vfio: Fix virqfd release race

vfoi-pci supports a mechanism like KVM's irqfd for unmasking an
interrupt through an eventfd.  There are two ways to shutdown this
interface: 1) close the eventfd, 2) ioctl (such as disabling the
interrupt).  Both of these do the release through a workqueue,
which can result in a segfault if two jobs get queued for the same
virqfd.

Fix this by protecting the pointer to these virqfds by a spinlock.
The vfio pci device will therefore no longer have a reference to it
once the release job is queued under lock.  On the ioctl side, we
still flush the workqueue to ensure that any outstanding releases
are completed.

Signed-off-by: Alex Williamson <alex.williamson@redhat.com>

1 files changed, 56 insertions, 20 deletions

diff --git a/drivers/vfio/pci/vfio_pci_intrs.c b/drivers/vfio/pci/vfio_pci_intrs.c
index 211a4920b88a..d8dedc7d3910 100644
--- a/drivers/vfio/pci/vfio_pci_intrs.c
+++ b/drivers/vfio/pci/vfio_pci_intrs.c
@@ -76,9 +76,24 @@ static int virqfd_wakeup(wait_queue_t *wait, unsigned mode, int sync, void *key)
                         schedule_work(&virqfd->inject);
         }
 
-        if (flags & POLLHUP)
-                /* The eventfd is closing, detach from VFIO */
-                virqfd_deactivate(virqfd);
+        if (flags & POLLHUP) {
+                unsigned long flags;
+                spin_lock_irqsave(&virqfd->vdev->irqlock, flags);
+
+                /*
+                 * The eventfd is closing, if the virqfd has not yet been
+                 * queued for release, as determined by testing whether the
+                 * vdev pointer to it is still valid, queue it now.  As
+                 * with kvm irqfds, we know we won't race against the virqfd
+                 * going away because we hold wqh->lock to get here.
+                 */
+                if (*(virqfd->pvirqfd) == virqfd) {
+                        *(virqfd->pvirqfd) = NULL;
+                        virqfd_deactivate(virqfd);
+                }
+
+                spin_unlock_irqrestore(&virqfd->vdev->irqlock, flags);
+        }
 
         return 0;
 }
@@ -93,7 +108,6 @@ static void virqfd_ptable_queue_proc(struct file *file,
 static void virqfd_shutdown(struct work_struct *work)
 {
         struct virqfd *virqfd = container_of(work, struct virqfd, shutdown);
-        struct virqfd **pvirqfd = virqfd->pvirqfd;
         u64 cnt;
 
         eventfd_ctx_remove_wait_queue(virqfd->eventfd, &virqfd->wait, &cnt);
@@ -101,7 +115,6 @@ static void virqfd_shutdown(struct work_struct *work)
         eventfd_ctx_put(virqfd->eventfd);
 
         kfree(virqfd);
-        *pvirqfd = NULL;
 }
 
 static void virqfd_inject(struct work_struct *work)
@@ -122,15 +135,11 @@ static int virqfd_enable(struct vfio_pci_device *vdev,
         int ret = 0;
         unsigned int events;
 
-        if (*pvirqfd)
-                return -EBUSY;
-
         virqfd = kzalloc(sizeof(*virqfd), GFP_KERNEL);
         if (!virqfd)
                 return -ENOMEM;
 
         virqfd->pvirqfd = pvirqfd;
-        *pvirqfd = virqfd;
         virqfd->vdev = vdev;
         virqfd->handler = handler;
         virqfd->thread = thread;
@@ -154,6 +163,23 @@ static int virqfd_enable(struct vfio_pci_device *vdev,
         virqfd->eventfd = ctx;
 
         /*
+         * virqfds can be released by closing the eventfd or directly
+         * through ioctl.  These are both done through a workqueue, so
+         * we update the pointer to the virqfd under lock to avoid
+         * pushing multiple jobs to release the same virqfd.
+         */
+        spin_lock_irq(&vdev->irqlock);
+
+        if (*pvirqfd) {
+                spin_unlock_irq(&vdev->irqlock);
+                ret = -EBUSY;
+                goto fail;
+        }
+        *pvirqfd = virqfd;
+
+        spin_unlock_irq(&vdev->irqlock);
+
+        /*
          * Install our own custom wake-up handling so we are notified via
          * a callback whenever someone signals the underlying eventfd.
          */
@@ -187,19 +213,29 @@ fail:
                 fput(file);
 
         kfree(virqfd);
-        *pvirqfd = NULL;
 
         return ret;
 }
 
-static void virqfd_disable(struct virqfd *virqfd)
+static void virqfd_disable(struct vfio_pci_device *vdev,
+                           struct virqfd **pvirqfd)
 {
-        if (!virqfd)
-                return;
+        unsigned long flags;
+
+        spin_lock_irqsave(&vdev->irqlock, flags);
+
+        if (*pvirqfd) {
+                virqfd_deactivate(*pvirqfd);
+                *pvirqfd = NULL;
+        }
 
-        virqfd_deactivate(virqfd);
+        spin_unlock_irqrestore(&vdev->irqlock, flags);
 
-        /* Block until we know all outstanding shutdown jobs have completed. */
+        /*
+         * Block until we know all outstanding shutdown jobs have completed.
+         * Even if we don't queue the job, flush the wq to be sure it's
+         * been released.
+         */
         flush_workqueue(vfio_irqfd_cleanup_wq);
 }
 
@@ -392,8 +428,8 @@ static int vfio_intx_set_signal(struct vfio_pci_device *vdev, int fd)
 static void vfio_intx_disable(struct vfio_pci_device *vdev)
 {
         vfio_intx_set_signal(vdev, -1);
-        virqfd_disable(vdev->ctx[0].unmask);
-        virqfd_disable(vdev->ctx[0].mask);
+        virqfd_disable(vdev, &vdev->ctx[0].unmask);
+        virqfd_disable(vdev, &vdev->ctx[0].mask);
         vdev->irq_type = VFIO_PCI_NUM_IRQS;
         vdev->num_ctx = 0;
         kfree(vdev->ctx);
@@ -539,8 +575,8 @@ static void vfio_msi_disable(struct vfio_pci_device *vdev, bool msix)
         vfio_msi_set_block(vdev, 0, vdev->num_ctx, NULL, msix);
 
         for (i = 0; i < vdev->num_ctx; i++) {
-                virqfd_disable(vdev->ctx[i].unmask);
-                virqfd_disable(vdev->ctx[i].mask);
+                virqfd_disable(vdev, &vdev->ctx[i].unmask);
+                virqfd_disable(vdev, &vdev->ctx[i].mask);
         }
 
         if (msix) {
@@ -577,7 +613,7 @@ static int vfio_pci_set_intx_unmask(struct vfio_pci_device *vdev,
                                              vfio_send_intx_eventfd, NULL,
                                              &vdev->ctx[0].unmask, fd);
 
-                virqfd_disable(vdev->ctx[0].unmask);
+                virqfd_disable(vdev, &vdev->ctx[0].unmask);
         }
 
         return 0;