2023-03-13 16:55:31

by Zheng Wang

[permalink] [raw]
Subject: [PATCH net v2] 9p/xen : Fix use after free bug in xen_9pfs_front_remove due to race condition

In xen_9pfs_front_probe, it calls xen_9pfs_front_alloc_dataring
to init priv->rings and bound &ring->work with p9_xen_response.

When it calls xen_9pfs_front_event_handler to handle IRQ requests,
it will finally call schedule_work to start the work.

When we call xen_9pfs_front_remove to remove the driver, there
may be a sequence as follows:

Fix it by finishing the work before cleanup in xen_9pfs_front_free.

Note that, this bug is found by static analysis, which might be
false positive.

CPU0 CPU1

|p9_xen_response
xen_9pfs_front_remove|
xen_9pfs_front_free|
kfree(priv) |
//free priv |
|p9_tag_lookup
|//use priv->client

Fixes: 71ebd71921e4 ("xen/9pfs: connect to the backend")
Signed-off-by: Zheng Wang <[email protected]>
---
v2:
- fix type error of ring found by kernel test robot
---
net/9p/trans_xen.c | 5 +++++
1 file changed, 5 insertions(+)

diff --git a/net/9p/trans_xen.c b/net/9p/trans_xen.c
index c64050e839ac..83764431c066 100644
--- a/net/9p/trans_xen.c
+++ b/net/9p/trans_xen.c
@@ -274,12 +274,17 @@ static const struct xenbus_device_id xen_9pfs_front_ids[] = {
static void xen_9pfs_front_free(struct xen_9pfs_front_priv *priv)
{
int i, j;
+ struct xen_9pfs_dataring *ring = NULL;

write_lock(&xen_9pfs_lock);
list_del(&priv->list);
write_unlock(&xen_9pfs_lock);

for (i = 0; i < priv->num_rings; i++) {
+ /*cancel work*/
+ ring = &priv->rings[i];
+ cancel_work_sync(&ring->work);
+
if (!priv->rings[i].intf)
break;
if (priv->rings[i].irq > 0)
--
2.25.1