mrb_str_modify segfault

[jbreeden]

EDIT: Found a minimal reproduction

[jared:~/projects/mruby] git checkout upstream/master
[jared:~/projects/mruby] git pull upstream master
[jared:~/projects/mruby] rake clean default
[jared:~/projects/mruby] mirb
mirb - Embeddable Interactive Ruby Shell

> count = 1
 => 1
> singular = 'thing'
 => "thing"
> "#{count} #{singular}#{'s' unless count == 1}"
Segmentation fault: 11
[jared:~/projects/mruby] 

Previous description follows below

---

Getting a segfault in mrb_str_modify. Not 100% on what's causing the error yet. To capture what I can for now....

git bisect

leads me to this commit:

[jared:~/projects/mruby] git status
HEAD detached at d8fc09c
You are currently bisecting, started from branch 'cf5ca03'.
....
[jared:~/projects/mruby] git show head
commit d8fc09cf927aad33138b54bc79d286ac0636cc06
Author: Yukihiro "Matz" Matsumoto <matz@ruby-lang.org>
Date:   Wed Feb 17 11:02:29 2016 +0900

    push value for NULL AST when value is required; fix #3110

Tests & Output

Previous ruby/spec results were ok (relatively few crashes).

When moving from d8fc09^ to d8fc09 I start seeing this instead.

Most of the tests fail with the same segfault. Seems to be something in the mspec HTML formatter, since they all fail just before printing the summary for the tests to the output file.

Any tips for debugging?

Still trying to come up with a minimal reproduction...

The valgrind output shows me where in the c code the error happens. Is there a quick way to dump the ruby stack trace at that point? (edit: mrb_raise seems to generate a stack trace pretty well :)

[jbreeden]

Looks like a null pointer is being passed in as s

When I add this:

MRB_API void
mrb_str_modify(mrb_state *mrb, struct RString *s)
{
  fprintf(stderr, "mrb_str_modify\n");
  fflush(stderr);
  fprintf(stderr, "s: %p\n", s);
  fflush(stderr);
  fprintf(stderr, "RSTR_PTR %%p: %p\n", RSTR_PTR(s));
  fflush(stderr);
  fprintf(stderr, "RSTR_PTR %%s: %s\n", RSTR_PTR(s));
  fflush(stderr);

The first 3 test files all fail with the same final output:

mrb_str_modify
s: 0x0
==35178== Invalid read of size 1
==35178==    at 0x10001FE40: mrb_str_modify (string.c:664)
==35178==    by 0x10002020E: mrb_str_concat (string.c:766)
==35178==    by 0x100034039: mrb_vm_exec (vm.c:2217)
==35178==    by 0x100035026: mrb_top_run (vm.c:2448)
==35178==    by 0x100063418: load_exec (parse.y:5637)
==35178==    by 0x10000161B: main (mruby.c:226)
==35178==  Address 0x1 is not stack'd, malloc'd or (recently) free'd
==35178== 
==35178== 
==35178== Process terminating with default action of signal 11 (SIGSEGV)
==35178==  Access not within mapped region at address 0x1
==35178==    at 0x10001FE40: mrb_str_modify (string.c:664)
==35178==    by 0x10002020E: mrb_str_concat (string.c:766)
==35178==    by 0x100034039: mrb_vm_exec (vm.c:2217)
==35178==    by 0x100035026: mrb_top_run (vm.c:2448)
==35178==    by 0x100063418: load_exec (parse.y:5637)
==35178==    by 0x10000161B: main (mruby.c:226)
==35178==  If you believe this happened as a result of a stack
==35178==  overflow in your program's main thread (unlikely but
==35178==  possible), you can try to increase the size of the
==35178==  main thread stack using the --main-stacksize= flag.
==35178==  The main thread stack size used in this run was 8388608.
==35178== 
==35178== HEAP SUMMARY:
==35178==     in use at exit: 1,924,475 bytes in 22,230 blocks
==35178==   total heap usage: 147,147 allocs, 124,917 frees, 48,801,309 bytes allocated
==35178== 
==35178== LEAK SUMMARY:
==35178==    definitely lost: 4,096 bytes in 1 blocks
==35178==    indirectly lost: 0 bytes in 0 blocks
==35178==      possibly lost: 0 bytes in 0 blocks
==35178==    still reachable: 1,885,686 bytes in 21,808 blocks
==35178==         suppressed: 34,693 bytes in 421 blocks
==35178== Rerun with --leak-check=full to see details of leaked memory
==35178== 
==35178== For counts of detected and suppressed errors, rerun with: -v
==35178== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)

[jbreeden]

Awesome! Thanks!