linux-kernel - Re: [PATCH] infiniband: fix a subtle race condition

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]

Date:   Thu, 14 Jun 2018 08:34:46 +0300
From:   Leon Romanovsky <leon@...nel.org>
To:     Cong Wang <xiyou.wangcong@...il.com>
Cc:     linux-kernel@...r.kernel.org, linux-rdma@...r.kernel.org,
        Doug Ledford <dledford@...hat.com>,
        Jason Gunthorpe <jgg@...lanox.com>
Subject: Re: [PATCH] infiniband: fix a subtle race condition

On Wed, Jun 13, 2018 at 04:49:47PM -0700, Cong Wang wrote:
> In ucma_event_handler() we lock the mutex like this:
>
> mutex_lock(&ctx->file->mut);
> ...
> mutex_unlock(&ctx->file->mut);
>
> which seems correct, but we could translate it into this:
>
> f = ctx->file;
> mutex_lock(&f->mut);
> ...
> f = ctx->file;
> mutex_unlock(&f->mut);
>
> as the compiler does. And, because ucma_event_handler() is
> called in a workqueue so it could race with ucma_migrate_id(),
> so the following race condition could happen:
>
> CPU0			CPU1
> f = ctx->file;
> 			ucma_lock_files(f, new_file);
> 			ctx->file = new_file
> 			ucma_lock_files(f, new_file);
> mutex_lock(&f->mut); // still the old file!
> ...
> f = ctx->file; // now the new one!!
> mutex_unlock(&f->mut); // unlock new file!
>
> Fix this by reading ctx->file once before mutex_lock(), so we
> won't unlock a different mutex any more.

Hi Cong,

If the compiler optimizes the first line (mutex_lock) as you wrote,
it will reuse "f" for the second line (mutex_unlock) too.

You need to ensure that ucma_modify_id() doesn't run in parallel to
anything that uses "ctx->file" directly and indirectly.

Thanks

Download attachment "signature.asc" of type "application/pgp-signature" (802 bytes)