Re: snd_card_disconnect race (sound/core/init.c)

[Takashi Iwai <tiwai@suse.de>]

At Fri, 25 Mar 2011 10:40:50 -0700,
Russ Dill wrote:
> 
> On Thu, Mar 24, 2011 at 10:23 AM, Russ Dill <russ.dill@gmail.com> wrote:
> > On Thu, Mar 24, 2011 at 1:58 AM, Takashi Iwai <tiwai@suse.de> wrote:
> >> At Thu, 24 Mar 2011 01:19:43 -0700,
> >> Russ Dill wrote:
> >>>
> >>> On Wed, Mar 23, 2011 at 11:43 PM, Takashi Iwai <tiwai@suse.de> wrote:
> >>> > At Wed, 23 Mar 2011 18:40:58 -0700,
> >>> > Russ Dill wrote:
> >>> >>
> >>> >> With slub poisoning on, I'm seeing an oops in snd_disconnect_release
> >>> >> with slub poison (6b6b6b6b). Investigating, snd_card_disconnect looks
> >>> >> to be the possible culprit:
> >>> >>
> >>> >> 333         spin_lock(&card->files_lock);
> >>> >> 334         mfile = card->files;
> >>> >> 335         while (mfile) {
> >>> >> 336                 file = mfile->file;
> >>> >> 337
> >>> >> 338                 /* it's critical part, use endless loop */
> >>> >> 339                 /* we have no room to fail */
> >>> >> 340                 mfile->disconnected_f_op = mfile->file->f_op;
> >>> >> 341
> >>> >> 342                 spin_lock(&shutdown_lock);
> >>> >> 343                 list_add(&mfile->shutdown_list, &shutdown_files);
> >>> >> 344                 spin_unlock(&shutdown_lock);
> >>> >> 345
> >>> >> 346                 mfile->file->f_op = &snd_shutdown_f_ops;
> >>> >> 347                 fops_get(mfile->file->f_op);
> >>> >> 348
> >>> >> 349                 mfile = mfile->next;
> >>> >> 350         }
> >>> >> 351         spin_unlock(&card->files_lock);
> >>> >>
> >>> >> I'm not aware of any locking that would prevent the original release
> >>> >> function running concurrently if it were started before line 346 was
> >>> >> executed. The original release functions (such as snd_hwdep_release)
> >>> >> call snd_card_remove_file which frees the mfile object which would
> >>> >> have been just added to the shutdown_files list leading to a use of
> >>> >> freed (or in my case poisoned) memory later on down the line when the
> >>> >> shutdown_files gets walked again.
> >>> >>
> >>> >> It seems that 118dd6bf (ALSA: Clean up snd_monitor_file management,
> >>> >> v2.6.30) may have either introduced this problem or made it worse
> >>> >> since snd_card_file_remove previously removed items from the
> >>> >> shutdown_files list before freeing them.
> >>> >>
> >>> >> I'm currently experiencing these crashes on 2.6.32.26, but I can't
> >>> >> find any changes that would cause them not to occur on later kernel
> >>> >> versions.
> >>> >
> >>> > Which device are you using?  USB-audio had a known problem at
> >>> > disconnection, for example, and it was fixed recently.
> >>> >
> >>>
> >>> I'm using USB-audio. I tried back-porting a couple of patches, to no avail:
> >>>
> >>> ALSA: usb-audio: fix oops due to cleanup race when disconnecting
> >>> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=382225e6
> >>>
> >>> ALSA: usb - Release capture substream URBs properly
> >>> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=76195fb0
> >>>
> >>> Backing out 118dd6bf fixes the problem. The app that is running when
> >>> this crash occurs does full duplex streaming and segfaults while the
> >>> USB-audio device is going away. I'd have to check, but I think its
> >>> actually segfaulting it response to a disconnect of another piece of
> >>> the composite device the USB-audio device is part of, so the release
> >>> may actually be happening at the time the snd_card_disconnect is
> >>> occurring. I don't see any locking that would prevent bad things from
> >>> happening in this case, and the shutdown_files list is definitely
> >>> poisoned.
> >>
> >> OK, what about the patch below?
> >>
> >>
> >> Takashi
> >>
> >> ---
> >> From: Takashi Iwai <tiwai@suse.de>
> >> Subject: [PATCH] ALSA: Fix yet another race in disconnection
> >>
> >> This patch fixes a race between snd_card_file_remove() and
> >> snd_card_disconnect().  When the card is added to shutdown_files list
> >> in snd_card_disconnect(), but it's freed in snd_card_file_remove() at
> >> the same time, the shutdown_files list gets corrupted.  The list member
> >> must be freed in snd_card_file_remove() as well.
> >>
> >> Reported-by: Russ Dill <russ.dill@gmail.com>
> >> Signed-off-by: Takashi Iwai <tiwai@suse.de>
> >> ---
> >>  sound/core/init.c |    4 ++++
> >>  1 files changed, 4 insertions(+), 0 deletions(-)
> >>
> >> diff --git a/sound/core/init.c b/sound/core/init.c
> >> index 3e65da2..a0080aa 100644
> >> --- a/sound/core/init.c
> >> +++ b/sound/core/init.c
> >> @@ -848,6 +848,7 @@ int snd_card_file_add(struct snd_card *card, struct file *file)
> >>                return -ENOMEM;
> >>        mfile->file = file;
> >>        mfile->disconnected_f_op = NULL;
> >> +       INIT_LIST_HEAD(&mfile->shutdown_list);
> >>        spin_lock(&card->files_lock);
> >>        if (card->shutdown) {
> >>                spin_unlock(&card->files_lock);
> >> @@ -883,6 +884,9 @@ int snd_card_file_remove(struct snd_card *card, struct file *file)
> >>        list_for_each_entry(mfile, &card->files_list, list) {
> >>                if (mfile->file == file) {
> >>                        list_del(&mfile->list);
> >> +                       spin_lock(&shutdown_lock);
> >> +                       list_del(&mfile->shutdown_list);
> >> +                       spin_unlock(&shutdown_lock);
> >>                        if (mfile->disconnected_f_op)
> >>                                fops_put(mfile->disconnected_f_op);
> >>                        found = mfile;
> >> --
> >> 1.7.4.1
> >
> > It looks sane, I'll run it through my torture (about 24hr) testing and
> > let you know.
> >
> 
> Passes testing and looks good! Thanks.

Thanks for testing.  I merged the patch now.


Takashi