CVS log for src/tests/net/net/t_tcp.c

Up to [cvs.NetBSD.org] / src / tests / net / net

Request diff between arbitrary revisions

Keyword substitution: kv
Default branch: MAIN

Pull up following revision(s) (requested by rin in ticket #813):

	tests/kernel/kqueue/t_empty.c: revision 1.2
	tests/net/net/t_tcp.c: revision 1.13

tests: Fix false positives due to race b/w connect(2) and accept(2)

For kernel/kqueue/t_empty and net/net/t_tcp, there were no sync ops
b/w connect(2) and accept(2) for non-blocking socket pair on host
(rump is not used).

As a result, accept(2) can fail immediately with EAGAIN, when
kernel-side routines for connect(2) and accept(2) are processed in
different CPU cores.
1-sec sleep(3) between two syscalls seems to mitigate this problem
as far as I can see, although this should not be a perfect solution...

Thanks ozaki-r@ for discussion.

tests: Fix false positives due to race b/w connect(2) and accept(2)

For kernel/kqueue/t_empty and net/net/t_tcp, there were no sync ops
b/w connect(2) and accept(2) for non-blocking socket pair on host
(rump is not used).

As a result, accept(2) can fail immediately with EAGAIN, when
kernel-side routines for connect(2) and accept(2) are processed in
different CPU cores.

1-sec sleep(3) between two syscalls seems to mitigate this problem
as far as I can see, although this should not be a perfect solution...

Thanks ozaki-r@ for discussion.

Fix (a kind of) violation of strict aliasing rule.

Due to the rule, "sin" and "sin6" can be treated as restrict pointers.
Compilers seem to be confused by structure copy for those pointed by
them before assignments.

For aarch64eb, GCC 9 and 10 compile t_tcp.c rev 1.11 into a code, where
fetch for "sin6->sin6_port" is preceding the structure copy "ss = bs".
This results in failure of connect(2) with EADDRNOOTAVAIL.

Mostly merge changes from HEAD upto 20200411

- use accept4 instead of paccept for everyone.
- add test for accept preserving non-block
- comment on FreeBSD and Linux behavior.

Use the same variable name for the accepted socket as with the AF_LOCAL test.
Call getpeereid on the accepted socket.

ensure that getpeereid does not succeed on tcp sockets.

explain what's going on before we fix it.

add getpeereid tests for non-unix sockets, returns garbage...

add tests for 4->6 connections.

add v6 tests

make it work on linux, be pickier about errors, and correct variable type.

Rebase to HEAD as of a few days ago.

sync with head.

for a reference, the tree before this commit was tagged
as yamt-pagecache-tag8.

this commit was splitted into small chunks to avoid
a limitation of cvs.  ("Protocol error: too many arguments")

file t_tcp.c was added on branch tls-maxphys on 2014-08-20 00:04:52 +0000

file t_tcp.c was added on branch yamt-pagecache on 2014-05-22 11:42:22 +0000

CID 1107548: resource leak

more tests

new test to check if non-blocking sockets are reset to blocking on the
accepted file descriptor.