USB: cdc-acm: fix racy tty buffer accesses

A recent change that started reporting break events to the line
discipline caused the tty-buffer insertions to no longer be serialised
by inserting events also from the completion handler for the interrupt
endpoint.

Completion calls for distinct endpoints are not guaranteed to be
serialised. For example, in case a host-controller driver uses
bottom-half completion, the interrupt and bulk-in completion handlers
can end up running in parallel on two CPUs (high-and low-prio tasklets,
respectively) thereby breaking the tty layer's single producer
assumption.

Fix this by holding the read lock also when inserting characters from
the bulk endpoint.

Fixes: fe2ccf0b6859 ("cdc-acm: fix BREAK rx code path adding necessary calls")
Cc: stable@vger.kernel.org
Acked-by: Oliver Neukum <oneukum@suse.com>
Signed-off-by: Johan Hovold <johan@kernel.org>
Link: https://lore.kernel.org/r/20210929090937.7410-2-johan@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/drivers/usb/class/cdc-acm.c b/drivers/usb/class/cdc-acm.c
index 4e2f1552f4b78386eae859abb2571087490f5750..c7a1736720e71b4f9c2f29632d213f95ce530e08 100644 (file)
--- a/drivers/usb/class/cdc-acm.c
+++ b/drivers/usb/class/cdc-acm.c
@@ -475,11 +475,16 @@ static int acm_submit_read_urbs(struct acm *acm, gfp_t mem_flags)
 
 static void acm_process_read_urb(struct acm *acm, struct urb *urb)
 {
+       unsigned long flags;
+
        if (!urb->actual_length)
                return;
 
+       spin_lock_irqsave(&acm->read_lock, flags);
        tty_insert_flip_string(&acm->port, urb->transfer_buffer,
                        urb->actual_length);
+       spin_unlock_irqrestore(&acm->read_lock, flags);
+
        tty_flip_buffer_push(&acm->port);
 }