Lockdep is unhappy if two locks from the same class are held.
Fix the below warning by making __vsock_release() non-recursive -- this
patch is kind of ugly, but it looks to me there is not a better way to
deal with the problem here.
============================================
WARNING: possible recursive locking detected
5.2.0+ #6 Not tainted
--------------------------------------------
a.out/1020 is trying to acquire lock:
0000000074731a98 (sk_lock-AF_VSOCK){+.+.}, at: hvs_release+0x10/0x120 [hv_sock]
but task is already holding lock:
0000000014ff8397 (sk_lock-AF_VSOCK){+.+.}, at: __vsock_release+0x2e/0xf0 [vsock]
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock(sk_lock-AF_VSOCK);
lock(sk_lock-AF_VSOCK);
*** DEADLOCK ***
May be due to missing lock nesting notation
2 locks held by a.out/1020:
#0: 00000000f8bceaa7 (&sb->s_type->i_mutex_key#10){+.+.}, at: __sock_release+0x2d/0xa0
#1: 0000000014ff8397 (sk_lock-AF_VSOCK){+.+.}, at: __vsock_release+0x2e/0xf0 [vsock]
stack backtrace:
CPU: 7 PID: 1020 Comm: a.out Not tainted 5.2.0+ #6
Call Trace:
dump_stack+0x67/0x90
__lock_acquire.cold.66+0x14d/0x1f8
lock_acquire+0xb5/0x1c0
lock_sock_nested+0x6d/0x90
hvs_release+0x10/0x120 [hv_sock]
__vsock_release+0x24/0xf0 [vsock]
__vsock_release+0xa0/0xf0 [vsock]
vsock_release+0x12/0x30 [vsock]
__sock_release+0x37/0xa0
sock_close+0x14/0x20
__fput+0xc1/0x250
task_work_run+0x98/0xc0
do_exit+0x3dd/0xc60
do_group_exit+0x47/0xc0
get_signal+0x169/0xc60
do_signal+0x30/0x710
exit_to_usermode_loop+0x50/0xa0
do_syscall_64+0x1fc/0x220
entry_SYSCALL_64_after_hwframe+0x49/0xbe
Signed-off-by: Dexuan Cui <[email protected]>
---
net/vmw_vsock/af_vsock.c | 33 ++++++++++++++++++++++++++++++++-
net/vmw_vsock/hyperv_transport.c | 2 +-
2 files changed, 33 insertions(+), 2 deletions(-)
diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
index ab47bf3..420f605 100644
--- a/net/vmw_vsock/af_vsock.c
+++ b/net/vmw_vsock/af_vsock.c
@@ -638,6 +638,37 @@ struct sock *__vsock_create(struct net *net,
}
EXPORT_SYMBOL_GPL(__vsock_create);
+static void __vsock_release2(struct sock *sk)
+{
+ if (sk) {
+ struct sk_buff *skb;
+ struct vsock_sock *vsk;
+
+ vsk = vsock_sk(sk);
+
+ /* The release call is supposed to use lock_sock_nested()
+ * rather than lock_sock(), if a lock should be acquired.
+ */
+ transport->release(vsk);
+
+ /* Use the nested version to avoid the warning
+ * "possible recursive locking detected".
+ */
+ lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
+ sock_orphan(sk);
+ sk->sk_shutdown = SHUTDOWN_MASK;
+
+ while ((skb = skb_dequeue(&sk->sk_receive_queue)))
+ kfree_skb(skb);
+
+ /* This sk can not be a listener, so it's unnecessary
+ * to call vsock_dequeue_accept().
+ */
+ release_sock(sk);
+ sock_put(sk);
+ }
+}
+
static void __vsock_release(struct sock *sk)
{
if (sk) {
@@ -659,7 +690,7 @@ static void __vsock_release(struct sock *sk)
/* Clean up any sockets that never were accepted. */
while ((pending = vsock_dequeue_accept(sk)) != NULL) {
- __vsock_release(pending);
+ __vsock_release2(pending);
sock_put(pending);
}
diff --git a/net/vmw_vsock/hyperv_transport.c b/net/vmw_vsock/hyperv_transport.c
index 9d864eb..4b126b2 100644
--- a/net/vmw_vsock/hyperv_transport.c
+++ b/net/vmw_vsock/hyperv_transport.c
@@ -559,7 +559,7 @@ static void hvs_release(struct vsock_sock *vsk)
struct sock *sk = sk_vsock(vsk);
bool remove_sock;
- lock_sock(sk);
+ lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
remove_sock = hvs_close_lock_held(vsk);
release_sock(sk);
if (remove_sock)
--
1.8.3.1
From: Dexuan Cui <[email protected]>
Date: Tue, 20 Aug 2019 03:14:22 +0000
> +static void __vsock_release2(struct sock *sk)
Do not duplicate an entire function just to adjust some aspect of the
lock debugging, please find a cleaner and more minimal way to
implement this fix.
On Tue, Aug 20, 2019 at 03:14:22AM +0000, Dexuan Cui wrote:
> Lockdep is unhappy if two locks from the same class are held.
>
> Fix the below warning by making __vsock_release() non-recursive -- this
> patch is kind of ugly, but it looks to me there is not a better way to
> deal with the problem here.
>
> ============================================
> WARNING: possible recursive locking detected
> 5.2.0+ #6 Not tainted
> --------------------------------------------
> a.out/1020 is trying to acquire lock:
> 0000000074731a98 (sk_lock-AF_VSOCK){+.+.}, at: hvs_release+0x10/0x120 [hv_sock]
>
> but task is already holding lock:
> 0000000014ff8397 (sk_lock-AF_VSOCK){+.+.}, at: __vsock_release+0x2e/0xf0 [vsock]
>
> other info that might help us debug this:
> Possible unsafe locking scenario:
>
> CPU0
> ----
> lock(sk_lock-AF_VSOCK);
> lock(sk_lock-AF_VSOCK);
>
> *** DEADLOCK ***
>
> May be due to missing lock nesting notation
>
> 2 locks held by a.out/1020:
> #0: 00000000f8bceaa7 (&sb->s_type->i_mutex_key#10){+.+.}, at: __sock_release+0x2d/0xa0
> #1: 0000000014ff8397 (sk_lock-AF_VSOCK){+.+.}, at: __vsock_release+0x2e/0xf0 [vsock]
>
> stack backtrace:
> CPU: 7 PID: 1020 Comm: a.out Not tainted 5.2.0+ #6
> Call Trace:
> dump_stack+0x67/0x90
> __lock_acquire.cold.66+0x14d/0x1f8
> lock_acquire+0xb5/0x1c0
> lock_sock_nested+0x6d/0x90
> hvs_release+0x10/0x120 [hv_sock]
> __vsock_release+0x24/0xf0 [vsock]
> __vsock_release+0xa0/0xf0 [vsock]
> vsock_release+0x12/0x30 [vsock]
> __sock_release+0x37/0xa0
> sock_close+0x14/0x20
> __fput+0xc1/0x250
> task_work_run+0x98/0xc0
> do_exit+0x3dd/0xc60
> do_group_exit+0x47/0xc0
> get_signal+0x169/0xc60
> do_signal+0x30/0x710
> exit_to_usermode_loop+0x50/0xa0
> do_syscall_64+0x1fc/0x220
> entry_SYSCALL_64_after_hwframe+0x49/0xbe
>
> Signed-off-by: Dexuan Cui <[email protected]>
> ---
> net/vmw_vsock/af_vsock.c | 33 ++++++++++++++++++++++++++++++++-
> net/vmw_vsock/hyperv_transport.c | 2 +-
> 2 files changed, 33 insertions(+), 2 deletions(-)
>
> diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
> index ab47bf3..420f605 100644
> --- a/net/vmw_vsock/af_vsock.c
> +++ b/net/vmw_vsock/af_vsock.c
> @@ -638,6 +638,37 @@ struct sock *__vsock_create(struct net *net,
> }
> EXPORT_SYMBOL_GPL(__vsock_create);
>
> +static void __vsock_release2(struct sock *sk)
> +{
> + if (sk) {
> + struct sk_buff *skb;
> + struct vsock_sock *vsk;
> +
> + vsk = vsock_sk(sk);
> +
> + /* The release call is supposed to use lock_sock_nested()
> + * rather than lock_sock(), if a lock should be acquired.
> + */
> + transport->release(vsk);
> +
> + /* Use the nested version to avoid the warning
> + * "possible recursive locking detected".
> + */
> + lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
What about using lock_sock_nested() in the __vsock_release() without
define this new function?
> + sock_orphan(sk);
> + sk->sk_shutdown = SHUTDOWN_MASK;
> +
> + while ((skb = skb_dequeue(&sk->sk_receive_queue)))
> + kfree_skb(skb);
> +
> + /* This sk can not be a listener, so it's unnecessary
> + * to call vsock_dequeue_accept().
> + */
> + release_sock(sk);
> + sock_put(sk);
> + }
> +}
> +
> static void __vsock_release(struct sock *sk)
> {
> if (sk) {
> @@ -659,7 +690,7 @@ static void __vsock_release(struct sock *sk)
>
> /* Clean up any sockets that never were accepted. */
> while ((pending = vsock_dequeue_accept(sk)) != NULL) {
> - __vsock_release(pending);
> + __vsock_release2(pending);
> sock_put(pending);
> }
>
> diff --git a/net/vmw_vsock/hyperv_transport.c b/net/vmw_vsock/hyperv_transport.c
> index 9d864eb..4b126b2 100644
> --- a/net/vmw_vsock/hyperv_transport.c
> +++ b/net/vmw_vsock/hyperv_transport.c
> @@ -559,7 +559,7 @@ static void hvs_release(struct vsock_sock *vsk)
> struct sock *sk = sk_vsock(vsk);
> bool remove_sock;
>
> - lock_sock(sk);
> + lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
Should we update also other transports?
Thanks,
Stefano
> From: Stefano Garzarella <[email protected]>
> Sent: Thursday, August 22, 2019 3:25 AM
> > [...snipped...]
> > --- a/net/vmw_vsock/hyperv_transport.c
> > +++ b/net/vmw_vsock/hyperv_transport.c
> > @@ -559,7 +559,7 @@ static void hvs_release(struct vsock_sock *vsk)
> > struct sock *sk = sk_vsock(vsk);
> > bool remove_sock;
> >
> > - lock_sock(sk);
> > + lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
>
> Should we update also other transports?
>
> Stefano
Hi Stefano,
Sorry for the late reply! I'll post a v2 shortly.
As I checked, hyperv socket and virtio socket need to be fixed.
The vmci socket code doesn't acquire the sock lock in the release
callback, so it doesn't need any fix.
Thanks,
-- Dexuan