Before setting channel->rescind in vmbus_rescind_cleanup(), we should make
sure the channel callback won't run any more, otherwise a high-level
driver like pci_hyperv, which may be infinitely waiting for the host VSP's
response and notices the channel has been rescinded, can't safely give
up: e.g., in hv_pci_protocol_negotiation() -> wait_for_response(), it's
unsafe to exit from wait_for_response() and proceed with the on-stack
variable "comp_pkt" popped. The issue was originally spotted by
Michael Kelley <[email protected]>.
In vmbus_close_internal(), the patch also minimizes the range protected by
disabling/enabling channel->callback_event: we don't really need that for
the whole function.
Signed-off-by: Dexuan Cui <[email protected]>
Cc: [email protected]
Cc: K. Y. Srinivasan <[email protected]>
Cc: Stephen Hemminger <[email protected]>
Cc: Michael Kelley <[email protected]>
---
drivers/hv/channel.c | 40 ++++++++++++++++++++++++----------------
drivers/hv/channel_mgmt.c | 6 ++++++
include/linux/hyperv.h | 2 ++
3 files changed, 32 insertions(+), 16 deletions(-)
diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
index ba0a092..c394922 100644
--- a/drivers/hv/channel.c
+++ b/drivers/hv/channel.c
@@ -558,11 +558,8 @@ static void reset_channel_cb(void *arg)
channel->onchannel_callback = NULL;
}
-static int vmbus_close_internal(struct vmbus_channel *channel)
+void vmbus_reset_channel_cb(struct vmbus_channel *channel)
{
- struct vmbus_channel_close_channel *msg;
- int ret;
-
/*
* vmbus_on_event(), running in the per-channel tasklet, can race
* with vmbus_close_internal() in the case of SMP guest, e.g., when
@@ -572,6 +569,29 @@ static int vmbus_close_internal(struct vmbus_channel *channel)
*/
tasklet_disable(&channel->callback_event);
+ channel->sc_creation_callback = NULL;
+
+ /* Stop the callback asap */
+ if (channel->target_cpu != get_cpu()) {
+ put_cpu();
+ smp_call_function_single(channel->target_cpu, reset_channel_cb,
+ channel, true);
+ } else {
+ reset_channel_cb(channel);
+ put_cpu();
+ }
+
+ /* Re-enable tasklet for use on re-open */
+ tasklet_enable(&channel->callback_event);
+}
+
+static int vmbus_close_internal(struct vmbus_channel *channel)
+{
+ struct vmbus_channel_close_channel *msg;
+ int ret;
+
+ vmbus_reset_channel_cb(channel);
+
/*
* In case a device driver's probe() fails (e.g.,
* util_probe() -> vmbus_open() returns -ENOMEM) and the device is
@@ -585,16 +605,6 @@ static int vmbus_close_internal(struct vmbus_channel *channel)
}
channel->state = CHANNEL_OPEN_STATE;
- channel->sc_creation_callback = NULL;
- /* Stop callback and cancel the timer asap */
- if (channel->target_cpu != get_cpu()) {
- put_cpu();
- smp_call_function_single(channel->target_cpu, reset_channel_cb,
- channel, true);
- } else {
- reset_channel_cb(channel);
- put_cpu();
- }
/* Send a closing message */
@@ -639,8 +649,6 @@ static int vmbus_close_internal(struct vmbus_channel *channel)
get_order(channel->ringbuffer_pagecount * PAGE_SIZE));
out:
- /* re-enable tasklet for use on re-open */
- tasklet_enable(&channel->callback_event);
return ret;
}
diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c
index ecc2bd2..9536b93 100644
--- a/drivers/hv/channel_mgmt.c
+++ b/drivers/hv/channel_mgmt.c
@@ -895,6 +895,12 @@ static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
}
/*
+ * Before setting channel->rescind in vmbus_rescind_cleanup(), we
+ * should make sure the channel callback is not running any more.
+ */
+ vmbus_reset_channel_cb(channel);
+
+ /*
* Now wait for offer handling to complete.
*/
vmbus_rescind_cleanup(channel);
diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
index 3a3012f..5389012 100644
--- a/include/linux/hyperv.h
+++ b/include/linux/hyperv.h
@@ -1046,6 +1046,8 @@ extern int vmbus_establish_gpadl(struct vmbus_channel *channel,
extern int vmbus_teardown_gpadl(struct vmbus_channel *channel,
u32 gpadl_handle);
+void vmbus_reset_channel_cb(struct vmbus_channel *channel);
+
extern int vmbus_recvpacket(struct vmbus_channel *channel,
void *buffer,
u32 bufferlen,
--
2.7.4
From: Dexuan Cui Sent: Thursday, July 12, 2018 10:53 PM
>
> Before setting channel->rescind in vmbus_rescind_cleanup(), we should make
> sure the channel callback won't run any more, otherwise a high-level
> driver like pci_hyperv, which may be infinitely waiting for the host VSP's
> response and notices the channel has been rescinded, can't safely give
> up: e.g., in hv_pci_protocol_negotiation() -> wait_for_response(), it's
> unsafe to exit from wait_for_response() and proceed with the on-stack
> variable "comp_pkt" popped. The issue was originally spotted by
> Michael Kelley <[email protected]>.
>
> In vmbus_close_internal(), the patch also minimizes the range protected by
> disabling/enabling channel->callback_event: we don't really need that for
> the whole function.
>
> Signed-off-by: Dexuan Cui <[email protected]>
> Cc: [email protected]
> Cc: K. Y. Srinivasan <[email protected]>
> Cc: Stephen Hemminger <[email protected]>
> Cc: Michael Kelley <[email protected]>
Reviewed-by: Michael Kelley <[email protected]>