packet_sendmsg_spkt sleeping from invalid context
Subsystems (3)
Hi,
I don't know if it has been reported already.
I get the following warning on boot, with latest upstream tree:
[ 32.776502] sched: BUG: sleeping function called from invalid context at mm/slab.c:3032
[ 32.802173] sched: in_atomic(): 1, irqs_disabled(): 0, pid: 3555, name: dhclient3
[ 32.821141] 1 lock held by dhclient3/3555:
[ 32.821147] #0: (rcu_read_lock){.+.+.+}, at: [<ffffffff815d177d>] packet_sendmsg_spkt+0x7d/0x2c0
[ 32.821174] Pid: 3555, comm: dhclient3 Tainted: G W 2.6.32-tip+ #134
[ 32.821181] Call Trace:
[ 32.821194] [<ffffffff810811c3>] ? __debug_show_held_locks+0x13/0x30
[ 32.821207] [<ffffffff8103d6b8>] __might_sleep+0x118/0x140
[ 32.821219] [<ffffffff81110c23>] kmem_cache_alloc+0x173/0x190
[ 32.821231] [<ffffffff815483d9>] __alloc_skb+0x49/0x170
[ 32.821241] [<ffffffff81542238>] sock_wmalloc+0x38/0x80
[ 32.821250] [<ffffffff815d182b>] packet_sendmsg_spkt+0x12b/0x2c0
[ 32.821260] [<ffffffff815d177d>] ? packet_sendmsg_spkt+0x7d/0x2c0
[ 32.821272] [<ffffffff8153ded7>] sock_sendmsg+0x127/0x140
[ 32.821285] [<ffffffff8106f4a0>] ? autoremove_wake_function+0x0/0x40
[ 32.821297] [<ffffffff810f467b>] ? might_fault+0x7b/0xd0
[ 32.821306] [<ffffffff810f467b>] ? might_fault+0x7b/0xd0
[ 32.821318] [<ffffffff815401aa>] ? move_addr_to_kernel+0x6a/0x70
[ 32.821328] [<ffffffff8154029f>] sys_sendto+0xef/0x120
[ 32.821340] [<ffffffff81131479>] ? mntput_no_expire+0x29/0x110
[ 32.821355] [<ffffffff810027db>] system_call_fastpath+0x16/0x1bQuoted 18 lines · Frederic Weisbecker
Hi,
I don't know if it has been reported already.
I get the following warning on boot, with latest upstream tree:[ 32.776502] sched: BUG: sleeping function called from invalid context at mm/slab.c:3032 [ 32.802173] sched: in_atomic(): 1, irqs_disabled(): 0, pid: 3555, name: dhclient3 [ 32.821141] 1 lock held by dhclient3/3555: [ 32.821147] #0: (rcu_read_lock){.+.+.+}, at: [<ffffffff815d177d>] packet_sendmsg_spkt+0x7d/0x2c0 [ 32.821174] Pid: 3555, comm: dhclient3 Tainted: G W 2.6.32-tip+ #134 [ 32.821181] Call Trace: [ 32.821194] [<ffffffff810811c3>] ? __debug_show_held_locks+0x13/0x30 [ 32.821207] [<ffffffff8103d6b8>] __might_sleep+0x118/0x140 [ 32.821219] [<ffffffff81110c23>] kmem_cache_alloc+0x173/0x190 [ 32.821231] [<ffffffff815483d9>] __alloc_skb+0x49/0x170 [ 32.821241] [<ffffffff81542238>] sock_wmalloc+0x38/0x80 [ 32.821250] [<ffffffff815d182b>] packet_sendmsg_spkt+0x12b/0x2c0 [ 32.821260] [<ffffffff815d177d>] ? packet_sendmsg_spkt+0x7d/0x2c0
Quoted 9 lines
[ 32.821272] [<ffffffff8153ded7>] sock_sendmsg+0x127/0x140 [ 32.821285] [<ffffffff8106f4a0>] ? autoremove_wake_function+0x0/0x40 [ 32.821297] [<ffffffff810f467b>] ? might_fault+0x7b/0xd0 [ 32.821306] [<ffffffff810f467b>] ? might_fault+0x7b/0xd0 [ 32.821318] [<ffffffff815401aa>] ? move_addr_to_kernel+0x6a/0x70 [ 32.821328] [<ffffffff8154029f>] sys_sendto+0xef/0x120 [ 32.821340] [<ffffffff81131479>] ? mntput_no_expire+0x29/0x110 [ 32.821355] [<ffffffff810027db>] system_call_fastpath+0x16/0x1b
Thanks for the report Frederic.
We could partly revert the original commit, but as we wanted to avoid touching device refcount, and af_packet might be the only real abuser, we could try following patch instead.
Thanks
[PATCH] packet: dont call sleeping function while holding rcu_read_lock()
commit 654d1f8a019dfa06d (packet: less dev_put() calls) introduced a problem, calling a potentially sleeping function from a rcu_read_lock() protected section.
Fix this by releasing lock before the sock_wmalloc() call. After skb allocation, we redo device lookup and appropriate tests.
--- net/packet/af_packet.c | 36 +++++++++++++----------------------- 1 file changed, 13 insertions(+), 23 deletions(-)
net/packet/af_packet.c +13 -23
@@ -415,7 +415,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, {struct sock *sk = sock->sk;struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name; -struct sk_buff *skb; +struct sk_buff *skb =NULL ;struct net_device *dev;__be16 proto =0 ;int err;
@@ -437,6 +437,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,*/ saddr->spkt_device[13 ] =0 ; +retry:rcu_read_lock (); dev =dev_get_by_name_rcu (sock_net (sk), saddr->spkt_device); err = -ENODEV ;
@@ -456,27 +457,21 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,if (len > dev->mtu + dev->hard_header_len) goto out_unlock; - err = -ENOBUFS ; - skb =sock_wmalloc (sk, len +LL_RESERVED_SPACE (dev),0 ,GFP_KERNEL ); - -/* -* If the write buffer is full, then tough. At this level the user -* gets to deal with the problem - do your own algorithmic backoffs. -* That's far more flexible. -*/ +if (!skb) { +size_t reserved =LL_RESERVED_SPACE (dev); -if (skb ==NULL ) - goto out_unlock; - -/* -* Fill it in -*/ +rcu_read_unlock (); + skb =sock_wmalloc (sk, len + reserved,0 ,GFP_KERNEL ); +if (skb ==NULL ) +return -ENOBUFS ; +skb_reserve (skb, reserved); + goto retry; + }/* FIXME: Save some space for broken drivers that write a * hard header at transmission time by themselves. PPP is the * notable one here. This should really be fixed at the driver level. */ -skb_reserve (skb,LL_RESERVED_SPACE (dev));skb_reset_network_header (skb);/* Try to align data part correctly */
@@ -494,20 +489,15 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, skb->priority = sk->sk_priority; skb->mark = sk->sk_mark;if (err) - goto out_free; - -/* -* Now send it -*/ + goto out_unlock;dev_queue_xmit (skb);rcu_read_unlock ();return len; -out_free: -kfree_skb (skb); out_unlock:rcu_read_unlock (); +kfree_skb (skb);return err; }
Quoted 94 lines · Eric Dumazet
Thanks for the report Frederic.
We could partly revert the original commit, but as we wanted to avoid touching device refcount, and
af_packetmight be the only real abuser, we could try following patch instead.Thanks
[PATCH] packet: dont call sleeping function while holding
rcu_read_lock()commit
654d1f8a019dfa06d(packet: lessdev_put()calls) introduced a problem, calling a potentially sleeping function from arcu_read_lock()protected section.Fix this by releasing lock before the
sock_wmalloc()call. After skb allocation, we redo device lookup and appropriate tests.Reported-by Frederic Weisbecker <[email protected]>Signed-off-by Eric Dumazet <[email protected]>--- net/packet/af_packet.c | 36 +++++++++++++----------------------- 1 file changed, 13 insertions(+), 23 deletions(-)net/packet/af_packet.c +12
-21@@ -415,7 +415,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, {struct sock *sk = sock->sk;struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name; -struct sk_buff *skb; +struct sk_buff *skb =NULL ;struct net_device *dev;__be16 proto =0 ;int err;@@ -437,6 +437,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,*/ saddr->spkt_device[13 ] =0 ; +retry:rcu_read_lock (); dev =dev_get_by_name_rcu (sock_net (sk), saddr->spkt_device); err = -ENODEV ;@@ -456,27 +457,21 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,if (len > dev->mtu + dev->hard_header_len) goto out_unlock; - err = -ENOBUFS ; - skb =sock_wmalloc (sk, len +LL_RESERVED_SPACE (dev),0 ,GFP_KERNEL ); - -/* -* If the write buffer is full, then tough. At this level the user -* gets to deal with the problem - do your own algorithmic backoffs. -* That's far more flexible. -*/ +if (!skb) { +size_t reserved =LL_RESERVED_SPACE (dev); -if (skb ==NULL ) - goto out_unlock; - -/* -* Fill it in -*/ +rcu_read_unlock (); + skb =sock_wmalloc (sk, len + reserved,0 ,GFP_KERNEL ); +if (skb ==NULL ) +return -ENOBUFS ; +skb_reserve (skb, reserved); + goto retry; + }/* FIXME: Save some space for broken drivers that write a * hard header at transmission time by themselves. PPP is the * notable one here. This should really be fixed at the driver level. */ -skb_reserve (skb,LL_RESERVED_SPACE (dev));skb_reset_network_header (skb);/* Try to align data part correctly */ @@ -494,20 +489,15 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, skb->priority = sk->sk_priority; skb->mark = sk->sk_mark;if (err) - goto out_free; - -/* -* Now send it -*/ + goto out_unlock;dev_queue_xmit (skb);rcu_read_unlock ();
net/packet/af_packet.c static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, · line 503
Quoted 10 lines
net/packet/af_packet.c +1
-2return len; -out_free: -kfree_skb (skb); out_unlock:rcu_read_unlock (); +kfree_skb (skb);return err; }
Thanks, yeah it fixes the problem but unearthes a new one:
[ 32.428785] sched: BUG: sleeping function called from invalid context at mm/memory.c:3369
[ 32.454154] sched: in_atomic(): 1, irqs_disabled(): 0, pid: 3531, name: dhclient3
[ 32.472866] 1 lock held by dhclient3/3531:
[ 32.472872] #0: (rcu_read_lock){.+.+..}, at: [<ffffffff815d1a5e>] packet_sendmsg_spkt+0xce/0x340
[ 32.472900] Pid: 3531, comm: dhclient3 Tainted: G W 2.6.32-tip+ #135
[ 32.472906] Call Trace:
[ 32.472920] [<ffffffff81081403>] ? __debug_show_held_locks+0x13/0x30
[ 32.472933] [<ffffffff8103d6b8>] __might_sleep+0x118/0x140
[ 32.472944] [<ffffffff810f48cb>] might_fault+0x3b/0xd0
[ 32.472955] [<ffffffff81549c1e>] memcpy_fromiovec+0x6e/0xa0
[ 32.472965] [<ffffffff815d1c44>] packet_sendmsg_spkt+0x2b4/0x340
[ 32.472975] [<ffffffff815d1a5e>] ? packet_sendmsg_spkt+0xce/0x340
[ 32.472986] [<ffffffff8153e167>] sock_sendmsg+0x127/0x140
[ 32.472999] [<ffffffff8106f6e0>] ? autoremove_wake_function+0x0/0x40
[ 32.473009] [<ffffffff810f490b>] ? might_fault+0x7b/0xd0
[ 32.473019] [<ffffffff810f490b>] ? might_fault+0x7b/0xd0
[ 32.473030] [<ffffffff8154043a>] ? move_addr_to_kernel+0x6a/0x70
[ 32.473040] [<ffffffff8154052f>] sys_sendto+0xef/0x120
[ 32.473053] [<ffffffff81131709>] ? mntput_no_expire+0x29/0x110
[ 32.473067] [<ffffffff810027db>] system_call_fastpath+0x16/0x1bAnd I guess you need to protect dev until the packet is submitted. Looks tricky...
I've searched a kind of get_net_dev() but did not find any :)
Quoted 29 lines · Eric Dumazet
Quoted 18 lines · Frederic Weisbecker
Hi,
I don't know if it has been reported already.
I get the following warning on boot, with latest upstream tree:[ 32.776502] sched: BUG: sleeping function called from invalid context at mm/slab.c:3032 [ 32.802173] sched: in_atomic(): 1, irqs_disabled(): 0, pid: 3555, name: dhclient3 [ 32.821141] 1 lock held by dhclient3/3555: [ 32.821147] #0: (rcu_read_lock){.+.+.+}, at: [<ffffffff815d177d>] packet_sendmsg_spkt+0x7d/0x2c0 [ 32.821174] Pid: 3555, comm: dhclient3 Tainted: G W 2.6.32-tip+ #134 [ 32.821181] Call Trace: [ 32.821194] [<ffffffff810811c3>] ? __debug_show_held_locks+0x13/0x30 [ 32.821207] [<ffffffff8103d6b8>] __might_sleep+0x118/0x140 [ 32.821219] [<ffffffff81110c23>] kmem_cache_alloc+0x173/0x190 [ 32.821231] [<ffffffff815483d9>] __alloc_skb+0x49/0x170 [ 32.821241] [<ffffffff81542238>] sock_wmalloc+0x38/0x80 [ 32.821250] [<ffffffff815d182b>] packet_sendmsg_spkt+0x12b/0x2c0 [ 32.821260] [<ffffffff815d177d>] ? packet_sendmsg_spkt+0x7d/0x2c0Quoted 9 lines
[ 32.821272] [<ffffffff8153ded7>] sock_sendmsg+0x127/0x140 [ 32.821285] [<ffffffff8106f4a0>] ? autoremove_wake_function+0x0/0x40 [ 32.821297] [<ffffffff810f467b>] ? might_fault+0x7b/0xd0 [ 32.821306] [<ffffffff810f467b>] ? might_fault+0x7b/0xd0 [ 32.821318] [<ffffffff815401aa>] ? move_addr_to_kernel+0x6a/0x70 [ 32.821328] [<ffffffff8154029f>] sys_sendto+0xef/0x120 [ 32.821340] [<ffffffff81131479>] ? mntput_no_expire+0x29/0x110 [ 32.821355] [<ffffffff810027db>] system_call_fastpath+0x16/0x1b
Quoted 7 lines
Thanks for the report Frederic.
We could partly revert the original commit, but as we wanted to avoid touching device refcount, and
af_packetmight be the only real abuser, we could try following patch instead.Thanks
I also wonder. Are you using PREEMPT_RCU ?
That may explain why you haven't seen this issue because might_sleep() doesn't see you are in a rcu read locked section as preemption is not disabled, but it is illegal to voluntarily sleep in such area (although it's fine with preempt rcu) as doing so with non-prempt RCU config would barf.
I'm trying a patch to handle that.
Quoted 2 lines · Frederic Weisbecker
I also wonder. Are you using
PREEMPT_RCU?
Not at all :)
But yes, this is illegal to do the memcpy_fromiovec() in rcu_read_lock() context.
Quoted 7 lines
That may explain why you haven't seen this issue because
might_sleep()doesn't see you are in a rcu read locked section as preemption is not disabled, but it is illegal to voluntarily sleep in such area (although it's fine with preempt rcu) as doing so with non-prempt RCU config would barf.I'm trying a patch to handle that.
As you want, I also have a patch testing right now :)
Quoted 7 lines · Eric Dumazet
Quoted 2 lines · Frederic Weisbecker
I also wonder. Are you using
PREEMPT_RCU?Not at all :)
But yes, this is illegal to do the
memcpy_fromiovec()inrcu_read_lock()context.
I've just tested, and with rcu preempt it is mute, no warning :)
Quoted 9 lines
Quoted 7 lines
That may explain why you haven't seen this issue because
might_sleep()doesn't see you are in a rcu read locked section as preemption is not disabled, but it is illegal to voluntarily sleep in such area (although it's fine with preempt rcu) as doing so with non-prempt RCU config would barf.I'm trying a patch to handle that.
As you want, I also have a patch testing right now :)
But mine is to teach might_sleep() to handle rcu preempt case, not to fix this net dev thing.
But I'll happily test the fix you have :)
Thanks.
Quoted 21 lines · Frederic Weisbecker
Quoted 7 lines · Eric Dumazet
Quoted 2 lines · Frederic Weisbecker
I also wonder. Are you using
PREEMPT_RCU?Not at all :)
But yes, this is illegal to do the
memcpy_fromiovec()inrcu_read_lock()context.I've just tested, and with rcu preempt it is mute, no warning :)
Quoted 8 lines
Quoted 7 lines
That may explain why you haven't seen this issue because
might_sleep()doesn't see you are in a rcu read locked section as preemption is not disabled, but it is illegal to voluntarily sleep in such area (although it's fine with preempt rcu) as doing so with non-prempt RCU config would barf.I'm trying a patch to handle that.
Quoted 8 lines
Quoted 1 line
As you want, I also have a patch testing right now :)
But mine is to teach
might_sleep()to handle rcu preempt case, not to fix this net dev thing.But I'll happily test the fix you have :)
OK here it is, I tested it with PREEMPT_RCU as well
Thanks !
net/packet/af_packet.c +31 -40
@@ -415,7 +415,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, {struct sock *sk = sock->sk;struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name; -struct sk_buff *skb; +struct sk_buff *skb =NULL ;struct net_device *dev;__be16 proto =0 ;int err;
@@ -437,6 +437,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,*/ saddr->spkt_device[13 ] =0 ; +retry:rcu_read_lock (); dev =dev_get_by_name_rcu (sock_net (sk), saddr->spkt_device); err = -ENODEV ;
@@ -456,58 +457,48 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,if (len > dev->mtu + dev->hard_header_len) goto out_unlock; - err = -ENOBUFS ; - skb =sock_wmalloc (sk, len +LL_RESERVED_SPACE (dev),0 ,GFP_KERNEL ); - -/* -* If the write buffer is full, then tough. At this level the user -* gets to deal with the problem - do your own algorithmic backoffs. -* That's far more flexible. -*/ - -if (skb ==NULL ) - goto out_unlock; - -/* -* Fill it in -*/ - -/* FIXME: Save some space for broken drivers that write a -* hard header at transmission time by themselves. PPP is the -* notable one here. This should really be fixed at the driver level. -*/ -skb_reserve (skb,LL_RESERVED_SPACE (dev)); -skb_reset_network_header (skb); - -/* Try to align data part correctly */ -if (dev->header_ops) { - skb->data -= dev->hard_header_len; - skb->tail -= dev->hard_header_len; -if (len < dev->hard_header_len) -skb_reset_network_header (skb); +if (!skb) { +size_t reserved =LL_RESERVED_SPACE (dev); +unsigned int hhlen = dev->header_ops ? dev->hard_header_len :0 ; + +rcu_read_unlock (); + skb =sock_wmalloc (sk, len + reserved,0 ,GFP_KERNEL ); +if (skb ==NULL ) +return -ENOBUFS ; +skb_reserve (skb, reserved); +/* FIXME: Save some space for broken drivers that write a hard +* header at transmission time by themselves. PPP is the notable +* one here. This should really be fixed at the driver level. +*/ +skb_reset_network_header (skb); + +/* Try to align data part correctly */ +if (hhlen) { + skb->data -= hhlen; + skb->tail -= hhlen; +if (len < hhlen) +skb_reset_network_header (skb); + } + err =memcpy_fromiovec (skb_put (skb, len), msg->msg_iov, len); +if (err) + goto out_free; + goto retry; } -/* Returns -EFAULT on error */ - err =memcpy_fromiovec (skb_put (skb, len), msg->msg_iov, len); + skb->protocol = proto; skb->dev = dev; skb->priority = sk->sk_priority; skb->mark = sk->sk_mark; -if (err) - goto out_free; - -/* -* Now send it -*/ dev_queue_xmit (skb);rcu_read_unlock ();return len; -out_free: -kfree_skb (skb); out_unlock:rcu_read_unlock (); +out_free: +kfree_skb (skb);return err; }
Frederic Weisbecker
Tested-byQuoted 104 lines · Eric Dumazet
OK here it is, I tested it with
PREEMPT_RCUas wellThanks !
net/packet/af_packet.c +29
-38@@ -415,7 +415,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, {struct sock *sk = sock->sk;struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name; -struct sk_buff *skb; +struct sk_buff *skb =NULL ;struct net_device *dev;__be16 proto =0 ;int err;@@ -437,6 +437,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,*/ saddr->spkt_device[13 ] =0 ; +retry:rcu_read_lock (); dev =dev_get_by_name_rcu (sock_net (sk), saddr->spkt_device); err = -ENODEV ;@@ -456,58 +457,48 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,if (len > dev->mtu + dev->hard_header_len) goto out_unlock; - err = -ENOBUFS ; - skb =sock_wmalloc (sk, len +LL_RESERVED_SPACE (dev),0 ,GFP_KERNEL ); - -/* -* If the write buffer is full, then tough. At this level the user -* gets to deal with the problem - do your own algorithmic backoffs. -* That's far more flexible. -*/ - -if (skb ==NULL ) - goto out_unlock; - -/* -* Fill it in -*/ - -/* FIXME: Save some space for broken drivers that write a -* hard header at transmission time by themselves. PPP is the -* notable one here. This should really be fixed at the driver level. -*/ -skb_reserve (skb,LL_RESERVED_SPACE (dev)); -skb_reset_network_header (skb); - -/* Try to align data part correctly */ -if (dev->header_ops) { - skb->data -= dev->hard_header_len; - skb->tail -= dev->hard_header_len; -if (len < dev->hard_header_len) -skb_reset_network_header (skb); +if (!skb) { +size_t reserved =LL_RESERVED_SPACE (dev); +unsigned int hhlen = dev->header_ops ? dev->hard_header_len :0 ; + +rcu_read_unlock (); + skb =sock_wmalloc (sk, len + reserved,0 ,GFP_KERNEL ); +if (skb ==NULL ) +return -ENOBUFS ; +skb_reserve (skb, reserved); +/* FIXME: Save some space for broken drivers that write a hard +* header at transmission time by themselves. PPP is the notable +* one here. This should really be fixed at the driver level. +*/ +skb_reset_network_header (skb); + +/* Try to align data part correctly */ +if (hhlen) { + skb->data -= hhlen; + skb->tail -= hhlen; +if (len < hhlen) +skb_reset_network_header (skb); + } + err =memcpy_fromiovec (skb_put (skb, len), msg->msg_iov, len); +if (err) + goto out_free; + goto retry; } -/* Returns -EFAULT on error */ - err =memcpy_fromiovec (skb_put (skb, len), msg->msg_iov, len); + skb->protocol = proto; skb->dev = dev; skb->priority = sk->sk_priority; skb->mark = sk->sk_mark; -if (err) - goto out_free; - -/* -* Now send it -*/ dev_queue_xmit (skb);rcu_read_unlock ();return len;
net/packet/af_packet.c static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, · line 504
Quoted 10 lines
net/packet/af_packet.c +2
-2-out_free: -kfree_skb (skb); out_unlock:rcu_read_unlock (); +out_free: +kfree_skb (skb);return err; }
Thanks!
Quoted 2 lines · Frederic Weisbecker
Tested-by Frederic Weisbecker <[email protected]>
Thanks for testing Frederic, here is the official patch submission then :)
[PATCH] packet: dont call sleeping functions while holding rcu_read_lock()
commit 654d1f8a019dfa06d (packet: less dev_put() calls) introduced a problem, calling potentially sleeping functions from a rcu_read_lock() protected section.
Fix this by releasing lock before the sock_wmalloc()/memcpy_fromiovec() calls.
After skb allocation and copy from user space, we redo device lookup and appropriate tests.
--- net/packet/af_packet.c | 71 +++++++++++++++++---------------------- 1 file changed, 31 insertions(+), 40 deletions(-)
net/packet/af_packet.c +31 -40
@@ -415,7 +415,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock, {struct sock *sk = sock->sk;struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name; -struct sk_buff *skb; +struct sk_buff *skb =NULL ;struct net_device *dev;__be16 proto =0 ;int err;
@@ -437,6 +437,7 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,*/ saddr->spkt_device[13 ] =0 ; +retry:rcu_read_lock (); dev =dev_get_by_name_rcu (sock_net (sk), saddr->spkt_device); err = -ENODEV ;
@@ -456,58 +457,48 @@ static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,if (len > dev->mtu + dev->hard_header_len) goto out_unlock; - err = -ENOBUFS ; - skb =sock_wmalloc (sk, len +LL_RESERVED_SPACE (dev),0 ,GFP_KERNEL ); - -/* -* If the write buffer is full, then tough. At this level the user -* gets to deal with the problem - do your own algorithmic backoffs. -* That's far more flexible. -*/ - -if (skb ==NULL ) - goto out_unlock; - -/* -* Fill it in -*/ - -/* FIXME: Save some space for broken drivers that write a -* hard header at transmission time by themselves. PPP is the -* notable one here. This should really be fixed at the driver level. -*/ -skb_reserve (skb,LL_RESERVED_SPACE (dev)); -skb_reset_network_header (skb); - -/* Try to align data part correctly */ -if (dev->header_ops) { - skb->data -= dev->hard_header_len; - skb->tail -= dev->hard_header_len; -if (len < dev->hard_header_len) -skb_reset_network_header (skb); +if (!skb) { +size_t reserved =LL_RESERVED_SPACE (dev); +unsigned int hhlen = dev->header_ops ? dev->hard_header_len :0 ; + +rcu_read_unlock (); + skb =sock_wmalloc (sk, len + reserved,0 ,GFP_KERNEL ); +if (skb ==NULL ) +return -ENOBUFS ; +/* FIXME: Save some space for broken drivers that write a hard +* header at transmission time by themselves. PPP is the notable +* one here. This should really be fixed at the driver level. +*/ +skb_reserve (skb, reserved); +skb_reset_network_header (skb); + +/* Try to align data part correctly */ +if (hhlen) { + skb->data -= hhlen; + skb->tail -= hhlen; +if (len < hhlen) +skb_reset_network_header (skb); + } + err =memcpy_fromiovec (skb_put (skb, len), msg->msg_iov, len); +if (err) + goto out_free; + goto retry; } -/* Returns -EFAULT on error */ - err =memcpy_fromiovec (skb_put (skb, len), msg->msg_iov, len); + skb->protocol = proto; skb->dev = dev; skb->priority = sk->sk_priority; skb->mark = sk->sk_mark; -if (err) - goto out_free; - -/* -* Now send it -*/ dev_queue_xmit (skb);rcu_read_unlock ();return len; -out_free: -kfree_skb (skb); out_unlock:rcu_read_unlock (); +out_free: +kfree_skb (skb);return err; }
From: Eric Dumazet <[email protected]>
Date: Tue, 15 Dec 2009 16:47:03 +0100
Quoted 9 lines · Frederic Weisbecker
Quoted 3 lines
Tested-by Frederic Weisbecker <[email protected]>Thanks for testing Frederic, here is the official patch submission then :)
[PATCH] packet: dont call sleeping functions while holding
rcu_read_lock()
Applied, thanks.