Skip to contentSkip to conversation

sctp: avoid irq lock inversion while call sk->sk_data_ready()

PATCH In review Wei Yongjun 2 participants 7 messages Feb 3, 2010 – Feb 8, 2010

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sctp: avoid irq lock inversion while call sk->sk_data_ready()

sk->sk_data_ready() of sctp socket can be called from both BH and non-BH contexts, but the default sk->sk_data_ready(), sock_def_readable(), can not be used in this case. Therefore, we have to make a new function sctp_data_ready() to grab sk->sk_data_ready() with BH disabling.

============================[ INFO: possible irq lock inversion dependency detected ] 2.6.33-rc6 #129
---------------------------------------------------------
sctp_darn/1517 just changed the state of lock:
(clock-AF_INET){++.?..}, at: [<c06aab60>] sock_def_readable+0x20/0x80
but this lock took another, SOFTIRQ-unsafe lock in the past:
(slock-AF_INET){+.-...}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
1 lock held by sctp_darn/1517:
#0: (sk_lock-AF_INET){+.+.+.}, at: [<cdfe363d>] sctp_sendmsg+0x23d/0xc00 [sctp]

---
 include/net/sctp/sctp.h |    1 +
 net/sctp/endpointola.c  |    1 +
 net/sctp/socket.c       |   10 ++++++++++
 3 files changed, 12 insertions(+), 0 deletions(-)
include/net/sctp/sctp.h +1 -0
index 78740ec..fa6cde5 100644
@@ -128,6 +128,7 @@ extern int sctp_register_pf(struct sctp_pf *, sa_family_t);
 int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb);
 int sctp_inet_listen(struct socket *sock, int backlog);
 void sctp_write_space(struct sock *sk);
+void sctp_data_ready(struct sock *sk, int len);
 unsigned int sctp_poll(struct file *file, struct socket *sock,
 		poll_table *wait);
 void sctp_sock_rfree(struct sk_buff *skb);
net/sctp/endpointola.c +1 -0
index 905fda5..7ec09ba 100644
@@ -144,6 +144,7 @@ static struct sctp_endpoint *sctp_endpoint_init(struct sctp_endpoint *ep,
 	/* Use SCTP specific send buffer space queues.  */
 	ep->sndbuf_policy = sctp_sndbuf_policy;
 
+	sk->sk_data_ready = sctp_data_ready;
 	sk->sk_write_space = sctp_write_space;
 	sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
 
net/sctp/socket.c +10 -0
index 67fdac9..b437e2a 100644
@@ -6185,6 +6185,16 @@ do_nonblock:
 	goto out;
 }
 
+void sctp_data_ready(struct sock *sk, int len)
+{
+	read_lock_bh(&sk->sk_callback_lock);
+	if (sk_has_sleeper(sk))
+		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
+						POLLRDNORM | POLLRDBAND);
+	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
+	read_unlock_bh(&sk->sk_callback_lock);
+}
+
 /* If socket sndbuf has changed, wake up all per association waiters.  */
 void sctp_write_space(struct sock *sk)
 {
Signature · 1 line
1.6.5.2
#1 lore raw
Quoted 14 lines

sk->sk_data_ready() of sctp socket can be called from both BH and non-BH contexts, but the default sk->sk_data_ready(), sock_def_readable(), can not be used in this case. Therefore, we have to make a new function sctp_data_ready() to grab sk->sk_data_ready() with BH disabling.

============================> [ INFO: possible irq lock inversion dependency detected ] 2.6.33-rc6 #129
---------------------------------------------------------
sctp_darn/1517 just changed the state of lock:
(clock-AF_INET){++.?..}, at: [<c06aab60>] sock_def_readable+0x20/0x80
but this lock took another, SOFTIRQ-unsafe lock in the past:
(slock-AF_INET){+.-...}

Quoted 6 lines

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
1 lock held by sctp_darn/1517:
#0: (sk_lock-AF_INET){+.+.+.}, at: [<cdfe363d>] sctp_sendmsg+0x23d/0xc00 [sctp]

The full lockdep output message is:

============================[ INFO: possible irq lock inversion dependency detected ] 2.6.33-rc6 #129
---------------------------------------------------------
sctp_darn/1517 just changed the state of lock:
(clock-AF_INET){++.?..}, at: [<c06aab60>] sock_def_readable+0x20/0x80
but this lock took another, SOFTIRQ-unsafe lock in the past:
(slock-AF_INET){+.-...}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
1 lock held by sctp_darn/1517:
#0: (sk_lock-AF_INET){+.+.+.}, at: [<cdfe363d>] sctp_sendmsg+0x23d/0xc00 [sctp]

the shortest dependencies between 2nd lock and 1st lock:

 -> (slock-AF_INET){+.-...} ops: 0 {
    HARDIRQ-ON-W at:
                          [<c04717ff>] __lock_acquire+0x9af/0x1890
                          [<c047275f>] lock_acquire+0x7f/0xf0
                          [<c0762bbd>] _raw_spin_lock_bh+0x3d/0x50
                          [<c06aa723>] lock_sock_nested+0x33/0xf0
                          [<c06ed27a>] tcp_close+0x1a/0x390
                          [<c070eefb>] inet_release+0x3b/0x60
                          [<c06a8030>] sock_release+0x20/0x70
                          [<c06a8097>] sock_close+0x17/0x30
                          [<c04f687b>] __fput+0xfb/0x200
                          [<c04f699d>] fput+0x1d/0x30
                          [<c04f341c>] filp_close+0x4c/0x80
                          [<c04f34c7>] sys_close+0x77/0xc0
                          [<c0402d8c>] sysenter_do_call+0x12/0x32
    IN-SOFTIRQ-W at:
                          [<c04717e3>] __lock_acquire+0x993/0x1890
                          [<c047275f>] lock_acquire+0x7f/0xf0
                          [<c0762968>] _raw_spin_lock+0x38/0x50
                          [<c070601e>] udp_queue_rcv_skb+0xee/0x2c0
                          [<c07067ef>] __udp4_lib_rcv+0x1bf/0x6a0
                          [<c0706ce7>] udp_rcv+0x17/0x20
                          [<c06e194f>] ip_local_deliver_finish+0xdf/0x2c0
                          [<c06e1bbf>] ip_local_deliver+0x8f/0xa0
                          [<c06e12cb>] ip_rcv_finish+0xdb/0x3c0
                          [<c06e17b6>] ip_rcv+0x206/0x2c0
                          [<c06b79ef>] netif_receive_skb+0x34f/0x570
                          [<cc83ac5e>] pcnet32_poll+0x27e/0x7a0 [pcnet32]
                          [<c06b84b0>] net_rx_action+0x150/0x230
                          [<c0449520>] __do_softirq+0xa0/0x1c0
    INITIAL USE at:
                         [<c04711bf>] __lock_acquire+0x36f/0x1890
                         [<c047275f>] lock_acquire+0x7f/0xf0
                         [<c0762bbd>] _raw_spin_lock_bh+0x3d/0x50
                         [<c06aa723>] lock_sock_nested+0x33/0xf0
                         [<c06ed27a>] tcp_close+0x1a/0x390
                         [<c070eefb>] inet_release+0x3b/0x60
                         [<c06a8030>] sock_release+0x20/0x70
                         [<c06a8097>] sock_close+0x17/0x30
                         [<c04f687b>] __fput+0xfb/0x200
                         [<c04f699d>] fput+0x1d/0x30
                         [<c04f341c>] filp_close+0x4c/0x80
                         [<c04f34c7>] sys_close+0x77/0xc0
                         [<c0402d8c>] sysenter_do_call+0x12/0x32
  }
  ... key      at: [<c0f2d8d0>] af_family_slock_keys+0x10/0x140
  ... acquired at:
   [<c0471fc2>] __lock_acquire+0x1172/0x1890
   [<c047275f>] lock_acquire+0x7f/0xf0
   [<c0762c0d>] _raw_write_lock_bh+0x3d/0x50
   [<c06ab3ad>] sk_common_release+0x2d/0xb0
   [<cdfe2ff8>] sctp_close+0xe8/0x1f0 [sctp]
   [<c070eefb>] inet_release+0x3b/0x60
   [<c06a8030>] sock_release+0x20/0x70
   [<c06a8097>] sock_close+0x17/0x30
   [<c04f687b>] __fput+0xfb/0x200
   [<c04f699d>] fput+0x1d/0x30
   [<c04f341c>] filp_close+0x4c/0x80
   [<c04f34c7>] sys_close+0x77/0xc0
   [<c0402d8c>] sysenter_do_call+0x12/0x32

-> (clock-AF_INET){++.?..} ops: 0 {
HARDIRQ-ON-W at:

                        [<c04717ff>] __lock_acquire+0x9af/0x1890
                        [<c047275f>] lock_acquire+0x7f/0xf0
                        [<c0762c0d>] _raw_write_lock_bh+0x3d/0x50
                        [<c06ed359>] tcp_close+0xf9/0x390
                        [<c070eefb>] inet_release+0x3b/0x60
                        [<c06a8030>] sock_release+0x20/0x70
                        [<c06a8097>] sock_close+0x17/0x30
                        [<c04f687b>] __fput+0xfb/0x200
                        [<c04f699d>] fput+0x1d/0x30
                        [<c04f341c>] filp_close+0x4c/0x80
                        [<c04f34c7>] sys_close+0x77/0xc0
                        [<c0402d8c>] sysenter_do_call+0x12/0x32
   HARDIRQ-ON-R at:
                        [<c0470fd6>] __lock_acquire+0x186/0x1890
                        [<c047275f>] lock_acquire+0x7f/0xf0
                        [<c0762fa8>] _raw_read_lock+0x38/0x50
                        [<c06aaaac>] sock_def_write_space+0x1c/0xb0
                        [<c06ab33a>] sock_wfree+0x4a/0x60
                        [<c06af175>] skb_release_head_state+0x45/0xc0
                        [<c06aeea0>] __kfree_skb+0x10/0x90
                        [<c06b57a9>] net_tx_action+0x59/0x140
                        [<c0449520>] __do_softirq+0xa0/0x1c0
   IN-SOFTIRQ-R at:
                        [<c04717e3>] __lock_acquire+0x993/0x1890
                        [<c047275f>] lock_acquire+0x7f/0xf0
                        [<c0762fa8>] _raw_read_lock+0x38/0x50
                        [<c06aaaac>] sock_def_write_space+0x1c/0xb0
                        [<c06ab33a>] sock_wfree+0x4a/0x60
                        [<c06af175>] skb_release_head_state+0x45/0xc0
                        [<c06aeea0>] __kfree_skb+0x10/0x90
                        [<c06b57a9>] net_tx_action+0x59/0x140
                        [<c0449520>] __do_softirq+0xa0/0x1c0
   SOFTIRQ-ON-R at:
                        [<c0471824>] __lock_acquire+0x9d4/0x1890
                        [<c047275f>] lock_acquire+0x7f/0xf0
                        [<c0762fa8>] _raw_read_lock+0x38/0x50
                        [<c06aab60>] sock_def_readable+0x20/0x80
                        [<cdfdebc4>] sctp_ulpq_tail_event+0x134/0x210 [sctp]
                        [<cdfd253e>] sctp_side_effects+0x8ee/0x10c0 [sctp]
                        [<cdfd2dc0>] sctp_do_sm+0xb0/0x1c0 [sctp]
                        [<cdfe6bc2>] sctp_primitive_ABORT+0x42/0x50 [sctp]
                        [<cdfe3892>] sctp_sendmsg+0x492/0xc00 [sctp]
                        [<c070df9e>] inet_sendmsg+0x2e/0x60
                        [<c06a7527>] sock_sendmsg+0xe7/0x110
                        [<c06a7723>] sys_sendmsg+0x113/0x230
                        [<c06a956b>] sys_socketcall+0xeb/0x2a0
                        [<c0402d8c>] sysenter_do_call+0x12/0x32
   INITIAL USE at:
                       [<c04711bf>] __lock_acquire+0x36f/0x1890
                       [<c047275f>] lock_acquire+0x7f/0xf0
                       [<c0762c0d>] _raw_write_lock_bh+0x3d/0x50
                       [<c06ed359>] tcp_close+0xf9/0x390
                       [<c070eefb>] inet_release+0x3b/0x60
                       [<c06a8030>] sock_release+0x20/0x70
                       [<c06a8097>] sock_close+0x17/0x30
                       [<c04f687b>] __fput+0xfb/0x200
                       [<c04f699d>] fput+0x1d/0x30
                       [<c04f341c>] filp_close+0x4c/0x80
                       [<c04f34c7>] sys_close+0x77/0xc0
                       [<c0402d8c>] sysenter_do_call+0x12/0x32
 }
 ... key      at: [<c0f2da10>] af_callback_keys+0x10/0x128
 ... acquired at:
   [<c04735db>] check_usage_backwards+0x8b/0xd0
   [<c046fd9a>] mark_lock+0x1ba/0x5c0
   [<c0471824>] __lock_acquire+0x9d4/0x1890
   [<c047275f>] lock_acquire+0x7f/0xf0
   [<c0762fa8>] _raw_read_lock+0x38/0x50
   [<c06aab60>] sock_def_readable+0x20/0x80
   [<cdfdebc4>] sctp_ulpq_tail_event+0x134/0x210 [sctp]
   [<cdfd253e>] sctp_side_effects+0x8ee/0x10c0 [sctp]
   [<cdfd2dc0>] sctp_do_sm+0xb0/0x1c0 [sctp]
   [<cdfe6bc2>] sctp_primitive_ABORT+0x42/0x50 [sctp]
   [<cdfe3892>] sctp_sendmsg+0x492/0xc00 [sctp]
   [<c070df9e>] inet_sendmsg+0x2e/0x60
   [<c06a7527>] sock_sendmsg+0xe7/0x110
   [<c06a7723>] sys_sendmsg+0x113/0x230
   [<c06a956b>] sys_socketcall+0xeb/0x2a0
   [<c0402d8c>] sysenter_do_call+0x12/0x32


stack backtrace:
Pid: 1517, comm: sctp_darn Not tainted 2.6.33-rc6 #129
Call Trace:
 [<c075f994>] ? printk+0x1d/0x21
 [<c047353e>] print_irq_inversion_bug.clone.0+0xfe/0x110
 [<c04735db>] check_usage_backwards+0x8b/0xd0
 [<c046fd9a>] mark_lock+0x1ba/0x5c0
 [<c059e513>] ? string+0x33/0xe0
 [<c0473550>] ? check_usage_backwards+0x0/0xd0
 [<c0471824>] __lock_acquire+0x9d4/0x1890
 [<c059e513>] ? string+0x33/0xe0
 [<c0762a4b>] ? _raw_spin_lock_irqsave+0x1b/0x60
 [<c07630cf>] ? _raw_spin_unlock_irqrestore+0x4f/0x60
 [<c046f37b>] ? trace_hardirqs_off+0xb/0x10
 [<c07630cf>] ? _raw_spin_unlock_irqrestore+0x4f/0x60
 [<c044338f>] ? release_console_sem+0x1ef/0x240
 [<c047275f>] lock_acquire+0x7f/0xf0
 [<c06aab60>] ? sock_def_readable+0x20/0x80
 [<c0762fa8>] _raw_read_lock+0x38/0x50
 [<c06aab60>] ? sock_def_readable+0x20/0x80
 [<c06aab60>] sock_def_readable+0x20/0x80
 [<cdfdebc4>] sctp_ulpq_tail_event+0x134/0x210 [sctp]
 [<cdfd253e>] sctp_side_effects+0x8ee/0x10c0 [sctp]
 [<c0762a4b>] ? _raw_spin_lock_irqsave+0x1b/0x60
 [<cdfd2dc0>] sctp_do_sm+0xb0/0x1c0 [sctp]
 [<c044338f>] ? release_console_sem+0x1ef/0x240
 [<cdfe6bc2>] sctp_primitive_ABORT+0x42/0x50 [sctp]
 [<cdfe3892>] sctp_sendmsg+0x492/0xc00 [sctp]
 [<c070df9e>] inet_sendmsg+0x2e/0x60
 [<c06a7527>] sock_sendmsg+0xe7/0x110
 [<c046f37b>] ? trace_hardirqs_off+0xb/0x10
 [<c04d6810>] ? might_fault+0x50/0xa0
 [<c04d6810>] ? might_fault+0x50/0xa0
 [<c04d6856>] ? might_fault+0x96/0xa0
 [<c04d6810>] ? might_fault+0x50/0xa0
 [<c05a140d>] ? _copy_from_user+0x3d/0x130
 [<c06a7723>] sys_sendmsg+0x113/0x230
 [<c06aa6e7>] ? release_sock+0xd7/0xe0
 [<c04704eb>] ? trace_hardirqs_on+0xb/0x10
 [<c0449298>] ? local_bh_enable_ip+0x68/0xd0
 [<cdfe4b0c>] ? sctp_getsockopt+0x9c/0x1010 [sctp]
 [<c0472949>] ? lock_release_non_nested+0x59/0x2f0
 [<c04704eb>] ? trace_hardirqs_on+0xb/0x10
 [<c06093ce>] ? put_ldisc+0x3e/0xc0
 [<c04d6810>] ? might_fault+0x50/0xa0
 [<c04d6810>] ? might_fault+0x50/0xa0
 [<c06a956b>] sys_socketcall+0xeb/0x2a0
 [<c0402dbb>] ? sysenter_exit+0xf/0x16
 [<c0402d8c>] sysenter_do_call+0x12/0x32
#2 lore raw
In reply to Wei Yongjun
Quoted 5 lines · Wei Yongjun

sk->sk_data_ready() of sctp socket can be called from both BH and non-BH contexts, but the default sk->sk_data_ready(), sock_def_readable(), can not be used in this case. Therefore, we have to make a new function sctp_data_ready() to grab sk->sk_data_ready() with BH disabling.

Wouldn't the same inversion happen in TCP as well? TCP can call that function in _bh and user contexts as well.

-vlad

Quoted 66 lines

============================> [ INFO: possible irq lock inversion dependency detected ] 2.6.33-rc6 #129
---------------------------------------------------------
sctp_darn/1517 just changed the state of lock:
(clock-AF_INET){++.?..}, at: [<c06aab60>] sock_def_readable+0x20/0x80
but this lock took another, SOFTIRQ-unsafe lock in the past:
(slock-AF_INET){+.-...}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
1 lock held by sctp_darn/1517:
#0: (sk_lock-AF_INET){+.+.+.}, at: [<cdfe363d>] sctp_sendmsg+0x23d/0xc00 [sctp]

---
 include/net/sctp/sctp.h |    1 +
 net/sctp/endpointola.c  |    1 +
 net/sctp/socket.c       |   10 ++++++++++
 3 files changed, 12 insertions(+), 0 deletions(-)
include/net/sctp/sctp.h +1 -0
index 78740ec..fa6cde5 100644
@@ -128,6 +128,7 @@ extern int sctp_register_pf(struct sctp_pf *, sa_family_t);
 int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb);
 int sctp_inet_listen(struct socket *sock, int backlog);
 void sctp_write_space(struct sock *sk);
+void sctp_data_ready(struct sock *sk, int len);
 unsigned int sctp_poll(struct file *file, struct socket *sock,
 		poll_table *wait);
 void sctp_sock_rfree(struct sk_buff *skb);
net/sctp/endpointola.c +1 -0
index 905fda5..7ec09ba 100644
@@ -144,6 +144,7 @@ static struct sctp_endpoint *sctp_endpoint_init(struct sctp_endpoint *ep,
 	/* Use SCTP specific send buffer space queues.  */
 	ep->sndbuf_policy = sctp_sndbuf_policy;
 
+	sk->sk_data_ready = sctp_data_ready;
 	sk->sk_write_space = sctp_write_space;
 	sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
 
net/sctp/socket.c +10 -0
index 67fdac9..b437e2a 100644
@@ -6185,6 +6185,16 @@ do_nonblock:
 	goto out;
 }
 
+void sctp_data_ready(struct sock *sk, int len)
+{
+	read_lock_bh(&sk->sk_callback_lock);
+	if (sk_has_sleeper(sk))
+		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
+						POLLRDNORM | POLLRDBAND);
+	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
+	read_unlock_bh(&sk->sk_callback_lock);
+}
+
 /* If socket sndbuf has changed, wake up all per association waiters.  */
 void sctp_write_space(struct sock *sk)
 {
#3 lore raw
In reply to Wei Yongjun
Quoted 9 lines · Vlad Yasevich
Quoted 6 lines · Wei Yongjun

sk->sk_data_ready() of sctp socket can be called from both BH and non-BH contexts, but the default sk->sk_data_ready(), sock_def_readable(), can not be used in this case. Therefore, we have to make a new function sctp_data_ready() to grab sk->sk_data_ready() with BH disabling.

Quoted 3 lines

Wouldn't the same inversion happen in TCP as well? TCP can call that function in _bh and user contexts as well.

Not sure, but TCP does not call that function in user context at all.

Quoted 73 lines

-vlad

Quoted 67 lines

============================>> [ INFO: possible irq lock inversion dependency detected ] 2.6.33-rc6 #129
---------------------------------------------------------
sctp_darn/1517 just changed the state of lock:
(clock-AF_INET){++.?..}, at: [<c06aab60>] sock_def_readable+0x20/0x80
but this lock took another, SOFTIRQ-unsafe lock in the past:
(slock-AF_INET){+.-...}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
1 lock held by sctp_darn/1517:
#0: (sk_lock-AF_INET){+.+.+.}, at: [<cdfe363d>] sctp_sendmsg+0x23d/0xc00 [sctp]

---
 include/net/sctp/sctp.h |    1 +
 net/sctp/endpointola.c  |    1 +
 net/sctp/socket.c       |   10 ++++++++++
 3 files changed, 12 insertions(+), 0 deletions(-)
include/net/sctp/sctp.h +1 -0
index 78740ec..fa6cde5 100644
@@ -128,6 +128,7 @@ extern int sctp_register_pf(struct sctp_pf *, sa_family_t);
 int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb);
 int sctp_inet_listen(struct socket *sock, int backlog);
 void sctp_write_space(struct sock *sk);
+void sctp_data_ready(struct sock *sk, int len);
 unsigned int sctp_poll(struct file *file, struct socket *sock,
 		poll_table *wait);
 void sctp_sock_rfree(struct sk_buff *skb);
net/sctp/endpointola.c +1 -0
index 905fda5..7ec09ba 100644
@@ -144,6 +144,7 @@ static struct sctp_endpoint *sctp_endpoint_init(struct sctp_endpoint *ep,
 	/* Use SCTP specific send buffer space queues.  */
 	ep->sndbuf_policy = sctp_sndbuf_policy;
 
+	sk->sk_data_ready = sctp_data_ready;
 	sk->sk_write_space = sctp_write_space;
 	sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
 
net/sctp/socket.c +10 -0
index 67fdac9..b437e2a 100644
@@ -6185,6 +6185,16 @@ do_nonblock:
 	goto out;
 }
 
+void sctp_data_ready(struct sock *sk, int len)
+{
+	read_lock_bh(&sk->sk_callback_lock);
+	if (sk_has_sleeper(sk))
+		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
+						POLLRDNORM | POLLRDBAND);
+	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
+	read_unlock_bh(&sk->sk_callback_lock);
+}
+
 /* If socket sndbuf has changed, wake up all per association waiters.  */
 void sctp_write_space(struct sock *sk)
 {
#4 lore raw
In reply to Wei Yongjun
Quoted 10 lines · Wei Yongjun
Quoted 8 lines · Vlad Yasevich
Quoted 6 lines · Wei Yongjun

sk->sk_data_ready() of sctp socket can be called from both BH and non-BH contexts, but the default sk->sk_data_ready(), sock_def_readable(), can not be used in this case. Therefore, we have to make a new function sctp_data_ready() to grab sk->sk_data_ready() with BH disabling.

Quoted 5 lines
Quoted 3 lines

Wouldn't the same inversion happen in TCP as well? TCP can call that function in _bh and user contexts as well.

Not sure, but TCP does not call that function in user context at all.

It might as part of release_sock(). It call out to sk_backlog_rcv, which calls tcp_v4_do_rcv() (the backlog handler). That handler will eventually call tcp_data_queue() which will call sk_data_ready(). I don't see the code paths that disable BH there.

-vlad

Quoted 78 lines
Quoted 57 lines

-vlad

Quoted 53 lines

============================>>> [ INFO: possible irq lock inversion dependency detected ] 2.6.33-rc6 #129
---------------------------------------------------------
sctp_darn/1517 just changed the state of lock:
(clock-AF_INET){++.?..}, at: [<c06aab60>] sock_def_readable+0x20/0x80
but this lock took another, SOFTIRQ-unsafe lock in the past:
(slock-AF_INET){+.-...}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
1 lock held by sctp_darn/1517:
#0: (sk_lock-AF_INET){+.+.+.}, at: [<cdfe363d>] sctp_sendmsg+0x23d/0xc00 [sctp]

---
 include/net/sctp/sctp.h |    1 +
 net/sctp/endpointola.c  |    1 +
 net/sctp/socket.c       |   10 ++++++++++
 3 files changed, 12 insertions(+), 0 deletions(-)
include/net/sctp/sctp.h +1 -0
index 78740ec..fa6cde5 100644
@@ -128,6 +128,7 @@ extern int sctp_register_pf(struct sctp_pf *, sa_family_t);
 int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb);
 int sctp_inet_listen(struct socket *sock, int backlog);
 void sctp_write_space(struct sock *sk);
+void sctp_data_ready(struct sock *sk, int len);
 unsigned int sctp_poll(struct file *file, struct socket *sock,
 		poll_table *wait);
 void sctp_sock_rfree(struct sk_buff *skb);
net/sctp/endpointola.c +1 -0
index 905fda5..7ec09ba 100644
@@ -144,6 +144,7 @@ static struct sctp_endpoint *sctp_endpoint_init(struct sctp_endpoint *ep,
 	/* Use SCTP specific send buffer space queues.  */
 	ep->sndbuf_policy = sctp_sndbuf_policy;
 
+	sk->sk_data_ready = sctp_data_ready;
 	sk->sk_write_space = sctp_write_space;
 	sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
 
net/sctp/socket.c +0 -0
index 67fdac9..b437e2a 100644
@@ -6185,6 +6185,16 @@ do_nonblock:
 	goto out;
 }
 
Quoted 16 lines
Quoted 4 lines
net/sctp/socket.c +3 -0
+void sctp_data_ready(struct sock *sk, int len)
+{
+	read_lock_bh(&sk->sk_callback_lock);

net/sctp/socket.c do_nonblock: · line 6200

Quoted 10 lines
net/sctp/socket.c +7 -0
+	if (sk_has_sleeper(sk))
+		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
+						POLLRDNORM | POLLRDBAND);
+	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
+	read_unlock_bh(&sk->sk_callback_lock);
+}
+
 /* If socket sndbuf has changed, wake up all per association waiters.  */
 void sctp_write_space(struct sock *sk)
 {

--
To unsubscribe from this list: send the line "unsubscribe linux-sctp" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#5 lore raw
In reply to Wei Yongjun
Quoted 10 lines · Wei Yongjun
Quoted 8 lines · Vlad Yasevich
Quoted 6 lines · Wei Yongjun

sk->sk_data_ready() of sctp socket can be called from both BH and non-BH contexts, but the default sk->sk_data_ready(), sock_def_readable(), can not be used in this case. Therefore, we have to make a new function sctp_data_ready() to grab sk->sk_data_ready() with BH disabling.

Quoted 6 lines
Quoted 3 lines

Wouldn't the same inversion happen in TCP as well? TCP can call that function in _bh and user contexts as well.

Not sure, but TCP does not call that function in user context at all.

Wei

Can you trigger this problem with this patch applied?

net/sctp/socket.c +1 -1
index 3a95fcb..dabdc50 100644
@@ -3717,9 +3717,9 @@ SCTP_STATIC int sctp_init_sock(struct sock *sk)
 	sp->hmac = NULL;
 
 	SCTP_DBG_OBJCNT_INC(sock);
-	percpu_counter_inc(&sctp_sockets_allocated);
 
 	local_bh_disable();
+	percpu_counter_inc(&sctp_sockets_allocated);
 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 	local_bh_enable();
 

-vlad

Quoted 73 lines
Quoted 72 lines

-vlad

Quoted 67 lines

============================>>> [ INFO: possible irq lock inversion dependency detected ] 2.6.33-rc6 #129
---------------------------------------------------------
sctp_darn/1517 just changed the state of lock:
(clock-AF_INET){++.?..}, at: [<c06aab60>] sock_def_readable+0x20/0x80
but this lock took another, SOFTIRQ-unsafe lock in the past:
(slock-AF_INET){+.-...}

and interrupts could create inverse lock ordering between them.

other info that might help us debug this:
1 lock held by sctp_darn/1517:
#0: (sk_lock-AF_INET){+.+.+.}, at: [<cdfe363d>] sctp_sendmsg+0x23d/0xc00 [sctp]

---
 include/net/sctp/sctp.h |    1 +
 net/sctp/endpointola.c  |    1 +
 net/sctp/socket.c       |   10 ++++++++++
 3 files changed, 12 insertions(+), 0 deletions(-)
include/net/sctp/sctp.h +1 -0
index 78740ec..fa6cde5 100644
@@ -128,6 +128,7 @@ extern int sctp_register_pf(struct sctp_pf *, sa_family_t);
 int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb);
 int sctp_inet_listen(struct socket *sock, int backlog);
 void sctp_write_space(struct sock *sk);
+void sctp_data_ready(struct sock *sk, int len);
 unsigned int sctp_poll(struct file *file, struct socket *sock,
 		poll_table *wait);
 void sctp_sock_rfree(struct sk_buff *skb);
net/sctp/endpointola.c +1 -0
index 905fda5..7ec09ba 100644
@@ -144,6 +144,7 @@ static struct sctp_endpoint *sctp_endpoint_init(struct sctp_endpoint *ep,
 	/* Use SCTP specific send buffer space queues.  */
 	ep->sndbuf_policy = sctp_sndbuf_policy;
 
+	sk->sk_data_ready = sctp_data_ready;
 	sk->sk_write_space = sctp_write_space;
 	sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
 
net/sctp/socket.c +10 -0
index 67fdac9..b437e2a 100644
@@ -6185,6 +6185,16 @@ do_nonblock:
 	goto out;
 }
 
+void sctp_data_ready(struct sock *sk, int len)
+{
+	read_lock_bh(&sk->sk_callback_lock);
+	if (sk_has_sleeper(sk))
+		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
+						POLLRDNORM | POLLRDBAND);
+	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
+	read_unlock_bh(&sk->sk_callback_lock);
+}
+
 /* If socket sndbuf has changed, wake up all per association waiters.  */
 void sctp_write_space(struct sock *sk)
 {
#6 lore raw
In reply to Wei Yongjun
Quoted 32 lines · Vlad Yasevich
Quoted 12 lines · Wei Yongjun
Quoted 10 lines · Vlad Yasevich
Quoted 7 lines · Wei Yongjun

sk->sk_data_ready() of sctp socket can be called from both BH and non-BH contexts, but the default sk->sk_data_ready(), sock_def_readable(), can not be used in this case. Therefore, we have to make a new function sctp_data_ready() to grab sk->sk_data_ready() with BH disabling.

Quoted 7 lines
Quoted 4 lines

Wouldn't the same inversion happen in TCP as well? TCP can call that function in _bh and user contexts as well.

Not sure, but TCP does not call that function in user context at all.

Wei

Can you trigger this problem with this patch applied?

net/sctp/socket.c +0 -0
index 3a95fcb..dabdc50 100644
@@ -3717,9 +3717,9 @@ SCTP_STATIC int sctp_init_sock(struct sock *sk)
 	sp->hmac = NULL;

net/sctp/socket.c SCTP_STATIC int sctp_init_sock(struct sock *sk)

Quoted 10 lines
net/sctp/socket.c +1 -2
 
 	SCTP_DBG_OBJCNT_INC(sock);
-	percpu_counter_inc(&sctp_sockets_allocated);
 
 	local_bh_disable();
+	percpu_counter_inc(&sctp_sockets_allocated);
 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 	local_bh_enable();
 
-vlad

This patch did not change anything, the lockdep INFO still exists.

This lockdep INFO happend when I try to make a user abort through sctp_primitive_ABORT.