mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH] misc: vmw_vmci: wait for callbacks in event unsubscribe
@ 2026-10-27 16:58 Greg Kroah-Hartman
  0 siblings, 0 replies; only message in thread
From: Greg Kroah-Hartman @ 2026-10-27 16:58 UTC (permalink / raw)
  To: bryan-bt.tan, vishnu.dasa
  Cc: linux-kernel, Josef Bacik, stable,
	Broadcom internal kernel review list, Arnd Bergmann,
	Greg Kroah-Hartman

From: Josef Bacik <josef@toxicpanda.com>

event_deliver() calls the callbacks of the subscribers inside an RCU read
side critical section. vmci_event_unsubscribe() used to call
synchronize_rcu() before it freed the subscription. This did more than
to protect the struct vmci_subscription: when the function returned, no
callback was running any more, and its callers rely on that.

vmci_transport_free_resources() unsubscribes from
VMCI_EVENT_QP_PEER_DETACH and frees the struct vmci_transport, which is
the callback data, a few lines later:

  CPU0                                CPU1
  event_deliver()
    rcu_read_lock()
    vmci_transport_peer_detach_cb()
                                      vmci_event_unsubscribe()
                                        list_del_rcu(&s->node)
                                        kvfree_rcu_mightsleep(s)
                                      kfree(transport)
      spin_lock_bh(&trans->lock)

The event is sent when the peer of the queue pair detaches, so the other
end of a vsock connection chooses the time. vmci_transport_exit() and
vmci_guest_remove_device() unsubscribe callbacks which are gone with
their module.

kvfree_rcu_mightsleep() only waits for a grace period if it cannot
allocate memory. Bring the synchronize_rcu() back. The cleanup work of
the vmci transport now waits for one grace period per socket again.

Fixes: 6da3f33770e0 ("misc: vmw_vmci: Switch to kvfree_rcu() API")
Cc: stable <stable@kernel.org>
Cc: Bryan Tan <bryan-bt.tan@broadcom.com>
Cc: Vishnu Dasa <vishnu.dasa@broadcom.com>
Cc: Broadcom internal kernel review list <bcm-kernel-feedback-list@broadcom.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
---
 drivers/misc/vmw_vmci/vmci_event.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/misc/vmw_vmci/vmci_event.c b/drivers/misc/vmw_vmci/vmci_event.c
index c3ef3b98c432..57d3a63678ac 100644
--- a/drivers/misc/vmw_vmci/vmci_event.c
+++ b/drivers/misc/vmw_vmci/vmci_event.c
@@ -213,7 +213,12 @@ int vmci_event_unsubscribe(u32 sub_id)
 	if (!s)
 		return VMCI_ERROR_NOT_FOUND;
 
-	kvfree_rcu_mightsleep(s);
+	/*
+	 * Wait for event_deliver(): the caller is going to free what the
+	 * callback works on.
+	 */
+	synchronize_rcu();
+	kfree(s);
 
 	return VMCI_SUCCESS;
 }
-- 
2.56.0


^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2026-10-27 16:58 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-10-27 16:58 [PATCH] misc: vmw_vmci: wait for callbacks in event unsubscribe Greg Kroah-Hartman

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®