[compiler-rt] r271046 - [sanitizers] introduce __sanitizer_set_report_fd so that we can re-route the sanitizer logging to another fd from inside the process

Tue May 31 14:12:47 PDT 2016

On Tue, May 31, 2016 at 1:49 PM, Mehdi Amini <mehdi.amini at apple.com> wrote:

>
> On May 31, 2016, at 1:43 PM, Kostya Serebryany <kcc at google.com> wrote:
>
>
>
> On Tue, May 31, 2016 at 1:38 PM, Mehdi Amini <mehdi.amini at apple.com>
> wrote:
>
>>
>> On May 31, 2016, at 1:22 PM, Kostya Serebryany <kcc at google.com> wrote:
>>
>>
>>
>> On Tue, May 31, 2016 at 1:16 PM, Mehdi Amini <mehdi.amini at apple.com>
>> wrote:
>>
>>> Committed in r271323.
>>>
>>> Thanks!
>>
>>> I wonder what is the rational for the discrepancy between Darwin and
>>> Linux?
>>>
>>
>> Don't get me started :)
>>
>>
>> Sorry :)
>>
>> http://reviews.llvm.org/D7203
>> https://groups.google.com/forum/#!topic/address-sanitizer/FGYgD_P_884
>>
>>
>> I see some rational to abort, I don't see some strong rational to *not*
>> abort on every platform.
>>
>
> Abort is slow
>
>
> An ASAN failure is not supposed to be on your critical path, right?
>
No :)
One of the modes of operation is to run a binary for a very short period of
time (5-10 seconds)
and kill it if it did not crash with asan report.
abort() may break this workflow as abort() will take several seconds to
complete on a large binary

Another problem, is that we want to run asan's own tests in the default
mode and we can't cheaply run them with abort.
This got broken on Mac since on Mac we now run the tests in non-default
mode.

> (Disabling it for testing using a build setting/env-var seems appropriate
> if it is the only concern...)
>
> and verbose
>
>
> Being verbose on crashes by default seems like a good thing to me?
>

Not too verbose.
Asan provides all the details needed to investigate memory bugs, any extra
info will distract.

>
> , it often involves lots of extra stuff that you don't need when you see
> an asan report.
>
>
> Is core dump the concern? You mentioned in the thread that is is possible
> to tweak this when ASAN is enabled though.
>

Core is disabled with asan by default, that's not a problem.

BTW: choosing the default is hard.

--kcc

>
> --
> Mehdi
>
>
>
> --kcc
>
>
>>
>>
>> --
>> Mehdi
>>
>>
>>
>>
>> --kcc
>>
>>
>>>
>>> --
>>> Mehdi
>>>
>>> On May 31, 2016, at 1:11 PM, Kostya Serebryany <kcc at google.com> wrote:
>>>
>>> Ah, I think this is caused by different defaults.
>>> On linux by default asan dies with _exit and on Mac it dies with abort()
>>> This difference was introduced by Kuba after a long resistance from my
>>> side :)
>>> The fix should be pretty simple: add abort_on_error=0 to %tool_options
>>> in the test.
>>> Could you please check if this helps on Mac?
>>> If so, feel free to submit.
>>>
>>> Thanks!
>>> --kcc
>>>
>>>
>>> On Tue, May 31, 2016 at 1:01 PM, Mehdi Amini <mehdi.amini at apple.com>
>>> wrote:
>>>
>>>> Hi Kostya,
>>>>
>>>> We see a bot failure with the test case you added below:
>>>> http://lab.llvm.org:8080/green/job/clang-stage1-cmake-RA-globalisel_check/2068/console
>>>>
>>>> After investigating, it seems that they trigger an ASAN failure, which
>>>> is returning a negative error code. The `not` command tool interprets it as
>>>> a crash, and you need to pass the --crash option to it.
>>>>
>>>> Don't you observe this on your setup? I wonder why they are passing?
>>>>
>>>> --
>>>> Mehdi
>>>>
>>>>
>>>>
>>>>
>>>> > On May 27, 2016, at 2:23 PM, Kostya Serebryany via llvm-commits <
>>>> llvm-commits at lists.llvm.org> wrote:
>>>> >
>>>> > Author: kcc
>>>> > Date: Fri May 27 16:23:05 2016
>>>> > New Revision: 271046
>>>> >
>>>> > URL: http://llvm.org/viewvc/llvm-project?rev=271046&view=rev
>>>> > Log:
>>>> > [sanitizers] introduce __sanitizer_set_report_fd so that we can
>>>> re-route the sanitizer logging to another fd from inside the process
>>>> >
>>>> > Added:
>>>> >
>>>> compiler-rt/trunk/test/sanitizer_common/TestCases/Posix/sanitizer_set_report_fd_test.cc
>>>> > Modified:
>>>> >    compiler-rt/trunk/include/sanitizer/common_interface_defs.h
>>>> >    compiler-rt/trunk/lib/asan/asan_posix.cc
>>>> >    compiler-rt/trunk/lib/sanitizer_common/sanitizer_common.cc
>>>> >
>>>> > Modified: compiler-rt/trunk/include/sanitizer/common_interface_defs.h
>>>> > URL:
>>>> http://llvm.org/viewvc/llvm-project/compiler-rt/trunk/include/sanitizer/common_interface_defs.h?rev=271046&r1=271045&r2=271046&view=diff
>>>> >
>>>> ==============================================================================
>>>> > --- compiler-rt/trunk/include/sanitizer/common_interface_defs.h
>>>> (original)
>>>> > +++ compiler-rt/trunk/include/sanitizer/common_interface_defs.h Fri
>>>> May 27 16:23:05 2016
>>>> > @@ -41,6 +41,9 @@ extern "C" {
>>>> >
>>>> >   // Tell the tools to write their reports to "path.<pid>" instead of
>>>> stderr.
>>>> >   void __sanitizer_set_report_path(const char *path);
>>>> > +  // Tell the tools to write their reports to the provided file
>>>> descriptor
>>>> > +  // (casted to void *).
>>>> > +  void __sanitizer_set_report_fd(void *fd);
>>>> >
>>>> >   // Notify the tools that the sandbox is going to be turned on. The
>>>> reserved
>>>> >   // parameter will be used in the future to hold a structure with
>>>> functions
>>>> >
>>>> > Modified: compiler-rt/trunk/lib/asan/asan_posix.cc
>>>> > URL:
>>>> http://llvm.org/viewvc/llvm-project/compiler-rt/trunk/lib/asan/asan_posix.cc?rev=271046&r1=271045&r2=271046&view=diff
>>>> >
>>>> ==============================================================================
>>>> > --- compiler-rt/trunk/lib/asan/asan_posix.cc (original)
>>>> > +++ compiler-rt/trunk/lib/asan/asan_posix.cc Fri May 27 16:23:05 2016
>>>> > @@ -36,8 +36,9 @@ namespace __asan {
>>>> > void AsanOnDeadlySignal(int signo, void *siginfo, void *context) {
>>>> >   ScopedDeadlySignal signal_scope(GetCurrentThread());
>>>> >   int code = (int)((siginfo_t*)siginfo)->si_code;
>>>> > -  // Write the first message using the bullet-proof write.
>>>> > -  if (18 != internal_write(2, "ASAN:DEADLYSIGNAL\n", 18)) Die();
>>>> > +  // Write the first message using fd=2, just in case.
>>>> > +  // It may actually fail to write in case stderr is closed.
>>>> > +  internal_write(2, "ASAN:DEADLYSIGNAL\n", 18);
>>>> >   SignalContext sig = SignalContext::Create(siginfo, context);
>>>> >
>>>> >   // Access at a reasonable offset above SP, or slightly below it (to
>>>> account
>>>> >
>>>> > Modified: compiler-rt/trunk/lib/sanitizer_common/sanitizer_common.cc
>>>> > URL:
>>>> http://llvm.org/viewvc/llvm-project/compiler-rt/trunk/lib/sanitizer_common/sanitizer_common.cc?rev=271046&r1=271045&r2=271046&view=diff
>>>> >
>>>> ==============================================================================
>>>> > --- compiler-rt/trunk/lib/sanitizer_common/sanitizer_common.cc
>>>> (original)
>>>> > +++ compiler-rt/trunk/lib/sanitizer_common/sanitizer_common.cc Fri
>>>> May 27 16:23:05 2016
>>>> > @@ -496,6 +496,11 @@ void __sanitizer_set_report_path(const c
>>>> >   report_file.SetReportPath(path);
>>>> > }
>>>> >
>>>> > +void __sanitizer_set_report_fd(void *fd) {
>>>> > +  report_file.fd = reinterpret_cast<uptr>(fd);
>>>> > +  report_file.fd_pid = internal_getpid();
>>>> > +}
>>>> > +
>>>> > void __sanitizer_report_error_summary(const char *error_summary) {
>>>> >   Printf("%s\n", error_summary);
>>>> > }
>>>> >
>>>> > Added:
>>>> compiler-rt/trunk/test/sanitizer_common/TestCases/Posix/sanitizer_set_report_fd_test.cc
>>>> > URL:
>>>> http://llvm.org/viewvc/llvm-project/compiler-rt/trunk/test/sanitizer_common/TestCases/Posix/sanitizer_set_report_fd_test.cc?rev=271046&view=auto
>>>> >
>>>> ==============================================================================
>>>> > ---
>>>> compiler-rt/trunk/test/sanitizer_common/TestCases/Posix/sanitizer_set_report_fd_test.cc
>>>> (added)
>>>> > +++
>>>> compiler-rt/trunk/test/sanitizer_common/TestCases/Posix/sanitizer_set_report_fd_test.cc
>>>> Fri May 27 16:23:05 2016
>>>> > @@ -0,0 +1,37 @@
>>>> > +// Test __sanitizer_set_report_fd:
>>>> > +// RUN: %clangxx -O2 %s -o %t
>>>> > +// RUN: not %run %t 2>&1   | FileCheck %s
>>>> > +// RUN: not %run %t stdout | FileCheck %s
>>>> > +// RUN: not %run %t %t-out && FileCheck < %t-out %s
>>>> > +
>>>> > +// REQUIRES: stable-runtime
>>>> > +// FIXME: implement SEGV handler in other sanitizers, not just asan.
>>>> > +// XFAIL: msan
>>>> > +// XFAIL: lsan
>>>> > +// XFAIL: tsan
>>>> > +
>>>> > +#include <sanitizer/common_interface_defs.h>
>>>> > +#include <stdio.h>
>>>> > +#include <string.h>
>>>> > +#include <stdlib.h>
>>>> > +#include <sys/types.h>
>>>> > +#include <sys/stat.h>
>>>> > +#include <fcntl.h>
>>>> > +#include <assert.h>
>>>> > +
>>>> > +volatile int *null = 0;
>>>> > +
>>>> > +int main(int argc, char **argv) {
>>>> > +  if (argc == 2) {
>>>> > +    if (!strcmp(argv[1], "stdout")) {
>>>> > +      __sanitizer_set_report_fd(reinterpret_cast<void*>(1));
>>>> > +    } else {
>>>> > +      int fd = open(argv[1], O_CREAT | O_WRONLY | O_TRUNC, S_IRWXU);
>>>> > +      assert(fd > 0);
>>>> > +      __sanitizer_set_report_fd(reinterpret_cast<void*>(fd));
>>>> > +    }
>>>> > +  }
>>>> > +  *null = 0;
>>>> > +}
>>>> > +
>>>> > +// CHECK: ERROR: {{.*}} SEGV on unknown address
>>>> >
>>>> >
>>>> > _______________________________________________
>>>> > llvm-commits mailing list
>>>> > llvm-commits at lists.llvm.org
>>>> > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>>>
>>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160531/52f50d3a/attachment.html>