Hi, On Fri, Jan 15, 2016 at 05:06:13PM +0800, Fei, Jie/费 杰 wrote: > Hi! > > I'm still not sure what kind of tests are needed so I continued writing > patches. > Could you give me some advice on these patches?
First of all, there has to be clear what is being tested.
Taking your clone-f.test as an example, are you testing just the fact
of following both threads after clone syscall, or maybe something beyond
that?
> On 01/13/2016 04:43 PM, Fei Jie wrote:
> > tests: add clone-f.test.
> >
> > Check how strace -f follows clone syscall.
> >
> > * tests/clone-f.c: New file.
> > * tests/clone-f.test: New test.
> > * tests/Makefile.am: (check_PROGRAMS): Add clone-f.
> > (TESTS): Add clone-f.test.
> > * tests/.gitignore: Add clone-f.
> > ---
> > tests/.gitignore | 1 +
> > tests/Makefile.am | 2 ++
> > tests/clone-f.c | 44 ++++++++++++++++++++++++++++++++++++++++++++
> > tests/clone-f.test | 11 +++++++++++
> > 4 files changed, 58 insertions(+)
> > create mode 100644 tests/clone-f.c
> > create mode 100755 tests/clone-f.test
> >
> > diff --git a/tests/.gitignore b/tests/.gitignore
> > index cfe1e9f..7aa2449 100644
> > --- a/tests/.gitignore
> > +++ b/tests/.gitignore
> > @@ -12,6 +12,7 @@ bpf
> > caps
> > clock_nanosleep
> > clock_xettime
> > +clone-f
> > epoll_create1
> > eventfd
> > execve
> > diff --git a/tests/Makefile.am b/tests/Makefile.am
> > index 33f76cb..80292c4 100644
> > --- a/tests/Makefile.am
> > +++ b/tests/Makefile.am
> > @@ -58,6 +58,7 @@ check_PROGRAMS = \
> > caps \
> > clock_nanosleep \
> > clock_xettime \
> > + clone-f \
> > epoll_create1 \
> > eventfd \
> > execve \
> > @@ -194,6 +195,7 @@ TESTS = \
> > caps.test \
> > clock_nanosleep.test \
> > clock_xettime.test \
> > + clone-f.test \
> > dumpio.test \
> > epoll_create1.test \
> > eventfd.test \
> > diff --git a/tests/clone-f.c b/tests/clone-f.c
> > new file mode 100644
> > index 0000000..df97b54
> > --- /dev/null
> > +++ b/tests/clone-f.c
> > @@ -0,0 +1,44 @@
> > +#include "tests.h"
> > +#include <stdio.h>
> > +#include <string.h>
> > +#include <unistd.h>
> > +#include <stdlib.h>
> > +#include <sched.h>
> > +#include <signal.h>
> > +
> > +#define CHILD_STACK_SIZE 16384
> > +
> > +static int
> > +logit(const char *const str)
> > +{
> > + return pwrite(-1, str, strlen(str), 0) >= 0;
My original idea of using pwrite for text marks in the log haven't passed
the test of real world: I stumbled upon a system where pwrite is not
implemented using pwrite64 syscall. I had to replace pwrite with chdir
which is portable (see commit v4.11-125-g6833d61).
> > +}
> > +
> > +void
> > +child(void)
> > +{
> > + logit("child");
> > +}
> > +
> > +int main()
> > +{
> > + logit("parent");
> > +
> > + void *child_stack;
> > + if ((child_stack = (void *) malloc(CHILD_STACK_SIZE)) == NULL) {
> > + printf("cannot allocate stack for child!\n");
> > + }
> > +
> > + pid_t child_pid = clone(&child, child_stack + CHILD_STACK_SIZE,
> > CLONE_VM, NULL);
On some architectures supported by strace, stack grows upwards.
clone behavior noticeably depends on the clone flags. Is there any
particular reason to use these flags in the test?
> > + if (child_pid < 0) {
> > + perror_msg_and_fail("clone");
> > + }
> > +
> > + free(child_stack);
> > +
> > + pid_t pid = getpid();
> > + printf("%-5d pwrite64(-1, \"parent\", 6, 0) = -1 EBADF (%m)\n"
> > + "%-5d pwrite64(-1, \"child\", 5, 0) = -1 EBADF (%m)\n",
> > + pid, child_pid);
There is no guarantee of the parent being first, or visa versa.
I explicitly added synchronization into fork-f.test and vfork-f.test
to make the output predictable.
> > + return 0;
> > +}
> > diff --git a/tests/clone-f.test b/tests/clone-f.test
> > new file mode 100755
> > index 0000000..439426d
> > --- /dev/null
> > +++ b/tests/clone-f.test
> > @@ -0,0 +1,11 @@
> > +#!/bin/sh
> > +
> > +. "${srcdir=.}/init.sh"
> > +
> > +OUT="$LOG.out"
> > +run_prog > /dev/null
> > +run_strace -a32 -epwrite64 -esignal=none -f -qq $args >"$OUT"
> > +match_diff "$LOG" "$OUT"
> > +rm -f "$OUT"
> > +
> > +exit 0
--
ldv
pgpHeEQVswVXc.pgp
Description: PGP signature
------------------------------------------------------------------------------ Site24x7 APM Insight: Get Deep Visibility into Application Performance APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month Monitor end-to-end web transactions and take corrective actions now Troubleshoot faster and improve end-user experience. Signup Now! http://pubads.g.doubleclick.net/gampad/clk?id=267308311&iu=/4140
_______________________________________________ Strace-devel mailing list Strace-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/strace-devel
