Date:   Wed, 20 Apr 2022 16:23:02 +0200
From:   Jann Horn <jannh@...gle.com>
To:     "Jason A. Donenfeld" <Jason@...c4.com>
Cc:     LKML <linux-kernel@...r.kernel.org>,
        Linux Crypto Mailing List <linux-crypto@...r.kernel.org>,
        Alexander Graf <graf@...zon.com>,
        Dominik Brodowski <linux@...inikbrodowski.net>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        "Theodore Ts'o" <tytso@....edu>,
        Colm MacCarthaigh <colmmacc@...zon.com>
Subject: Re: [PATCH] random: add fork_event sysctl for polling VM forks

On Wed, Apr 20, 2022 at 3:25 PM Jason A. Donenfeld <Jason@...c4.com> wrote:
>
> Hey again,
>
> On Wed, Apr 20, 2022 at 02:15:45AM +0200, Jason A. Donenfeld wrote:
> > Hi Jann,
> >
> > On Tue, Apr 19, 2022 at 9:45 PM Jann Horn <jannh@...gle.com> wrote:
> > > AFAIK this also means that if you make an epoll watch for
> > > /proc/sys/kernel/random/fork_event, and then call poll() *on the epoll
> > > fd* for some reason, that will probably already consume the event; and
> > > if you then try to actually receive the epoll event via epoll_wait(),
> > > it'll already be gone (because epoll tries to re-poll the "ready"
> > > files to figure out what state those files are at now). Similarly if
> > > you try to create an epoll watch for an FD that already has an event
> > > pending: Installing the watch will call the ->poll handler once,
> > > resetting the file's state, and the following epoll_wait() will call
> > > ->poll again and think the event is already gone. See the call paths
> > > to vfs_poll() in fs/eventpoll.c.
> > >
> > > Maybe we don't care about such exotic usage, and are willing to accept
> > > the UAPI inconsistency and slight epoll breakage of plumbing
> > > edge-triggered polling through APIs designed for level-triggered
> > > polling. IDK.
> >
> > Hmm, I see. The thing is, this is _already_ what's done for
> > domainname/hostname. It's how the sysctl poll handler was "designed".
> > So our options here are:
> >
> > a) Remove this quirky behavior from domainname/hostname and start
> > over. This would potentially break userspace, but maybe nobody uses
> > this? No idea, but sounds risky.
> >
> > b) Apply this commit as-is, because it's using the API as the API was
> > designed, and call it a day.
> >
> > c) Apply this commit as-is, because it's using the API as the API was
> > designed, and then later try to fix up the epoll behavior on this.
> >
> > Of these, (a) seems like a non-starter. (c) is most appealing, but it
> > sounds like it might not actually be possible?
> >
> > Jason
>
> I actually tried to verify your concern but didn't have success doing
> so.

My point is that when you run this code:

$ cat edgepoll.c
#include <time.h>
#include <stdio.h>
#include <fcntl.h>
#include <err.h>
#include <unistd.h>
#include <poll.h>
#include <sys/epoll.h>

#define SYSCHK(x) ({          \
  typeof(x) __res = (x);      \
  if (__res == (typeof(x))-1) \
    err(1, "SYSCHK(" #x ")"); \
  __res;                      \
})

int main(void) {
  int epfd = SYSCHK(epoll_create1(0));
  int hostname_fd = SYSCHK(open("/proc/sys/kernel/hostname", O_RDONLY));
  struct epoll_event event = { .events = EPOLLERR, .data = { .u32 = 1234 } };
  SYSCHK(epoll_ctl(epfd, EPOLL_CTL_ADD, hostname_fd, &event));

  while (1) {
    struct pollfd pollfds[1] = { { .fd = epfd, .events = POLLIN } };
    int poll_res = poll(pollfds, 1, -1);
    if (poll_res == -1) {
      perror("poll() error");
      continue;
    }
    if (poll_res == 0) {
      printf("poll(): no events ready (can't happen, we're using
timeout=-1)\n");
      continue;
    }
    struct epoll_event events[1];
    int epoll_res = epoll_wait(epfd, events, 1, 0);
    if (epoll_res == -1) {
      perror("epoll error");
      continue;
    }
    if (epoll_res == 0) {
      printf("spurious epoll readiness\n");
      continue;
    }
    printf("got epoll fd readiness: events=0x%x, u32=%u\n",
events[0].events, events[0].data.u32);
  }
}
$ gcc -o edgepoll edgepoll.c
$ ./edgepoll

and then change the hostname, you'll just get "spurious epoll
readiness" logged - simply calling poll() on the epoll FD resets the
state of the hostname file that is being polled, so when we then try
to receive the epoll event with epoll_wait(), the event is gone.


And the other case is this:

$ cat edgepoll2.c
#include <time.h>
#include <stdio.h>
#include <fcntl.h>
#include <err.h>
#include <unistd.h>
#include <poll.h>
#include <sys/epoll.h>

#define SYSCHK(x) ({          \
  typeof(x) __res = (x);      \
  if (__res == (typeof(x))-1) \
    err(1, "SYSCHK(" #x ")"); \
  __res;                      \
})

int main(void) {
  int epfd = SYSCHK(epoll_create1(0));
  int hostname_fd = SYSCHK(open("/proc/sys/kernel/hostname", O_RDONLY));
  printf("opened hostname fd, sleeping\n");
  sleep(10);
  printf("done sleeping\n");
  struct epoll_event event = { .events = EPOLLERR, .data = { .u32 = 1234 } };
  SYSCHK(epoll_ctl(epfd, EPOLL_CTL_ADD, hostname_fd, &event));

  struct epoll_event events[1];
  int epoll_res = SYSCHK(epoll_wait(epfd, events, 1, 0));
  if (epoll_res == 0)
    errx(1, "no epoll events ready");
  printf("got epoll fd readiness: events=0x%x, u32=%u\n",
events[0].events, events[0].data.u32);
}
$ gcc -o edgepoll2 edgepoll2.c
$ ./edgepoll2
opened hostname fd, sleeping
done sleeping
edgepoll2: no epoll events ready
$

If you change the hostname when "opened hostname fd, sleeping" is
printed, it'll still say "edgepoll2: no epoll events ready", because
the EPOLL_CTL_ADD basically consumed the event.

Powered by blists - more mailing lists