net: add accounting for socket backlog
Subsystems (4)
PATCH v2 1/1 net: add accounting for socket backlog
We got system OOM while running some UDP netperf testing on the loopback device. The case is multiple senders sent stream UDP packets to a single receiver via loopback on local host. Of course, the receiver is not able to handle all the packets in time. But we surprisingly found that these packets were not discarded due to the receiver's sk->sk_rcvbuf limit. Instead, they are kept queuing to sk->sk_backlog and finally ate up all the memory. We believe this is a secure hole that a none privileged user can crash the system.
The root cause for this problem is, when the receiver is doing __release_sock() (i.e. after userspace recv, kernel udp_recvmsg -> skb_free_datagram_locked -> release_sock), it moves skbs from backlog to sk_receive_queue with the softirq enabled. In the above case, multiple busy senders will almost make it an endless loop. The skbs in the backlog end up eat all the system memory.
The patch fixed this problem by adding accounting for the socket backlog. So that the backlog size can be restricted by protocol's choice (i.e. UDP).
---
V2: remove atomic operation for sk_backlog.len
limit UDP backlog size to 2*sk->sk_rcvbuf
include/net/sock.h +12 -0
@@ -253,6 +253,7 @@ struct sock {struct {struct sk_buff *head;struct sk_buff *tail; +int len; } sk_backlog;wait_queue_head_t *sk_sleep;struct dst_entry *sk_dst_cache;
@@ -583,11 +584,22 @@ static inline void sk_add_backlog(struct sock *sk, struct sk_buff *skb) sk->sk_backlog.tail->next = skb; sk->sk_backlog.tail = skb; } + sk->sk_backlog.len += skb->truesize; skb->next =NULL ; } +static inline int sk_add_backlog_limited (struct sock *sk,struct sk_buff *skb) +{ +if (sk->sk_backlog.len >=2 * sk->sk_rcvbuf) +return -ENOBUFS ; + +sk_add_backlog (sk, skb); +return 0 ; +} +static inline int sk_backlog_rcv (struct sock *sk,struct sk_buff *skb) { + sk->sk_backlog.len -= skb->truesize;return sk->sk_backlog_rcv (sk, skb); }
net/core/sock.c +1 -0
@@ -1138,6 +1138,7 @@ struct sock *sk_clone(const struct sock *sk, const gfp_t priority)sock_lock_init (newsk);bh_lock_sock (newsk); newsk->sk_backlog.head = newsk->sk_backlog.tail =NULL ; + newsk->sk_backlog.len =0 ;atomic_set (&newsk->sk_rmem_alloc,0 );/*
net/ipv4/udp.c +4 -2
@@ -1372,8 +1372,10 @@ int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)bh_lock_sock (sk);if (!sock_owned_by_user (sk)) rc =__udp_queue_rcv_skb (sk, skb); -else -sk_add_backlog (sk, skb); +else if (sk_add_backlog_limited (sk, skb)) { +bh_unlock_sock (sk); + goto drop; + }bh_unlock_sock (sk);return rc;
net/ipv6/udp.c +16 -10
@@ -584,16 +584,19 @@ static void flush_stack(struct sock **stack, unsigned int count,bh_lock_sock (sk);if (!sock_owned_by_user (sk))udpv6_queue_rcv_skb (sk, skb1); -else -sk_add_backlog (sk, skb1); +else if (sk_add_backlog_limited (sk, skb1)) { +bh_unlock_sock (sk); + goto drop; + }bh_unlock_sock (sk); - }else { -atomic_inc (&sk->sk_drops); -UDP6_INC_STATS_BH (sock_net (sk), -UDP_MIB_RCVBUFERRORS ,IS_UDPLITE (sk)); -UDP6_INC_STATS_BH (sock_net (sk), -UDP_MIB_INERRORS ,IS_UDPLITE (sk)); +continue ; } +drop: +atomic_inc (&sk->sk_drops); +UDP6_INC_STATS_BH (sock_net (sk), +UDP_MIB_RCVBUFERRORS ,IS_UDPLITE (sk)); +UDP6_INC_STATS_BH (sock_net (sk), +UDP_MIB_INERRORS ,IS_UDPLITE (sk)); } }/*
@@ -756,8 +759,11 @@ int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,bh_lock_sock (sk);if (!sock_owned_by_user (sk))udpv6_queue_rcv_skb (sk, skb); -else -sk_add_backlog (sk, skb); +else if (sk_add_backlog_limited (sk, skb)) { +bh_unlock_sock (sk); +sock_put (sk); + goto discard; + }bh_unlock_sock (sk);sock_put (sk);return 0 ;
David Miller
maintainerFrom: Zhu Yi <[email protected]>
Date: Fri, 26 Feb 2010 17:27:44 +0800
Quoted 16 lines
We got system OOM while running some UDP netperf testing on the loopback device. The case is multiple senders sent stream UDP packets to a single receiver via loopback on local host. Of course, the receiver is not able to handle all the packets in time. But we surprisingly found that these packets were not discarded due to the receiver's
sk->sk_rcvbuflimit. Instead, they are kept queuing tosk->sk_backlogand finally ate up all the memory. We believe this is a secure hole that a none privileged user can crash the system.The root cause for this problem is, when the receiver is doing
__release_sock()(i.e. after userspace recv, kerneludp_recvmsg->skb_free_datagram_locked->release_sock), it moves skbs from backlog tosk_receive_queuewith the softirq enabled. In the above case, multiple busy senders will almost make it an endless loop. The skbs in the backlog end up eat all the system memory.
Quoted 8 lines
The patch fixed this problem by adding accounting for the socket backlog. So that the backlog size can be restricted by protocol's choice (i.e. UDP).
Reported-by Alex Shi <[email protected]>Cc David Miller <[email protected]>Cc Eric Dumazet <[email protected]>Signed-off-by Zhu Yi <[email protected]>
So remind me why TCP, or any other non-UDP protocol, won't intrinsically have this problem too?
It seems pretty trivial to do with any protocol, especially remotely, with a packet generator. The code in TCP, for example, which queues to the backlog, doesn't care about sequence numbers or anything like that.
So you could spray a machine with the same TCP frame over and over again, as fast as possible, as long as it matches the socket identity.
And in this way fill up the backlog endlessly and OOM the system.
Quoted 2 lines · David Miller
So remind me why TCP, or any other non-UDP protocol, won't intrinsically have this problem too?
If TCP ACKs are not received, the (closed) remote window prevents the TCP sender to send more frames.
Quoted 9 lines
It seems pretty trivial to do with any protocol, especially remotely, with a packet generator. The code in TCP, for example, which queues to the backlog, doesn't care about sequence numbers or anything like that.
So you could spray a machine with the same TCP frame over and over again, as fast as possible, as long as it matches the socket identity.
And in this way fill up the backlog endlessly and OOM the system.
Yeah, I only considered about the normal case, that is the TCP frames are built and managed in the kernel. If a user does frame generation himself, yes, the same problem could happen potentially for all protocols using backlog.
Thanks,
-yi
David Miller
maintainerFrom: Zhu Yi <[email protected]>
Date: Mon, 01 Mar 2010 10:08:48 +0800
Quoted 2 lines
Yeah, I only considered about the normal case, that is the TCP frames are built and managed in the kernel.
You're not even considering the kernel case completely. It's just as easy to modify the kernel to maliciously send frames in this way.
Quoted 2 lines
If a user does frame generation himself, yes, the same problem could happen potentially for all protocols using backlog.
We need the protection for every protocol, please implement your changes this way.
Thanks.
Quoted 30 lines · Zhu Yi
We got system OOM while running some UDP netperf testing on the loopback device. The case is multiple senders sent stream UDP packets to a single receiver via loopback on local host. Of course, the receiver is not able to handle all the packets in time. But we surprisingly found that these packets were not discarded due to the receiver's
sk->sk_rcvbuflimit. Instead, they are kept queuing tosk->sk_backlogand finally ate up all the memory. We believe this is a secure hole that a none privileged user can crash the system.The root cause for this problem is, when the receiver is doing
__release_sock()(i.e. after userspace recv, kerneludp_recvmsg->skb_free_datagram_locked->release_sock), it moves skbs from backlog tosk_receive_queuewith the softirq enabled. In the above case, multiple busy senders will almost make it an endless loop. The skbs in the backlog end up eat all the system memory.The patch fixed this problem by adding accounting for the socket backlog. So that the backlog size can be restricted by protocol's choice (i.e. UDP).
Reported-by Alex Shi <[email protected]>Cc David Miller <[email protected]>Cc Eric Dumazet <[email protected]>Signed-off-by Zhu Yi <[email protected]>---
V2: remove atomic operation for sk_backlog.lenlimit UDP backlog size to 2*sk->sk_rcvbuf
Review
Quoted 6 lines
(quoted diff) +2
-0+ static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) { + sk->sk_backlog.len -= skb->truesize; return sk->sk_backlog_rcv(sk, skb); }
I am afraid sk_backlog_rcv() is not always called with lock held, and not always called to process backlog (see TCP ucopy.prequeue)
If you take a look at __release_sock() for example, we make the backlog private to the process before handling it (outside of lock_sock())
Therefore, I suggest doing the 'substraction' outside of sk_backlog_rcv().
net/core/sock.c +1 -0
@@ -1520,6 +1520,7 @@ static void __release_sock(struct sock *sk) do { sk->sk_backlog.head = sk->sk_backlog.tail = NULL; + sk->sk_backlog.len = 0; bh_unlock_sock(sk); do {
Ah, I see __release_sock() is already doing a preemption check, please ignore my previous comment, when I said "__release_sock() could run forever with no preemption, even with a limit on backlog"
Thanks
Quoted 16 lines · Eric Dumazet
I am afraid
sk_backlog_rcv()is not always called with lock held, and not always called to process backlog (see TCP ucopy.prequeue)If you take a look at
__release_sock()for example, we make the backlog private to the process before handling it (outside oflock_sock())Therefore, I suggest doing the 'substraction' outside of
sk_backlog_rcv().net/core/sock.c +0
-0
net/core/sock.c static void __release_sock(struct sock *sk) · line 1526
Quoted 7 lines
net/core/sock.c +1
-0@@ -1520,6 +1520,7 @@ static void __release_sock(struct sock *sk) do { sk->sk_backlog.head = sk->sk_backlog.tail = NULL; + sk->sk_backlog.len = 0; bh_unlock_sock(sk); do {
Thinking again about this, doing this zero initialization at the very end of __release_sock() solves the problem of potential infinite loop in __release_sock(). Since producer will hit the backlog limit.
Thanks
net/core/sock.c +5 -0
@@ -1542,6 +1542,11 @@ static void __release_sock(struct sock *sk)bh_lock_sock (sk); }while ((skb = sk->sk_backlog.head) !=NULL ); +/* +* Doing this zeroing at the end of this function guarantee we can not +* loop forever while a wild producer attempts to flood us +*/ + sk->sk_backlog.len =0 ; }/**
Quoted 18 lines · Eric Dumazet
Thinking again about this, doing this zero initialization at the very end of
__release_sock()solves the problem of potential infinite loop in__release_sock(). Since producer will hit the backlog limit.Thanks
net/core/sock.c +2
-0@@ -1542,6 +1542,11 @@ static void __release_sock(struct sock *sk)bh_lock_sock (sk); }while ((skb = sk->sk_backlog.head) !=NULL ); +/* +* Doing this zeroing at the end of this function guarantee we can not
net/core/sock.c static void __release_sock(struct sock *sk) · line 1550
Quoted 4 lines
net/core/sock.c +3
-0+* loop forever while a wild producer attempts to flood us +*/ + sk->sk_backlog.len =0 ; }
Very good point. I'll add your sob in my next patch.
Thanks,
-yi