Skip to contentSkip to conversation

bhash2 and WARN_ON() for inconsistent sk saddr.

RFC In review Kuniyuki Iwashima 4 participants 16 messages Oct 29, 2022 – Nov 10, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

RFC 1/1 bhash2 and WARN_ON() for inconsistent sk saddr.

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

For the very first implementation of bhash2, there was a similar report hitting the same WARN_ON(). The fix was to update the bhash2 bucket when the kernel changes sk->sk_rcv_saddr from INADDR_ANY. Then, syzbot has a repro, so we can indeed confirm that it no longer triggers the warning on the latest kernel.

https://lore.kernel.org/netdev/[email protected]/

However, Mat reported at that time that there were at least two variants, the latter being the same as mine.

https://lore.kernel.org/netdev/[email protected]/
https://lore.kernel.org/netdev/[email protected]/

This week I started looking into this issue and finally figured out why we could not catch all cases with a single repro.

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket. So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
  	inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

Here I think the s2 case also _should_ trigger WARN_ON(). The issue seems to be fixed, but it's just because we forgot to _fix_ the source address in error paths after inet6?_hash_connect() in tcp_v[46]_connect(). (Of course, DCCP as well).

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 	 * This unhashes the socket and releases the local port, if necessary.
 	 */
 	dccp_set_state(sk, DCCP_CLOSED);
+	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+		inet_reset_saddr(sk);
 	ip_rt_put(rt);
 	sk->sk_route_caps = 0;
 	inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
 	dccp_set_state(sk, DCCP_CLOSED);
+	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+		inet_reset_saddr(sk);
 	__sk_dst_reset(sk);
 failure:
 	inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 	 * if necessary.
 	 */
 	tcp_set_state(sk, TCP_CLOSE);
+	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+		inet_reset_saddr(sk);
 	ip_rt_put(rt);
 	sk->sk_route_caps = 0;
 	inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -0
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
 	tcp_set_state(sk, TCP_CLOSE);
+	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+		inet_reset_saddr(sk);
 failure:
 	inet->inet_dport = 0;
 	sk->sk_route_caps = 0;

---8<---

#1 lore raw

Hello Kuniyuki Iwashima,

Quoted 22 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

For the very first implementation of bhash2, there was a similar report hitting the same WARN_ON(). The fix was to update the bhash2 bucket when the kernel changes sk->sk_rcv_saddr from INADDR_ANY. Then, syzbot has a repro, so we can indeed confirm that it no longer triggers the warning on the latest kernel.

https://lore.kernel.org/netdev/[email protected]/

Quoted 9 lines

However, Mat reported at that time that there were at least two variants, the latter being the same as mine.

https://lore.kernel.org/netdev/[email protected]/
https://lore.kernel.org/netdev/[email protected]/

This week I started looking into this issue and finally figured out why we could not catch all cases with a single repro.

Provide another C repro for analysis. See the attachment.

Quoted 66 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 	 * This unhashes the socket and releases the local port, if necessary.
 	 */
 	dccp_set_state(sk, DCCP_CLOSED);
+	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+		inet_reset_saddr(sk);
 	ip_rt_put(rt);
 	sk->sk_route_caps = 0;
 	inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
 	dccp_set_state(sk, DCCP_CLOSED);
+	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+		inet_reset_saddr(sk);
 	__sk_dst_reset(sk);
 failure:
 	inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 	 * if necessary.
 	 */
 	tcp_set_state(sk, TCP_CLOSE);
+	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+		inet_reset_saddr(sk);
 	ip_rt_put(rt);
 	sk->sk_route_caps = 0;
 	inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
 	tcp_set_state(sk, TCP_CLOSE);
+	if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+		inet_reset_saddr(sk);
 failure:
 	inet->inet_dport = 0;
 	sk->sk_route_caps = 0;
---8<---
Quoted 2 lines

.

#define _GNU_SOURCE

#include <endian.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/syscall.h>
#include <sys/types.h>
#include <unistd.h>

uint64_t r[1] = {0xffffffffffffffff};

int main(void)

{
		syscall(__NR_mmap, 0x1ffff000ul, 0x1000ul, 0ul, 0x32ul, -1, 0ul);
	syscall(__NR_mmap, 0x20000000ul, 0x1000000ul, 7ul, 0x32ul, -1, 0ul);
	syscall(__NR_mmap, 0x21000000ul, 0x1000ul, 0ul, 0x32ul, -1, 0ul);
				intptr_t res = 0;
	res = syscall(__NR_socket, 0xaul, 1ul, 0);
	if (res != -1)
		r[0] = res;

*(uint16_t*)0x20000000 = 0xa;
*(uint16_t*)0x20000002 = htobe16(0x4e22);
*(uint32_t*)0x20000004 = htobe32(0);
memset((void*)0x20000008, 0, 16);

*(uint32_t*)0x20000018 = 0;
	syscall(__NR_bind, r[0], 0x20000000ul, 0x1cul);

*(uint16_t*)0x200000c0 = 0xa;
*(uint16_t*)0x200000c2 = htobe16(0x4e24);
*(uint32_t*)0x200000c4 = htobe32(0x285);
memset((void*)0x200000c8, 0, 16);

*(uint32_t*)0x200000d8 = 3;
	syscall(__NR_connect, r[0], 0x200000c0ul, 0x1cul);
	syscall(__NR_listen, r[0], 0);
	return 0;
}
#2 lore raw
From:   "Ziyang Xuan (William)" <[email protected]>
Date:   Tue, 1 Nov 2022 15:08:15 +0800
Quoted 27 lines

Hello Kuniyuki Iwashima,

Quoted 25 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

For the very first implementation of bhash2, there was a similar report hitting the same WARN_ON(). The fix was to update the bhash2 bucket when the kernel changes sk->sk_rcv_saddr from INADDR_ANY. Then, syzbot has a repro, so we can indeed confirm that it no longer triggers the warning on the latest kernel.

https://lore.kernel.org/netdev/[email protected]/

However, Mat reported at that time that there were at least two variants, the latter being the same as mine.

Quoted 8 lines
Quoted 6 lines

https://lore.kernel.org/netdev/[email protected]/
https://lore.kernel.org/netdev/[email protected]/

This week I started looking into this issue and finally figured out why we could not catch all cases with a single repro.

Provide another C repro for analysis. See the attachment.

Thanks for another variant.

Your repro also fails to connect() by RST, which resets saddr without updating bhash2 bucket, and then listen() hits the WARN_ON().

I meant to say if there was no difference in failure paths we should have caught all places where we need fixes with a single repro.

Once we know the root cause, it's not so hard to generate variants.

Anyway, I'll post a patch for consistent error handling and later another patch to fix the root cause when I find a solid way.

#3 lore raw
In reply to Kuniyuki Iwashima
Quoted 155 lines · Kuniyuki Iwashima

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

   WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
   ...
   inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
   inet_listen (net/ipv4/af_inet.c:228)
   __sys_listen (net/socket.c:1810)
   __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
   do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
   entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

For the very first implementation of bhash2, there was a similar report hitting the same WARN_ON(). The fix was to update the bhash2 bucket when the kernel changes sk->sk_rcv_saddr from INADDR_ANY. Then, syzbot has a repro, so we can indeed confirm that it no longer triggers the warning on the latest kernel.

   https://lore.kernel.org/netdev/[email protected]/

However, Mat reported at that time that there were at least two variants, the latter being the same as mine.

   https://lore.kernel.org/netdev/[email protected]/
   https://lore.kernel.org/netdev/[email protected]/

This week I started looking into this issue and finally figured out why we could not catch all cases with a single repro.

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

   # sysctl -w net.ipv4.tcp_syn_retries=1
   net.ipv4.tcp_syn_retries = 1
   # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                    ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

   [ 1096.845905] ------------[ cut here ]------------
   [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket. So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

   if (!inet_csk(sk)->icsk_bind_hash)
   	inet_bind_hash(sk, tb, tb2, port);
   WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
   WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

Here I think the s2 case also _should_ trigger WARN_ON(). The issue seems to be fixed, but it's just because we forgot to _fix_ the source address in error paths after inet6?_hash_connect() in tcp_v[46]_connect(). (Of course, DCCP as well).

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

     inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

Thanks for the repro script and the detailed analysis on the issue. iiuc, this is limited to the sk that was bind() to ADDR_ANY:port-1234 (or !SOCK_BINDADDR_LOCK). Does it make sense to avoid adding the sk with ADDR_ANY:port-1234 to bhash2 at all? From inet_use_bhash2_on_bind(), it does not seem ADDR_ANY will use bhash2. or I have missed some cases?

#4 lore raw
Quoted 163 lines · Martin KaFai Lau
Quoted 161 lines · Kuniyuki Iwashima

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

   WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
   ...
   inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
   inet_listen (net/ipv4/af_inet.c:228)
   __sys_listen (net/socket.c:1810)
   __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
   do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
   entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

For the very first implementation of bhash2, there was a similar report hitting the same WARN_ON(). The fix was to update the bhash2 bucket when the kernel changes sk->sk_rcv_saddr from INADDR_ANY. Then, syzbot has a repro, so we can indeed confirm that it no longer triggers the warning on the latest kernel.

   https://lore.kernel.org/netdev/[email protected]/

However, Mat reported at that time that there were at least two variants, the latter being the same as mine.

   https://lore.kernel.org/netdev/[email protected]/
   https://lore.kernel.org/netdev/[email protected]/

This week I started looking into this issue and finally figured out why we could not catch all cases with a single repro.

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

   # sysctl -w net.ipv4.tcp_syn_retries=1
   net.ipv4.tcp_syn_retries = 1
   # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                    ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

   [ 1096.845905] ------------[ cut here ]------------
   [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket. So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

   if (!inet_csk(sk)->icsk_bind_hash)
      inet_bind_hash(sk, tb, tb2, port);
   WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
   WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

Here I think the s2 case also _should_ trigger WARN_ON(). The issue seems to be fixed, but it's just because we forgot to _fix_ the source address in error paths after inet6?_hash_connect() in tcp_v[46]_connect(). (Of course, DCCP as well).

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

     inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

Quoted 8 lines
Quoted 1 line

What do you think is the best way to fix the issue?

Thanks for the repro script and the detailed analysis on the issue. iiuc, this is limited to the sk that was bind() to ADDR_ANY:port-1234 (or !SOCK_BINDADDR_LOCK). Does it make sense to avoid adding the sk with ADDR_ANY:port-1234 to bhash2 at all? From inet_use_bhash2_on_bind(), it does not seem ADDR_ANY will use bhash2. or I have missed some cases?

Thanks for your analysis and your repro instructions, Kuniyuki. And apologies for being absent from bhash2 these past ~6 weeks, I've been on PTO traveling and should have communicated that better. Thanks Martin for taking care of bhash2 during this time.

For avoiding adding sockets with ADDR_ANY to the bhash2 hashtable, I think the issue is that other sockets need to detect whether there's a bind conflict against an ADDR_ANY socket on that port, so if ADDR_ANY is not hashed to bhash2, then on binds, we would have to iterate through the regular bhash table to check against ADDR_ANY, where the bhash table could be very long if there are many sockets bound to that port.

I intend to look deeply into this issue this weekend when I fly back home and have my dev environment accessible. Kuniyuki, you mentioned that the next LTS will likely be v6.1 which has bhash2 - do you know what the deadline for this is for getting the fix in by?

#5 lore raw
From:   Joanne Koong <[email protected]>
Date:   Wed, 2 Nov 2022 00:57:00 +0000
Quoted 164 lines · Martin KaFai Lau
Quoted 163 lines · Kuniyuki Iwashima
Quoted 161 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

   WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
   ...
   inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
   inet_listen (net/ipv4/af_inet.c:228)
   __sys_listen (net/socket.c:1810)
   __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
   do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
   entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

For the very first implementation of bhash2, there was a similar report hitting the same WARN_ON(). The fix was to update the bhash2 bucket when the kernel changes sk->sk_rcv_saddr from INADDR_ANY. Then, syzbot has a repro, so we can indeed confirm that it no longer triggers the warning on the latest kernel.

   https://lore.kernel.org/netdev/[email protected]/

However, Mat reported at that time that there were at least two variants, the latter being the same as mine.

   https://lore.kernel.org/netdev/[email protected]/
   https://lore.kernel.org/netdev/[email protected]/

This week I started looking into this issue and finally figured out why we could not catch all cases with a single repro.

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

   # sysctl -w net.ipv4.tcp_syn_retries=1
   net.ipv4.tcp_syn_retries = 1
   # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                    > > >  ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

   [ 1096.845905] ------------[ cut here ]------------
   [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket. So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

   if (!inet_csk(sk)->icsk_bind_hash)
      inet_bind_hash(sk, tb, tb2, port);
   WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
   WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

Here I think the s2 case also _should_ trigger WARN_ON(). The issue seems to be fixed, but it's just because we forgot to _fix_ the source address in error paths after inet6?_hash_connect() in tcp_v[46]_connect(). (Of course, DCCP as well).

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

     inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

     File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

Quoted 5 lines
Quoted 5 lines
Quoted 1 line

What do you think is the best way to fix the issue?

Thanks for the repro script and the detailed analysis on the issue. iiuc, this is limited to the sk that was bind() to ADDR_ANY:port-1234 (or !SOCK_BINDADDR_LOCK).

Yes, it's limited to !SOCK_BINDADDR_LOCK && SOCK_BINDPORT_LOCK, when we don't reset snum but saddr. We seem to need like ->put_addr() which never fails even in the failure path.

Quoted 9 lines
Quoted 4 lines

Does it make sense to avoid adding the sk with
ADDR_ANY:port-1234 to bhash2 at all? From inet_use_bhash2_on_bind(), it does not seem ADDR_ANY will use bhash2. or I have missed some cases?

Thanks for your analysis and your repro instructions, Kuniyuki. And apologies for being absent from bhash2 these past ~6 weeks, I've been on PTO traveling and should have communicated that better. Thanks Martin for taking care of bhash2 during this time.

No worries :)

Quoted 7 lines

For avoiding adding sockets with ADDR_ANY to the bhash2 hashtable, I think the issue is that other sockets need to detect whether there's a bind conflict against an ADDR_ANY socket on that port, so if ADDR_ANY is not hashed to bhash2, then on binds, we would have to iterate through the regular bhash table to check against ADDR_ANY, where the bhash table could be very long if there are many sockets bound to that port.

Right, inet_bhash2_addr_any_conflict() will not work then and it means we cannot enjoy the very merit of bhash2.

Quoted 4 lines

I intend to look deeply into this issue this weekend when I fly back home and have my dev environment accessible. Kuniyuki, you mentioned that the next LTS will likely be v6.1 which has bhash2 - do you know what the deadline for this is for getting the fix in by?

The bhash2 is already in the Linus' tree, so it will be released next month until someone sends a revert, and we can send a fix anytime.

I think there is no deadline, but it should be better to send a fix in this cycle. My concern is that if the fix is big, it could be queued for v6.2 and a bug could remain in LTS.

If bhash2 has the inconsistent state, we can bind a socket to a would-conflict-address.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

from socket import *

s1 = socket()
s1.bind(('0', 1000))
s1.connect(('localhost', 10))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s1

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 1000)>

---
0.0.0.0:1000 -> slot A -> tb A (should have 0.0.0.0:1000)

10.0.2.15:1000 -> slot B -> tb B

                                `-> 0.0.0.0:1000

---

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

s2 = socket()
s2.bind(('10.0.2.16', 1000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.16', 1000)>

---
0.0.0.0:1000 -> slot A -> tb A (no conflict)
10.0.2.16:1000 -> slot C -> tb C (no conflict)
---

I don't know the real use cases for bind(INADDR_ANY, port) + connect(), but I think this is not acceptable for LTS?

#6 lore raw
Quoted 10 lines · Kuniyuki Iwashima
Quoted 7 lines

For avoiding adding sockets with ADDR_ANY to the bhash2 hashtable, I think the issue is that other sockets need to detect whether there's a bind conflict against an ADDR_ANY socket on that port, so if ADDR_ANY is not hashed to bhash2, then on binds, we would have to iterate through the regular bhash table to check against ADDR_ANY, where the bhash table could be very long if there are many sockets bound to that port.

Right, inet_bhash2_addr_any_conflict() will not work then and it means we cannot enjoy the very merit of bhash2.

One thought is to have the ADDR_ANY sk linked at either end of the bhash (head or tail) so that it can get to the ADDRY_ANY sk faster in bhash when checking conflict. However, only the 'struct hlist_node owner' alone may not be good enough. Just an idea.

#7 lore raw
In reply to Kuniyuki Iwashima
Quoted 15 lines · Kuniyuki Iwashima

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

[...]

Quoted 5 lines

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

Since the s2 address mismatch case is addressed by your patch
https://lore.kernel.org/netdev/[email protected]/,
I will focus my comments here on the s3 case.

Quoted 36 lines
  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

I'm on the head of net-next/master (commit bf46390f39c686d62afeae9845860e63886d63b) and trying to repro this locally, but the warning isn't showing up for me after following the steps above. Not sure why.

Quoted 3 lines

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket.

To summarize, the path you are talking about is tcp_v4_connect() in kernel/linux/net/ipv4/tcp_ipv4.c where the sk originally has saddr INADDR_ANY, the sk gets assigned a new address, that new address gets updated in the bhash2 table, and then when inet_hash_connect() is called, it fails which brings us to the "goto failure". In the failure goto, we call "tcp_set_state(sk, TCP_CLOSE)" but in the case where "SOCK_BINDPORT_LOCK" is held, "inet_put_port(sk)" is *not* called, which means the sk will still be in the bhash2 table with the new address.

Quoted 8 lines

So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
        inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

[...]

Quoted 4 lines

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

This is the case where in __inet_stream_connect(), the call to "sk->sk_prot->connect()" succeeds but then the connection is closed by RST/timeout/ICMP error, so then the "goto sock_error" is triggered, correct?

Quoted 57 lines

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

To summarize your analysis, there is an issue right now where for sockets that are binded on address INADDR_ANY, we need to handle the error case where if a connection fails and SOCK_BINDPORT_LOCK is held, the new address it was assigned needs to be taken out of bhash2 and the original address (INADDR_ANY) needs to be re-added to bhash2. There are two of these error cases we need to handle, as you mentioned above - 1) in dccp/tcp_v4_connect() where the connect call fails and 2) in __inet_stream_connect() where the connect call succeeds but the connection is closed by a RST/timeout/ICMP error.

I think the simplest solution is to modify inet_bhash2_update_saddr() so that we don't free the inet_bind2_bucket() for INADDR_ANY/port (if it is empty after we update the saddr to the new addr) *until* the connect succeeds. When the connect succeeds, then we can check whether the inet_bind2_bucket for INADDR_ANY is empty, and if it is, then do the freeing for it.

What are your thoughts on this?

Thank you.

Quoted 58 lines

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * This unhashes the socket and releases the local port, if necessary.
         */
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        __sk_dst_reset(sk);
 failure:
        inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * if necessary.
         */
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
 failure:
        inet->inet_dport = 0;
        sk->sk_route_caps = 0;
---8<---
#8 lore raw
From:   Joanne Koong <[email protected]>
Date:   Mon, 7 Nov 2022 14:20:46 -0800
Quoted 68 lines · Kuniyuki Iwashima
Quoted 15 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

[...]

Quoted 5 lines

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

Since the s2 address mismatch case is addressed by your patch
https://lore.kernel.org/netdev/[email protected]/,
I will focus my comments here on the s3 case.

Quoted 41 lines
  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

I'm on the head of net-next/master (commit bf46390f39c686d62afeae9845860e63886d63b) and trying to repro this locally, but the warning isn't showing up for me after following the steps above. Not sure why.

Hmm... it reproduced on top of the commit. I'm testing on QEMU and login to serial console which outputs syslog in the same stream, so you may want to check /var/log/messages or something.

---8<---

# sysctl -w net.ipv4.tcp_syn_retries=1
# python3
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

from socket import *

s = socket()
s.bind(('0', 10000))
s.connect(('0.0.0.1', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TimeoutError: [Errno 110] Connection timed out

Quoted 1 line
Quoted 1 line
Quoted 1 line

s

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[   96.598308] ------------[ cut here ]------------
[   96.598598] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

s = socket()
s.bind(('0', 10001))
s.connect(('localhost', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[  139.157193] ------------[ cut here ]------------
[  139.157528] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

---8<---

Quoted 5 lines
Quoted 4 lines

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket.

Quoted 9 lines

To summarize, the path you are talking about is tcp_v4_connect() in kernel/linux/net/ipv4/tcp_ipv4.c where the sk originally has saddr INADDR_ANY, the sk gets assigned a new address, that new address gets updated in the bhash2 table, and then when inet_hash_connect() is called, it fails which brings us to the "goto failure". In the failure goto, we call "tcp_set_state(sk, TCP_CLOSE)" but in the case where "SOCK_BINDPORT_LOCK" is held, "inet_put_port(sk)" is *not* called, which means the sk will still be in the bhash2 table with the new address.

Correct.

More precisely, 3 functions after inet_hash_connect() can cause the same issue.

- ip_route_newports
- tcp_fastopen_defer_connect
- tcp_connect

Quoted 11 lines
Quoted 9 lines

So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
        inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

[...]

Quoted 1 line
Quoted 8 lines
Quoted 3 lines

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

This is the case where in __inet_stream_connect(), the call to "sk->sk_prot->connect()" succeeds but then the connection is closed by RST/timeout/ICMP error, so then the "goto sock_error" is triggered, correct?

Yes.

Quoted 76 lines
Quoted 58 lines

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

To summarize your analysis, there is an issue right now where for sockets that are binded on address INADDR_ANY, we need to handle the error case where if a connection fails and SOCK_BINDPORT_LOCK is held, the new address it was assigned needs to be taken out of bhash2 and the original address (INADDR_ANY) needs to be re-added to bhash2. There are two of these error cases we need to handle, as you mentioned above - 1) in dccp/tcp_v4_connect() where the connect call fails and 2) in __inet_stream_connect() where the connect call succeeds but the connection is closed by a RST/timeout/ICMP error.

Quoted 8 lines

I think the simplest solution is to modify inet_bhash2_update_saddr() so that we don't free the inet_bind2_bucket() for INADDR_ANY/port (if it is empty after we update the saddr to the new addr) *until* the connect succeeds. When the connect succeeds, then we can check whether the inet_bind2_bucket for INADDR_ANY is empty, and if it is, then do the freeing for it.

What are your thoughts on this?

I was thinking the same, but this scenario will break it ?

connect() <-- unblocking socket
return -EINPROGRESS

receive SYN+ACK, send back ACK, and set state to TCP_ESTABLISEHD

free the old INADDR_ANY bucket

get RST and set state to TCP_CLOSE

connect()

    goto sock_error and ->disconect() fail to restore the bucket
Quoted 61 lines

Thank you.

Quoted 58 lines

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * This unhashes the socket and releases the local port, if necessary.
         */
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        __sk_dst_reset(sk);
 failure:
        inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * if necessary.
         */
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
 failure:
        inet->inet_dport = 0;
        sk->sk_route_caps = 0;
---8<---
#9 lore raw
Quoted 252 lines · Kuniyuki Iwashima
From:   Joanne Koong <[email protected]>
Date:   Mon, 7 Nov 2022 14:20:46 -0800
Quoted 68 lines · Kuniyuki Iwashima
Quoted 15 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

[...]

Quoted 5 lines

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

Since the s2 address mismatch case is addressed by your patch
https://lore.kernel.org/netdev/[email protected]/,
I will focus my comments here on the s3 case.

Quoted 41 lines
  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

I'm on the head of net-next/master (commit bf46390f39c686d62afeae9845860e63886d63b) and trying to repro this locally, but the warning isn't showing up for me after following the steps above. Not sure why.

Hmm... it reproduced on top of the commit. I'm testing on QEMU and login to serial console which outputs syslog in the same stream, so you may want to check /var/log/messages or something.

---8<---

# sysctl -w net.ipv4.tcp_syn_retries=1
# python3
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

from socket import *

s = socket()
s.bind(('0', 10000))
s.connect(('0.0.0.1', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TimeoutError: [Errno 110] Connection timed out

Quoted 1 line
Quoted 1 line
Quoted 1 line

s

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[   96.598308] ------------[ cut here ]------------
[   96.598598] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

s = socket()
s.bind(('0', 10001))
s.connect(('localhost', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[  139.157193] ------------[ cut here ]------------
[  139.157528] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

---8<---

Quoted 5 lines
Quoted 4 lines

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket.

Quoted 9 lines

To summarize, the path you are talking about is tcp_v4_connect() in kernel/linux/net/ipv4/tcp_ipv4.c where the sk originally has saddr INADDR_ANY, the sk gets assigned a new address, that new address gets updated in the bhash2 table, and then when inet_hash_connect() is called, it fails which brings us to the "goto failure". In the failure goto, we call "tcp_set_state(sk, TCP_CLOSE)" but in the case where "SOCK_BINDPORT_LOCK" is held, "inet_put_port(sk)" is *not* called, which means the sk will still be in the bhash2 table with the new address.

Correct.

More precisely, 3 functions after inet_hash_connect() can cause the same issue.

- ip_route_newports
- tcp_fastopen_defer_connect
- tcp_connect

Quoted 11 lines
Quoted 9 lines

So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
        inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

[...]

Quoted 1 line
Quoted 8 lines
Quoted 3 lines

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

This is the case where in __inet_stream_connect(), the call to "sk->sk_prot->connect()" succeeds but then the connection is closed by RST/timeout/ICMP error, so then the "goto sock_error" is triggered, correct?

Yes.

Quoted 76 lines
Quoted 58 lines

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

To summarize your analysis, there is an issue right now where for sockets that are binded on address INADDR_ANY, we need to handle the error case where if a connection fails and SOCK_BINDPORT_LOCK is held, the new address it was assigned needs to be taken out of bhash2 and the original address (INADDR_ANY) needs to be re-added to bhash2. There are two of these error cases we need to handle, as you mentioned above - 1) in dccp/tcp_v4_connect() where the connect call fails and 2) in __inet_stream_connect() where the connect call succeeds but the connection is closed by a RST/timeout/ICMP error.

Quoted 8 lines

I think the simplest solution is to modify inet_bhash2_update_saddr() so that we don't free the inet_bind2_bucket() for INADDR_ANY/port (if it is empty after we update the saddr to the new addr) *until* the connect succeeds. When the connect succeeds, then we can check whether the inet_bind2_bucket for INADDR_ANY is empty, and if it is, then do the freeing for it.

What are your thoughts on this?

I was thinking the same, but this scenario will break it ?

connect() <-- unblocking socket
return -EINPROGRESS

Quoted 9 lines

receive SYN+ACK, send back ACK, and set state to TCP_ESTABLISEHD

free the old INADDR_ANY bucket

get RST and set state to TCP_CLOSE

connect()

    goto sock_error and ->disconect() fail to restore the bucket

I don't think this scenario will break it because after a successful connect call, a disconnect will remove the socket from the bhash (and bhash2) table altogether since the socket will *not* have SOCK_BINDPORT_LOCK set.

This is the code path I'm looking at:

connect call is successful and the three way handshake has completed -> tcp_v4_syn_recv_sock() is called where we create the new socket tcp_v4_syn_recv_sock() -> tcp_create_openreq_child() -> inet_csk_clone_lock() -> sk_clone_lock()

In sk_clone_lock(), we create a newsk where newsk->sk_userlocks will *not* have the SOCK_BINDPORT_LOCK flag set (newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;).

If the connection gets closed by a RST/timeout/ICMP error, then this socket will go through disconnect() where in the disconnect call, it will remove the socket from the bhash and bhash2 tables. For example, for tcp, this code path is tcp_disconnect() -> tcp_set_state(sk, TCP_CLOSE) -> inet_put_port() - this is also verified by the "WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);" at the end of tcp_disconnect().

What are your thoughts?

Quoted 62 lines
Quoted 62 lines

Thank you.

Quoted 58 lines

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * This unhashes the socket and releases the local port, if necessary.
         */
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        __sk_dst_reset(sk);
 failure:
        inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * if necessary.
         */
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
 failure:
        inet->inet_dport = 0;
        sk->sk_route_caps = 0;
---8<---
#10 lore raw
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 09:40:22 -0800
Quoted 255 lines · Kuniyuki Iwashima
Quoted 254 lines
From:   Joanne Koong <[email protected]>
Date:   Mon, 7 Nov 2022 14:20:46 -0800
Quoted 68 lines · Kuniyuki Iwashima
Quoted 15 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

[...]

Quoted 5 lines

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

Since the s2 address mismatch case is addressed by your patch
https://lore.kernel.org/netdev/[email protected]/,
I will focus my comments here on the s3 case.

Quoted 41 lines
  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

I'm on the head of net-next/master (commit bf46390f39c686d62afeae9845860e63886d63b) and trying to repro this locally, but the warning isn't showing up for me after following the steps above. Not sure why.

Hmm... it reproduced on top of the commit. I'm testing on QEMU and login to serial console which outputs syslog in the same stream, so you may want to check /var/log/messages or something.

---8<---

# sysctl -w net.ipv4.tcp_syn_retries=1
# python3
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

from socket import *

s = socket()
s.bind(('0', 10000))
s.connect(('0.0.0.1', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TimeoutError: [Errno 110] Connection timed out

Quoted 1 line
Quoted 1 line
Quoted 1 line

s

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[   96.598308] ------------[ cut here ]------------
[   96.598598] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

s = socket()
s.bind(('0', 10001))
s.connect(('localhost', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[  139.157193] ------------[ cut here ]------------
[  139.157528] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

---8<---

Quoted 5 lines
Quoted 4 lines

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket.

Quoted 9 lines

To summarize, the path you are talking about is tcp_v4_connect() in kernel/linux/net/ipv4/tcp_ipv4.c where the sk originally has saddr INADDR_ANY, the sk gets assigned a new address, that new address gets updated in the bhash2 table, and then when inet_hash_connect() is called, it fails which brings us to the "goto failure". In the failure goto, we call "tcp_set_state(sk, TCP_CLOSE)" but in the case where "SOCK_BINDPORT_LOCK" is held, "inet_put_port(sk)" is *not* called, which means the sk will still be in the bhash2 table with the new address.

Correct.

More precisely, 3 functions after inet_hash_connect() can cause the same issue.

- ip_route_newports
- tcp_fastopen_defer_connect
- tcp_connect

Quoted 11 lines
Quoted 9 lines

So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
        inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

[...]

Quoted 1 line
Quoted 8 lines
Quoted 3 lines

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

This is the case where in __inet_stream_connect(), the call to "sk->sk_prot->connect()" succeeds but then the connection is closed by RST/timeout/ICMP error, so then the "goto sock_error" is triggered, correct?

Yes.

Quoted 76 lines
Quoted 58 lines

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

To summarize your analysis, there is an issue right now where for sockets that are binded on address INADDR_ANY, we need to handle the error case where if a connection fails and SOCK_BINDPORT_LOCK is held, the new address it was assigned needs to be taken out of bhash2 and the original address (INADDR_ANY) needs to be re-added to bhash2. There are two of these error cases we need to handle, as you mentioned above - 1) in dccp/tcp_v4_connect() where the connect call fails and 2) in __inet_stream_connect() where the connect call succeeds but the connection is closed by a RST/timeout/ICMP error.

Quoted 8 lines

I think the simplest solution is to modify inet_bhash2_update_saddr() so that we don't free the inet_bind2_bucket() for INADDR_ANY/port (if it is empty after we update the saddr to the new addr) *until* the connect succeeds. When the connect succeeds, then we can check whether the inet_bind2_bucket for INADDR_ANY is empty, and if it is, then do the freeing for it.

What are your thoughts on this?

I was thinking the same, but this scenario will break it ?

connect() <-- unblocking socket
return -EINPROGRESS

receive SYN+ACK, send back ACK, and set state to TCP_ESTABLISEHD

Quoted 7 lines
Quoted 7 lines

free the old INADDR_ANY bucket

get RST and set state to TCP_CLOSE

connect()

    goto sock_error and ->disconect() fail to restore the bucket

I meant all of the above happen at the client side, so the server-side code is orthogonal.

But I got your point. I've misunderstood "When the connect succeeds". I was thinking that we'll free bhash2 bucket in the fast path where we send back SYN+ACK to the server. OTOH, you meant it to be when the connect() syscall returns 0, right ?

Then, I think it's doable, like

1) inet_bhash2_update_saddr() increments addrany_tb2->connecting

2) when connect() succeeds,

     if (dec_and_test(&addrany_tb2->connecting) && is_empty(addrany_tb2->owners))
         free(addrany_tb2);

3) while close()ing,

     tb2 = ...
     addrany_tb2 = ...

     if ((sk->sk_user_locks & SOCK_BINDPORT_LOCK) &&
         !(sk->sk_user_locks & SOCK_BINDADDR_LOCK) &&
         tb2 != addrany_tb2 &&
         sock->state != SS_CONNECTED &&
         dec_and_test(&addrany_tb2->connecting) &&
         is_empty(addrany_tb2->owners))
         free(addrany_tb2)

More consideration would be needed for inet_shutdown() etc though.

Quoted 88 lines

I don't think this scenario will break it because after a successful connect call, a disconnect will remove the socket from the bhash (and bhash2) table altogether since the socket will *not* have SOCK_BINDPORT_LOCK set.

This is the code path I'm looking at:

connect call is successful and the three way handshake has completed -> tcp_v4_syn_recv_sock() is called where we create the new socket tcp_v4_syn_recv_sock() -> tcp_create_openreq_child() -> inet_csk_clone_lock() -> sk_clone_lock()

In sk_clone_lock(), we create a newsk where newsk->sk_userlocks will *not* have the SOCK_BINDPORT_LOCK flag set (newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;).

If the connection gets closed by a RST/timeout/ICMP error, then this socket will go through disconnect() where in the disconnect call, it will remove the socket from the bhash and bhash2 tables. For example, for tcp, this code path is tcp_disconnect() -> tcp_set_state(sk, TCP_CLOSE) -> inet_put_port() - this is also verified by the "WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);" at the end of tcp_disconnect().

What are your thoughts?

Quoted 62 lines
Quoted 62 lines

Thank you.

Quoted 58 lines

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * This unhashes the socket and releases the local port, if necessary.
         */
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        __sk_dst_reset(sk);
 failure:
        inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * if necessary.
         */
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
 failure:
        inet->inet_dport = 0;
        sk->sk_route_caps = 0;
---8<---
#11 lore raw
Quoted 296 lines · Kuniyuki Iwashima
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 09:40:22 -0800
Quoted 255 lines · Kuniyuki Iwashima
Quoted 254 lines
From:   Joanne Koong <[email protected]>
Date:   Mon, 7 Nov 2022 14:20:46 -0800
Quoted 68 lines · Kuniyuki Iwashima
Quoted 15 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

[...]

Quoted 5 lines

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

Since the s2 address mismatch case is addressed by your patch
https://lore.kernel.org/netdev/[email protected]/,
I will focus my comments here on the s3 case.

Quoted 41 lines
  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

I'm on the head of net-next/master (commit bf46390f39c686d62afeae9845860e63886d63b) and trying to repro this locally, but the warning isn't showing up for me after following the steps above. Not sure why.

Hmm... it reproduced on top of the commit. I'm testing on QEMU and login to serial console which outputs syslog in the same stream, so you may want to check /var/log/messages or something.

---8<---

# sysctl -w net.ipv4.tcp_syn_retries=1
# python3
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

from socket import *

s = socket()
s.bind(('0', 10000))
s.connect(('0.0.0.1', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TimeoutError: [Errno 110] Connection timed out

Quoted 1 line
Quoted 1 line
Quoted 1 line

s

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[   96.598308] ------------[ cut here ]------------
[   96.598598] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

s = socket()
s.bind(('0', 10001))
s.connect(('localhost', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[  139.157193] ------------[ cut here ]------------
[  139.157528] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

---8<---

Quoted 5 lines
Quoted 4 lines

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket.

Quoted 9 lines

To summarize, the path you are talking about is tcp_v4_connect() in kernel/linux/net/ipv4/tcp_ipv4.c where the sk originally has saddr INADDR_ANY, the sk gets assigned a new address, that new address gets updated in the bhash2 table, and then when inet_hash_connect() is called, it fails which brings us to the "goto failure". In the failure goto, we call "tcp_set_state(sk, TCP_CLOSE)" but in the case where "SOCK_BINDPORT_LOCK" is held, "inet_put_port(sk)" is *not* called, which means the sk will still be in the bhash2 table with the new address.

Correct.

More precisely, 3 functions after inet_hash_connect() can cause the same issue.

- ip_route_newports
- tcp_fastopen_defer_connect
- tcp_connect

Quoted 11 lines
Quoted 9 lines

So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
        inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

[...]

Quoted 1 line
Quoted 8 lines
Quoted 3 lines

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

This is the case where in __inet_stream_connect(), the call to "sk->sk_prot->connect()" succeeds but then the connection is closed by RST/timeout/ICMP error, so then the "goto sock_error" is triggered, correct?

Yes.

Quoted 76 lines
Quoted 58 lines

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

To summarize your analysis, there is an issue right now where for sockets that are binded on address INADDR_ANY, we need to handle the error case where if a connection fails and SOCK_BINDPORT_LOCK is held, the new address it was assigned needs to be taken out of bhash2 and the original address (INADDR_ANY) needs to be re-added to bhash2. There are two of these error cases we need to handle, as you mentioned above - 1) in dccp/tcp_v4_connect() where the connect call fails and 2) in __inet_stream_connect() where the connect call succeeds but the connection is closed by a RST/timeout/ICMP error.

Quoted 8 lines

I think the simplest solution is to modify inet_bhash2_update_saddr() so that we don't free the inet_bind2_bucket() for INADDR_ANY/port (if it is empty after we update the saddr to the new addr) *until* the connect succeeds. When the connect succeeds, then we can check whether the inet_bind2_bucket for INADDR_ANY is empty, and if it is, then do the freeing for it.

What are your thoughts on this?

I was thinking the same, but this scenario will break it ?

connect() <-- unblocking socket
return -EINPROGRESS

receive SYN+ACK, send back ACK, and set state to TCP_ESTABLISEHD

Quoted 7 lines
Quoted 7 lines

free the old INADDR_ANY bucket

get RST and set state to TCP_CLOSE

connect()

    goto sock_error and ->disconect() fail to restore the bucket

I meant all of the above happen at the client side, so the server-side code is orthogonal.

But I got your point. I've misunderstood "When the connect succeeds". I was thinking that we'll free bhash2 bucket in the fast path where we send back SYN+ACK to the server. OTOH, you meant it to be when the connect() syscall returns 0, right ?

Then, I think it's doable, like

1) inet_bhash2_update_saddr() increments addrany_tb2->connecting

2) when connect() succeeds,

     if (dec_and_test(&addrany_tb2->connecting) && is_empty(addrany_tb2->owners))
         free(addrany_tb2);

3) while close()ing,

     tb2 = ...
     addrany_tb2 = ...

     if ((sk->sk_user_locks & SOCK_BINDPORT_LOCK) &&
         !(sk->sk_user_locks & SOCK_BINDADDR_LOCK) &&
         tb2 != addrany_tb2 &&
         sock->state != SS_CONNECTED &&
         dec_and_test(&addrany_tb2->connecting) &&
         is_empty(addrany_tb2->owners))
         free(addrany_tb2)
Quoted 1 line

More consideration would be needed for inet_shutdown() etc though.

Sorry, my terminology was ambiguous :) By "when the connect succeeds", I meant when the call to tcp_connect() succeeds at the end of tcp_v4_connect().

There are 2 cases you brought up regarding connection failures on binded INADDR_ANY sockets that might lead to a bhash2 mismatch:

#1) in dccp/tcp_v4_connect() where an error leads to the "goto failure" #2) in __inet_stream_connect() where an error leads to the "goto sock_error"

Regarding #2, I don't think this leads to a bhash2 mismatch. The "goto sock_error" triggers disconnect() and the disconnect() call will always remove the socket from the bhash and bhash2 tables since the socket when it was cloned had the SOCK_BINDPORT_LOCK flag unset.

For #1, I think the simplest way to handle this is to not free the inet_bind2_bucket for INADDR_ANY/port when we call inet_bhash2_update_saddr(). Then, at the end of the dccp/tcp_v4/6_connect() call, if it is successful, we free the INADDR_ANY/port inet_bind2_bucket if it's empty. And if that is not successful and there was a failure somewhere, then in the "goto failure", we remove the socket from the new_addr bind bucket / clean up the new_addr bind bucket and re-add the socket back to the INADDR_ANY bind bucket.

What are your thoughts?

Quoted 90 lines
Quoted 90 lines

I don't think this scenario will break it because after a successful connect call, a disconnect will remove the socket from the bhash (and bhash2) table altogether since the socket will *not* have SOCK_BINDPORT_LOCK set.

This is the code path I'm looking at:

connect call is successful and the three way handshake has completed -> tcp_v4_syn_recv_sock() is called where we create the new socket tcp_v4_syn_recv_sock() -> tcp_create_openreq_child() -> inet_csk_clone_lock() -> sk_clone_lock()

In sk_clone_lock(), we create a newsk where newsk->sk_userlocks will *not* have the SOCK_BINDPORT_LOCK flag set (newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;).

If the connection gets closed by a RST/timeout/ICMP error, then this socket will go through disconnect() where in the disconnect call, it will remove the socket from the bhash and bhash2 tables. For example, for tcp, this code path is tcp_disconnect() -> tcp_set_state(sk, TCP_CLOSE) -> inet_put_port() - this is also verified by the "WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);" at the end of tcp_disconnect().

What are your thoughts?

Quoted 62 lines
Quoted 62 lines

Thank you.

Quoted 58 lines

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * This unhashes the socket and releases the local port, if necessary.
         */
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        __sk_dst_reset(sk);
 failure:
        inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * if necessary.
         */
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
 failure:
        inet->inet_dport = 0;
        sk->sk_route_caps = 0;
---8<---
#12 lore raw
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 14:02:11 -0800
Quoted 306 lines · Kuniyuki Iwashima
Quoted 296 lines
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 09:40:22 -0800
Quoted 255 lines · Kuniyuki Iwashima
Quoted 254 lines
From:   Joanne Koong <[email protected]>
Date:   Mon, 7 Nov 2022 14:20:46 -0800
Quoted 68 lines · Kuniyuki Iwashima
Quoted 15 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

[...]

Quoted 5 lines

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

Since the s2 address mismatch case is addressed by your patch
https://lore.kernel.org/netdev/[email protected]/,
I will focus my comments here on the s3 case.

Quoted 41 lines
  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

I'm on the head of net-next/master (commit bf46390f39c686d62afeae9845860e63886d63b) and trying to repro this locally, but the warning isn't showing up for me after following the steps above. Not sure why.

Hmm... it reproduced on top of the commit. I'm testing on QEMU and login to serial console which outputs syslog in the same stream, so you may want to check /var/log/messages or something.

---8<---

# sysctl -w net.ipv4.tcp_syn_retries=1
# python3
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

from socket import *

s = socket()
s.bind(('0', 10000))
s.connect(('0.0.0.1', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TimeoutError: [Errno 110] Connection timed out

Quoted 1 line
Quoted 1 line
Quoted 1 line

s

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[   96.598308] ------------[ cut here ]------------
[   96.598598] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

s = socket()
s.bind(('0', 10001))
s.connect(('localhost', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[  139.157193] ------------[ cut here ]------------
[  139.157528] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

---8<---

Quoted 5 lines
Quoted 4 lines

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket.

Quoted 9 lines

To summarize, the path you are talking about is tcp_v4_connect() in kernel/linux/net/ipv4/tcp_ipv4.c where the sk originally has saddr INADDR_ANY, the sk gets assigned a new address, that new address gets updated in the bhash2 table, and then when inet_hash_connect() is called, it fails which brings us to the "goto failure". In the failure goto, we call "tcp_set_state(sk, TCP_CLOSE)" but in the case where "SOCK_BINDPORT_LOCK" is held, "inet_put_port(sk)" is *not* called, which means the sk will still be in the bhash2 table with the new address.

Correct.

More precisely, 3 functions after inet_hash_connect() can cause the same issue.

- ip_route_newports
- tcp_fastopen_defer_connect
- tcp_connect

Quoted 11 lines
Quoted 9 lines

So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
        inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

[...]

Quoted 1 line
Quoted 8 lines
Quoted 3 lines

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

This is the case where in __inet_stream_connect(), the call to "sk->sk_prot->connect()" succeeds but then the connection is closed by RST/timeout/ICMP error, so then the "goto sock_error" is triggered, correct?

Yes.

Quoted 76 lines
Quoted 58 lines

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

To summarize your analysis, there is an issue right now where for sockets that are binded on address INADDR_ANY, we need to handle the error case where if a connection fails and SOCK_BINDPORT_LOCK is held, the new address it was assigned needs to be taken out of bhash2 and the original address (INADDR_ANY) needs to be re-added to bhash2. There are two of these error cases we need to handle, as you mentioned above - 1) in dccp/tcp_v4_connect() where the connect call fails and 2) in __inet_stream_connect() where the connect call succeeds but the connection is closed by a RST/timeout/ICMP error.

Quoted 8 lines

I think the simplest solution is to modify inet_bhash2_update_saddr() so that we don't free the inet_bind2_bucket() for INADDR_ANY/port (if it is empty after we update the saddr to the new addr) *until* the connect succeeds. When the connect succeeds, then we can check whether the inet_bind2_bucket for INADDR_ANY is empty, and if it is, then do the freeing for it.

What are your thoughts on this?

I was thinking the same, but this scenario will break it ?

connect() <-- unblocking socket
return -EINPROGRESS

receive SYN+ACK, send back ACK, and set state to TCP_ESTABLISEHD

Quoted 7 lines
Quoted 7 lines

free the old INADDR_ANY bucket

get RST and set state to TCP_CLOSE

connect()

    goto sock_error and ->disconect() fail to restore the bucket

I meant all of the above happen at the client side, so the server-side code is orthogonal.

But I got your point. I've misunderstood "When the connect succeeds". I was thinking that we'll free bhash2 bucket in the fast path where we send back SYN+ACK to the server. OTOH, you meant it to be when the connect() syscall returns 0, right ?

Then, I think it's doable, like

1) inet_bhash2_update_saddr() increments addrany_tb2->connecting

2) when connect() succeeds,

     if (dec_and_test(&addrany_tb2->connecting) && is_empty(addrany_tb2->owners))
         free(addrany_tb2);

3) while close()ing,

     tb2 = ...
     addrany_tb2 = ...

     if ((sk->sk_user_locks & SOCK_BINDPORT_LOCK) &&
         !(sk->sk_user_locks & SOCK_BINDADDR_LOCK) &&
         tb2 != addrany_tb2 &&
         sock->state != SS_CONNECTED &&
         dec_and_test(&addrany_tb2->connecting) &&
         is_empty(addrany_tb2->owners))
         free(addrany_tb2)
Quoted 1 line

More consideration would be needed for inet_shutdown() etc though.

Sorry, my terminology was ambiguous :) By "when the connect succeeds", I meant when the call to tcp_connect() succeeds at the end of tcp_v4_connect().

There are 2 cases you brought up regarding connection failures on binded INADDR_ANY sockets that might lead to a bhash2 mismatch:

Quoted 7 lines

#1) in dccp/tcp_v4_connect() where an error leads to the "goto failure" #2) in __inet_stream_connect() where an error leads to the "goto sock_error"

Regarding #2, I don't think this leads to a bhash2 mismatch. The "goto sock_error" triggers disconnect() and the disconnect() call will always remove the socket from the bhash and bhash2 tables since the socket when it was cloned had the SOCK_BINDPORT_LOCK flag unset.

I think client sockets are not cloned. The clone code is to create a child socket from a listener when receiving an ACK for SYN+ACK or earlier fastopen stage.

When a client socket receives a SYN+ACK,

tcp_v4_rcv
tcp_v4_do_rcv
tcp_rcv_state_process
tcp_rcv_synsent_state_process
tcp_finish_connect <-- just set TCP_ESTABLISHED

there's no clone, so tcp_disconnect() will not unhash sk from bhash2 if the sk is bound to INADDR_ANY.

Even if #1 succeeds, #2 could fail after, thus we have to take care of #2 at the same time.

So, I think the only place we can free the addrany_tb2 is just after "sock->state = SS_CONNECTED;" in __inet_stream_connect().

Quoted 102 lines

For #1, I think the simplest way to handle this is to not free the inet_bind2_bucket for INADDR_ANY/port when we call inet_bhash2_update_saddr(). Then, at the end of the dccp/tcp_v4/6_connect() call, if it is successful, we free the INADDR_ANY/port inet_bind2_bucket if it's empty. And if that is not successful and there was a failure somewhere, then in the "goto failure", we remove the socket from the new_addr bind bucket / clean up the new_addr bind bucket and re-add the socket back to the INADDR_ANY bind bucket.

What are your thoughts?

Quoted 90 lines
Quoted 90 lines

I don't think this scenario will break it because after a successful connect call, a disconnect will remove the socket from the bhash (and bhash2) table altogether since the socket will *not* have SOCK_BINDPORT_LOCK set.

This is the code path I'm looking at:

connect call is successful and the three way handshake has completed -> tcp_v4_syn_recv_sock() is called where we create the new socket tcp_v4_syn_recv_sock() -> tcp_create_openreq_child() -> inet_csk_clone_lock() -> sk_clone_lock()

In sk_clone_lock(), we create a newsk where newsk->sk_userlocks will *not* have the SOCK_BINDPORT_LOCK flag set (newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;).

If the connection gets closed by a RST/timeout/ICMP error, then this socket will go through disconnect() where in the disconnect call, it will remove the socket from the bhash and bhash2 tables. For example, for tcp, this code path is tcp_disconnect() -> tcp_set_state(sk, TCP_CLOSE) -> inet_put_port() - this is also verified by the "WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);" at the end of tcp_disconnect().

What are your thoughts?

Quoted 62 lines
Quoted 62 lines

Thank you.

Quoted 58 lines

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * This unhashes the socket and releases the local port, if necessary.
         */
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        __sk_dst_reset(sk);
 failure:
        inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * if necessary.
         */
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
 failure:
        inet->inet_dport = 0;
        sk->sk_route_caps = 0;
---8<---
#13 lore raw
Quoted 312 lines · Kuniyuki Iwashima
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 14:02:11 -0800
Quoted 309 lines · Kuniyuki Iwashima
Quoted 296 lines
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 09:40:22 -0800
Quoted 255 lines · Kuniyuki Iwashima
Quoted 254 lines
From:   Joanne Koong <[email protected]>
Date:   Mon, 7 Nov 2022 14:20:46 -0800
Quoted 68 lines · Kuniyuki Iwashima
Quoted 15 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

[...]

Quoted 5 lines

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

Since the s2 address mismatch case is addressed by your patch
https://lore.kernel.org/netdev/[email protected]/,
I will focus my comments here on the s3 case.

Quoted 41 lines
  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

I'm on the head of net-next/master (commit bf46390f39c686d62afeae9845860e63886d63b) and trying to repro this locally, but the warning isn't showing up for me after following the steps above. Not sure why.

Hmm... it reproduced on top of the commit. I'm testing on QEMU and login to serial console which outputs syslog in the same stream, so you may want to check /var/log/messages or something.

---8<---

# sysctl -w net.ipv4.tcp_syn_retries=1
# python3
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

from socket import *

s = socket()
s.bind(('0', 10000))
s.connect(('0.0.0.1', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TimeoutError: [Errno 110] Connection timed out

Quoted 1 line
Quoted 1 line
Quoted 1 line

s

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[   96.598308] ------------[ cut here ]------------
[   96.598598] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

s = socket()
s.bind(('0', 10001))
s.connect(('localhost', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[  139.157193] ------------[ cut here ]------------
[  139.157528] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

---8<---

Quoted 5 lines
Quoted 4 lines

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket.

Quoted 9 lines

To summarize, the path you are talking about is tcp_v4_connect() in kernel/linux/net/ipv4/tcp_ipv4.c where the sk originally has saddr INADDR_ANY, the sk gets assigned a new address, that new address gets updated in the bhash2 table, and then when inet_hash_connect() is called, it fails which brings us to the "goto failure". In the failure goto, we call "tcp_set_state(sk, TCP_CLOSE)" but in the case where "SOCK_BINDPORT_LOCK" is held, "inet_put_port(sk)" is *not* called, which means the sk will still be in the bhash2 table with the new address.

Correct.

More precisely, 3 functions after inet_hash_connect() can cause the same issue.

- ip_route_newports
- tcp_fastopen_defer_connect
- tcp_connect

Quoted 11 lines
Quoted 9 lines

So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
        inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

[...]

Quoted 1 line
Quoted 8 lines
Quoted 3 lines

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

This is the case where in __inet_stream_connect(), the call to "sk->sk_prot->connect()" succeeds but then the connection is closed by RST/timeout/ICMP error, so then the "goto sock_error" is triggered, correct?

Yes.

Quoted 76 lines
Quoted 58 lines

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

To summarize your analysis, there is an issue right now where for sockets that are binded on address INADDR_ANY, we need to handle the error case where if a connection fails and SOCK_BINDPORT_LOCK is held, the new address it was assigned needs to be taken out of bhash2 and the original address (INADDR_ANY) needs to be re-added to bhash2. There are two of these error cases we need to handle, as you mentioned above - 1) in dccp/tcp_v4_connect() where the connect call fails and 2) in __inet_stream_connect() where the connect call succeeds but the connection is closed by a RST/timeout/ICMP error.

Quoted 8 lines

I think the simplest solution is to modify inet_bhash2_update_saddr() so that we don't free the inet_bind2_bucket() for INADDR_ANY/port (if it is empty after we update the saddr to the new addr) *until* the connect succeeds. When the connect succeeds, then we can check whether the inet_bind2_bucket for INADDR_ANY is empty, and if it is, then do the freeing for it.

What are your thoughts on this?

I was thinking the same, but this scenario will break it ?

connect() <-- unblocking socket
return -EINPROGRESS

receive SYN+ACK, send back ACK, and set state to TCP_ESTABLISEHD

Quoted 7 lines
Quoted 7 lines

free the old INADDR_ANY bucket

get RST and set state to TCP_CLOSE

connect()

    goto sock_error and ->disconect() fail to restore the bucket

I meant all of the above happen at the client side, so the server-side code is orthogonal.

But I got your point. I've misunderstood "When the connect succeeds". I was thinking that we'll free bhash2 bucket in the fast path where we send back SYN+ACK to the server. OTOH, you meant it to be when the connect() syscall returns 0, right ?

Then, I think it's doable, like

1) inet_bhash2_update_saddr() increments addrany_tb2->connecting

2) when connect() succeeds,

     if (dec_and_test(&addrany_tb2->connecting) && is_empty(addrany_tb2->owners))
         free(addrany_tb2);

3) while close()ing,

     tb2 = ...
     addrany_tb2 = ...

     if ((sk->sk_user_locks & SOCK_BINDPORT_LOCK) &&
         !(sk->sk_user_locks & SOCK_BINDADDR_LOCK) &&
         tb2 != addrany_tb2 &&
         sock->state != SS_CONNECTED &&
         dec_and_test(&addrany_tb2->connecting) &&
         is_empty(addrany_tb2->owners))
         free(addrany_tb2)
Quoted 1 line

More consideration would be needed for inet_shutdown() etc though.

Sorry, my terminology was ambiguous :) By "when the connect succeeds", I meant when the call to tcp_connect() succeeds at the end of tcp_v4_connect().

There are 2 cases you brought up regarding connection failures on binded INADDR_ANY sockets that might lead to a bhash2 mismatch:

#1) in dccp/tcp_v4_connect() where an error leads to the "goto failure" #2) in __inet_stream_connect() where an error leads to the "goto sock_error"

Quoted 8 lines
Quoted 4 lines

Regarding #2, I don't think this leads to a bhash2 mismatch. The "goto sock_error" triggers disconnect() and the disconnect() call will always remove the socket from the bhash and bhash2 tables since the socket when it was cloned had the SOCK_BINDPORT_LOCK flag unset.

I think client sockets are not cloned. The clone code is to create a child socket from a listener when receiving an ACK for SYN+ACK or earlier fastopen stage.

Gotcha, thanks for clarifying.

This edge case only happens when this situation occurs:
A socket is bound to INADDR_ANY on a specified port -> socket makes a connect() call -> we run into one of these scenarios: the error case in dccp/tcp_v4/6_connect() or error case in __inet_stream_connect() or call to inet_shutdown()) -> in handling that error case, we remove socket from bhash2's new address/port table, and try to add it to the bhash2's INADDR_ANY/port table -> there is a "not enough memory" error in the case where the INADDR_ANY/port table got previously freed and we try to re-allocate it with kmem_cache_alloc()

I think the best solution is to unbind the socket if this situation happens, because

1) The socket will have to be unbinded in other error cases.

After looking at the code some more, I believe there's an existing bug when resetting the socket address back to INADDR_ANY after an error occurs (while BIND_PORTLOCK is set). The socket's address *cannot* simply be reset back to 0 without removing it from the bhash table, without first checking whether this leads to a bind conflict. For example, there could be this scenario: socket 1 binds to INADDR_ANY on port 443, there is no bind conflict (for example, no other sockets are binded to port 443), the connect succeeds and socket1 is now assigned address 1234 on port 443, then socket2 connects to port 443 on some other address, there's an error that causes socket1's connection to close and socket1's address is now reset back to INADDR_ANY. However, socket1's address cannot be reset back to INADDR_ANY on port 443 because now there is a conflict with socket2 if socket1's address is INADDR_ANY.

Thus, we need to check whether there would be a bind conflict before resetting the socket's address back to INADDR_ANY. And if there is, then we'd need to unbind socket1.

I think this fix should be in a separate patchset from any bhash2 changes.

2) If the "-ENOMEM" case is hit in the kmem_cache_alloc(), I believe this only happens if the system is out of memory, in which case I think the socket will most likely be ineffective anyways (eg system reboots)

3) This situation is a very rare corner case. I don't think it's worth providing the optimal solution for, by making the common case more complex (for example, not including INADDR_ANY entries in bhash2 and putting them at the front of the list in bhash, as suggested by Martin in a previous reply, which seems like the best solution if we were to fix this optimally)

I propose that we do the following:
in the error cases in dccp/tcp_v4/6_connect(), __inet_stream_connect() and inet_shutdown() where the socket's address needs to get reset back to INADDR_ANY, we remove the socket (from socket's new address + port) from the bhash2 table. Then we add the socket to the bhash2 table using INADDR_ANY + port; in the unlikely case where this operation fails due to -ENOMEM, then we unbind the socket.

What are your thoughts?

Quoted 121 lines

When a client socket receives a SYN+ACK,

tcp_v4_rcv
tcp_v4_do_rcv
tcp_rcv_state_process
tcp_rcv_synsent_state_process
tcp_finish_connect <-- just set TCP_ESTABLISHED

there's no clone, so tcp_disconnect() will not unhash sk from bhash2 if the sk is bound to INADDR_ANY.

Even if #1 succeeds, #2 could fail after, thus we have to take care of #2 at the same time.

So, I think the only place we can free the addrany_tb2 is just after "sock->state = SS_CONNECTED;" in __inet_stream_connect().

Quoted 102 lines

For #1, I think the simplest way to handle this is to not free the inet_bind2_bucket for INADDR_ANY/port when we call inet_bhash2_update_saddr(). Then, at the end of the dccp/tcp_v4/6_connect() call, if it is successful, we free the INADDR_ANY/port inet_bind2_bucket if it's empty. And if that is not successful and there was a failure somewhere, then in the "goto failure", we remove the socket from the new_addr bind bucket / clean up the new_addr bind bucket and re-add the socket back to the INADDR_ANY bind bucket.

What are your thoughts?

Quoted 90 lines
Quoted 90 lines

I don't think this scenario will break it because after a successful connect call, a disconnect will remove the socket from the bhash (and bhash2) table altogether since the socket will *not* have SOCK_BINDPORT_LOCK set.

This is the code path I'm looking at:

connect call is successful and the three way handshake has completed -> tcp_v4_syn_recv_sock() is called where we create the new socket tcp_v4_syn_recv_sock() -> tcp_create_openreq_child() -> inet_csk_clone_lock() -> sk_clone_lock()

In sk_clone_lock(), we create a newsk where newsk->sk_userlocks will *not* have the SOCK_BINDPORT_LOCK flag set (newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;).

If the connection gets closed by a RST/timeout/ICMP error, then this socket will go through disconnect() where in the disconnect call, it will remove the socket from the bhash and bhash2 tables. For example, for tcp, this code path is tcp_disconnect() -> tcp_set_state(sk, TCP_CLOSE) -> inet_put_port() - this is also verified by the "WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);" at the end of tcp_disconnect().

What are your thoughts?

Quoted 62 lines
Quoted 62 lines

Thank you.

Quoted 58 lines

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * This unhashes the socket and releases the local port, if necessary.
         */
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        __sk_dst_reset(sk);
 failure:
        inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * if necessary.
         */
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
 failure:
        inet->inet_dport = 0;
        sk->sk_route_caps = 0;
---8<---
#14 lore raw
From:   Joanne Koong <[email protected]>
Date:   Wed, 9 Nov 2022 19:00:00 -0800
Quoted 354 lines · Kuniyuki Iwashima
Quoted 312 lines
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 14:02:11 -0800
Quoted 309 lines · Kuniyuki Iwashima
Quoted 296 lines
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 09:40:22 -0800
Quoted 255 lines · Kuniyuki Iwashima
Quoted 254 lines
From:   Joanne Koong <[email protected]>
Date:   Mon, 7 Nov 2022 14:20:46 -0800
Quoted 68 lines · Kuniyuki Iwashima
Quoted 15 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

[...]

Quoted 5 lines

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

Since the s2 address mismatch case is addressed by your patch
https://lore.kernel.org/netdev/[email protected]/,
I will focus my comments here on the s3 case.

Quoted 41 lines
  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

I'm on the head of net-next/master (commit bf46390f39c686d62afeae9845860e63886d63b) and trying to repro this locally, but the warning isn't showing up for me after following the steps above. Not sure why.

Hmm... it reproduced on top of the commit. I'm testing on QEMU and login to serial console which outputs syslog in the same stream, so you may want to check /var/log/messages or something.

---8<---

# sysctl -w net.ipv4.tcp_syn_retries=1
# python3
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

from socket import *

s = socket()
s.bind(('0', 10000))
s.connect(('0.0.0.1', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TimeoutError: [Errno 110] Connection timed out

Quoted 1 line
Quoted 1 line
Quoted 1 line

s

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[   96.598308] ------------[ cut here ]------------
[   96.598598] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

s = socket()
s.bind(('0', 10001))
s.connect(('localhost', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[  139.157193] ------------[ cut here ]------------
[  139.157528] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

---8<---

Quoted 5 lines
Quoted 4 lines

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket.

Quoted 9 lines

To summarize, the path you are talking about is tcp_v4_connect() in kernel/linux/net/ipv4/tcp_ipv4.c where the sk originally has saddr INADDR_ANY, the sk gets assigned a new address, that new address gets updated in the bhash2 table, and then when inet_hash_connect() is called, it fails which brings us to the "goto failure". In the failure goto, we call "tcp_set_state(sk, TCP_CLOSE)" but in the case where "SOCK_BINDPORT_LOCK" is held, "inet_put_port(sk)" is *not* called, which means the sk will still be in the bhash2 table with the new address.

Correct.

More precisely, 3 functions after inet_hash_connect() can cause the same issue.

- ip_route_newports
- tcp_fastopen_defer_connect
- tcp_connect

Quoted 11 lines
Quoted 9 lines

So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
        inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

[...]

Quoted 1 line
Quoted 8 lines
Quoted 3 lines

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

This is the case where in __inet_stream_connect(), the call to "sk->sk_prot->connect()" succeeds but then the connection is closed by RST/timeout/ICMP error, so then the "goto sock_error" is triggered, correct?

Yes.

Quoted 76 lines
Quoted 58 lines

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

To summarize your analysis, there is an issue right now where for sockets that are binded on address INADDR_ANY, we need to handle the error case where if a connection fails and SOCK_BINDPORT_LOCK is held, the new address it was assigned needs to be taken out of bhash2 and the original address (INADDR_ANY) needs to be re-added to bhash2. There are two of these error cases we need to handle, as you mentioned above - 1) in dccp/tcp_v4_connect() where the connect call fails and 2) in __inet_stream_connect() where the connect call succeeds but the connection is closed by a RST/timeout/ICMP error.

Quoted 8 lines

I think the simplest solution is to modify inet_bhash2_update_saddr() so that we don't free the inet_bind2_bucket() for INADDR_ANY/port (if it is empty after we update the saddr to the new addr) *until* the connect succeeds. When the connect succeeds, then we can check whether the inet_bind2_bucket for INADDR_ANY is empty, and if it is, then do the freeing for it.

What are your thoughts on this?

I was thinking the same, but this scenario will break it ?

connect() <-- unblocking socket
return -EINPROGRESS

receive SYN+ACK, send back ACK, and set state to TCP_ESTABLISEHD

Quoted 7 lines
Quoted 7 lines

free the old INADDR_ANY bucket

get RST and set state to TCP_CLOSE

connect()

    goto sock_error and ->disconect() fail to restore the bucket

I meant all of the above happen at the client side, so the server-side code is orthogonal.

But I got your point. I've misunderstood "When the connect succeeds". I was thinking that we'll free bhash2 bucket in the fast path where we send back SYN+ACK to the server. OTOH, you meant it to be when the connect() syscall returns 0, right ?

Then, I think it's doable, like

1) inet_bhash2_update_saddr() increments addrany_tb2->connecting

2) when connect() succeeds,

     if (dec_and_test(&addrany_tb2->connecting) && is_empty(addrany_tb2->owners))
         free(addrany_tb2);

3) while close()ing,

     tb2 = ...
     addrany_tb2 = ...

     if ((sk->sk_user_locks & SOCK_BINDPORT_LOCK) &&
         !(sk->sk_user_locks & SOCK_BINDADDR_LOCK) &&
         tb2 != addrany_tb2 &&
         sock->state != SS_CONNECTED &&
         dec_and_test(&addrany_tb2->connecting) &&
         is_empty(addrany_tb2->owners))
         free(addrany_tb2)
Quoted 1 line

More consideration would be needed for inet_shutdown() etc though.

Sorry, my terminology was ambiguous :) By "when the connect succeeds", I meant when the call to tcp_connect() succeeds at the end of tcp_v4_connect().

There are 2 cases you brought up regarding connection failures on binded INADDR_ANY sockets that might lead to a bhash2 mismatch:

#1) in dccp/tcp_v4_connect() where an error leads to the "goto failure" #2) in __inet_stream_connect() where an error leads to the "goto sock_error"

Quoted 8 lines
Quoted 4 lines

Regarding #2, I don't think this leads to a bhash2 mismatch. The "goto sock_error" triggers disconnect() and the disconnect() call will always remove the socket from the bhash and bhash2 tables since the socket when it was cloned had the SOCK_BINDPORT_LOCK flag unset.

I think client sockets are not cloned. The clone code is to create a child socket from a listener when receiving an ACK for SYN+ACK or earlier fastopen stage.

Gotcha, thanks for clarifying.

This edge case only happens when this situation occurs:
A socket is bound to INADDR_ANY on a specified port -> socket makes a connect() call -> we run into one of these scenarios: the error case in dccp/tcp_v4/6_connect() or error case in __inet_stream_connect() or call to inet_shutdown()) -> in handling that error case, we remove socket from bhash2's new address/port table, and try to add it to the bhash2's INADDR_ANY/port table -> there is a "not enough memory" error in the case where the INADDR_ANY/port table got previously freed and we try to re-allocate it with kmem_cache_alloc()

I think the best solution is to unbind the socket if this situation happens, because

1) The socket will have to be unbinded in other error cases.

After looking at the code some more, I believe there's an existing bug when resetting the socket address back to INADDR_ANY after an error occurs (while BIND_PORTLOCK is set). The socket's address *cannot* simply be reset back to 0 without removing it from the bhash table, without first checking whether this leads to a bind conflict. For example, there could be this scenario: socket 1 binds to INADDR_ANY on port 443, there is no bind conflict (for example, no other sockets are binded to port 443), the connect succeeds and socket1 is now assigned address 1234 on port 443, then socket2 connects to port 443 on some other address, there's an error that causes socket1's connection to close and socket1's address is now reset back to INADDR_ANY. However, socket1's address cannot be reset back to INADDR_ANY on port 443 because now there is a conflict with socket2 if socket1's address is INADDR_ANY.

Quoted 5 lines

Thus, we need to check whether there would be a bind conflict before resetting the socket's address back to INADDR_ANY. And if there is, then we'd need to unbind socket1.

I think this fix should be in a separate patchset from any bhash2 changes.

Right, I was wondering why this bug was acceptable. And yes, this should be a separate series.

Quoted 4 lines

2) If the "-ENOMEM" case is hit in the kmem_cache_alloc(), I believe this only happens if the system is out of memory, in which case I think the socket will most likely be ineffective anyways (eg system reboots)

Exactly.

Quoted 16 lines

3) This situation is a very rare corner case. I don't think it's worth providing the optimal solution for, by making the common case more complex (for example, not including INADDR_ANY entries in bhash2 and putting them at the front of the list in bhash, as suggested by Martin in a previous reply, which seems like the best solution if we were to fix this optimally)

I propose that we do the following:
in the error cases in dccp/tcp_v4/6_connect(), __inet_stream_connect() and inet_shutdown() where the socket's address needs to get reset back to INADDR_ANY, we remove the socket (from socket's new address + port) from the bhash2 table. Then we add the socket to the bhash2 table using INADDR_ANY + port; in the unlikely case where this operation fails due to -ENOMEM, then we unbind the socket.

Quoted 1 line

What are your thoughts?

I think it's reasonable and a good point of compromise. I'll post a fix that way.

Thank you!

Quoted 121 lines
Quoted 121 lines

When a client socket receives a SYN+ACK,

tcp_v4_rcv
tcp_v4_do_rcv
tcp_rcv_state_process
tcp_rcv_synsent_state_process
tcp_finish_connect <-- just set TCP_ESTABLISHED

there's no clone, so tcp_disconnect() will not unhash sk from bhash2 if the sk is bound to INADDR_ANY.

Even if #1 succeeds, #2 could fail after, thus we have to take care of #2 at the same time.

So, I think the only place we can free the addrany_tb2 is just after "sock->state = SS_CONNECTED;" in __inet_stream_connect().

Quoted 102 lines

For #1, I think the simplest way to handle this is to not free the inet_bind2_bucket for INADDR_ANY/port when we call inet_bhash2_update_saddr(). Then, at the end of the dccp/tcp_v4/6_connect() call, if it is successful, we free the INADDR_ANY/port inet_bind2_bucket if it's empty. And if that is not successful and there was a failure somewhere, then in the "goto failure", we remove the socket from the new_addr bind bucket / clean up the new_addr bind bucket and re-add the socket back to the INADDR_ANY bind bucket.

What are your thoughts?

Quoted 90 lines
Quoted 90 lines

I don't think this scenario will break it because after a successful connect call, a disconnect will remove the socket from the bhash (and bhash2) table altogether since the socket will *not* have SOCK_BINDPORT_LOCK set.

This is the code path I'm looking at:

connect call is successful and the three way handshake has completed -> tcp_v4_syn_recv_sock() is called where we create the new socket tcp_v4_syn_recv_sock() -> tcp_create_openreq_child() -> inet_csk_clone_lock() -> sk_clone_lock()

In sk_clone_lock(), we create a newsk where newsk->sk_userlocks will *not* have the SOCK_BINDPORT_LOCK flag set (newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;).

If the connection gets closed by a RST/timeout/ICMP error, then this socket will go through disconnect() where in the disconnect call, it will remove the socket from the bhash and bhash2 tables. For example, for tcp, this code path is tcp_disconnect() -> tcp_set_state(sk, TCP_CLOSE) -> inet_put_port() - this is also verified by the "WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);" at the end of tcp_disconnect().

What are your thoughts?

Quoted 62 lines
Quoted 62 lines

Thank you.

Quoted 58 lines

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * This unhashes the socket and releases the local port, if necessary.
         */
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        __sk_dst_reset(sk);
 failure:
        inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * if necessary.
         */
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
 failure:
        inet->inet_dport = 0;
        sk->sk_route_caps = 0;
---8<---
#15 lore raw
Quoted 391 lines · Kuniyuki Iwashima
From:   Joanne Koong <[email protected]>
Date:   Wed, 9 Nov 2022 19:00:00 -0800
Quoted 354 lines · Kuniyuki Iwashima
Quoted 312 lines
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 14:02:11 -0800
Quoted 309 lines · Kuniyuki Iwashima
Quoted 296 lines
From:   Joanne Koong <[email protected]>
Date:   Tue, 8 Nov 2022 09:40:22 -0800
Quoted 255 lines · Kuniyuki Iwashima
Quoted 254 lines
From:   Joanne Koong <[email protected]>
Date:   Mon, 7 Nov 2022 14:20:46 -0800
Quoted 68 lines · Kuniyuki Iwashima
Quoted 15 lines

Hi,

I want to discuss bhash2 and WARN_ON() being fired every day this month on my syzkaller instance without repro.

  WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port (net/ipv4/inet_connection_sock.c:548 (discriminator 1))
  ...
  inet_csk_listen_start (net/ipv4/inet_connection_sock.c:1205)
  inet_listen (net/ipv4/af_inet.c:228)
  __sys_listen (net/socket.c:1810)
  __x64_sys_listen (net/socket.c:1819 net/socket.c:1817 net/socket.c:1817)
  do_syscall_64 (arch/x86/entry/common.c:50 arch/x86/entry/common.c:80)
  entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:120)

[...]

Quoted 5 lines

Please see the source addresses of s2/s3 below after connect() fails. The s2 case is another variant of the first syzbot report, which has been already _fixed_. And the s3 case is a repro for the issue that Mat and I saw.

Since the s2 address mismatch case is addressed by your patch
https://lore.kernel.org/netdev/[email protected]/,
I will focus my comments here on the s3 case.

Quoted 41 lines
  # sysctl -w net.ipv4.tcp_syn_retries=1
  net.ipv4.tcp_syn_retries = 1
  # python3
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines

from socket import *

s1 = socket()
s1.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s1.bind(('0.0.0.0', 10000))
s1.connect(('127.0.0.1', 10000))

s2 = socket()
s2.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s2.bind(('0.0.0.0', 10000))
s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2.connect(('127.0.0.1', 10000))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

OSError: [Errno 99] Cannot assign requested address

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s2

<socket.socket fd=4, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 10000)>

                                                                                                   ^^^ ???
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s3 = socket()
s3.setsockopt(SOL_SOCKET, SO_REUSEADDR, 1)
s3.bind(('0.0.0.0', 10000))
s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

TimeoutError: [Errno 110] Connection timed out

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

s3

<socket.socket fd=5, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

We can fire the WARN_ON() by s3.listen().

Quoted 8 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line

s3.listen()

  [ 1096.845905] ------------[ cut here ]------------
  [ 1096.846290] WARNING: CPU: 0 PID: 209 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

I'm on the head of net-next/master (commit bf46390f39c686d62afeae9845860e63886d63b) and trying to repro this locally, but the warning isn't showing up for me after following the steps above. Not sure why.

Hmm... it reproduced on top of the commit. I'm testing on QEMU and login to serial console which outputs syslog in the same stream, so you may want to check /var/log/messages or something.

---8<---

# sysctl -w net.ipv4.tcp_syn_retries=1
# python3
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

from socket import *

s = socket()
s.bind(('0', 10000))
s.connect(('0.0.0.1', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TimeoutError: [Errno 110] Connection timed out

Quoted 1 line
Quoted 1 line
Quoted 1 line

s

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[   96.598308] ------------[ cut here ]------------
[   96.598598] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

...

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

s = socket()
s.bind(('0', 10001))
s.connect(('localhost', 1))

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s.listen(32)

[  139.157193] ------------[ cut here ]------------
[  139.157528] WARNING: CPU: 0 PID: 214 at net/ipv4/inet_connection_sock.c:548 inet_csk_get_port+0x6bb/0x9e0

---8<---

Quoted 5 lines
Quoted 4 lines

In the s3 case, connect() resets sk->sk_rcv_saddr to INADDR_ANY without updating the bhash2 bucket; OTOH sk has the correct non-NULL bhash bucket.

Quoted 9 lines

To summarize, the path you are talking about is tcp_v4_connect() in kernel/linux/net/ipv4/tcp_ipv4.c where the sk originally has saddr INADDR_ANY, the sk gets assigned a new address, that new address gets updated in the bhash2 table, and then when inet_hash_connect() is called, it fails which brings us to the "goto failure". In the failure goto, we call "tcp_set_state(sk, TCP_CLOSE)" but in the case where "SOCK_BINDPORT_LOCK" is held, "inet_put_port(sk)" is *not* called, which means the sk will still be in the bhash2 table with the new address.

Correct.

More precisely, 3 functions after inet_hash_connect() can cause the same issue.

- ip_route_newports
- tcp_fastopen_defer_connect
- tcp_connect

Quoted 11 lines
Quoted 9 lines

So, when we call listen() for s3, inet_csk_get_port() does not call inet_bind_hash() and the WARN_ON() for bhash2 fires.

  if (!inet_csk(sk)->icsk_bind_hash)
        inet_bind_hash(sk, tb, tb2, port);
  WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
  WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);

[...]

Quoted 1 line
Quoted 8 lines
Quoted 3 lines

In the s3 case, connect() falls into a different path. We reach the sock_error label in __inet_stream_connect() and call sk_prot->disconnect(), which resets sk->sk_rcv_saddr.

This is the case where in __inet_stream_connect(), the call to "sk->sk_prot->connect()" succeeds but then the connection is closed by RST/timeout/ICMP error, so then the "goto sock_error" is triggered, correct?

Yes.

Quoted 76 lines
Quoted 58 lines

Then, there could be two subsequent issues.

* listen() leaks a newly allocated bhash2 bucket

* In inet_put_port(), inet_bhashfn_portaddr() computes a wrong hash for

    inet_csk(sk)->icsk_bind2_hash, resulting in list corruption.

We can fix these easily but it still leaves inconsistent sockets in bhash2, so we need to fix the root cause.

As a side note, this issue only happens when we bind() only port before connect(). If we do not bind() port, tcp_set_state(sk, TCP_CLOSE) calls inet_put_port() and unhashes the sk from bhash2.

At first, I applied the patch below so that both sk2 and sk3 trigger WARN_ON(). Then, I tried two approaches:

* Fix up the bhash2 entry when calling sk_rcv_saddr

* Change the bhash2 entry only when connect() succeeds

The former does not work when we run out of memory. When we change saddr from INADDR_ANY, inet_bhash2_update_saddr() could free (INADDR_ANY, port) bhash2 bucket. Then, we possibly could not allocate it again when restoring saddr in the failure path.

The latter does not work when a sk is in non-blocking mode. In this case, a user might not call the second connect() to fix up the bhash2 bucket.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

s4 = socket()
s4.bind(('0.0.0.0', 10000))
s4.setblocking(False)
s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

BlockingIOError: [Errno 115] Operation now in progress

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('10.0.2.15', 10000)>

Also, the former approach does not work for the non-blocking case. Let's say the second connect() fails. What if we fail to allocate an INADDR_ANY bhash2 bucket? We have to change saddr to INADDR_ANY but cannot.... but the connect() actually failed....??

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4.connect(('0.0.0.1', 1))

Traceback (most recent call last):

    File "<stdin>", line 1, in <module>

ConnectionRefusedError: [Errno 111] Connection refused

Quoted 1 line
Quoted 1 line
Quoted 1 line

s4

<socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('0.0.0.0', 10000)>

Quoted 7 lines

Now, I'm thinking bhash2 bucket needs a refcnt not to be freed while refcnt is greater than 1. And we need to change the conflict logic so that the kernel ignores empty bhash2 bucket. Such changes could be big for the net tree, but the next LTS will likely be v6.1 which has bhash2.

What do you think is the best way to fix the issue?

To summarize your analysis, there is an issue right now where for sockets that are binded on address INADDR_ANY, we need to handle the error case where if a connection fails and SOCK_BINDPORT_LOCK is held, the new address it was assigned needs to be taken out of bhash2 and the original address (INADDR_ANY) needs to be re-added to bhash2. There are two of these error cases we need to handle, as you mentioned above - 1) in dccp/tcp_v4_connect() where the connect call fails and 2) in __inet_stream_connect() where the connect call succeeds but the connection is closed by a RST/timeout/ICMP error.

Quoted 8 lines

I think the simplest solution is to modify inet_bhash2_update_saddr() so that we don't free the inet_bind2_bucket() for INADDR_ANY/port (if it is empty after we update the saddr to the new addr) *until* the connect succeeds. When the connect succeeds, then we can check whether the inet_bind2_bucket for INADDR_ANY is empty, and if it is, then do the freeing for it.

What are your thoughts on this?

I was thinking the same, but this scenario will break it ?

connect() <-- unblocking socket
return -EINPROGRESS

receive SYN+ACK, send back ACK, and set state to TCP_ESTABLISEHD

Quoted 7 lines
Quoted 7 lines

free the old INADDR_ANY bucket

get RST and set state to TCP_CLOSE

connect()

    goto sock_error and ->disconect() fail to restore the bucket

I meant all of the above happen at the client side, so the server-side code is orthogonal.

But I got your point. I've misunderstood "When the connect succeeds". I was thinking that we'll free bhash2 bucket in the fast path where we send back SYN+ACK to the server. OTOH, you meant it to be when the connect() syscall returns 0, right ?

Then, I think it's doable, like

1) inet_bhash2_update_saddr() increments addrany_tb2->connecting

2) when connect() succeeds,

     if (dec_and_test(&addrany_tb2->connecting) && is_empty(addrany_tb2->owners))
         free(addrany_tb2);

3) while close()ing,

     tb2 = ...
     addrany_tb2 = ...

     if ((sk->sk_user_locks & SOCK_BINDPORT_LOCK) &&
         !(sk->sk_user_locks & SOCK_BINDADDR_LOCK) &&
         tb2 != addrany_tb2 &&
         sock->state != SS_CONNECTED &&
         dec_and_test(&addrany_tb2->connecting) &&
         is_empty(addrany_tb2->owners))
         free(addrany_tb2)
Quoted 1 line

More consideration would be needed for inet_shutdown() etc though.

Sorry, my terminology was ambiguous :) By "when the connect succeeds", I meant when the call to tcp_connect() succeeds at the end of tcp_v4_connect().

There are 2 cases you brought up regarding connection failures on binded INADDR_ANY sockets that might lead to a bhash2 mismatch:

#1) in dccp/tcp_v4_connect() where an error leads to the "goto failure" #2) in __inet_stream_connect() where an error leads to the "goto sock_error"

Quoted 8 lines
Quoted 4 lines

Regarding #2, I don't think this leads to a bhash2 mismatch. The "goto sock_error" triggers disconnect() and the disconnect() call will always remove the socket from the bhash and bhash2 tables since the socket when it was cloned had the SOCK_BINDPORT_LOCK flag unset.

I think client sockets are not cloned. The clone code is to create a child socket from a listener when receiving an ACK for SYN+ACK or earlier fastopen stage.

Gotcha, thanks for clarifying.

This edge case only happens when this situation occurs:
A socket is bound to INADDR_ANY on a specified port -> socket makes a connect() call -> we run into one of these scenarios: the error case in dccp/tcp_v4/6_connect() or error case in __inet_stream_connect() or call to inet_shutdown()) -> in handling that error case, we remove socket from bhash2's new address/port table, and try to add it to the bhash2's INADDR_ANY/port table -> there is a "not enough memory" error in the case where the INADDR_ANY/port table got previously freed and we try to re-allocate it with kmem_cache_alloc()

I think the best solution is to unbind the socket if this situation happens, because

1) The socket will have to be unbinded in other error cases.

After looking at the code some more, I believe there's an existing bug when resetting the socket address back to INADDR_ANY after an error occurs (while BIND_PORTLOCK is set). The socket's address *cannot* simply be reset back to 0 without removing it from the bhash table, without first checking whether this leads to a bind conflict. For example, there could be this scenario: socket 1 binds to INADDR_ANY on port 443, there is no bind conflict (for example, no other sockets are binded to port 443), the connect succeeds and socket1 is now assigned address 1234 on port 443, then socket2 connects to port 443 on some other address, there's an error that causes socket1's connection to close and socket1's address is now reset back to INADDR_ANY. However, socket1's address cannot be reset back to INADDR_ANY on port 443 because now there is a conflict with socket2 if socket1's address is INADDR_ANY.

Quoted 5 lines

Thus, we need to check whether there would be a bind conflict before resetting the socket's address back to INADDR_ANY. And if there is, then we'd need to unbind socket1.

I think this fix should be in a separate patchset from any bhash2 changes.

Right, I was wondering why this bug was acceptable. And yes, this should be a separate series.

Quoted 4 lines

2) If the "-ENOMEM" case is hit in the kmem_cache_alloc(), I believe this only happens if the system is out of memory, in which case I think the socket will most likely be ineffective anyways (eg system reboots)

Exactly.

Quoted 16 lines

3) This situation is a very rare corner case. I don't think it's worth providing the optimal solution for, by making the common case more complex (for example, not including INADDR_ANY entries in bhash2 and putting them at the front of the list in bhash, as suggested by Martin in a previous reply, which seems like the best solution if we were to fix this optimally)

I propose that we do the following:
in the error cases in dccp/tcp_v4/6_connect(), __inet_stream_connect() and inet_shutdown() where the socket's address needs to get reset back to INADDR_ANY, we remove the socket (from socket's new address + port) from the bhash2 table. Then we add the socket to the bhash2 table using INADDR_ANY + port; in the unlikely case where this operation fails due to -ENOMEM, then we unbind the socket.

Quoted 7 lines
Quoted 1 line

What are your thoughts?

I think it's reasonable and a good point of compromise. I'll post a fix that way.

Thank you!

Awesome! Thank you Kuniyuki!

Quoted 122 lines
Quoted 122 lines
Quoted 122 lines

When a client socket receives a SYN+ACK,

tcp_v4_rcv
tcp_v4_do_rcv
tcp_rcv_state_process
tcp_rcv_synsent_state_process
tcp_finish_connect <-- just set TCP_ESTABLISHED

there's no clone, so tcp_disconnect() will not unhash sk from bhash2 if the sk is bound to INADDR_ANY.

Even if #1 succeeds, #2 could fail after, thus we have to take care of #2 at the same time.

So, I think the only place we can free the addrany_tb2 is just after "sock->state = SS_CONNECTED;" in __inet_stream_connect().

Quoted 102 lines

For #1, I think the simplest way to handle this is to not free the inet_bind2_bucket for INADDR_ANY/port when we call inet_bhash2_update_saddr(). Then, at the end of the dccp/tcp_v4/6_connect() call, if it is successful, we free the INADDR_ANY/port inet_bind2_bucket if it's empty. And if that is not successful and there was a failure somewhere, then in the "goto failure", we remove the socket from the new_addr bind bucket / clean up the new_addr bind bucket and re-add the socket back to the INADDR_ANY bind bucket.

What are your thoughts?

Quoted 90 lines
Quoted 90 lines

I don't think this scenario will break it because after a successful connect call, a disconnect will remove the socket from the bhash (and bhash2) table altogether since the socket will *not* have SOCK_BINDPORT_LOCK set.

This is the code path I'm looking at:

connect call is successful and the three way handshake has completed -> tcp_v4_syn_recv_sock() is called where we create the new socket tcp_v4_syn_recv_sock() -> tcp_create_openreq_child() -> inet_csk_clone_lock() -> sk_clone_lock()

In sk_clone_lock(), we create a newsk where newsk->sk_userlocks will *not* have the SOCK_BINDPORT_LOCK flag set (newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;).

If the connection gets closed by a RST/timeout/ICMP error, then this socket will go through disconnect() where in the disconnect call, it will remove the socket from the bhash and bhash2 tables. For example, for tcp, this code path is tcp_disconnect() -> tcp_set_state(sk, TCP_CLOSE) -> inet_put_port() - this is also verified by the "WARN_ON(inet->inet_num && !icsk->icsk_bind_hash);" at the end of tcp_disconnect().

What are your thoughts?

Quoted 62 lines
Quoted 62 lines

Thank you.

Quoted 58 lines

Thank you.

---8<---

net/dccp/ipv4.c +2 -0
index 713b7b8dad7e..40640c26680e 100644
@@ -157,6 +157,8 @@ int dccp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * This unhashes the socket and releases the local port, if necessary.
         */
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/dccp/ipv6.c +2 -0
index e57b43006074..626166cb6d7e 100644
@@ -985,6 +985,8 @@ static int dccp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        dccp_set_state(sk, DCCP_CLOSED);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        __sk_dst_reset(sk);
 failure:
        inet->inet_dport = 0;
net/ipv4/tcp_ipv4.c +2 -0
index 7a250ef9d1b7..834245da1e95 100644
@@ -343,6 +343,8 @@ int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
         * if necessary.
         */
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
        ip_rt_put(rt);
        sk->sk_route_caps = 0;
        inet->inet_dport = 0;
net/ipv6/tcp_ipv6.c +2 -1
index 2a3f9296df1e..81b396e5cf79 100644
@@ -359,6 +359,8 @@ static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
 
 late_failure:
        tcp_set_state(sk, TCP_CLOSE);
+       if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
+               inet_reset_saddr(sk);
 failure:
        inet->inet_dport = 0;
        sk->sk_route_caps = 0;
---8<---