From: Hans Verkuil <hverkuil+cisco@kernel.org>
To: Sangyun Kim <sangyun.kim@snu.ac.kr>, Mike Isely <isely@pobox.com>,
Mauro Carvalho Chehab <mchehab@kernel.org>,
Hans Verkuil <hverkuil@kernel.org>,
Edward Adam Davis <eadavis@qq.com>
Cc: linux-media@vger.kernel.org, linux-kernel@vger.kernel.org,
stable@vger.kernel.org
Subject: Re: [PATCH] media: pvrusb2: fix disconnect and teardown races
Date: Tue, 5 May 2026 15:14:17 +0200 [thread overview]
Message-ID: <d5f15e0c-b406-4940-81e0-3c957086ba77@kernel.org> (raw)
In-Reply-To: <20260420060621.1627352-1-sangyun.kim@snu.ac.kr>
On 4/20/26 08:06, Sangyun Kim wrote:
> pvr2_context_disconnect() queues a notification to the pvrusb2-context
> kthread before it stores mp->disconnect_flag:
>
> pvr2_hdw_disconnect(mp->hdw);
> if (!pvr2_context_shutok())
> pvr2_context_notify(mp);
> mp->disconnect_flag = !0;
>
> The context thread only destroys a context when disconnect_flag is set
> and mc_first is NULL. If the notification wakes the thread before the
> flag store becomes visible, the thread dequeues the context, runs
> pvr2_context_check() with disconnect_flag still observed as 0, decides
> that the destroy condition is not met yet, and goes back to sleep.
> Nothing wakes the thread again once the flag is finally stored, so the
> context stays on the global exist list forever and
> pvr2_context_global_done() blocks module unload. commit 0a0b79ea55de
> ("media: pvrusb2: fix uaf in pvr2_context_set_notify") made this
> liveness failure easier to hit by moving the notify earlier in the
> disconnect path.
>
> The same teardown sequence still contains a use-after-free.
> pvr2_context_exit() inspects disconnect_flag after releasing mp->mutex
> and may then call pvr2_context_notify(mp) after the context thread has
> already freed the object via pvr2_context_destroy()/kfree(). The hdw
> completion callback registered through pvr2_hdw_initialize() can race
> the same way. Reordering the disconnect path alone closes the unload
> hang, but it still leaves late notifiers able to touch freed memory.
>
> Fix both problems together:
>
> - Split pvr2_context_set_notify() into a locked helper
> (pvr2_context_set_notify_locked()) and a wrapper that acquires
> pvr2_context_mutex. This lets callers update several pieces of
> related state inside a single critical section without relocking.
>
> - In pvr2_context_disconnect(), set disconnect_flag and enqueue the
> thread notification under pvr2_context_mutex. The context thread
> manipulates the notify list under the same mutex, so when it observes
> the queued entry it is guaranteed to observe disconnect_flag = 1 as
> well and the destroy condition evaluates correctly. This eliminates
> the original notify-before-flag liveness hole.
>
> - Add a per-context refcount_t. pvr2_context_create() initialises it to
> 1 (creator reference). pvr2_channel_init() and pvr2_channel_done()
> take and drop a reference around each channel's lifetime.
> pvr2_context_disconnect() takes a temporary reference across its body
> so the context cannot be freed while disconnect is still touching it.
> pvr2_context_destroy() no longer calls kfree() directly; it drops its
> reference via pvr2_context_put(), and whichever caller drops the last
> reference performs the actual kfree. This keeps the object alive
> until disconnect and the final channel teardown finish, regardless of
> how the context thread, channel close, and USB disconnect paths
> interleave.
>
> - Add a destroying_flag that pvr2_context_destroy() sets under
> pvr2_context_mutex before unlinking the context from the notify and
> exist lists. pvr2_context_set_notify_locked() refuses to re-enqueue a
> context whose destroying_flag is set, so a late notifier arriving
> after destroy has started cannot resurrect the context on the notify
> list. The dequeue path (fl == 0) still proceeds unconditionally
> because destroy itself must be able to remove any still-queued entry.
>
> - Update pvr2_context_exit() to enqueue through
> pvr2_context_set_notify_locked() after releasing mp->mutex. The
> caller (channel close or disconnect) always holds a reference, so the
> object is stable across the mp->mutex / pvr2_context_mutex hand-off
> and a concurrent destroy cannot free it under us. If destroy has
> already won the race, destroying_flag short-circuits the enqueue into
> a no-op.
>
> Lock ordering: pvr2_context_mutex is only acquired after mp->mutex is
> released; no path holds pvr2_context_mutex while acquiring mp->mutex,
> so no AB/BA deadlock is introduced. wake_up() on
> pvr2_context_sync_data is moved outside pvr2_context_mutex in every
> path that grew a new locked section, matching the existing style.
Huge commit log, weird patch. All the hallmarks of AI slop.
Rejected.
Regards,
Hans
>
> Fixes: 0a0b79ea55de ("media: pvrusb2: fix uaf in pvr2_context_set_notify")
> Cc: stable@vger.kernel.org
> Signed-off-by: Sangyun Kim <sangyun.kim@snu.ac.kr>
> ---
> drivers/media/usb/pvrusb2/pvrusb2-context.c | 56 ++++++++++++++++++---
> drivers/media/usb/pvrusb2/pvrusb2-context.h | 3 ++
> 2 files changed, 51 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/media/usb/pvrusb2/pvrusb2-context.c b/drivers/media/usb/pvrusb2/pvrusb2-context.c
> index 93f5da65ead9..fb9bdbf5c886 100644
> --- a/drivers/media/usb/pvrusb2/pvrusb2-context.c
> +++ b/drivers/media/usb/pvrusb2/pvrusb2-context.c
> @@ -27,11 +27,19 @@ static int pvr2_context_cleaned_flag;
> static struct task_struct *pvr2_context_thread_ptr;
>
>
> -static void pvr2_context_set_notify(struct pvr2_context *mp, int fl)
> +static void pvr2_context_put(struct pvr2_context *mp)
> +{
> + if (refcount_dec_and_test(&mp->refcount))
> + kfree(mp);
> +}
> +
> +static int pvr2_context_set_notify_locked(struct pvr2_context *mp, int fl)
> {
> int signal_flag = 0;
> - mutex_lock(&pvr2_context_mutex);
> +
> if (fl) {
> + if (mp->destroying_flag)
> + return 0;
> if (!mp->notify_flag) {
> signal_flag = (pvr2_context_notify_first == NULL);
> mp->notify_prev = pvr2_context_notify_last;
> @@ -59,6 +67,15 @@ static void pvr2_context_set_notify(struct pvr2_context *mp, int fl)
> }
> }
> }
> + return signal_flag;
> +}
> +
> +static void pvr2_context_set_notify(struct pvr2_context *mp, int fl)
> +{
> + int signal_flag = 0;
> +
> + mutex_lock(&pvr2_context_mutex);
> + signal_flag = pvr2_context_set_notify_locked(mp, fl);
> mutex_unlock(&pvr2_context_mutex);
> if (signal_flag) wake_up(&pvr2_context_sync_data);
> }
> @@ -66,10 +83,13 @@ static void pvr2_context_set_notify(struct pvr2_context *mp, int fl)
>
> static void pvr2_context_destroy(struct pvr2_context *mp)
> {
> + int signal_flag = 0;
> +
> pvr2_trace(PVR2_TRACE_CTXT,"pvr2_context %p (destroy)",mp);
> pvr2_hdw_destroy(mp->hdw);
> - pvr2_context_set_notify(mp, 0);
> mutex_lock(&pvr2_context_mutex);
> + mp->destroying_flag = !0;
> + pvr2_context_set_notify_locked(mp, 0);
> if (mp->exist_next) {
> mp->exist_next->exist_prev = mp->exist_prev;
> } else {
> @@ -83,10 +103,12 @@ static void pvr2_context_destroy(struct pvr2_context *mp)
> if (!pvr2_context_exist_first) {
> /* Trigger wakeup on control thread in case it is waiting
> for an exit condition. */
> - wake_up(&pvr2_context_sync_data);
> + signal_flag = !0;
> }
> mutex_unlock(&pvr2_context_mutex);
> - kfree(mp);
> + if (signal_flag)
> + wake_up(&pvr2_context_sync_data);
> + pvr2_context_put(mp);
> }
>
>
> @@ -209,6 +231,7 @@ struct pvr2_context *pvr2_context_create(
> pvr2_trace(PVR2_TRACE_CTXT,"pvr2_context %p (create)",mp);
> mp->setup_func = setup_func;
> mutex_init(&mp->mutex);
> + refcount_set(&mp->refcount, 1);
> mutex_lock(&pvr2_context_mutex);
> mp->exist_prev = pvr2_context_exist_last;
> mp->exist_next = NULL;
> @@ -256,25 +279,41 @@ static void pvr2_context_enter(struct pvr2_context *mp)
> static void pvr2_context_exit(struct pvr2_context *mp)
> {
> int destroy_flag = 0;
> + int signal_flag = 0;
> if (!(mp->mc_first || !mp->disconnect_flag)) {
> destroy_flag = !0;
> }
> mutex_unlock(&mp->mutex);
> - if (destroy_flag) pvr2_context_notify(mp);
> + if (destroy_flag) {
> + mutex_lock(&pvr2_context_mutex);
> + signal_flag = pvr2_context_set_notify_locked(mp, !0);
> + mutex_unlock(&pvr2_context_mutex);
> + if (signal_flag)
> + wake_up(&pvr2_context_sync_data);
> + }
> }
>
>
> void pvr2_context_disconnect(struct pvr2_context *mp)
> {
> + int signal_flag = 0;
> +
> + refcount_inc(&mp->refcount);
> pvr2_hdw_disconnect(mp->hdw);
> - if (!pvr2_context_shutok())
> - pvr2_context_notify(mp);
> + mutex_lock(&pvr2_context_mutex);
> mp->disconnect_flag = !0;
> + if (!pvr2_context_shutok())
> + signal_flag = pvr2_context_set_notify_locked(mp, !0);
> + mutex_unlock(&pvr2_context_mutex);
> + if (signal_flag)
> + wake_up(&pvr2_context_sync_data);
> + pvr2_context_put(mp);
> }
>
>
> void pvr2_channel_init(struct pvr2_channel *cp,struct pvr2_context *mp)
> {
> + refcount_inc(&mp->refcount);
> pvr2_context_enter(mp);
> cp->hdw = mp->hdw;
> cp->mc_head = mp;
> @@ -318,6 +357,7 @@ void pvr2_channel_done(struct pvr2_channel *cp)
> }
> cp->hdw = NULL;
> pvr2_context_exit(mp);
> + pvr2_context_put(mp);
> }
>
>
> diff --git a/drivers/media/usb/pvrusb2/pvrusb2-context.h b/drivers/media/usb/pvrusb2/pvrusb2-context.h
> index 5840b2ce8f1e..4e06530eccb8 100644
> --- a/drivers/media/usb/pvrusb2/pvrusb2-context.h
> +++ b/drivers/media/usb/pvrusb2/pvrusb2-context.h
> @@ -7,6 +7,7 @@
> #define __PVRUSB2_CONTEXT_H
>
> #include <linux/mutex.h>
> +#include <linux/refcount.h>
> #include <linux/usb.h>
> #include <linux/workqueue.h>
>
> @@ -33,9 +34,11 @@ struct pvr2_context {
> struct pvr2_hdw *hdw;
> struct pvr2_context_stream video_stream;
> struct mutex mutex;
> + refcount_t refcount;
> int notify_flag;
> int initialized_flag;
> int disconnect_flag;
> + int destroying_flag;
>
> /* Called after pvr2_context initialization is complete */
> void (*setup_func)(struct pvr2_context *);
prev parent reply other threads:[~2026-05-05 13:14 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-04-20 6:06 [PATCH] media: pvrusb2: fix disconnect and teardown races Sangyun Kim
2026-05-05 13:14 ` Hans Verkuil [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=d5f15e0c-b406-4940-81e0-3c957086ba77@kernel.org \
--to=hverkuil+cisco@kernel.org \
--cc=eadavis@qq.com \
--cc=hverkuil@kernel.org \
--cc=isely@pobox.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-media@vger.kernel.org \
--cc=mchehab@kernel.org \
--cc=sangyun.kim@snu.ac.kr \
--cc=stable@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox