2019-10-04 22:28:20

by Bjorn Andersson

Subject: [PATCH v2 5/6] rpmsg: glink: Don't send pending rx_done during remove

Attempting to transmit rx_done messages after the GLINK instance is
being torn down will cause use after free and memory leaks. So cancel
the intent_work and free up the pending intents.

With this there are no concurrent accessors of the channel left during
qcom_glink_native_remove() and there is therefor no need to hold the
spinlock during this operation - which would prohibit the use of
cancel_work_sync() in the release function. So remove this.

Fixes: 1d2ea36eead9 ("rpmsg: glink: Add rx done command")
Cc: [email protected]
Tested-by: Srinivas Kandagatla <[email protected]>
Signed-off-by: Bjorn Andersson <[email protected]>
---

Changes since v1:
- Drop the locking of idr_lock in qcom_glink_native_remove()

drivers/rpmsg/qcom_glink_native.c | 14 ++++++++++++--
1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/drivers/rpmsg/qcom_glink_native.c b/drivers/rpmsg/qcom_glink_native.c
index 89e02baea2d0..4117818db6a1 100644
--- a/drivers/rpmsg/qcom_glink_native.c
+++ b/drivers/rpmsg/qcom_glink_native.c
@@ -241,11 +241,23 @@ static void qcom_glink_channel_release(struct kref *ref)
{
struct glink_channel *channel = container_of(ref, struct glink_channel,
refcount);
+ struct glink_core_rx_intent *intent;
struct glink_core_rx_intent *tmp;
unsigned long flags;
int iid;

+ /* cancel pending rx_done work */
+ cancel_work_sync(&channel->intent_work);
+
spin_lock_irqsave(&channel->intent_lock, flags);
+ /* Free all non-reuse intents pending rx_done work */
+ list_for_each_entry_safe(intent, tmp, &channel->done_intents, node) {
+ if (!intent->reuse) {
+ kfree(intent->data);
+ kfree(intent);
+ }
+ }
+
idr_for_each_entry(&channel->liids, tmp, iid) {
kfree(tmp->data);
kfree(tmp);
@@ -1634,7 +1646,6 @@ void qcom_glink_native_remove(struct qcom_glink *glink)
if (ret)
dev_warn(glink->dev, "Can't remove GLINK devices: %d\n", ret);

- spin_lock_irqsave(&glink->idr_lock, flags);
/* Release any defunct local channels, waiting for close-ack */
idr_for_each_entry(&glink->lcids, channel, cid)
kref_put(&channel->refcount, qcom_glink_channel_release);
@@ -1645,7 +1656,6 @@ void qcom_glink_native_remove(struct qcom_glink *glink)

idr_destroy(&glink->lcids);
idr_destroy(&glink->rcids);
- spin_unlock_irqrestore(&glink->idr_lock, flags);
mbox_free_channel(glink->mbox_chan);
}
EXPORT_SYMBOL_GPL(qcom_glink_native_remove);
--
2.18.0

2019-10-09 01:09:38

by Chris Lew

Subject: Re: [PATCH v2 5/6] rpmsg: glink: Don't send pending rx_done during remove

On 10/4/2019 3:27 PM, Bjorn Andersson wrote:
> Attempting to transmit rx_done messages after the GLINK instance is
> being torn down will cause use after free and memory leaks. So cancel
> the intent_work and free up the pending intents.
>
> With this there are no concurrent accessors of the channel left during
> qcom_glink_native_remove() and there is therefor no need to hold the
> spinlock during this operation - which would prohibit the use of
> cancel_work_sync() in the release function. So remove this.
>
> Fixes: 1d2ea36eead9 ("rpmsg: glink: Add rx done command")
> Cc: [email protected]
> Tested-by: Srinivas Kandagatla <[email protected]>
> Signed-off-by: Bjorn Andersson <[email protected]>
> ---

Acked-By: Chris Lew <[email protected]>

--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project