]> www.infradead.org Git - linux.git/commit
af_unix: Stop recv(MSG_PEEK) at consumed OOB skb.
authorKuniyuki Iwashima <kuniyu@amazon.com>
Tue, 25 Jun 2024 01:36:37 +0000 (18:36 -0700)
committerPaolo Abeni <pabeni@redhat.com>
Thu, 27 Jun 2024 10:05:00 +0000 (12:05 +0200)
commitb94038d841a91d0e3f59cfe4d073e210910366ee
treeb869399ee5e9b3d3f41a9a84a21406d17807321a
parentd098d77232c375cb2cded4a7099f0a763016ee0d
af_unix: Stop recv(MSG_PEEK) at consumed OOB skb.

After consuming OOB data, recv() reading the preceding data must break at
the OOB skb regardless of MSG_PEEK.

Currently, MSG_PEEK does not stop recv() for AF_UNIX, and the behaviour is
not compliant with TCP.

  >>> from socket import *
  >>> c1, c2 = socketpair(AF_UNIX)
  >>> c1.send(b'hello', MSG_OOB)
  5
  >>> c1.send(b'world')
  5
  >>> c2.recv(1, MSG_OOB)
  b'o'
  >>> c2.recv(9, MSG_PEEK)  # This should return b'hell'
  b'hellworld'              # even with enough buffer.

Let's fix it by returning NULL for consumed skb and unlinking it only if
MSG_PEEK is not specified.

This patch also adds test cases that add recv(MSG_PEEK) before each recv().

Without fix:

  #  RUN           msg_oob.peek.oob_ahead_break ...
  # msg_oob.c:134:oob_ahead_break:AF_UNIX :hellworld
  # msg_oob.c:135:oob_ahead_break:Expected:hell
  # msg_oob.c:137:oob_ahead_break:Expected ret[0] (9) == expected_len (4)
  # oob_ahead_break: Test terminated by assertion
  #          FAIL  msg_oob.peek.oob_ahead_break
  not ok 13 msg_oob.peek.oob_ahead_break

With fix:

  #  RUN           msg_oob.peek.oob_ahead_break ...
  #            OK  msg_oob.peek.oob_ahead_break
  ok 13 msg_oob.peek.oob_ahead_break

Fixes: 314001f0bf92 ("af_unix: Add OOB support")
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
net/unix/af_unix.c
tools/testing/selftests/net/af_unix/msg_oob.c