[PATCH] posix: Add posix_spawn extension to setup resource limits (BZ 31049)

[Adhemerval Zanella Netto]

On 06/06/25 03:44, Florian Weimer wrote:
> * Adhemerval Zanella:
> 
>> This patch adds two new functions:
>>
>>   int posix_spawnattr_setrlimit_np (posix_spawnattr_t *restrict __attr,
>>  				    int resource,
>>                                     const struct rlimit *rlim);
>>
>>   int posix_spawnattr_getrlimit_np (posix_spawnattr_t *__restrict __attr,
>>                                     int resource,
>>                                     struct rlimit *rlim)
>>
>> Along with a new POSIX_SPAWN_SETRLIMIT flag.  It allows to setup the
>> new process resource limits, similar to call setlimit/prlimit before
>> execve.  It work for both posix_spawn and pidfd_spawn.
> 
> Why is the separate flag is needed?

Mainly because it was already added by a libc implementation [1] and it
allows to use the same attribute to spawn different process with and
without limits.

> 
>> The resources limits are applied after the effective user and group IDs
>> change (POSIX_SPAWN_RESETIDS) and before the file actions.  Only LFS
>> supported is provided (non-LFS prototypes are not exported or
>> implemented).
> 
> It would be more general to treat this as a file action, so that the
> limits can be applied after opening files.
> 

I think it would really matter if the RLIMIT_NOFILE is less than the
number of file descriptors set by the file actions.  And in this case,
setting the limit *after* won't trigger a potential failure for bogus
values, and I am not sure which is better (allow all file operation
or enforce RLIMIT_NOFILE is set, I am more inclined for the latter).

Also, resources are *not* file operation so I really think we should
not tie them together. And make them a file operation would make it
even harder to get consensus on a future standardization. 

> Thanks,
> Florian
> 


[1] https://justine.lol/cosmopolitan/documentation.html#posix_spawnattr_setrlimit_np