| Message ID | 20220630020805.74658-1-jasowang@redhat.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | [V2] virtio-net: fix the race between refill work and close | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Guessing tree name failed - patch did not apply |
On Thu, 30 Jun 2022 10:08:04 +0800, Jason Wang <jasowang@redhat.com> wrote: > We try using cancel_delayed_work_sync() to prevent the work from > enabling NAPI. This is insufficient since we don't disable the source > of the refill work scheduling. This means an NAPI poll callback after > cancel_delayed_work_sync() can schedule the refill work then can > re-enable the NAPI that leads to use-after-free [1]. Can you explain in more detail how this happened? napi_disable() is normally called after cancel_delayed_work_sync(). This ensures that all napi callbacks will end, and the new napi_disable() will wait. There will be no re-enable napi. So I guess the use-after-free is caused by refill_work being called after dev/vi/napi is released. In this way, we can just call cancel_delayed_work_sync() after napi_disalbe(). Thanks. > > Since the work can enable NAPI, we can't simply disable NAPI before > calling cancel_delayed_work_sync(). So fix this by introducing a > dedicated boolean to control whether or not the work could be > scheduled from NAPI. > > [1] > ================================================================== > BUG: KASAN: use-after-free in refill_work+0x43/0xd4 > Read of size 2 at addr ffff88810562c92e by task kworker/2:1/42 > > CPU: 2 PID: 42 Comm: kworker/2:1 Not tainted 5.19.0-rc1+ #480 > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014 > Workqueue: events refill_work > Call Trace: > <TASK> > dump_stack_lvl+0x34/0x44 > print_report.cold+0xbb/0x6ac > ? _printk+0xad/0xde > ? refill_work+0x43/0xd4 > kasan_report+0xa8/0x130 > ? refill_work+0x43/0xd4 > refill_work+0x43/0xd4 > process_one_work+0x43d/0x780 > worker_thread+0x2a0/0x6f0 > ? process_one_work+0x780/0x780 > kthread+0x167/0x1a0 > ? kthread_exit+0x50/0x50 > ret_from_fork+0x22/0x30 > </TASK> > ... > > Fixes: b2baed69e605c ("virtio_net: set/cancel work on ndo_open/ndo_stop") > Signed-off-by: Jason Wang <jasowang@redhat.com> > --- > drivers/net/virtio_net.c | 38 ++++++++++++++++++++++++++++++++++++-- > 1 file changed, 36 insertions(+), 2 deletions(-) > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c > index db05b5e930be..21bf1e5c81ef 100644 > --- a/drivers/net/virtio_net.c > +++ b/drivers/net/virtio_net.c > @@ -251,6 +251,12 @@ struct virtnet_info { > /* Does the affinity hint is set for virtqueues? */ > bool affinity_hint_set; > > + /* Is refill work enabled? */ > + bool refill_work_enabled; > + > + /* The lock to synchronize the access to refill_work_enabled */ > + spinlock_t refill_lock; > + > /* CPU hotplug instances for online & dead */ > struct hlist_node node; > struct hlist_node node_dead; > @@ -348,6 +354,20 @@ static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) > return p; > } > > +static void enable_refill_work(struct virtnet_info *vi) > +{ > + spin_lock(&vi->refill_lock); > + vi->refill_work_enabled = true; > + spin_unlock(&vi->refill_lock); > +} > + > +static void disable_refill_work(struct virtnet_info *vi) > +{ > + spin_lock(&vi->refill_lock); > + vi->refill_work_enabled = false; > + spin_unlock(&vi->refill_lock); > +} > + > static void virtqueue_napi_schedule(struct napi_struct *napi, > struct virtqueue *vq) > { > @@ -1527,8 +1547,12 @@ static int virtnet_receive(struct receive_queue *rq, int budget, > } > > if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { > - if (!try_fill_recv(vi, rq, GFP_ATOMIC)) > - schedule_delayed_work(&vi->refill, 0); > + if (!try_fill_recv(vi, rq, GFP_ATOMIC)) { > + spin_lock(&vi->refill_lock); > + if (vi->refill_work_enabled) > + schedule_delayed_work(&vi->refill, 0); > + spin_unlock(&vi->refill_lock); > + } > } > > u64_stats_update_begin(&rq->stats.syncp); > @@ -1651,6 +1675,8 @@ static int virtnet_open(struct net_device *dev) > struct virtnet_info *vi = netdev_priv(dev); > int i, err; > > + enable_refill_work(vi); > + > for (i = 0; i < vi->max_queue_pairs; i++) { > if (i < vi->curr_queue_pairs) > /* Make sure we have some buffers: if oom use wq. */ > @@ -2033,6 +2059,8 @@ static int virtnet_close(struct net_device *dev) > struct virtnet_info *vi = netdev_priv(dev); > int i; > > + /* Make sure NAPI doesn't schedule refill work */ > + disable_refill_work(vi); > /* Make sure refill_work doesn't re-enable napi! */ > cancel_delayed_work_sync(&vi->refill); > > @@ -2776,6 +2804,9 @@ static void virtnet_freeze_down(struct virtio_device *vdev) > netif_tx_lock_bh(vi->dev); > netif_device_detach(vi->dev); > netif_tx_unlock_bh(vi->dev); > + /* Make sure NAPI doesn't schedule refill work */ > + disable_refill_work(vi); > + /* Make sure refill_work doesn't re-enable napi! */ > cancel_delayed_work_sync(&vi->refill); > > if (netif_running(vi->dev)) { > @@ -2799,6 +2830,8 @@ static int virtnet_restore_up(struct virtio_device *vdev) > > virtio_device_ready(vdev); > > + enable_refill_work(vi); > + > if (netif_running(vi->dev)) { > for (i = 0; i < vi->curr_queue_pairs; i++) > if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) > @@ -3548,6 +3581,7 @@ static int virtnet_probe(struct virtio_device *vdev) > vdev->priv = vi; > > INIT_WORK(&vi->config_work, virtnet_config_changed_work); > + spin_lock_init(&vi->refill_lock); > > /* If we can receive ANY GSO packets, we must allocate large ones. */ > if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || > -- > 2.25.1 >
On Thu, 30 Jun 2022 10:08:04 +0800 Jason Wang wrote: > +static void enable_refill_work(struct virtnet_info *vi) > +{ > + spin_lock(&vi->refill_lock); > + vi->refill_work_enabled = true; > + spin_unlock(&vi->refill_lock); > +} > + > +static void disable_refill_work(struct virtnet_info *vi) > +{ > + spin_lock(&vi->refill_lock); > + vi->refill_work_enabled = false; > + spin_unlock(&vi->refill_lock); > +} > + > static void virtqueue_napi_schedule(struct napi_struct *napi, > struct virtqueue *vq) > { > @@ -1527,8 +1547,12 @@ static int virtnet_receive(struct receive_queue *rq, int budget, > } > > if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { > - if (!try_fill_recv(vi, rq, GFP_ATOMIC)) > - schedule_delayed_work(&vi->refill, 0); > + if (!try_fill_recv(vi, rq, GFP_ATOMIC)) { > + spin_lock(&vi->refill_lock); > + if (vi->refill_work_enabled) > + schedule_delayed_work(&vi->refill, 0); > + spin_unlock(&vi->refill_lock); Are you sure you can use the basic spin_lock() flavor in all cases? Isn't the disable/enable called from a different context than this thing here? The entire delayed work construct seems a little risky because the work may go to sleep after disabling napi, causing large latency spikes. I guess you must have a good reason no to simply reschedule the NAPI and keep retrying with GFP_ATOMIC... Please add the target tree name to the subject.
On Thu, Jun 30, 2022 at 10:51 AM Jakub Kicinski <kuba@kernel.org> wrote: > > On Thu, 30 Jun 2022 10:08:04 +0800 Jason Wang wrote: > > +static void enable_refill_work(struct virtnet_info *vi) > > +{ > > + spin_lock(&vi->refill_lock); > > + vi->refill_work_enabled = true; > > + spin_unlock(&vi->refill_lock); > > +} > > + > > +static void disable_refill_work(struct virtnet_info *vi) > > +{ > > + spin_lock(&vi->refill_lock); > > + vi->refill_work_enabled = false; > > + spin_unlock(&vi->refill_lock); > > +} > > + > > static void virtqueue_napi_schedule(struct napi_struct *napi, > > struct virtqueue *vq) > > { > > @@ -1527,8 +1547,12 @@ static int virtnet_receive(struct receive_queue *rq, int budget, > > } > > > > if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { > > - if (!try_fill_recv(vi, rq, GFP_ATOMIC)) > > - schedule_delayed_work(&vi->refill, 0); > > + if (!try_fill_recv(vi, rq, GFP_ATOMIC)) { > > + spin_lock(&vi->refill_lock); > > + if (vi->refill_work_enabled) > > + schedule_delayed_work(&vi->refill, 0); > > + spin_unlock(&vi->refill_lock); > > Are you sure you can use the basic spin_lock() flavor in all cases? > Isn't the disable/enable called from a different context than this > thing here? This function will only be called in bh so it's safe. > > The entire delayed work construct seems a little risky because the work > may go to sleep after disabling napi, causing large latency spikes. Yes, but it only happens on OOM. > I guess you must have a good reason no to simply reschedule the NAPI > and keep retrying with GFP_ATOMIC... Less pressure on the memory allocator on OOM probably, but it looks like an independent issue that might be optimized in the future. > > Please add the target tree name to the subject. Ok Thanks >
On Thu, Jun 30, 2022 at 10:22 AM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote: > > On Thu, 30 Jun 2022 10:08:04 +0800, Jason Wang <jasowang@redhat.com> wrote: > > We try using cancel_delayed_work_sync() to prevent the work from > > enabling NAPI. This is insufficient since we don't disable the source > > of the refill work scheduling. This means an NAPI poll callback after > > cancel_delayed_work_sync() can schedule the refill work then can > > re-enable the NAPI that leads to use-after-free [1]. > > > Can you explain in more detail how this happened? > > napi_disable() is normally called after cancel_delayed_work_sync(). This ensures > that all napi callbacks will end, and the new napi_disable() will wait. > There will be no re-enable napi. An rx interrupt that may come between after the cancel_delayed_work() but before the napi_disable(). It schedules a refill_work that may run after the napi_disable() in virtnet_close(). > > So I guess the use-after-free is caused by refill_work being called after > dev/vi/napi is released. In this way, we can just call > cancel_delayed_work_sync() after napi_disalbe(). So the refill_work can re-enable the NAPI when it is run after napi_disable() in this case. Thanks > > Thanks. > > > > > Since the work can enable NAPI, we can't simply disable NAPI before > > calling cancel_delayed_work_sync(). So fix this by introducing a > > dedicated boolean to control whether or not the work could be > > scheduled from NAPI. > > > > [1] > > ================================================================== > > BUG: KASAN: use-after-free in refill_work+0x43/0xd4 > > Read of size 2 at addr ffff88810562c92e by task kworker/2:1/42 > > > > CPU: 2 PID: 42 Comm: kworker/2:1 Not tainted 5.19.0-rc1+ #480 > > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014 > > Workqueue: events refill_work > > Call Trace: > > <TASK> > > dump_stack_lvl+0x34/0x44 > > print_report.cold+0xbb/0x6ac > > ? _printk+0xad/0xde > > ? refill_work+0x43/0xd4 > > kasan_report+0xa8/0x130 > > ? refill_work+0x43/0xd4 > > refill_work+0x43/0xd4 > > process_one_work+0x43d/0x780 > > worker_thread+0x2a0/0x6f0 > > ? process_one_work+0x780/0x780 > > kthread+0x167/0x1a0 > > ? kthread_exit+0x50/0x50 > > ret_from_fork+0x22/0x30 > > </TASK> > > ... > > > > Fixes: b2baed69e605c ("virtio_net: set/cancel work on ndo_open/ndo_stop") > > Signed-off-by: Jason Wang <jasowang@redhat.com> > > --- > > drivers/net/virtio_net.c | 38 ++++++++++++++++++++++++++++++++++++-- > > 1 file changed, 36 insertions(+), 2 deletions(-) > > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c > > index db05b5e930be..21bf1e5c81ef 100644 > > --- a/drivers/net/virtio_net.c > > +++ b/drivers/net/virtio_net.c > > @@ -251,6 +251,12 @@ struct virtnet_info { > > /* Does the affinity hint is set for virtqueues? */ > > bool affinity_hint_set; > > > > + /* Is refill work enabled? */ > > + bool refill_work_enabled; > > + > > + /* The lock to synchronize the access to refill_work_enabled */ > > + spinlock_t refill_lock; > > + > > /* CPU hotplug instances for online & dead */ > > struct hlist_node node; > > struct hlist_node node_dead; > > @@ -348,6 +354,20 @@ static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) > > return p; > > } > > > > +static void enable_refill_work(struct virtnet_info *vi) > > +{ > > + spin_lock(&vi->refill_lock); > > + vi->refill_work_enabled = true; > > + spin_unlock(&vi->refill_lock); > > +} > > + > > +static void disable_refill_work(struct virtnet_info *vi) > > +{ > > + spin_lock(&vi->refill_lock); > > + vi->refill_work_enabled = false; > > + spin_unlock(&vi->refill_lock); > > +} > > + > > static void virtqueue_napi_schedule(struct napi_struct *napi, > > struct virtqueue *vq) > > { > > @@ -1527,8 +1547,12 @@ static int virtnet_receive(struct receive_queue *rq, int budget, > > } > > > > if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { > > - if (!try_fill_recv(vi, rq, GFP_ATOMIC)) > > - schedule_delayed_work(&vi->refill, 0); > > + if (!try_fill_recv(vi, rq, GFP_ATOMIC)) { > > + spin_lock(&vi->refill_lock); > > + if (vi->refill_work_enabled) > > + schedule_delayed_work(&vi->refill, 0); > > + spin_unlock(&vi->refill_lock); > > + } > > } > > > > u64_stats_update_begin(&rq->stats.syncp); > > @@ -1651,6 +1675,8 @@ static int virtnet_open(struct net_device *dev) > > struct virtnet_info *vi = netdev_priv(dev); > > int i, err; > > > > + enable_refill_work(vi); > > + > > for (i = 0; i < vi->max_queue_pairs; i++) { > > if (i < vi->curr_queue_pairs) > > /* Make sure we have some buffers: if oom use wq. */ > > @@ -2033,6 +2059,8 @@ static int virtnet_close(struct net_device *dev) > > struct virtnet_info *vi = netdev_priv(dev); > > int i; > > > > + /* Make sure NAPI doesn't schedule refill work */ > > + disable_refill_work(vi); > > /* Make sure refill_work doesn't re-enable napi! */ > > cancel_delayed_work_sync(&vi->refill); > > > > @@ -2776,6 +2804,9 @@ static void virtnet_freeze_down(struct virtio_device *vdev) > > netif_tx_lock_bh(vi->dev); > > netif_device_detach(vi->dev); > > netif_tx_unlock_bh(vi->dev); > > + /* Make sure NAPI doesn't schedule refill work */ > > + disable_refill_work(vi); > > + /* Make sure refill_work doesn't re-enable napi! */ > > cancel_delayed_work_sync(&vi->refill); > > > > if (netif_running(vi->dev)) { > > @@ -2799,6 +2830,8 @@ static int virtnet_restore_up(struct virtio_device *vdev) > > > > virtio_device_ready(vdev); > > > > + enable_refill_work(vi); > > + > > if (netif_running(vi->dev)) { > > for (i = 0; i < vi->curr_queue_pairs; i++) > > if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) > > @@ -3548,6 +3581,7 @@ static int virtnet_probe(struct virtio_device *vdev) > > vdev->priv = vi; > > > > INIT_WORK(&vi->config_work, virtnet_config_changed_work); > > + spin_lock_init(&vi->refill_lock); > > > > /* If we can receive ANY GSO packets, we must allocate large ones. */ > > if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || > > -- > > 2.25.1 > > >
On Thu, 30 Jun 2022 14:07:52 +0800, Jason Wang <jasowang@redhat.com> wrote: > On Thu, Jun 30, 2022 at 10:22 AM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote: > > > > On Thu, 30 Jun 2022 10:08:04 +0800, Jason Wang <jasowang@redhat.com> wrote: > > > We try using cancel_delayed_work_sync() to prevent the work from > > > enabling NAPI. This is insufficient since we don't disable the source > > > of the refill work scheduling. This means an NAPI poll callback after > > > cancel_delayed_work_sync() can schedule the refill work then can > > > re-enable the NAPI that leads to use-after-free [1]. > > > > > > Can you explain in more detail how this happened? > > > > napi_disable() is normally called after cancel_delayed_work_sync(). This ensures > > that all napi callbacks will end, and the new napi_disable() will wait. > > There will be no re-enable napi. > > An rx interrupt that may come between after the cancel_delayed_work() > but before the napi_disable(). It schedules a refill_work that may run > after the napi_disable() in virtnet_close(). Yes > > > > > So I guess the use-after-free is caused by refill_work being called after > > dev/vi/napi is released. In this way, we can just call > > cancel_delayed_work_sync() after napi_disalbe(). > > So the refill_work can re-enable the NAPI when it is run after > napi_disable() in this case. Since napi_disable() has been called in virtnet_close(), it will get stuck when napi_disable() in refill_work(). I think use-after-free is because vi/napi etc. have been released, refill_work() going to access again causes an exception. napi will not be re-enable. I would like to call cancel_delayed_work_sync() after napi_disable() to solve this problem. But this also has a problem, refill_work() can get stuck on napi_disable() and cannot exit. In this way, we want napi_disable() to check that the current state is disabled and exit directly. Thanks. > > Thanks > > > > > > Thanks. > > > > > > > > Since the work can enable NAPI, we can't simply disable NAPI before > > > calling cancel_delayed_work_sync(). So fix this by introducing a > > > dedicated boolean to control whether or not the work could be > > > scheduled from NAPI. > > > > > > [1] > > > ================================================================== > > > BUG: KASAN: use-after-free in refill_work+0x43/0xd4 > > > Read of size 2 at addr ffff88810562c92e by task kworker/2:1/42 > > > > > > CPU: 2 PID: 42 Comm: kworker/2:1 Not tainted 5.19.0-rc1+ #480 > > > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014 > > > Workqueue: events refill_work > > > Call Trace: > > > <TASK> > > > dump_stack_lvl+0x34/0x44 > > > print_report.cold+0xbb/0x6ac > > > ? _printk+0xad/0xde > > > ? refill_work+0x43/0xd4 > > > kasan_report+0xa8/0x130 > > > ? refill_work+0x43/0xd4 > > > refill_work+0x43/0xd4 > > > process_one_work+0x43d/0x780 > > > worker_thread+0x2a0/0x6f0 > > > ? process_one_work+0x780/0x780 > > > kthread+0x167/0x1a0 > > > ? kthread_exit+0x50/0x50 > > > ret_from_fork+0x22/0x30 > > > </TASK> > > > ... > > > > > > Fixes: b2baed69e605c ("virtio_net: set/cancel work on ndo_open/ndo_stop") > > > Signed-off-by: Jason Wang <jasowang@redhat.com> > > > --- > > > drivers/net/virtio_net.c | 38 ++++++++++++++++++++++++++++++++++++-- > > > 1 file changed, 36 insertions(+), 2 deletions(-) > > > > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c > > > index db05b5e930be..21bf1e5c81ef 100644 > > > --- a/drivers/net/virtio_net.c > > > +++ b/drivers/net/virtio_net.c > > > @@ -251,6 +251,12 @@ struct virtnet_info { > > > /* Does the affinity hint is set for virtqueues? */ > > > bool affinity_hint_set; > > > > > > + /* Is refill work enabled? */ > > > + bool refill_work_enabled; > > > + > > > + /* The lock to synchronize the access to refill_work_enabled */ > > > + spinlock_t refill_lock; > > > + > > > /* CPU hotplug instances for online & dead */ > > > struct hlist_node node; > > > struct hlist_node node_dead; > > > @@ -348,6 +354,20 @@ static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) > > > return p; > > > } > > > > > > +static void enable_refill_work(struct virtnet_info *vi) > > > +{ > > > + spin_lock(&vi->refill_lock); > > > + vi->refill_work_enabled = true; > > > + spin_unlock(&vi->refill_lock); > > > +} > > > + > > > +static void disable_refill_work(struct virtnet_info *vi) > > > +{ > > > + spin_lock(&vi->refill_lock); > > > + vi->refill_work_enabled = false; > > > + spin_unlock(&vi->refill_lock); > > > +} > > > + > > > static void virtqueue_napi_schedule(struct napi_struct *napi, > > > struct virtqueue *vq) > > > { > > > @@ -1527,8 +1547,12 @@ static int virtnet_receive(struct receive_queue *rq, int budget, > > > } > > > > > > if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { > > > - if (!try_fill_recv(vi, rq, GFP_ATOMIC)) > > > - schedule_delayed_work(&vi->refill, 0); > > > + if (!try_fill_recv(vi, rq, GFP_ATOMIC)) { > > > + spin_lock(&vi->refill_lock); > > > + if (vi->refill_work_enabled) > > > + schedule_delayed_work(&vi->refill, 0); > > > + spin_unlock(&vi->refill_lock); > > > + } > > > } > > > > > > u64_stats_update_begin(&rq->stats.syncp); > > > @@ -1651,6 +1675,8 @@ static int virtnet_open(struct net_device *dev) > > > struct virtnet_info *vi = netdev_priv(dev); > > > int i, err; > > > > > > + enable_refill_work(vi); > > > + > > > for (i = 0; i < vi->max_queue_pairs; i++) { > > > if (i < vi->curr_queue_pairs) > > > /* Make sure we have some buffers: if oom use wq. */ > > > @@ -2033,6 +2059,8 @@ static int virtnet_close(struct net_device *dev) > > > struct virtnet_info *vi = netdev_priv(dev); > > > int i; > > > > > > + /* Make sure NAPI doesn't schedule refill work */ > > > + disable_refill_work(vi); > > > /* Make sure refill_work doesn't re-enable napi! */ > > > cancel_delayed_work_sync(&vi->refill); > > > > > > @@ -2776,6 +2804,9 @@ static void virtnet_freeze_down(struct virtio_device *vdev) > > > netif_tx_lock_bh(vi->dev); > > > netif_device_detach(vi->dev); > > > netif_tx_unlock_bh(vi->dev); > > > + /* Make sure NAPI doesn't schedule refill work */ > > > + disable_refill_work(vi); > > > + /* Make sure refill_work doesn't re-enable napi! */ > > > cancel_delayed_work_sync(&vi->refill); > > > > > > if (netif_running(vi->dev)) { > > > @@ -2799,6 +2830,8 @@ static int virtnet_restore_up(struct virtio_device *vdev) > > > > > > virtio_device_ready(vdev); > > > > > > + enable_refill_work(vi); > > > + > > > if (netif_running(vi->dev)) { > > > for (i = 0; i < vi->curr_queue_pairs; i++) > > > if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) > > > @@ -3548,6 +3581,7 @@ static int virtnet_probe(struct virtio_device *vdev) > > > vdev->priv = vi; > > > > > > INIT_WORK(&vi->config_work, virtnet_config_changed_work); > > > + spin_lock_init(&vi->refill_lock); > > > > > > /* If we can receive ANY GSO packets, we must allocate large ones. */ > > > if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || > > > -- > > > 2.25.1 > > > > > >
On Thu, Jun 30, 2022 at 2:07 PM Jason Wang <jasowang@redhat.com> wrote: > > On Thu, Jun 30, 2022 at 10:51 AM Jakub Kicinski <kuba@kernel.org> wrote: > > > > On Thu, 30 Jun 2022 10:08:04 +0800 Jason Wang wrote: > > > +static void enable_refill_work(struct virtnet_info *vi) > > > +{ > > > + spin_lock(&vi->refill_lock); > > > + vi->refill_work_enabled = true; > > > + spin_unlock(&vi->refill_lock); > > > +} > > > + > > > +static void disable_refill_work(struct virtnet_info *vi) > > > +{ > > > + spin_lock(&vi->refill_lock); > > > + vi->refill_work_enabled = false; > > > + spin_unlock(&vi->refill_lock); > > > +} > > > + > > > static void virtqueue_napi_schedule(struct napi_struct *napi, > > > struct virtqueue *vq) > > > { > > > @@ -1527,8 +1547,12 @@ static int virtnet_receive(struct receive_queue *rq, int budget, > > > } > > > > > > if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { > > > - if (!try_fill_recv(vi, rq, GFP_ATOMIC)) > > > - schedule_delayed_work(&vi->refill, 0); > > > + if (!try_fill_recv(vi, rq, GFP_ATOMIC)) { > > > + spin_lock(&vi->refill_lock); > > > + if (vi->refill_work_enabled) > > > + schedule_delayed_work(&vi->refill, 0); > > > + spin_unlock(&vi->refill_lock); > > > > Are you sure you can use the basic spin_lock() flavor in all cases? > > Isn't the disable/enable called from a different context than this > > thing here? > > This function will only be called in bh so it's safe. Ok, so it looks like we should use the bh variant in close. Otherwise we may have a deadlock. Will fix it. Thanks > > > > > The entire delayed work construct seems a little risky because the work > > may go to sleep after disabling napi, causing large latency spikes. > > Yes, but it only happens on OOM. > > > I guess you must have a good reason no to simply reschedule the NAPI > > and keep retrying with GFP_ATOMIC... > > Less pressure on the memory allocator on OOM probably, but it looks > like an independent issue that might be optimized in the future. > > > > > Please add the target tree name to the subject. > > Ok > > Thanks > > >
On Thu, Jun 30, 2022 at 2:26 PM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote: > > On Thu, 30 Jun 2022 14:07:52 +0800, Jason Wang <jasowang@redhat.com> wrote: > > On Thu, Jun 30, 2022 at 10:22 AM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote: > > > > > > On Thu, 30 Jun 2022 10:08:04 +0800, Jason Wang <jasowang@redhat.com> wrote: > > > > We try using cancel_delayed_work_sync() to prevent the work from > > > > enabling NAPI. This is insufficient since we don't disable the source > > > > of the refill work scheduling. This means an NAPI poll callback after > > > > cancel_delayed_work_sync() can schedule the refill work then can > > > > re-enable the NAPI that leads to use-after-free [1]. > > > > > > > > > Can you explain in more detail how this happened? > > > > > > napi_disable() is normally called after cancel_delayed_work_sync(). This ensures > > > that all napi callbacks will end, and the new napi_disable() will wait. > > > There will be no re-enable napi. > > > > An rx interrupt that may come between after the cancel_delayed_work() > > but before the napi_disable(). It schedules a refill_work that may run > > after the napi_disable() in virtnet_close(). > > Yes > > > > > > > > > > So I guess the use-after-free is caused by refill_work being called after > > > dev/vi/napi is released. In this way, we can just call > > > cancel_delayed_work_sync() after napi_disalbe(). > > > > So the refill_work can re-enable the NAPI when it is run after > > napi_disable() in this case. > > > Since napi_disable() has been called in virtnet_close(), it will get stuck when > napi_disable() in refill_work(). Right because e.g NAPIF_STATE_SCHED has been set by napi_disable() before. > I think use-after-free is because vi/napi etc. > have been released, refill_work() going to access again causes an exception. Yes, this is the use-after-free I mentioned above. > > napi will not be re-enable. > > I would like to call cancel_delayed_work_sync() after napi_disable() > to solve this problem. But this also has a problem, refill_work() can get stuck > on napi_disable() and cannot exit. In this way, we want napi_disable() to check > that the current state is disabled and exit directly. Not sure this is a good design and it doesn't fit for -stable. I think the design of NAPI is to pair napi_enable() and napi_disable() instead of allowing napi_disable() to be called twice. Thanks > > Thanks. > > > > > > Thanks > > > > > > > > > > Thanks. > > > > > > > > > > > Since the work can enable NAPI, we can't simply disable NAPI before > > > > calling cancel_delayed_work_sync(). So fix this by introducing a > > > > dedicated boolean to control whether or not the work could be > > > > scheduled from NAPI. > > > > > > > > [1] > > > > ================================================================== > > > > BUG: KASAN: use-after-free in refill_work+0x43/0xd4 > > > > Read of size 2 at addr ffff88810562c92e by task kworker/2:1/42 > > > > > > > > CPU: 2 PID: 42 Comm: kworker/2:1 Not tainted 5.19.0-rc1+ #480 > > > > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014 > > > > Workqueue: events refill_work > > > > Call Trace: > > > > <TASK> > > > > dump_stack_lvl+0x34/0x44 > > > > print_report.cold+0xbb/0x6ac > > > > ? _printk+0xad/0xde > > > > ? refill_work+0x43/0xd4 > > > > kasan_report+0xa8/0x130 > > > > ? refill_work+0x43/0xd4 > > > > refill_work+0x43/0xd4 > > > > process_one_work+0x43d/0x780 > > > > worker_thread+0x2a0/0x6f0 > > > > ? process_one_work+0x780/0x780 > > > > kthread+0x167/0x1a0 > > > > ? kthread_exit+0x50/0x50 > > > > ret_from_fork+0x22/0x30 > > > > </TASK> > > > > ... > > > > > > > > Fixes: b2baed69e605c ("virtio_net: set/cancel work on ndo_open/ndo_stop") > > > > Signed-off-by: Jason Wang <jasowang@redhat.com> > > > > --- > > > > drivers/net/virtio_net.c | 38 ++++++++++++++++++++++++++++++++++++-- > > > > 1 file changed, 36 insertions(+), 2 deletions(-) > > > > > > > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c > > > > index db05b5e930be..21bf1e5c81ef 100644 > > > > --- a/drivers/net/virtio_net.c > > > > +++ b/drivers/net/virtio_net.c > > > > @@ -251,6 +251,12 @@ struct virtnet_info { > > > > /* Does the affinity hint is set for virtqueues? */ > > > > bool affinity_hint_set; > > > > > > > > + /* Is refill work enabled? */ > > > > + bool refill_work_enabled; > > > > + > > > > + /* The lock to synchronize the access to refill_work_enabled */ > > > > + spinlock_t refill_lock; > > > > + > > > > /* CPU hotplug instances for online & dead */ > > > > struct hlist_node node; > > > > struct hlist_node node_dead; > > > > @@ -348,6 +354,20 @@ static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) > > > > return p; > > > > } > > > > > > > > +static void enable_refill_work(struct virtnet_info *vi) > > > > +{ > > > > + spin_lock(&vi->refill_lock); > > > > + vi->refill_work_enabled = true; > > > > + spin_unlock(&vi->refill_lock); > > > > +} > > > > + > > > > +static void disable_refill_work(struct virtnet_info *vi) > > > > +{ > > > > + spin_lock(&vi->refill_lock); > > > > + vi->refill_work_enabled = false; > > > > + spin_unlock(&vi->refill_lock); > > > > +} > > > > + > > > > static void virtqueue_napi_schedule(struct napi_struct *napi, > > > > struct virtqueue *vq) > > > > { > > > > @@ -1527,8 +1547,12 @@ static int virtnet_receive(struct receive_queue *rq, int budget, > > > > } > > > > > > > > if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { > > > > - if (!try_fill_recv(vi, rq, GFP_ATOMIC)) > > > > - schedule_delayed_work(&vi->refill, 0); > > > > + if (!try_fill_recv(vi, rq, GFP_ATOMIC)) { > > > > + spin_lock(&vi->refill_lock); > > > > + if (vi->refill_work_enabled) > > > > + schedule_delayed_work(&vi->refill, 0); > > > > + spin_unlock(&vi->refill_lock); > > > > + } > > > > } > > > > > > > > u64_stats_update_begin(&rq->stats.syncp); > > > > @@ -1651,6 +1675,8 @@ static int virtnet_open(struct net_device *dev) > > > > struct virtnet_info *vi = netdev_priv(dev); > > > > int i, err; > > > > > > > > + enable_refill_work(vi); > > > > + > > > > for (i = 0; i < vi->max_queue_pairs; i++) { > > > > if (i < vi->curr_queue_pairs) > > > > /* Make sure we have some buffers: if oom use wq. */ > > > > @@ -2033,6 +2059,8 @@ static int virtnet_close(struct net_device *dev) > > > > struct virtnet_info *vi = netdev_priv(dev); > > > > int i; > > > > > > > > + /* Make sure NAPI doesn't schedule refill work */ > > > > + disable_refill_work(vi); > > > > /* Make sure refill_work doesn't re-enable napi! */ > > > > cancel_delayed_work_sync(&vi->refill); > > > > > > > > @@ -2776,6 +2804,9 @@ static void virtnet_freeze_down(struct virtio_device *vdev) > > > > netif_tx_lock_bh(vi->dev); > > > > netif_device_detach(vi->dev); > > > > netif_tx_unlock_bh(vi->dev); > > > > + /* Make sure NAPI doesn't schedule refill work */ > > > > + disable_refill_work(vi); > > > > + /* Make sure refill_work doesn't re-enable napi! */ > > > > cancel_delayed_work_sync(&vi->refill); > > > > > > > > if (netif_running(vi->dev)) { > > > > @@ -2799,6 +2830,8 @@ static int virtnet_restore_up(struct virtio_device *vdev) > > > > > > > > virtio_device_ready(vdev); > > > > > > > > + enable_refill_work(vi); > > > > + > > > > if (netif_running(vi->dev)) { > > > > for (i = 0; i < vi->curr_queue_pairs; i++) > > > > if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) > > > > @@ -3548,6 +3581,7 @@ static int virtnet_probe(struct virtio_device *vdev) > > > > vdev->priv = vi; > > > > > > > > INIT_WORK(&vi->config_work, virtnet_config_changed_work); > > > > + spin_lock_init(&vi->refill_lock); > > > > > > > > /* If we can receive ANY GSO packets, we must allocate large ones. */ > > > > if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || > > > > -- > > > > 2.25.1 > > > > > > > > > >
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index db05b5e930be..21bf1e5c81ef 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -251,6 +251,12 @@ struct virtnet_info { /* Does the affinity hint is set for virtqueues? */ bool affinity_hint_set; + /* Is refill work enabled? */ + bool refill_work_enabled; + + /* The lock to synchronize the access to refill_work_enabled */ + spinlock_t refill_lock; + /* CPU hotplug instances for online & dead */ struct hlist_node node; struct hlist_node node_dead; @@ -348,6 +354,20 @@ static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) return p; } +static void enable_refill_work(struct virtnet_info *vi) +{ + spin_lock(&vi->refill_lock); + vi->refill_work_enabled = true; + spin_unlock(&vi->refill_lock); +} + +static void disable_refill_work(struct virtnet_info *vi) +{ + spin_lock(&vi->refill_lock); + vi->refill_work_enabled = false; + spin_unlock(&vi->refill_lock); +} + static void virtqueue_napi_schedule(struct napi_struct *napi, struct virtqueue *vq) { @@ -1527,8 +1547,12 @@ static int virtnet_receive(struct receive_queue *rq, int budget, } if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { - if (!try_fill_recv(vi, rq, GFP_ATOMIC)) - schedule_delayed_work(&vi->refill, 0); + if (!try_fill_recv(vi, rq, GFP_ATOMIC)) { + spin_lock(&vi->refill_lock); + if (vi->refill_work_enabled) + schedule_delayed_work(&vi->refill, 0); + spin_unlock(&vi->refill_lock); + } } u64_stats_update_begin(&rq->stats.syncp); @@ -1651,6 +1675,8 @@ static int virtnet_open(struct net_device *dev) struct virtnet_info *vi = netdev_priv(dev); int i, err; + enable_refill_work(vi); + for (i = 0; i < vi->max_queue_pairs; i++) { if (i < vi->curr_queue_pairs) /* Make sure we have some buffers: if oom use wq. */ @@ -2033,6 +2059,8 @@ static int virtnet_close(struct net_device *dev) struct virtnet_info *vi = netdev_priv(dev); int i; + /* Make sure NAPI doesn't schedule refill work */ + disable_refill_work(vi); /* Make sure refill_work doesn't re-enable napi! */ cancel_delayed_work_sync(&vi->refill); @@ -2776,6 +2804,9 @@ static void virtnet_freeze_down(struct virtio_device *vdev) netif_tx_lock_bh(vi->dev); netif_device_detach(vi->dev); netif_tx_unlock_bh(vi->dev); + /* Make sure NAPI doesn't schedule refill work */ + disable_refill_work(vi); + /* Make sure refill_work doesn't re-enable napi! */ cancel_delayed_work_sync(&vi->refill); if (netif_running(vi->dev)) { @@ -2799,6 +2830,8 @@ static int virtnet_restore_up(struct virtio_device *vdev) virtio_device_ready(vdev); + enable_refill_work(vi); + if (netif_running(vi->dev)) { for (i = 0; i < vi->curr_queue_pairs; i++) if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) @@ -3548,6 +3581,7 @@ static int virtnet_probe(struct virtio_device *vdev) vdev->priv = vi; INIT_WORK(&vi->config_work, virtnet_config_changed_work); + spin_lock_init(&vi->refill_lock); /* If we can receive ANY GSO packets, we must allocate large ones. */ if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
We try using cancel_delayed_work_sync() to prevent the work from enabling NAPI. This is insufficient since we don't disable the source of the refill work scheduling. This means an NAPI poll callback after cancel_delayed_work_sync() can schedule the refill work then can re-enable the NAPI that leads to use-after-free [1]. Since the work can enable NAPI, we can't simply disable NAPI before calling cancel_delayed_work_sync(). So fix this by introducing a dedicated boolean to control whether or not the work could be scheduled from NAPI. [1] ================================================================== BUG: KASAN: use-after-free in refill_work+0x43/0xd4 Read of size 2 at addr ffff88810562c92e by task kworker/2:1/42 CPU: 2 PID: 42 Comm: kworker/2:1 Not tainted 5.19.0-rc1+ #480 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014 Workqueue: events refill_work Call Trace: <TASK> dump_stack_lvl+0x34/0x44 print_report.cold+0xbb/0x6ac ? _printk+0xad/0xde ? refill_work+0x43/0xd4 kasan_report+0xa8/0x130 ? refill_work+0x43/0xd4 refill_work+0x43/0xd4 process_one_work+0x43d/0x780 worker_thread+0x2a0/0x6f0 ? process_one_work+0x780/0x780 kthread+0x167/0x1a0 ? kthread_exit+0x50/0x50 ret_from_fork+0x22/0x30 </TASK> ... Fixes: b2baed69e605c ("virtio_net: set/cancel work on ndo_open/ndo_stop") Signed-off-by: Jason Wang <jasowang@redhat.com> --- drivers/net/virtio_net.c | 38 ++++++++++++++++++++++++++++++++++++-- 1 file changed, 36 insertions(+), 2 deletions(-)