linux.git/net/smc, branch v6.14

net: better track kernel sockets lifetime

2025-02-22T00:00:58+00:00

While kernel sockets are dismantled during pernet_operations->exit(),
their freeing can be delayed by any tx packets still held in qdisc
or device queues, due to skb_set_owner_w() prior calls.

This then trigger the following warning from ref_tracker_dir_exit() [1]

To fix this, make sure that kernel sockets own a reference on net->passive.

Add sk_net_refcnt_upgrade() helper, used whenever a kernel socket
is converted to a refcounted one.

[1]

[  136.263918][   T35] ref_tracker: net notrefcnt@ffff8880638f01e0 has 1/2 users at
[  136.263918][   T35]      sk_alloc+0x2b3/0x370
[  136.263918][   T35]      inet6_create+0x6ce/0x10f0
[  136.263918][   T35]      __sock_create+0x4c0/0xa30
[  136.263918][   T35]      inet_ctl_sock_create+0xc2/0x250
[  136.263918][   T35]      igmp6_net_init+0x39/0x390
[  136.263918][   T35]      ops_init+0x31e/0x590
[  136.263918][   T35]      setup_net+0x287/0x9e0
[  136.263918][   T35]      copy_net_ns+0x33f/0x570
[  136.263918][   T35]      create_new_namespaces+0x425/0x7b0
[  136.263918][   T35]      unshare_nsproxy_namespaces+0x124/0x180
[  136.263918][   T35]      ksys_unshare+0x57d/0xa70
[  136.263918][   T35]      __x64_sys_unshare+0x38/0x40
[  136.263918][   T35]      do_syscall_64+0xf3/0x230
[  136.263918][   T35]      entry_SYSCALL_64_after_hwframe+0x77/0x7f
[  136.263918][   T35]
[  136.343488][   T35] ref_tracker: net notrefcnt@ffff8880638f01e0 has 1/2 users at
[  136.343488][   T35]      sk_alloc+0x2b3/0x370
[  136.343488][   T35]      inet6_create+0x6ce/0x10f0
[  136.343488][   T35]      __sock_create+0x4c0/0xa30
[  136.343488][   T35]      inet_ctl_sock_create+0xc2/0x250
[  136.343488][   T35]      ndisc_net_init+0xa7/0x2b0
[  136.343488][   T35]      ops_init+0x31e/0x590
[  136.343488][   T35]      setup_net+0x287/0x9e0
[  136.343488][   T35]      copy_net_ns+0x33f/0x570
[  136.343488][   T35]      create_new_namespaces+0x425/0x7b0
[  136.343488][   T35]      unshare_nsproxy_namespaces+0x124/0x180
[  136.343488][   T35]      ksys_unshare+0x57d/0xa70
[  136.343488][   T35]      __x64_sys_unshare+0x38/0x40
[  136.343488][   T35]      do_syscall_64+0xf3/0x230
[  136.343488][   T35]      entry_SYSCALL_64_after_hwframe+0x77/0x7f

Fixes: 0cafd77dcd03 ("net: add a refcount tracker for kernel sockets")
Reported-by: syzbot+30a19e01a97420719891@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/netdev/67b72aeb.050a0220.14d86d.0283.GAE@google.com/T/#u
Signed-off-by: Eric Dumazet 
Reviewed-by: Kuniyuki Iwashima 
Link: https://patch.msgid.link/20250220131854.4048077-1-edumazet@google.com
Signed-off-by: Jakub Kicinski

net/smc: fix data error when recvmsg with MSG_PEEK flag

2025-01-14T02:59:00+00:00

When recvmsg with MSG_PEEK flag, the data will be copied to
user's buffer without advancing consume cursor and without
reducing the length of rx available data. Once the expected
peek length is larger than the value of bytes_to_rcv, in the
loop of do while in smc_rx_recvmsg, the first loop will copy
bytes_to_rcv bytes of data from the position local_tx_ctrl.cons,
the second loop will copy the min(bytes_to_rcv, read_remaining)
bytes from the position local_tx_ctrl.cons again because of the
lacking of process with advancing consume cursor and reducing
the length of available data. So do the subsequent loops. The
data copied in the second loop and the subsequent loops will
result in data error, as it should not be copied if no more data
arrives and it should be copied from the position advancing
bytes_to_rcv bytes from the local_tx_ctrl.cons if more data arrives.

This issue can be reproduce by the following python script:
server.py:
import socket
import time
server_ip = '0.0.0.0'
server_port = 12346
server_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server_socket.bind((server_ip, server_port))
server_socket.listen(1)
print('Server is running and listening for connections...')
conn, addr = server_socket.accept()
print('Connected by', addr)
while True:
    data = conn.recv(1024)
    if not data:
        break
    print('Received request:', data.decode())
    conn.sendall(b'Hello, client!\n')
    time.sleep(5)
    conn.sendall(b'Hello, again!\n')
conn.close()

client.py:
import socket
server_ip = ''
server_port = 12346
resp=b'Hello, client!\nHello, again!\n'
client_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
client_socket.connect((server_ip, server_port))
request = 'Hello, server!'
client_socket.sendall(request.encode())
peek_data = client_socket.recv(len(resp),
    socket.MSG_PEEK | socket.MSG_WAITALL)
print('Peeked data:', peek_data.decode())
client_socket.close()

Fixes: 952310ccf2d8 ("smc: receive data from RMBE")
Reported-by: D. Wythe 
Signed-off-by: Guangguan Wang 
Link: https://patch.msgid.link/20250104143201.35529-1-guangguan.wang@linux.alibaba.com
Signed-off-by: Jakub Kicinski

net/smc: delete pointless divide by one

2025-01-11T21:08:54+00:00

Here "buf" is a void pointer so sizeof(*buf) is one.  Doing a divide
by one makes the code less readable.  Delete it.

Signed-off-by: Dan Carpenter 
Reviewed-by: Simon Horman 
Reviewed-by: Kalesh AP 
Link: https://patch.msgid.link/ee1a790b-f874-4512-b3ae-9c45f99dc640@stanley.mountain
Signed-off-by: Jakub Kicinski

Merge git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net

2024-12-19T19:35:07+00:00

Cross-merge networking fixes after downstream PR (net-6.13-rc4).

No conflicts.

Adjacent changes:

drivers/net/ethernet/renesas/rswitch.h
  32fd46f5b69e ("net: renesas: rswitch: remove speed from gwca structure")
  922b4b955a03 ("net: renesas: rswitch: rework ts tags management")

Signed-off-by: Jakub Kicinski

net/smc: check return value of sock_recvmsg when draining clc data

2024-12-15T12:34:59+00:00

When receiving clc msg, the field length in smc_clc_msg_hdr indicates the
length of msg should be received from network and the value should not be
fully trusted as it is from the network. Once the value of length exceeds
the value of buflen in function smc_clc_wait_msg it may run into deadloop
when trying to drain the remaining data exceeding buflen.

This patch checks the return value of sock_recvmsg when draining data in
case of deadloop in draining.

Fixes: fb4f79264c0f ("net/smc: tolerate future SMCD versions")
Signed-off-by: Guangguan Wang 
Reviewed-by: Wen Gu 
Reviewed-by: D. Wythe 
Signed-off-by: David S. Miller

net/smc: check smcd_v2_ext_offset when receiving proposal msg

2024-12-15T12:34:59+00:00

When receiving proposal msg in server, the field smcd_v2_ext_offset in
proposal msg is from the remote client and can not be fully trusted.
Once the value of smcd_v2_ext_offset exceed the max value, there has
the chance to access wrong address, and crash may happen.

This patch checks the value of smcd_v2_ext_offset before using it.

Fixes: 5c21c4ccafe8 ("net/smc: determine accepted ISM devices")
Signed-off-by: Guangguan Wang 
Reviewed-by: Wen Gu 
Reviewed-by: D. Wythe 
Signed-off-by: David S. Miller

net/smc: check v2_ext_offset/eid_cnt/ism_gid_cnt when receiving proposal msg

2024-12-15T12:34:59+00:00

When receiving proposal msg in server, the fields v2_ext_offset/
eid_cnt/ism_gid_cnt in proposal msg are from the remote client
and can not be fully trusted. Especially the field v2_ext_offset,
once exceed the max value, there has the chance to access wrong
address, and crash may happen.

This patch checks the fields v2_ext_offset/eid_cnt/ism_gid_cnt
before using them.

Fixes: 8c3dca341aea ("net/smc: build and send V2 CLC proposal")
Signed-off-by: Guangguan Wang 
Reviewed-by: Wen Gu 
Reviewed-by: D. Wythe 
Signed-off-by: David S. Miller

net/smc: check iparea_offset and ipv6_prefixes_cnt when receiving proposal msg

2024-12-15T12:34:59+00:00

When receiving proposal msg in server, the field iparea_offset
and the field ipv6_prefixes_cnt in proposal msg are from the
remote client and can not be fully trusted. Especially the
field iparea_offset, once exceed the max value, there has the
chance to access wrong address, and crash may happen.

This patch checks iparea_offset and ipv6_prefixes_cnt before using them.

Fixes: e7b7a64a8493 ("smc: support variable CLC proposal messages")
Signed-off-by: Guangguan Wang 
Reviewed-by: Wen Gu 
Reviewed-by: D. Wythe 
Signed-off-by: David S. Miller

net/smc: check sndbuf_space again after NOSPACE flag is set in smc_poll

2024-12-15T12:34:59+00:00

When application sending data more than sndbuf_space, there have chances
application will sleep in epoll_wait, and will never be wakeup again. This
is caused by a race between smc_poll and smc_cdc_tx_handler.

application                                      tasklet
smc_tx_sendmsg(len > sndbuf_space)   |
epoll_wait for EPOLL_OUT,timeout=0   |
  smc_poll                           |
    if (!smc->conn.sndbuf_space)     |
                                     |  smc_cdc_tx_handler
                                     |    atomic_add sndbuf_space
                                     |    smc_tx_sndbuf_nonfull
                                     |      if (!test_bit SOCK_NOSPACE)
                                     |        do not sk_write_space;
      set_bit SOCK_NOSPACE;          |
    return mask=0;                   |

Application will sleep in epoll_wait as smc_poll returns 0. And
smc_cdc_tx_handler will not call sk_write_space because the SOCK_NOSPACE
has not be set. If there is no inflight cdc msg, sk_write_space will not be
called any more, and application will sleep in epoll_wait forever.
So check sndbuf_space again after NOSPACE flag is set to break the race.

Fixes: 8dce2786a290 ("net/smc: smc_poll improvements")
Signed-off-by: Guangguan Wang 
Suggested-by: Paolo Abeni 
Signed-off-by: David S. Miller

net/smc: protect link down work from execute after lgr freed

2024-12-15T12:34:59+00:00

link down work may be scheduled before lgr freed but execute
after lgr freed, which may result in crash. So it is need to
hold a reference before shedule link down work, and put the
reference after work executed or canceled.

The relevant crash call stack as follows:
 list_del corruption. prev->next should be ffffb638c9c0fe20,
    but was 0000000000000000
 ------------[ cut here ]------------
 kernel BUG at lib/list_debug.c:51!
 invalid opcode: 0000 [#1] SMP NOPTI
 CPU: 6 PID: 978112 Comm: kworker/6:119 Kdump: loaded Tainted: G #1
 Hardware name: Alibaba Cloud Alibaba Cloud ECS, BIOS 2221b89 04/01/2014
 Workqueue: events smc_link_down_work [smc]
 RIP: 0010:__list_del_entry_valid.cold+0x31/0x47
 RSP: 0018:ffffb638c9c0fdd8 EFLAGS: 00010086
 RAX: 0000000000000054 RBX: ffff942fb75e5128 RCX: 0000000000000000
 RDX: ffff943520930aa0 RSI: ffff94352091fc80 RDI: ffff94352091fc80
 RBP: 0000000000000000 R08: 0000000000000000 R09: ffffb638c9c0fc38
 R10: ffffb638c9c0fc30 R11: ffffffffa015eb28 R12: 0000000000000002
 R13: ffffb638c9c0fe20 R14: 0000000000000001 R15: ffff942f9cd051c0
 FS:  0000000000000000(0000) GS:ffff943520900000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 00007f4f25214000 CR3: 000000025fbae004 CR4: 00000000007706e0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 PKRU: 55555554
 Call Trace:
  rwsem_down_write_slowpath+0x17e/0x470
  smc_link_down_work+0x3c/0x60 [smc]
  process_one_work+0x1ac/0x350
  worker_thread+0x49/0x2f0
  ? rescuer_thread+0x360/0x360
  kthread+0x118/0x140
  ? __kthread_bind_mask+0x60/0x60
  ret_from_fork+0x1f/0x30

Fixes: 541afa10c126 ("net/smc: add smcr_port_err() and smcr_link_down() processing")
Signed-off-by: Guangguan Wang 
Reviewed-by: Tony Lu 
Signed-off-by: David S. Miller