Message ID | 20210508021448.7390-1-zhaogongyi@huawei.com |
---|---|
State | Accepted |
Headers | show |
Series | [v2] syscalls/bind03: Bugfix for running with option "-i" | expand |
Hi! I was looking at the test wondering why do you try to unlink socket that wasn't bind properly and it looks like kernel creates the socket even in a case that the bind fails, which is strange and I would call it a bug. Looking at the kernel code it indeed does mknod before it takes locks and checks the socket. Even more it turns out that there is in-flight patchset that is fixing this: https://lore.kernel.org/netdev/YDQAmH9zSsaqf+Dg@zeniv-ca.linux.org.uk/ Anyways for now moved the code initializing the sockets into a setup() function and the only call that had to stay in the run() function is the unlink(SNAME_B); with a comment that kernel is buggy. And we will have to write a regression test for this bug once upstream kernel is fixed. Also note that I use unlink() here since the test will fail otherwise once kernel is fixed and there is no socket to be unlinked.
Hi, > Hi! > I was looking at the test wondering why do you try to unlink socket that > wasn't bind properly and it looks like kernel creates the socket even in > a case that the bind fails, which is strange and I would call it a bug. > Looking at the kernel code it indeed does mknod before it takes locks > and checks the socket. Even more it turns out that there is in-flight > patchset that is fixing this: > https://lore.kernel.org/netdev/YDQAmH9zSsaqf+Dg@zeniv-ca.linux.org.uk/ > Anyways for now moved the code initializing the sockets into a setup() > function and the only call that had to stay in the run() function is the > unlink(SNAME_B); with a comment that kernel is buggy. > And we will have to write a regression test for this bug once upstream > kernel is fixed. > Also note that I use unlink() here since the test will fail otherwise > once kernel is fixed and there is no socket to be unlinked. FYI I created issue #817 for it, feel free to improve the description. https://github.com/linux-test-project/ltp/issues/814 Kind regards, Petr
diff --git a/testcases/kernel/syscalls/bind/bind03.c b/testcases/kernel/syscalls/bind/bind03.c index ebde19591..879ce1bc6 100644 --- a/testcases/kernel/syscalls/bind/bind03.c +++ b/testcases/kernel/syscalls/bind/bind03.c @@ -62,16 +62,14 @@ void run(void) */ TST_EXP_FAIL(bind(sock2, (struct sockaddr *)&sun1, sizeof(sun1)), EADDRINUSE, "bind() with bound pathname"); -} -static void cleanup(void) -{ + SAFE_UNLINK(sun1.sun_path); + SAFE_UNLINK(sun2.sun_path); SAFE_CLOSE(sock1); SAFE_CLOSE(sock2); } static struct tst_test test = { - .cleanup = cleanup, .test_all = run, .needs_tmpdir = 1, };
When run the test with option "-i 2", test will fail and report EADDRINUSE. Signed-off-by: Zhao Gongyi <zhaogongyi@huawei.com> --- v1->v2:replace unlink() with SAFE_UNLINK() testcases/kernel/syscalls/bind/bind03.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) -- 2.17.1