[PATCH AUTOSEL 5.17 101/149] mctp: make __mctp_dev_get() take a refcount hold

From: Sasha Levin
Date: Fri Apr 01 2022 - 10:51:51 EST


From: Matt Johnston <matt@xxxxxxxxxxxxxxxxxxxx>

[ Upstream commit dc121c0084910db985cf1c8ba6fce5d8c307cc02 ]

Previously there was a race that could allow the mctp_dev refcount
to hit zero:

rcu_read_lock();
mdev = __mctp_dev_get(dev);
// mctp_unregister() happens here, mdev->refs hits zero
mctp_dev_hold(dev);
rcu_read_unlock();

Now we make __mctp_dev_get() take the hold itself. It is safe to test
against the zero refcount because __mctp_dev_get() is called holding
rcu_read_lock and mctp_dev uses kfree_rcu().

Reported-by: Jakub Kicinski <kuba@xxxxxxxxxx>
Signed-off-by: Matt Johnston <matt@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: David S. Miller <davem@xxxxxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
net/mctp/device.c | 21 ++++++++++++++++++---
net/mctp/route.c | 5 ++++-
net/mctp/test/utils.c | 1 -
3 files changed, 22 insertions(+), 5 deletions(-)

diff --git a/net/mctp/device.c b/net/mctp/device.c
index ef2755f82f87..f86ef6d751bd 100644
--- a/net/mctp/device.c
+++ b/net/mctp/device.c
@@ -24,12 +24,25 @@ struct mctp_dump_cb {
size_t a_idx;
};

-/* unlocked: caller must hold rcu_read_lock */
+/* unlocked: caller must hold rcu_read_lock.
+ * Returned mctp_dev has its refcount incremented, or NULL if unset.
+ */
struct mctp_dev *__mctp_dev_get(const struct net_device *dev)
{
- return rcu_dereference(dev->mctp_ptr);
+ struct mctp_dev *mdev = rcu_dereference(dev->mctp_ptr);
+
+ /* RCU guarantees that any mdev is still live.
+ * Zero refcount implies a pending free, return NULL.
+ */
+ if (mdev)
+ if (!refcount_inc_not_zero(&mdev->refs))
+ return NULL;
+ return mdev;
}

+/* Returned mctp_dev does not have refcount incremented. The returned pointer
+ * remains live while rtnl_lock is held, as that prevents mctp_unregister()
+ */
struct mctp_dev *mctp_dev_get_rtnl(const struct net_device *dev)
{
return rtnl_dereference(dev->mctp_ptr);
@@ -123,6 +136,7 @@ static int mctp_dump_addrinfo(struct sk_buff *skb, struct netlink_callback *cb)
if (mdev) {
rc = mctp_dump_dev_addrinfo(mdev,
skb, cb);
+ mctp_dev_put(mdev);
// Error indicates full buffer, this
// callback will get retried.
if (rc < 0)
@@ -297,7 +311,7 @@ void mctp_dev_hold(struct mctp_dev *mdev)

void mctp_dev_put(struct mctp_dev *mdev)
{
- if (refcount_dec_and_test(&mdev->refs)) {
+ if (mdev && refcount_dec_and_test(&mdev->refs)) {
dev_put(mdev->dev);
kfree_rcu(mdev, rcu);
}
@@ -369,6 +383,7 @@ static size_t mctp_get_link_af_size(const struct net_device *dev,
if (!mdev)
return 0;
ret = nla_total_size(4); /* IFLA_MCTP_NET */
+ mctp_dev_put(mdev);
return ret;
}

diff --git a/net/mctp/route.c b/net/mctp/route.c
index e52cef750500..05fbd318eb98 100644
--- a/net/mctp/route.c
+++ b/net/mctp/route.c
@@ -786,7 +786,7 @@ int mctp_local_output(struct sock *sk, struct mctp_route *rt,
{
struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
struct mctp_skb_cb *cb = mctp_cb(skb);
- struct mctp_route tmp_rt;
+ struct mctp_route tmp_rt = {0};
struct mctp_sk_key *key;
struct net_device *dev;
struct mctp_hdr *hdr;
@@ -892,6 +892,7 @@ int mctp_local_output(struct sock *sk, struct mctp_route *rt,
mctp_route_release(rt);

dev_put(dev);
+ mctp_dev_put(tmp_rt.dev);

return rc;

@@ -1057,11 +1058,13 @@ static int mctp_pkttype_receive(struct sk_buff *skb, struct net_device *dev,

rt->output(rt, skb);
mctp_route_release(rt);
+ mctp_dev_put(mdev);

return NET_RX_SUCCESS;

err_drop:
kfree_skb(skb);
+ mctp_dev_put(mdev);
return NET_RX_DROP;
}

diff --git a/net/mctp/test/utils.c b/net/mctp/test/utils.c
index 7b7918702592..e03ba66bbe18 100644
--- a/net/mctp/test/utils.c
+++ b/net/mctp/test/utils.c
@@ -54,7 +54,6 @@ struct mctp_test_dev *mctp_test_create_dev(void)

rcu_read_lock();
dev->mdev = __mctp_dev_get(ndev);
- mctp_dev_hold(dev->mdev);
rcu_read_unlock();

return dev;
--
2.34.1