Thanks for taking a look and for the confirmation!

I agree that 1024+PATH_MAX would be a reasonable value here, good call.

I had similar thought about making the extra stack usage conditional,
but would rather keep it simple and clear-- as weighed against my possibly
wrong "expectation" that the difference won't be significant for folks.
I don't feel strongly about it and of course defer to your judgement :).

Patch making the discussed change is attached.

~Will


On Fri, Sep 15, 2017 at 9:17 AM, Rich Felker <dalias@libc.org> wrote:
> On Thu, Sep 14, 2017 at 03:39:35PM -0500, Will Dietz wrote:
>> Hi,
>>
>> I believe there is a bug in posix_spawn/execvpe, please take a look and confirm
>> or kindly let me know if I'm mistaken and accept my apologies :).
>>
>> It looks like __posix_spawnx calls clone() with a 1024-byte stack buffer
>> (allocated from its own stack), which is insufficient to handle stack
>> allocations performed
>> in execvpe which are something around a few bytes more than NAME_MAX+PATH_MAX.
>>
>> This path is taken when using posix_spawnp, and the problem exists on
>> 1.1.16 and latest git.
>>
>> For what it's worth I tracked this down from a crash in 'bison' when
>> invoking m4,
>> but I've had success reproducing it with the following demo program
>> and driver script:
>>
>> -------------------------------------------
>> #include <spawn.h>
>> #include <stdio.h>
>> #include <stdlib.h>
>> #include <sys/types.h>
>> #include <sys/wait.h>
>>
>> extern char **environ;
>>
>> int main() {
>>
>>   pid_t p;
>>   char *argv[] = {"sh", "-c", "echo Hello", NULL};
>>   int s, status;
>>   s = posix_spawnp(&p, "sh", NULL, NULL, argv, environ);
>>   if (s) {
>>     perror("posix_spawn");
>>     exit(1);
>>   }
>>
>>   s = waitpid(p, &status, 0);
>>
>>   printf("pid: %d, s: %d, status: %d\n", p, s, status);
>>
>>   return 0;
>> }
>> --------------
>>
>> And little shell script to create a suitably large PATH (mostly to
>> demonstrate what I mean, not for unmodified use):
>> ---------------
>> #!/bin/sh
>>
>> SLASH_100_As="/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
>> SUFFIX="/123456789012345678901234567" #1234567890" #1234567890"
>>
>> VAR="/bin:$SUFFIX"
>> for x in `seq 10`; do
>>   VAR="${SLASH_100_As}:$VAR"
>> done
>>
>> echo $VAR
>> echo $VAR|wc -c
>>
>> # Works fine with normal PATH
>> ~/cur/musl-spawn/test
>> ~/cur/musl-spawn/test
>>
>> # Crashes when PATH is ~1050 characters
>> PATH=$VAR \
>> ~/cur/musl-spawn/test
>> --------------
>>
>> Where "~/cur/musl-spawn/test" is the test program compiled against musl.
>>
>> I cannot speak regarding any security implications, but since this may
>> grant some measure of stack-scribbling-powers it seems to warrant
>> being given brief attention in this context.
>>
>> An easy fix is to bump the size of the 'char stack[1024]' in
>> src/process/posix_spawn.c to a suitable value-- 8096 is overkill but
>> does the trick, for example.
>>
>> Please let me know if I'm missing something or if details are not clear.
>
> It's very clear, and this seems pretty serious. 1024+PATH_MAX would
> probably be a safe limit. If we care about minimal stack usage when
> plain posix_spawn (not spawnp) is called, it could be something like
> "exec==execve ? 1024 : 1024+PATH_MAX", perhaps.
>
> Rich