Re: [PATCH v5 2/4] man/man5/tunables.conf: Document system-wide tunables config

[Alejandro Colomar <alx@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 3525 bytes --]

> Date: 2026-08-07 02:11:04+0200
> From: Alejandro Colomar <alx@kernel.org>
>
> > Date: 2026-08-07 02:07:00+0200
> > From: Alejandro Colomar <alx@kernel.org>
> >
> > Hi DJ,
> > 
> > > Date: 2026-08-06 18:15:15-0400
> > > From: DJ Delorie <dj@redhat.com>
> > >
> > > Alejandro Colomar <alx@kernel.org> writes:
> > > > You forgot to sign.
> > > 
> > > I will never remember that... :-P
> > > 
> > > >> +The resulting data is read when a new process is created by
> > > >> +.IR ld.so .
> > > >
> > > > Sorry for noticing this in v4; I forgot about it.  I see this unresolved
> > > > issue from earlier versions remains.
> > > 
> > > A program is a file on disk.  A process is a running memory image.
> > 
> > Yup.  fork(3) creates a process (produces a new PID).  execve(2)
> > replaces the image of the current process with a program read from disk
> > (or in short, executes a program in the current process).
> > 
> > > 
> > > Neither fork() nor exec() cause the tunables to be loaded, the process
> > > itself has to do that.  For dynamic ELF programs/processes built with
> > > glibc, ld.so does it before it passes control to the ELF image.  There
> > > are other types of programs/processes that do not use ld.so and thus do
> > > not read the tunables cache (which is called "ld.so.cache" for a reason ;-)
> > 
> > Hmmmmm, let's say you call execve(2) several times without any fork(2)s.
> > Assuming a that the program paths that you pass to execve(2) are all
> > dynamic ELF programs built with glibc, I guess for every time you call
> > execve(2), ld.so(8) will be run, and the tunables stored in ld.so.cache
> > will be used.  Is this correct?
> > 
> > 	#include <unistd.h>
> > 	int
> > 	main(int, char *argv[])
> > 	{
> > 		sleep(1);
> > 		execve("proc/self/exe", argv, NULL);

Dumb of me; missing leading '/'.  :)


Cheers,
Alex

> > 	}
> 
> Oh, I thought this would work as a recursive infinite-loop program.  It
> doesn't seem to work.  I guess there's some race condition?  :D
> 
> 	alx@devuan:~$ cd tmp/
> 	alx@devuan:~/tmp$ cat ex.c 
> 	#include <unistd.h>
> 	int
> 	main(int, char *argv[])
> 	{
> 		sleep(1);
> 		execve("proc/self/exe", argv, NULL);
> 	}
> 
> 	alx@devuan:~/tmp$ gcc -Wall -Wextra ex.c 
> 	alx@devuan:~/tmp$ time ./a.out 
> 
> 	real	0m1.003s
> 	user	0m0.003s
> 	sys	0m0.000s
> 
> 
> Cheers,
> Alex
> 
> > 
> > Let's also say you fork(2) several times, without any execve(2) calls.
> > I suppose that won't trigger any of this, right?
> > 
> > 	#include <unistd.h>
> > 	int
> > 	main(void)
> > 	{
> > 		for (int i = 0; i < 10; i++) {
> > 			sleep(1);
> > 			fork();
> > 		}
> > 	}
> > 
> > From my understanding, only the first program will trigger this more
> > than once, right?
> > 
> > If so, I'd use the same wording that execve(2) uses:
> > 
> > 	execve() executes the program referred to by path.
> > 
> > That is, execve(2) executes programs, and ld.so.cache is read when
> > a program is executed (in the current process).
> > 
> > > 
> > > I don't know how to say this clearly with fewer words...
> > > 
> > > >> +The syntax allows lines to start with the keyword
> > > >> +.I include
> > > >
> > > > s/I/B/  (since it's a literal, not a variable)
> > > 
> > > Fixed.
> > > 
> > 
> > 
> > Cheers,
> > Alex
> > 
> > -- 
> > <https://www.alejandro-colomar.es>
> 
> 
> 
> -- 
> <https://www.alejandro-colomar.es>



-- 
<https://www.alejandro-colomar.es>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]