linux-kernel - Re: [path][rfc] add PR

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <BANLkTimbB4rhbdTy07DuneSf=SydGYvbBA@mail.gmail.com>
Date:	Fri, 8 Apr 2011 16:52:05 -0400
From:	Bryan Donlan <bdonlan@...il.com>
To:	Stas Sergeev <stsp@...et.ru>
Cc:	Linux kernel <linux-kernel@...r.kernel.org>,
	Oleg Nesterov <oleg@...hat.com>
Subject: Re: [path][rfc] add PR_DETACH prctl command

2011/4/8 Stas Sergeev <stsp@...et.ru>:
> 08.04.2011 22:13, Bryan Donlan wrote:
>>
>> I can't comment on the patch itself, but, if your application knows it
>> might have to daemonize after spinning up threads, why not simply
>> fork() immediately on startup, and have the parent simply wait forever
>> for either the child to die or for a daemonize signal from the child?
>> If done early enough it shouldn't tie up too much memory, and it
>> wouldn't require any of these invasive kernel changes. As an added
>> bonus, it's portable to all unixen :)
>
> Yes, thats almost always true. Except when you deal with
> the vendor-provided poorly-coded drivers and libs, where
> you get the drivers initialized only via the lib. And the
> initialization process must be finished before the boot-up
> can continue, but that's not the whole story: only the
> process that initialized that lib, can then work with it.
> And of course that lib creates a dozen of threads...
> OK, you've got the idea. :)

Still, you can workaround this by either:
a) Load the vendor library via dlopen()
b) Use a separate launcher executable to handle the fork-and-wait

Either of these approaches are far simpler than patching the kernel,
more portable, and much, much easier to get right. In fact, here's a
quick and dirty implementation of the latter:

#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>

void sigusr1_handler(int unused)
{
        (void)unused;

        _exit(0);
}

int main(int argc, char **argv)
{
        int waitstatus;
        pid_t child, parent = getpid();

        (void)argc;

        signal(SIGUSR1, sigusr1_handler);

        child = fork();
        if (child < 0) {
                perror("Fork failed");
                return EXIT_FAILURE;
        }
        if (child == 0) {
                char envbuf[20];
                sprintf(envbuf, "%d", (int)parent);
                setenv("LAUNCHER_PID", envbuf, 1);

                execvp(argv[1], &argv[1]);
                perror("Launching child failed");
                _exit(EXIT_FAILURE);
        }

        wait(&waitstatus);

        if (WIFEXITED(waitstatus))
                _exit(WEXITSTATUS(waitstatus));

        _exit(256);
}

Just kill(atoi(getenv("LAUNCHER_PID")), SIGUSR1) to detach. Much
easier than doing some very racy things in the kernel, no? It's
certainly more obvious that this ought to be correct in the face of
races with its parent :)
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/