[LTP] [PATCH v3 3/3] syscalls/pipe2_03: Add new test for pipe2 O_DIRECT flag

[Yang Xu]

Hi Li


> Xu,
> 
> On Tue, Apr 21, 2020 at 6:24 PM Yang Xu <xuyang2018.jy@cn.fujitsu.com 
> <mailto:xuyang2018.jy@cn.fujitsu.com>> wrote:
> 
>     ...
>      >
>      >     +       if (pid == 0) {
>      >     +               (*test_func[n])();
>      >
>      >     +               tst_reap_children();
>      >     +               SAFE_CLOSE(fds[0]);
>      >     +               SAFE_CLOSE(fds[1]);
>      >
>      >
>      > I tend to move the cloese(fd) to the test process to nearby pipe2().
>      > Otherwise, it causes an unclear error like:
>      >
>      > # ./pipe2_03
>      > tst_test.c:1246: INFO: Timeout per run is 0h 05m 00s
>      > pipe2_03.c:48: PASS: Each write(2) uses a separate packet
>      > pipe2_03.c:61: PASS: Each read(2) reads a separate packet
>      > pipe2_03.c:81: PASS: write(higner than PIPE_BUF) split into
>     multiple packet
>      > pipe2_03.c:108: FAIL: the excess bytes in the packet is discarded by
>      > read, expect 1 got 1%A
>     Even I use one fork and exit(0), I don't meet this problem. Or, I miss
>     something?
> 
> 
> My test env is kernel-5.7.0-rc1+ x86_64, I haven't figured out why the 
> read() get more words from the pipe, but the error has gone one I move 
> the safe_close(fd) part to the test main process.
> 
OK.
> 
>     --- a/testcases/kernel/syscalls/pipe2/pipe2_03.c
>     +++ b/testcases/kernel/syscalls/pipe2/pipe2_03.c
>     @@ -59,6 +59,7 @@ static void check_peer_rw(void)
>                                               "Each read(2) doesn't read a
>     separate packet, return %ld", TST_RET);
>                       }
>                       tst_res(TPASS, "Each read(2) reads a separate
>     packet");
>     +                _exit(0);
>               }
>        }
> 
>     @@ -79,6 +80,7 @@ static void check_split(void)
>                                               "write(higner than PIPE_BUF)
>     split into multiple packet, return %ld", TST_RET);
>                       }
>                       tst_res(TPASS, "write(higner than PIPE_BUF) split
>     into
>     multiple packet");
>     +                _exit(0);
>               }
>        }
> 
>     @@ -107,21 +109,17 @@ static void check_discard(void)
>                                       tst_res(TFAIL,
>                                               "the excess bytes in the
>     packet
>     is discarded by read, expect 1 got %s", tmp_secondbuf);
>                       }
>     +                _exit(0);
>               }
>        }
> 
>        static void verify_pipe2(unsigned int n)
>        {
>     -       int pid;
>     -
>     -       pid = SAFE_FORK();
>     -       if (pid == 0) {
>     -               (*test_func[n])();
>     -               tst_reap_children();
>     -               SAFE_CLOSE(fds[0]);
>     -               SAFE_CLOSE(fds[1]);
>     -       }
>     +  (*test_func[n])();
>     +       SAFE_CLOSE(fds[0]);
>     +       SAFE_CLOSE(fds[1]);
>               tst_reap_children();
> 
> 
> In my opinion, it is not a good habit to wait for children's status 
> after the test function, we'd better do that at the behind of fork() in 
> the test main process. I'd only keep the  (*test_func[n])(); in this 
> verify_pipe2() but no more else.
Yes.
> -- 
> Regards,
> Li Wang