On Mon, 2021-11-01 at 17:00 -0400, Michael S. Tsirkin wrote:
> On Mon, Nov 01, 2021 at 04:38:13PM +0800, Xueming Li wrote:
> > When vhost-user device cleanup is executed and un-mmaps notifier
> > address, VM cpu thread writing the notifier fails by accessing invalid
> > address error.
> >
> > To avoid this concurrent issue, call RCU and wait for a memory flatview
> > update, then un-mmap notifiers in callback.
> >
> > Fixes: 44866521bd6e ("vhost-user: support registering external host
> > notifiers")
> > Cc: qemu-sta...@nongnu.org
> > Cc: Yuwei Zhang <zhangyuwei.9...@bytedance.com>
> > Signed-off-by: Xueming Li <xuemi...@nvidia.com>
> > ---
> > hw/virtio/vhost-user.c | 50 +++++++++++++++++++++-------------
> > include/hw/virtio/vhost-user.h | 2 ++
> > 2 files changed, 33 insertions(+), 19 deletions(-)
> >
> > diff --git a/hw/virtio/vhost-user.c b/hw/virtio/vhost-user.c
> > index c671719e9b..5adad4d029 100644
> > --- a/hw/virtio/vhost-user.c
> > +++ b/hw/virtio/vhost-user.c
> > @@ -25,6 +25,7 @@
> > #include "migration/migration.h"
> > #include "migration/postcopy-ram.h"
> > #include "trace.h"
> > +#include "exec/ramblock.h"
> >
> > #include <sys/ioctl.h>
> > #include <sys/socket.h>
> > @@ -1143,15 +1144,27 @@ static int vhost_user_set_vring_num(struct
> > vhost_dev *dev,
> > return vhost_set_vring(dev, VHOST_USER_SET_VRING_NUM, ring);
> > }
> >
> > -static void vhost_user_host_notifier_remove(struct vhost_dev *dev,
> > - int queue_idx)
> > +static void vhost_user_host_notifier_free(VhostUserHostNotifier *n)
> > {
> > - struct vhost_user *u = dev->opaque;
> > - VhostUserHostNotifier *n = &u->user->notifier[queue_idx];
> > - VirtIODevice *vdev = dev->vdev;
> > + assert(n && n->old_addr);
> > + munmap(n->old_addr, qemu_real_host_page_size);
> > + n->old_addr = NULL;
> > +}
> > +
> > +static void vhost_user_host_notifier_remove(VhostUserState *user,
> > + VirtIODevice *vdev, int
> > queue_idx)
> > +{
> > + VhostUserHostNotifier *n = &user->notifier[queue_idx];
> >
> > if (n->addr) {
> > - virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, false);
> > + if (vdev) {
> > + virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr,
> > false);
> > + }
> > + assert(n->addr);
> > + assert(!n->old_addr);
> > + n->old_addr = n->addr;
> > + n->addr = NULL;
> > + call_rcu(n, vhost_user_host_notifier_free, rcu);
> > }
> > }
> >
> > @@ -1190,8 +1203,9 @@ static int vhost_user_get_vring_base(struct vhost_dev
> > *dev,
> > .payload.state = *ring,
> > .hdr.size = sizeof(msg.payload.state),
> > };
> > + struct vhost_user *u = dev->opaque;
> >
> > - vhost_user_host_notifier_remove(dev, ring->index);
> > + vhost_user_host_notifier_remove(u->user, dev->vdev, ring->index);
> >
> > if (vhost_user_write(dev, &msg, NULL, 0) < 0) {
> > return -1;
> > @@ -1486,12 +1500,7 @@ static int
> > vhost_user_slave_handle_vring_host_notifier(struct vhost_dev *dev,
> >
> > n = &user->notifier[queue_idx];
> >
> > - if (n->addr) {
> > - virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, false);
> > - object_unparent(OBJECT(&n->mr));
> > - munmap(n->addr, page_size);
> > - n->addr = NULL;
> > - }
> > + vhost_user_host_notifier_remove(user, vdev, queue_idx);
> >
> > if (area->u64 & VHOST_USER_VRING_NOFD_MASK) {
> > return 0;
> > @@ -1510,9 +1519,12 @@ static int
> > vhost_user_slave_handle_vring_host_notifier(struct vhost_dev *dev,
> >
> > name = g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
> > user, queue_idx);
> > - if (!n->mr.ram) /* Don't init again after suspend. */
> > + if (!n->mr.ram) { /* Don't init again after suspend. */
> > memory_region_init_ram_device_ptr(&n->mr, OBJECT(vdev), name,
> > page_size, addr);
> > + } else {
> > + n->mr.ram_block->host = addr;
> > + }
> > g_free(name);
> >
> > if (virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, true)) {
> > @@ -2460,17 +2472,17 @@ bool vhost_user_init(VhostUserState *user,
> > CharBackend *chr, Error **errp)
> > void vhost_user_cleanup(VhostUserState *user)
> > {
> > int i;
> > + VhostUserHostNotifier *n;
> >
> > if (!user->chr) {
> > return;
> > }
> > memory_region_transaction_begin();
> > for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
> > - if (user->notifier[i].addr) {
> > - object_unparent(OBJECT(&user->notifier[i].mr));
> > - munmap(user->notifier[i].addr, qemu_real_host_page_size);
> > - user->notifier[i].addr = NULL;
> > - }
> > + n = &user->notifier[i];
> > + assert(!n->addr);
>
> I'm pretty confused as to why this assert holds.
> Add a comment?
Seems notifiers are removed by vhost_user_get_vring_base(), I used this
assert to catch exception. Will remove it.
>
> > + vhost_user_host_notifier_remove(user, NULL, i);
> > + object_unparent(OBJECT(&n->mr));
> > }
> > memory_region_transaction_commit();
> > user->chr = NULL;
>
> I'm also confused on why we can do unparent for notifiers which have
> never been set up. Won't n->mr be invalid then?
There is a parent check in object_unparent().
>
>
> > diff --git a/include/hw/virtio/vhost-user.h b/include/hw/virtio/vhost-user.h
> > index f6012b2078..03aa22d450 100644
> > --- a/include/hw/virtio/vhost-user.h
> > +++ b/include/hw/virtio/vhost-user.h
> > @@ -12,8 +12,10 @@
> > #include "hw/virtio/virtio.h"
> >
> > typedef struct VhostUserHostNotifier {
> > + struct rcu_head rcu;
> > MemoryRegion mr;
> > void *addr;
> > + void *old_addr;
>
> That's not a very clear name. Is this literally just
> "address for the rcu callback to unmap"?
> Maybe unmap_addr then?
LGTM, thanks!
>
> > } VhostUserHostNotifier;
> >
> > typedef struct VhostUserState {
> > --
> > 2.33.0
>
