Polling(2) /proc/mounts is broken?

Polling(2) /proc/mounts is broken?

[Tvrtko Ursulin]

[-- Attachment #1: Type: text/plain, Size: 815 bytes --]


Hi all,

So since we cannot get fanotify global mode we were told to poll(2) on
/proc/mounts and set up new mount marks dynamically.

Is the attached C program roughly how it should work?

Because on 2.6.37-rc5 it busy loops for me (POLLIN always reported), even
though if I remove the seek it gets EOF immediately after POLLIN.

Under 2.6.32 (Ubuntu flavour) I see the same behaviour.

Going further back to 2.6.24 (again Ubuntu flavour) and there it works, but it
set POLLERR instead of POLLIN in revents.

So I guess it regressed somewhere in between. I'll look to provide a fix
unless someone more familiar with this area beats me to it.

Tvrtko

Sophos Limited, The Pentagon, Abingdon Science Park, Abingdon, OX14 3YP, United Kingdom.
Company Reg No 2096520. VAT Reg No GB 991 2418 08.

[-- Attachment #2: pollmounts.c --]
[-- Type: text/x-csrc, Size: 759 bytes --]

#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>
#include <poll.h>
#include <unistd.h>
#include <stdio.h>

int main()
{
    int fd;
    struct pollfd ev;
    int ret;
    ssize_t bytes;
    char buf[4096];
    long unsigned int cnt = 0;


    fd = open("/proc/mounts", O_RDONLY);

    do {
        ev.events = POLLIN;
        ev.fd = fd;
        ev.revents = 0;

        ret = poll(&ev, 1, -1);
        lseek(fd, 0, SEEK_SET);
        if (ret == 1 && ev.revents == POLLIN) {
            printf("####  %lu ########################################\n", cnt);
            while ((bytes = read(fd, buf, sizeof(buf))) > 0)
                write(1, buf, bytes);
            cnt++;
        }
    } while (ret >= 0);

    return 0;
}

Re: Polling(2) /proc/mounts is broken?

[Tvrtko Ursulin]

On Friday 10 Dec 2010 15:32:48 Tvrtko Ursulin wrote:
> Hi all,
>
> So since we cannot get fanotify global mode we were told to poll(2) on
> /proc/mounts and set up new mount marks dynamically.
>
> Is the attached C program roughly how it should work?
>
> Because on 2.6.37-rc5 it busy loops for me (POLLIN always reported), even
> though if I remove the seek it gets EOF immediately after POLLIN.
>
> Under 2.6.32 (Ubuntu flavour) I see the same behaviour.
>
> Going further back to 2.6.24 (again Ubuntu flavour) and there it works, but
> it set POLLERR instead of POLLIN in revents.
>
> So I guess it regressed somewhere in between. I'll look to provide a fix
> unless someone more familiar with this area beats me to it.

It hasn't regressed but was fixed in 2.6.30. :)

http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=31b07093c44a7a442394d44423e21d783f5523b8

So it is incorrect to poll for POLLIN. In case of /proc/mounts POLLPRI should
be used and will be returned together with POLLERR.

Tvrtko

Sophos Limited, The Pentagon, Abingdon Science Park, Abingdon, OX14 3YP, United Kingdom.
Company Reg No 2096520. VAT Reg No GB 991 2418 08.