drop()
> or shrink_dcache_parent()...
>
> > Anyway, I'd like Mika to test the stupid "let's serialize the dentry
> > shrinking in check_submounts_and_drop()" to see if his problem goes
> > away. I agree that it's not the _proper_ fix, but we're damn late in
> > the rc series..
>
> That we are... FWIW, if the nastiness matches the description above,
> the right place to do something probably would be when those two
> suckers get positive return value from d_walk() along with an empty
> shrink list. I wonder if we should do down_read() in shrink_dentry_list()
> and down_write();up_write() in that case in shrink_dcache_parent() and
> check_submounts_and_drop(). How about the following?
As the matter of fact, let's try this instead - retry the same sucker
immediately in case if trylocks fail. Comments?
diff --git a/fs/dcache.c b/fs/dcache.c
index 42ae01e..d58d4cc 100644
--- a/fs/dcache.c
+++ b/fs/dcache.c
@@ -798,6 +798,7 @@ static void shrink_dentry_list(struct list_head *list)
while (!list_empty(list)) {
dentry = list_entry(list->prev, struct dentry, d_lru);
+again:
spin_lock(&dentry->d_lock);
/*
* The dispose list is isolated and dentries are not accounted
@@ -830,7 +831,8 @@ static void shrink_dentry_list(struct list_head *list)
*/
d_shrink_add(dentry, list);
spin_unlock(&dentry->d_lock);
- continue;
+ cpu_relax();
+ goto again;
}
/*
* We need to prune ancestors too. This is necessary to prevent
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
