linux-kernel - Re: [PATCH 0/9] [RFC v2] safely drop directory dentry on failed revalidate

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

Message-ID: <20130821054055.GN27005@ZenIV.linux.org.uk>
Date:	Wed, 21 Aug 2013 06:40:56 +0100
From:	Al Viro <viro@...IV.linux.org.uk>
To:	Miklos Szeredi <miklos@...redi.hu>
Cc:	rwheeler@...hat.com, avati@...hat.com, bfoster@...hat.com,
	dhowells@...hat.com, eparis@...hat.com,
	linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
	mszeredi@...e.cz, KONISHI Ryusuke <konishi.ryusuke@....ntt.co.jp>,
	Ian Kent <raven@...maw.net>
Subject: Re: [PATCH 0/9] [RFC v2] safely drop directory dentry on failed
 revalidate

On Thu, Aug 08, 2013 at 05:24:41PM +0200, Miklos Szeredi wrote:
> Here's a series for fixing issues with d_drop on a directory dentry with
> children and adding support for such dropped directories in fuse.
> 
> This one fixes a couple of issues I found with the previous incarnation and
> split out the filesystem parts into individual patches.

OK in general, but I'm not happy with the proliferation of such iterators in
fs/dcache.c ;-/  

We have have_submounts(), select_parent() and d_genocide(), with one more
such sucker added to the pile.  Sure, most of the callers of have_submounts()
are gone after that patchset, but we still have several left:
fs/autofs4/dev-ioctl.c:542:             err = have_submounts(path.dentry);
fs/autofs4/root.c:381:                  if (have_submounts(dentry)) {
fs/autofs4/waitq.c:338:         if (have_submounts(dentry))
fs/nilfs2/super.c:1010: if (have_submounts(root_dentry))

The thing is, the same race with mount() does, AFAICS, apply at least to
autofs4 call sites.

	And I would like to understand what nilfs one is trying to do...
Unless I'm seriously misreading that code, it's *not* on any kind of a hot
path, so I really wonder why don't we simply do shrink_dcache_parent() +
check if d_count has dropped to 1, without trying to look for submounts
first - if we have any, shrink_dcache_parent() is simply going to leave
us with d_count > 1 and that's it.  Actually, it's cheaper that way -
no need to walk the tree twice.  Moreover, checking for d_count == 1
case first is also pointless - in that case we have no children at all
and shrink_dcache_parent() will return immediately.  Could nilfs folks
explain why (and if) we need to bother with all that complexity there?

	We are really getting too many tree walkers in fs/dcache.c and
all that duplication is the prime breeding ground for bugs ;-/
--
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/