linux-kernel - Re: [PATCH] fs: avoid locking sb_lock in grab_super

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:	Tue, 24 Feb 2015 13:41:22 +0300
From:	Konstantin Khlebnikov <khlebnikov@...dex-team.ru>
To:	Al Viro <viro@...IV.linux.org.uk>,
	Andrew Morton <akpm@...ux-foundation.org>
CC:	linux-mm@...ck.org, linux-kernel@...r.kernel.org,
	linux-fsdevel@...r.kernel.org, Dave Chinner <david@...morbit.com>
Subject: Re: [PATCH] fs: avoid locking sb_lock in grab_super_passive()

On 21.02.2015 02:50, Al Viro wrote:
> On Fri, Feb 20, 2015 at 03:07:31PM -0800, Andrew Morton wrote:
>
>> - It no longer "acquires a reference".  All it does is to acquire an rwsem.
>>
>> - What the heck is a "passive reference" anyway?  It appears to be
>>    the situation where we increment s_count without incrementing s_active.
>
> Reference to struct super_block that guarantees only that its memory won't
> be freed until we drop it.
>
>>    After your patch, this superblock state no longer exists(?),
>
> Yes, it does.  The _only_ reason why that patch isn't outright bogus is that
> we do only down_read_trylock() on ->s_umount - try to pull off the same thing
> with down_read() and you'll get a nasty race.

I don't get this. What the problem with down_read(sb->s_umount)?

For grab_super_passive()/trylock_super() caller guarantees memory
wouldn't be freed and we check tsb activeness after grabbing shared
lock. And while we hold that lock it'll stay active.

It have to use down_read_trylock() just because it works in in atomic
context when writeback calls it. No?

Check for activeness actually is a quite confusing.
It seems checking for MS_BORN and MS_ACTIVE should be enough:

  bool trylock_super(struct super_block *sb)
  {
         if (down_read_trylock(&sb->s_umount)) {
-               if (!hlist_unhashed(&sb->s_instances) &&
-                   sb->s_root && (sb->s_flags & MS_BORN))
+               if ((sb->s_flags & MS_BORN) && (sb->s_flags & MS_ACTIVE))
                         return true;
                 up_read(&sb->s_umount);
         }

> Take a look at e.g.
> get_super().  Or user_get_super().  Or iterate_supers()/iterate_supers_type(),
> where we don't return such references, but pass them to a callback instead.
> In all those cases we end up with passive reference taken, ->s_umount
> taken shared (_NOT_ with trylock) and fs checked for being still alive.
> Then it's guaranteed to stay alive until we do drop_super().
>
> I agree that the name blows, BTW - something like try_get_super() might have
> been more descriptive, but with this change it actually becomes a bad name
> as well, since after it we need a different way to release the obtained ref;
> not the same as after get_super().  Your variant might be OK, but I'd
> probably make it trylock_super(), to match the verb-object order of the
> rest of identifiers in that area...
>
>> so
>>    perhaps the entire "passive reference" concept and any references to
>>    it can be expunged from the kernel.
>
> Nope.
>


-- 
Konstantin
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/