RDMA/rxe: fix UDP tunnel socket leak on rxe_newlink failure
Subsystems (2)
PATCH 1/1 RDMA/rxe: fix UDP tunnel socket leak on rxe_newlink failure
rxe_newlink() calls rxe_net_init() to set up the per-net-namespace UDP tunnel sockets before calling rxe_net_add() to create and register the rxe device. rxe_net_init() takes a reference on the IPv4 and IPv6 tunnel sockets, creating them if they do not already exist.
If rxe_net_add() subsequently fails, rxe_newlink() returns the error without releasing those socket references. The normal teardown path, rxe_net_del(), is never reached because rxe_net_add() has already deallocated the rxe device on failure. As a result the tunnel socket references are leaked and the per-net sockets bound to UDP port 4791 can never be released.
Release the references on the error path by adding rxe_net_uninit(), which performs the same socket teardown as rxe_net_del(). The shared release logic is factored into a helper to avoid duplication.
--- drivers/infiniband/sw/rxe/rxe.c | 1 + drivers/infiniband/sw/rxe/rxe_net.c | 32 ++++++++++++++++++++--------- drivers/infiniband/sw/rxe/rxe_net.h | 1 + 3 files changed, 24 insertions(+), 10 deletions(-)
drivers/infiniband/sw/rxe/rxe.c +1 -0
@@ -243,6 +243,7 @@ static int rxe_newlink(const char *ibdev_name, struct net_device *ndev) err =rxe_net_add (ibdev_name, ndev);if (err) {rxe_err ("failed to add %s\n" , ndev->name); +rxe_net_uninit (ndev); goto err; } err:
drivers/infiniband/sw/rxe/rxe_net.c +22 -10
@@ -642,18 +642,10 @@ static void rxe_sock_put(struct sock *sk, } } -void rxe_net_del (struct ib_device *dev) +/* release the per-net-namespace tunnel socket references held for @net */ +static void rxe_release_sockets (struct net *net) { - struct rxe_dev *rxe = container_of(dev, struct rxe_dev, ib_dev); -struct net_device *ndev;struct sock *sk; -struct net *net; - - ndev =rxe_ib_device_get_netdev (&rxe->ib_dev); -if (!ndev) -return ; - - net =dev_net (ndev); sk =rxe_ns_pernet_sk4 (net);if (sk)
@@ -662,6 +654,26 @@ void rxe_net_del(struct ib_device *dev) sk =rxe_ns_pernet_sk6 (net);if (sk)rxe_sock_put (sk, rxe_ns_pernet_set_sk6, net); +} + +/* release the socket references taken by a successful rxe_net_init() when a +* later step of device creation fails and rxe_net_del() will not be called +*/ +void rxe_net_uninit (struct net_device *ndev) +{ +rxe_release_sockets (dev_net (ndev)); +} + +void rxe_net_del (struct ib_device *dev) +{ + struct rxe_dev *rxe = container_of(dev, struct rxe_dev, ib_dev); +struct net_device *ndev; + + ndev =rxe_ib_device_get_netdev (&rxe->ib_dev); +if (!ndev) +return ; + +rxe_release_sockets (dev_net (ndev));dev_put (ndev); }
drivers/infiniband/sw/rxe/rxe_net.h +1 -0
@@ -16,6 +16,7 @@ void rxe_net_del(struct ib_device *dev);int rxe_register_notifier (void );int rxe_net_init (struct net_device *ndev); +void rxe_net_uninit (struct net_device *ndev);void rxe_net_exit (void );#endif /* RXE_NET_H */
Signature · 1 line
2.53.0
Quoted 29 lines · Purushothaman Ramalingam
rxe_newlink()callsrxe_net_init()to set up the per-net-namespace UDP tunnel sockets before callingrxe_net_add()to create and register the rxe device.rxe_net_init()takes a reference on the IPv4 and IPv6 tunnel sockets, creating them if they do not already exist.If
rxe_net_add()subsequently fails,rxe_newlink()returns the error without releasing those socket references. The normal teardown path,rxe_net_del(), is never reached becauserxe_net_add()has already deallocated the rxe device on failure. As a result the tunnel socket references are leaked and the per-net sockets bound to UDP port 4791 can never be released.Release the references on the error path by adding
rxe_net_uninit(), which performs the same socket teardown asrxe_net_del(). The shared release logic is factored into a helper to avoid duplication.Fixes f1327abd6abe ("RDMA/rxe: Support RDMA link creation and destruction per net namespace")Signed-off-by Purushothaman Ramalingam <[email protected]>--- drivers/infiniband/sw/rxe/rxe.c | 1 + drivers/infiniband/sw/rxe/rxe_net.c | 32 ++++++++++++++++++++--------- drivers/infiniband/sw/rxe/rxe_net.h | 1 + 3 files changed, 24 insertions(+), 10 deletions(-)drivers/infiniband/sw/rxe/rxe.c +0
-0
drivers/infiniband/sw/rxe/rxe.c static int rxe_newlink(const char *ibdev_name, struct net_device *ndev) · line 247
Quoted 5 lines
drivers/infiniband/sw/rxe/rxe.c +1
-0@@ -243,6 +243,7 @@ static int rxe_newlink(const char *ibdev_name, struct net_device *ndev) err =rxe_net_add (ibdev_name, ndev);if (err) {rxe_err ("failed to add %s\n" , ndev->name); +rxe_net_uninit (ndev); goto err;
To avoid introducing an unnecessary single-line wrapper helper (rxe_net_uninit), we have refactored the cleanup logic by directly exporting the core rxe_release_sockets(struct net *net) function. Since the tunnel sockets are managed per network namespace, passing struct net * aligns perfectly with the network subsystem's conventions.
Additionally, we have removed the now-redundant err: label from rxe_newlink() to keep the error path clean and idiomatic.
Here is the updated patch:
"
drivers/infiniband/sw/rxe/rxe.c +1 -1
@@ -243,7 +243,7 @@ static int rxe_newlink(const char *ibdev_name, struct net_device *ndev) err =rxe_net_add (ibdev_name, ndev);if (err) {rxe_err ("failed to add %s\n" , ndev->name); - goto err; +rxe_release_sockets (dev_net (ndev)); } err:return err;
drivers/infiniband/sw/rxe/rxe_net.c +14 -10
@@ -642,18 +642,10 @@ static void rxe_sock_put(struct sock *sk, } } -void rxe_net_del (struct ib_device *dev) +/* release the per-net-namespace tunnel socket references held for @net */ +void rxe_release_sockets (struct net *net) { - struct rxe_dev *rxe = container_of(dev, struct rxe_dev, ib_dev); -struct net_device *ndev;struct sock *sk; -struct net *net; - - ndev =rxe_ib_device_get_netdev (&rxe->ib_dev); -if (!ndev) -return ; - - net =dev_net (ndev); sk =rxe_ns_pernet_sk4 (net);if (sk)
@@ -662,6 +654,18 @@ void rxe_net_del(struct ib_device *dev) sk =rxe_ns_pernet_sk6 (net);if (sk)rxe_sock_put (sk, rxe_ns_pernet_set_sk6, net); +} + +void rxe_net_del (struct ib_device *dev) +{ + struct rxe_dev *rxe = container_of(dev, struct rxe_dev, ib_dev); +struct net_device *ndev; + + ndev =rxe_ib_device_get_netdev (&rxe->ib_dev); +if (!ndev) +return ; + +rxe_release_sockets (dev_net (ndev));dev_put (ndev); }
drivers/infiniband/sw/rxe/rxe_net.h +1 -0
@@ -16,6 +16,7 @@ void rxe_net_del(struct ib_device *dev);int rxe_register_notifier (void );int rxe_net_init (struct net_device *ndev); +void rxe_release_sockets (struct net *net);void rxe_net_exit (void );#endif /* RXE_NET_H */
"
Another commit is related to this issue and is currently on the waiting list:
https://patchwork.kernel.org/project/linux-rdma/patch/[email protected]/
The two commits conflict with each other. Once one of them is merged, the other will need to be updated accordingly.
Let's keep an eye on both commits and make any necessary changes after the merge decision is made.
Zhu Yanjun
Quoted 66 lines
drivers/infiniband/sw/rxe/rxe.c +0
-0} err:drivers/infiniband/sw/rxe/rxe_net.c +22
-10@@ -642,18 +642,10 @@ static void rxe_sock_put(struct sock *sk, } } -void rxe_net_del (struct ib_device *dev) +/* release the per-net-namespace tunnel socket references held for @net */ +static void rxe_release_sockets (struct net *net) { - struct rxe_dev *rxe = container_of(dev, struct rxe_dev, ib_dev); -struct net_device *ndev;struct sock *sk; -struct net *net; - - ndev =rxe_ib_device_get_netdev (&rxe->ib_dev); -if (!ndev) -return ; - - net =dev_net (ndev); sk =rxe_ns_pernet_sk4 (net);if (sk)@@ -662,6 +654,26 @@ void rxe_net_del(struct ib_device *dev) sk =rxe_ns_pernet_sk6 (net);if (sk)rxe_sock_put (sk, rxe_ns_pernet_set_sk6, net); +} + +/* release the socket references taken by a successful rxe_net_init() when a +* later step of device creation fails and rxe_net_del() will not be called +*/ +void rxe_net_uninit (struct net_device *ndev) +{ +rxe_release_sockets (dev_net (ndev)); +} + +void rxe_net_del (struct ib_device *dev) +{ + struct rxe_dev *rxe = container_of(dev, struct rxe_dev, ib_dev); +struct net_device *ndev; + + ndev =rxe_ib_device_get_netdev (&rxe->ib_dev); +if (!ndev) +return ; + +rxe_release_sockets (dev_net (ndev));dev_put (ndev); }drivers/infiniband/sw/rxe/rxe_net.h +1
-0@@ -16,6 +16,7 @@ void rxe_net_del(struct ib_device *dev);int rxe_register_notifier (void );int rxe_net_init (struct net_device *ndev); +void rxe_net_uninit (struct net_device *ndev);void rxe_net_exit (void );#endif /* RXE_NET_H */