linux-kernel - Re: processes hung after sys_renameat, and 'missing' processes

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:	Mon, 11 Jun 2012 18:02:53 +0200
From:	Miklos Szeredi <miklos@...redi.hu>
To:	Linus Torvalds <torvalds@...ux-foundation.org>
Cc:	Dave Jones <davej@...hat.com>, Al Viro <viro@...iv.linux.org.uk>,
	Linux Kernel <linux-kernel@...r.kernel.org>,
	Jan Kara <jack@...e.cz>
Subject: Re: processes hung after sys_renameat, and 'missing' processes

Linus Torvalds <torvalds@...ux-foundation.org> writes:

> On Thu, Jun 7, 2012 at 12:07 AM, Miklos Szeredi <miklos@...redi.hu> wrote:
>>
>> Yeah, see the comment in below patch for how it's supposed to work.  I
>> *think* it's correct.
>
> Ok, yes, that makes me happier.
>
> What would make me happier still is to get rid of the "save_parent"
> thing entirely, though.
>
> And I think you should be able to.
>
> Why?
>
> You already have the rule that:
>  - save_parent.mnt is always same as "path->mnt" (or NULL, if not saved)
>  - save_parent.dentry is the same as "dir" when you use it (you have a
> BUG_ON() for it not being the same)
>  - you currently use "save_parent.dentry != NULL" as a flag to say "do
> we have the valid state"
>
> So as far as I can tell, you should get rid of all the refcount games
> and the "struct path save_parent", and just replace the
> "save_parent.dentry != NULL" thing with a flag of whether you have a
> valid state.
>
> That would get rid of the whole
>
>         if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
>                 path_to_nameidata(path, nd);
>         } else {
>                 save_parent.dentry = nd->path.dentry;
>                 save_parent.mnt = mntget(path->mnt);
>                 nd->path.dentry = path->dentry;
>
>         }
>
> thing, and we could just have the old simple unconditional
>
>         path_to_nameidata(path, nd);
>
> back.
>
> And then you'd have
>
>         if (filp == ERR_PTR(-EOPENSTALE) && save_parent_flag && !retried) {
>                 dput(nd->path.dentry);
>                 nd->path.dentry = dget(dir);

But 'dir' may no longer be valid here since we dput it in
path_to_nameidata() earlier.

So, unfortunately, we do need to play those refcounting games.

The mntget() could be optimized away in theory, but it's tricky because
complete_walk() might do a path_put(nd->path) on failure and then we'd
be left with a dentry ref without a matching vfsmount ref.

Thanks,
Miklos
--
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/