From: Al Viro <viro@zeniv.linux.org.uk>
To: "Tobin C. Harding" <me@tobin.cc>
Cc: "Tobin C. Harding" <tobin@kernel.org>,
Andrew Morton <akpm@linux-foundation.org>,
Roman Gushchin <guro@fb.com>,
Alexander Viro <viro@ftp.linux.org.uk>,
Christoph Hellwig <hch@infradead.org>,
Pekka Enberg <penberg@cs.helsinki.fi>,
David Rientjes <rientjes@google.com>,
Joonsoo Kim <iamjoonsoo.kim@lge.com>,
Christopher Lameter <cl@linux.com>,
Matthew Wilcox <willy@infradead.org>,
Miklos Szeredi <mszeredi@redhat.com>,
Andreas Dilger <adilger@dilger.ca>,
Waiman Long <longman@redhat.com>, Tycho Andersen <tycho@tycho.ws>,
Theodore Ts'o <tytso@mit.edu>, Andi Kleen <ak@linux.intel.com>,
David Chinner <david@fromorbit.com>,
Nick Piggin <npiggin@gmail.com>, Rik van Riel <riel@redhat.com>,
Hugh Dickins <hughd@google.com>, Jonathan Corbet <corbet@lwn.net>,
linux-mm@kvack.org, linux-fsdevel@vger.kernel.org,
linux-kernel@vger.kernel.org,
Linus Torvalds <torvalds@linux-foundation.org>
Subject: Re: [RFC PATCH v3 14/15] dcache: Implement partial shrink via Slab Movable Objects
Date: Sat, 29 Jun 2019 05:08:44 +0100 [thread overview]
Message-ID: <20190629040844.GS17978@ZenIV.linux.org.uk> (raw)
In-Reply-To: <20190411210200.GH2217@ZenIV.linux.org.uk>
On Thu, Apr 11, 2019 at 10:02:00PM +0100, Al Viro wrote:
> Aaaarrgghhh... No, we can't. Look: we get one candidate dentry in isolate
> phase. We put it into shrink list. umount(2) comes and calls
> shrink_dcache_for_umount(), which calls shrink_dcache_parent(root).
> In the meanwhile, shrink_dentry_list() is run and does __dentry_kill() on
> that one dentry. Fine, it's gone - before shrink_dcache_parent() even
> sees it. Now shrink_dentry_list() holds a reference to its parent and
> is about to drop it in
> dentry = parent;
> while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
> dentry = dentry_kill(dentry);
> And dropped it will be, but... shrink_dcache_parent() has finished the
> scan, without finding *anything* with zero refcount - the thing that used
> to be on the shrink list was already gone before shrink_dcache_parent()
> has gotten there and the reference to parent was not dropped yet. So
> shrink_dcache_for_umount() plows past shrink_dcache_parent(), walks the
> tree and complains loudly about "busy" dentries (that parent we hadn't
> finished dropping), and then we proceed with filesystem shutdown.
> In the meanwhile, dentry_kill() finally gets to killing dentry and
> triggers an unexpected late call of ->d_iput() on a filesystem that
> has already been far enough into shutdown - far enough to destroy the
> data structures needed for that sucker.
>
> The reason we don't hit that problem with regular memory shrinker is
> this:
> unregister_shrinker(&s->s_shrink);
> fs->kill_sb(s);
> in deactivate_locked_super(). IOW, shrinker for this fs is gone
> before we get around to shutdown. And so are all normal sources
> of dentry eviction for that fs.
>
> Your earlier variants all suffer the same problem - picking a page
> shared by dentries from several superblocks can run into trouble
> if it overlaps with umount of one of those.
FWIW, I think I see a kinda-sorta sane solution. Namely, add
static void __dput_to_list(struct dentry *dentry, struct list_head *list)
{
if (dentry->d_flags & DCACHE_SHRINK_LIST) {
/* let the owner of the list it's on deal with it */
--dentry->d_lockref.count;
} else {
if (dentry->d_flags & DCACHE_LRU_LIST)
d_lru_del(dentry);
if (!--dentry->d_lockref.count)
d_shrink_add(parent, list);
}
}
and have
shrink_dentry_list() do this in the end of loop:
d_shrink_del(dentry);
parent = dentry->d_parent;
/* both dentry and parent are locked at that point */
if (parent != dentry) {
/*
* We need to prune ancestors too. This is necessary to
* prevent quadratic behavior of shrink_dcache_parent(),
* but is also expected to be beneficial in reducing
* dentry cache fragmentation.
*/
__dput_to_list(parent, list);
}
__dentry_kill(dentry);
}
instead of
d_shrink_del(dentry);
parent = dentry->d_parent;
__dentry_kill(dentry);
if (parent == dentry)
continue;
/*
* We need to prune ancestors too. This is necessary to prevent
* quadratic behavior of shrink_dcache_parent(), but is also
* expected to be beneficial in reducing dentry cache
* fragmentation.
*/
dentry = parent;
while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
dentry = dentry_kill(dentry);
}
we have there now. Linus, do you see any problems with that change? AFAICS,
that should avoid the problem described above. Moreover, it seems to allow
a fun API addition:
void dput_to_list(struct dentry *dentry, struct list_head *list)
{
rcu_read_lock();
if (likely(fast_dput(dentry))) {
rcu_read_unlock();
return;
}
rcu_read_unlock();
if (!retain_dentry(dentry))
__dput_to_list(dentry, list);
spin_unlock(&dentry->d_lock);
}
allowing to take an empty list, do a bunch of dput_to_list() (under spinlocks,
etc.), then, once we are in better locking conditions, shrink_dentry_list()
to take them all out. I can see applications for that in e.g. fs/namespace.c -
quite a bit of kludges with ->mnt_ex_mountpoint would be killable that way,
and there would be a chance to transfer the contribution to ->d_count of
mountpoint from struct mount to struct mountpoint (i.e. make any number of
mounts on the same mountpoint dentry contribute only 1 to its ->d_count,
not the number of such mounts).
next prev parent reply other threads:[~2019-06-29 4:09 UTC|newest]
Thread overview: 28+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-04-11 1:34 [RFC PATCH v3 00/15] Slab Movable Objects (SMO) Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 01/15] slub: Add isolate() and migrate() methods Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 02/15] tools/vm/slabinfo: Add support for -C and -M options Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 03/15] slub: Sort slab cache list Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 04/15] slub: Slab defrag core Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 05/15] tools/vm/slabinfo: Add remote node defrag ratio output Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 06/15] tools/vm/slabinfo: Add defrag_used_ratio output Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 07/15] tools/testing/slab: Add object migration test module Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 08/15] tools/testing/slab: Add object migration test suite Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 09/15] xarray: Implement migration function for objects Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 10/15] tools/testing/slab: Add XArray movable objects tests Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 11/15] slub: Enable moving objects to/from specific nodes Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 12/15] slub: Enable balancing slabs across nodes Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 13/15] dcache: Provide a dentry constructor Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 14/15] dcache: Implement partial shrink via Slab Movable Objects Tobin C. Harding
2019-04-11 2:33 ` Al Viro
2019-04-11 2:48 ` Tobin C. Harding
2019-04-11 4:47 ` Al Viro
2019-04-11 5:05 ` Tobin C. Harding
2019-04-11 20:01 ` Al Viro
2019-04-11 21:02 ` Al Viro
2019-06-29 4:08 ` Al Viro [this message]
2019-06-29 4:38 ` shrink_dentry_list() logics change (was Re: [RFC PATCH v3 14/15] dcache: Implement partial shrink via Slab Movable Objects) Al Viro
2019-06-29 19:06 ` Al Viro
2019-06-29 22:29 ` Al Viro
2019-06-29 22:34 ` Al Viro
2019-07-01 9:26 ` Tobin C. Harding
2019-04-11 1:34 ` [RFC PATCH v3 15/15] dcache: Add CONFIG_DCACHE_SMO Tobin C. Harding
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20190629040844.GS17978@ZenIV.linux.org.uk \
--to=viro@zeniv.linux.org.uk \
--cc=adilger@dilger.ca \
--cc=ak@linux.intel.com \
--cc=akpm@linux-foundation.org \
--cc=cl@linux.com \
--cc=corbet@lwn.net \
--cc=david@fromorbit.com \
--cc=guro@fb.com \
--cc=hch@infradead.org \
--cc=hughd@google.com \
--cc=iamjoonsoo.kim@lge.com \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=longman@redhat.com \
--cc=me@tobin.cc \
--cc=mszeredi@redhat.com \
--cc=npiggin@gmail.com \
--cc=penberg@cs.helsinki.fi \
--cc=riel@redhat.com \
--cc=rientjes@google.com \
--cc=tobin@kernel.org \
--cc=torvalds@linux-foundation.org \
--cc=tycho@tycho.ws \
--cc=tytso@mit.edu \
--cc=viro@ftp.linux.org.uk \
--cc=willy@infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).