Some fixes for vhost
cover Some fixes for vhost
Tiwei Bie (3):
vhost: do not realloc device and queues during running
vhost: fix vring addr handling during live migration
vhost: protect vring access done by application
lib/librte_vhost/vhost.c | 53 ++++++++++++++++++++++++++--------- lib/librte_vhost/vhost_user.c | 15 ++++++++-- 2 files changed, 53 insertions(+), 15 deletions(-)
Signature · 1 line
2.17.1
PATCH 1/3 vhost: do not realloc device and queues during running
When the device has been started, don't do the reallocation anymore. Otherwise the pointers used in application threads can be invalidated without proper protection. Instead of introducing a global lock to protect the change of device pointers which will hurt the performance, let's just do the reallocation during setup.
--- lib/librte_vhost/vhost_user.c | 3 +++ 1 file changed, 3 insertions(+)
lib/librte_vhost/vhost_user.c +3 -0
@@ -410,6 +410,9 @@ numa_realloc(struct virtio_net *dev, int index)struct batch_copy_elem *new_batch_copy_elems;int ret; +if (dev->flags &VIRTIO_DEV_RUNNING ) +return dev; + old_dev = dev; vq = old_vq = dev->virtqueue[index];
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH 1/3] vhost: do not realloc device and queues during running
Quoted 6 lines · Tiwei Bie
When the device has been started, don't do the reallocation anymore. Otherwise the pointers used in application threads can be invalidated without proper protection. Instead of introducing a global lock to protect the change of device pointers which will hurt the performance, let's just do the reallocation during setup.
Quoted 8 lines
Fixes af295ad4698c ("vhost: realloc device and queues to same numa node as vring desc")Reported-by Yinan Wang <[email protected]>Signed-off-by Tiwei Bie <[email protected]>--- lib/librte_vhost/vhost_user.c | 3 +++ 1 file changed, 3 insertions(+)
Thanks,
Maxime
PATCH 2/3 vhost: fix vring addr handling during live migration
In reply to Tiwei BieWhen live migration starts, QEMU will set ring addrs again for each virtqueue. In this case, we should try to translate ring addrs after we invalidating the ring, otherwise virtqueues can be enabled with the addrs untranslated. Besides, also leverage the access_ok flag in non-IOMMU case to prevent the data path accessing invalidated virtqueues.
--- lib/librte_vhost/vhost.c | 3 +-- lib/librte_vhost/vhost_user.c | 12 ++++++++++-- 2 files changed, 11 insertions(+), 4 deletions(-)
lib/librte_vhost/vhost.c +1 -2
@@ -358,7 +358,7 @@ vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq) {if (!(dev->features & (1ULL <<VIRTIO_F_IOMMU_PLATFORM ))) - goto out; +return -1 ;if (vq_is_packed (dev)) {if (vring_translate_packed (dev, vq) <0 )
@@ -367,7 +367,6 @@ vring_translate(struct virtio_net *dev, struct vhost_virtqueue *vq)if (vring_translate_split (dev, vq) <0 )return -1 ; } -out: vq->access_ok =1 ;return 0 ;
lib/librte_vhost/vhost_user.c +10 -2
@@ -622,6 +622,7 @@ translate_ring_addresses(struct virtio_net *dev, int vq_index)return dev; } + vq->access_ok =1 ;return dev; }
@@ -680,6 +681,7 @@ translate_ring_addresses(struct virtio_net *dev, int vq_index) } vq->log_guest_addr = addr->log_guest_addr; + vq->access_ok =1 ;VHOST_LOG_DEBUG (VHOST_CONFIG ,"(%d) mapped address desc: %p\n" , dev->vid, vq->desc);
@@ -704,6 +706,7 @@ vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,struct virtio_net *dev = *pdev;struct vhost_virtqueue *vq;struct vhost_vring_addr *addr = &msg->payload.addr; +bool access_ok;if (dev->mem ==NULL )return RTE_VHOST_MSG_RESULT_ERR ;
@@ -711,6 +714,8 @@ vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,/* addr->index refers to the queue index. The txq 1, rxq is 0. */ vq = dev->virtqueue[msg->payload.addr.index]; + access_ok = vq->access_ok; +/* * Rings addresses should not be interpreted as long as the ring is not * started and enabled
@@ -719,8 +724,9 @@ vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,vring_invalidate (dev, vq); -if (vq->enabled && (dev->features & - (1ULL <<VHOST_USER_F_PROTOCOL_FEATURES ))) { +if ((vq->enabled && (dev->features & + (1ULL <<VHOST_USER_F_PROTOCOL_FEATURES ))) || + access_ok) { dev =translate_ring_addresses (dev, msg->payload.addr.index);if (!dev)return RTE_VHOST_MSG_RESULT_ERR ;
@@ -1325,6 +1331,8 @@ vhost_user_get_vring_base(struct virtio_net **pdev, msg->size =sizeof (msg->payload.state); msg->fd_num =0 ; +vring_invalidate (dev, vq); +return RTE_VHOST_MSG_RESULT_REPLY ; }
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH 2/3] vhost: fix vring addr handling during live migration
Quoted 10 lines · Tiwei Bie
When live migration starts, QEMU will set ring addrs again for each virtqueue. In this case, we should try to translate ring addrs after we invalidating the ring, otherwise virtqueues can be enabled with the addrs untranslated. Besides, also leverage the
access_okflag in non-IOMMU case to prevent the data path accessing invalidated virtqueues.Fixes 5a4933e56be4 ("vhost: postpone ring address translations at kick time only")
Quoted 7 lines
Reported-by Yilong Lv <[email protected]>Signed-off-by Tiwei Bie <[email protected]>--- lib/librte_vhost/vhost.c | 3 +-- lib/librte_vhost/vhost_user.c | 12 ++++++++++-- 2 files changed, 11 insertions(+), 4 deletions(-)
Thanks,
Maxime
PATCH 3/3 vhost: protect vring access done by application
In reply to Tiwei BieBesides the enqueue/dequeue API, other APIs of the builtin net backend should also be protected.
--- lib/librte_vhost/vhost.c | 50 +++++++++++++++++++++++++++++++--------- 1 file changed, 39 insertions(+), 11 deletions(-)
lib/librte_vhost/vhost.c +39 -11
@@ -785,22 +785,33 @@ rte_vhost_avail_entries(int vid, uint16_t queue_id) {struct virtio_net *dev;struct vhost_virtqueue *vq; +uint16_t ret =0 ; dev =get_device (vid);if (!dev)return 0 ; vq = dev->virtqueue[queue_id]; -if (!vq->enabled) -return 0 ; -return *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx; +rte_spinlock_lock (&vq->access_lock); + +if (unlikely (!vq->enabled || vq->avail ==NULL )) + goto out; + + ret = *(volatile uint16_t *)&vq->avail->idx - vq->last_used_idx; + +out: +rte_spinlock_unlock (&vq->access_lock); +return ret; } -static inline void +static inline int vhost_enable_notify_split (struct virtio_net *dev,struct vhost_virtqueue *vq,int enable) { +if (vq->used ==NULL ) +return -1 ; +if (!(dev->features & (1ULL <<VIRTIO_RING_F_EVENT_IDX ))) {if (enable) vq->used->flags &= ~VRING_USED_F_NO_NOTIFY ;
@@ -810,17 +821,21 @@ vhost_enable_notify_split(struct virtio_net *dev,if (enable)vhost_avail_event (vq) = vq->last_avail_idx; } +return 0 ; } -static inline void +static inline int vhost_enable_notify_packed (struct virtio_net *dev,struct vhost_virtqueue *vq,int enable) {uint16_t flags; +if (vq->device_event ==NULL ) +return -1 ; +if (!enable) { vq->device_event->flags =VRING_EVENT_F_DISABLE ; -return ; +return 0 ; } flags =VRING_EVENT_F_ENABLE ;
@@ -833,6 +848,7 @@ vhost_enable_notify_packed(struct virtio_net *dev,rte_smp_wmb (); vq->device_event->flags = flags; +return 0 ; }int
@@ -840,18 +856,23 @@ rte_vhost_enable_guest_notification(int vid, uint16_t queue_id, int enable) {struct virtio_net *dev =get_device (vid);struct vhost_virtqueue *vq; +int ret;if (!dev)return -1 ; vq = dev->virtqueue[queue_id]; +rte_spinlock_lock (&vq->access_lock); +if (vq_is_packed (dev)) -vhost_enable_notify_packed (dev, vq, enable); + ret =vhost_enable_notify_packed (dev, vq, enable);else -vhost_enable_notify_split (dev, vq, enable); + ret =vhost_enable_notify_split (dev, vq, enable); -return 0 ; +rte_spinlock_unlock (&vq->access_lock); + +return ret; }void
@@ -890,6 +911,7 @@ rte_vhost_rx_queue_count(int vid, uint16_t qid) {struct virtio_net *dev;struct vhost_virtqueue *vq; +uint32_t ret =0 ; dev =get_device (vid);if (dev ==NULL )
@@ -905,10 +927,16 @@ rte_vhost_rx_queue_count(int vid, uint16_t qid)if (vq ==NULL )return 0 ; +rte_spinlock_lock (&vq->access_lock); +if (unlikely (vq->enabled ==0 || vq->avail ==NULL )) -return 0 ; + goto out; -return *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx; + ret = *((volatile uint16_t *)&vq->avail->idx) - vq->last_avail_idx; + +out: +rte_spinlock_unlock (&vq->access_lock); +return ret; }int rte_vhost_get_vdpa_device_id (int vid)
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH 3/3] vhost: protect vring access done by application
Quoted 12 lines · Tiwei Bie
Besides the enqueue/dequeue API, other APIs of the builtin net backend should also be protected.
Fixes a3688046995f ("vhost: protect active rings from async ring changes")Reported-by Peng He <[email protected]>Signed-off-by Tiwei Bie <[email protected]>--- lib/librte_vhost/vhost.c | 50 +++++++++++++++++++++++++++++++--------- 1 file changed, 39 insertions(+), 11 deletions(-)
Thanks,
Maxime
Maxime Coquelin
AppliedQuoted 9 lines · Tiwei Bie
Tiwei Bie (3):
vhost: do not realloc device and queues during running
vhost: fix vring addr handling during live migration
vhost: protect vring access done by applicationlib/librte_vhost/vhost.c | 53 ++++++++++++++++++++++++++--------- lib/librte_vhost/vhost_user.c | 15 ++++++++-- 2 files changed, 53 insertions(+), 15 deletions(-)
Applied to dpdk-next-virtio/master.
Thanks,
Maxime