From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from Galois.linutronix.de ([146.0.238.70]:50734 "EHLO Galois.linutronix.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751470AbeCNISk (ORCPT ); Wed, 14 Mar 2018 04:18:40 -0400 From: John Ogness To: Al Viro Cc: Linus Torvalds , linux-fsdevel , Christoph Hellwig , Thomas Gleixner , Peter Zijlstra , Sebastian Andrzej Siewior , Linux Kernel Mailing List , Eric Biederman Subject: Re: dcache: remove trylock loops (was Re: [BUG] lock_parent() breakage when used from shrink_dentry_list()) References: <20180223150928.GC30522@ZenIV.linux.org.uk> <20180223174216.GD30522@ZenIV.linux.org.uk> <20180223201317.GG30522@ZenIV.linux.org.uk> <20180224002248.GH30522@ZenIV.linux.org.uk> <20180225073950.GI30522@ZenIV.linux.org.uk> <87bmgbnhar.fsf_-_@linutronix.de> <20180312191351.GN30522@ZenIV.linux.org.uk> <87zi3bn1on.fsf@linutronix.de> <87lgevn0ss.fsf@linutronix.de> <20180313235930.GX30522@ZenIV.linux.org.uk> Date: Wed, 14 Mar 2018 09:18:21 +0100 In-Reply-To: <20180313235930.GX30522@ZenIV.linux.org.uk> (Al Viro's message of "Tue, 13 Mar 2018 23:59:30 +0000") Message-ID: <87a7vbvzn6.fsf@linutronix.de> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On 2018-03-14, Al Viro wrote: >>> + rcu_read_lock(); /* to protect parent */ >>> + spin_unlock(&dentry->d_lock); >>> + parent = READ_ONCE(dentry->d_parent); >> >> The preceeding line should be removed. We already have a "parent" >> from before we did the most recent trylock(). > > Nope. We have parent, yes, but it had been fetched outside of > rcu_read_lock(). So the object it used to point to might have been > already freed and we can't do this: > >>> + spin_lock(&parent->d_lock); When rcu_read_lock() is called, we are still holding dentry->d_lock. At that point dentry->d_parent cannot have changed and cannot have been freed. So the parent fetched outside of rcu_read_lock() is also protected from freeing inside that rcu_read_lock(). > Come to think of that, it might make sense to lift rcu_read_lock() all > the way out of that sucker. Agreed. > Objections? Below is the incremental I'd fold into that commit: > > diff --git a/fs/dcache.c b/fs/dcache.c > index f0e73c93182b..0d1dac750c0a 100644 > --- a/fs/dcache.c > +++ b/fs/dcache.c > @@ -1000,7 +1000,6 @@ static bool shrink_lock_dentry(struct dentry *dentry) > > inode = dentry->d_inode; > if (inode && unlikely(!spin_trylock(&inode->i_lock))) { > - rcu_read_lock(); /* to protect inode */ > spin_unlock(&dentry->d_lock); > spin_lock(&inode->i_lock); > spin_lock(&dentry->d_lock); > @@ -1009,16 +1008,14 @@ static bool shrink_lock_dentry(struct dentry *dentry) > /* changed inode means that somebody had grabbed it */ > if (unlikely(inode != dentry->d_inode)) > goto out; > - rcu_read_unlock(); > } > > parent = dentry->d_parent; > + /* parent will stay allocated until we drop rcu_read_lock */ I think this comment is not necessary since this function no longer deals with dropping rcu_read_lock. But if we keep it, it should be added for the inode above as well. > if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock))) > return true; > > - rcu_read_lock(); /* to protect parent */ > spin_unlock(&dentry->d_lock); > - parent = READ_ONCE(dentry->d_parent); > spin_lock(&parent->d_lock); > if (unlikely(parent != dentry->d_parent)) { > spin_unlock(&parent->d_lock); > @@ -1026,14 +1023,11 @@ static bool shrink_lock_dentry(struct dentry *dentry) > goto out; > } > spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); > - if (likely(!dentry->d_lockref.count)) { > - rcu_read_unlock(); > + if (likely(!dentry->d_lockref.count)) > return true; > - } > spin_unlock(&parent->d_lock); > out: > spin_unlock(&inode->i_lock); > - rcu_read_unlock(); > return false; > } > > @@ -1044,8 +1038,10 @@ static void shrink_dentry_list(struct list_head *list) > > dentry = list_entry(list->prev, struct dentry, d_lru); > spin_lock(&dentry->d_lock); > + rcu_read_lock(); > if (!shrink_lock_dentry(dentry)) { > bool can_free = false; > + rcu_read_unlock(); > d_shrink_del(dentry); > if (dentry->d_lockref.count < 0) > can_free = dentry->d_flags & DCACHE_MAY_FREE; > @@ -1054,6 +1050,7 @@ static void shrink_dentry_list(struct list_head *list) > dentry_free(dentry); > continue; > } > + rcu_read_unlock(); > d_shrink_del(dentry); > parent = dentry->d_parent; > __dentry_kill(dentry); John Ogness