diff mbox series

[v2] hurd: Make getrandom cache the server port

Message ID 20221202091351.74292-1-bugaevc@gmail.com
State New
Headers show
Series [v2] hurd: Make getrandom cache the server port | expand

Commit Message

Sergey Bugaev Dec. 2, 2022, 9:13 a.m. UTC
Previously, getrandom would, each time it's called, traverse the file
system to find /dev/urandom, fetch some random data from it, then throw
away that port. This is quite slow, while calls to getrandom are
genrally expected to be fast.

Additionally, this means that getrandom can not work when /dev/urandom
is unavailable, such as inside a chroot that lacks one. User programs
expect calls to getrandom to work inside a chroot if they first call
getrandom outside of the chroot.

In particular, this is known to break the OpenSSH server, and in that
case the issue is exacerbated by the API of arc4random, which prevents
it from properly reporting errors, forcing glibc to abort on failure.
This causes sshd to just die once it tries to generate a random number.

Caching the random server port, in a manner similar to how socket
server ports are cached, both improves the performance and works around
the chroot issue.

Tested on i686-gnu with the following program:

pthread_barrier_t barrier;

void *worker(void*) {
    pthread_barrier_wait(&barrier);
    uint32_t sum = 0;
    for (int i = 0; i < 10000; i++) {
        sum += arc4random();
    }
    return (void *)(uintptr_t) sum;
}

int main() {
    pthread_t threads[THREAD_COUNT];

    pthread_barrier_init(&barrier, NULL, THREAD_COUNT);

    for (int i = 0; i < THREAD_COUNT; i++) {
        pthread_create(&threads[i], NULL, worker, NULL);
    }
    for (int i = 0; i < THREAD_COUNT; i++) {
        void *retval;
        pthread_join(threads[i], &retval);
        printf("Thread %i: %lu\n", i, (unsigned long)(uintptr_t) retval);
    }

In my totally unscientific benchmark, with this patch, this completes
in about 7 seconds, whereas previously it took about 50 seconds. This
program was also used to test that getrandom () doesn't explode if the
random server dies, but instead reopens the /dev/urandom anew. I have
also verified that with this patch, OpenSSH can once again accept
connections properly.

Caveat: this new implementation does not respect the GRND_RANDOM flag
and always uses /dev/urandom to read random data. This does not seem to
be much of a problem, since there is only a single random server
implementation in the Hurd, and /dev/urandom is actually a symlink to
/dev/random:

$ showtrans /dev/*random
/dev/random: /hurd/random --seed-file /var/lib/random-seed
/dev/urandom: /hurd/symlink random

Signed-off-by: Sergey Bugaev <bugaevc@gmail.com>
---
 sysdeps/mach/hurd/getrandom.c | 101 ++++++++++++++++++++++++++++------
 1 file changed, 85 insertions(+), 16 deletions(-)

Comments

Samuel Thibault Dec. 2, 2022, 9:17 a.m. UTC | #1
Hello,

Sergey Bugaev, le ven. 02 déc. 2022 12:13:51 +0300, a ecrit:
> Caveat: this new implementation does not respect the GRND_RANDOM flag
> and always uses /dev/urandom to read random data.

It should be easy to fix that?

> +static file_t random_server, random_server_nonblock;

By adding a second series of file_t

> -  if (flags & GRND_RANDOM)
> -    random_source = "/dev/random";
> +  cached_server = (flags & GRND_NONBLOCK) ? &random_server_nonblock
> +                                          : &random_server;

and multiplexing here.

Thanks for your work on this,
Samuel
Sergey Bugaev Dec. 2, 2022, 1:18 p.m. UTC | #2
On Fri, Dec 2, 2022 at 12:17 PM Samuel Thibault <samuel.thibault@gnu.org> wrote:
>
> Hello,
>
> Sergey Bugaev, le ven. 02 déc. 2022 12:13:51 +0300, a ecrit:
> > Caveat: this new implementation does not respect the GRND_RANDOM flag
> > and always uses /dev/urandom to read random data.
>
> It should be easy to fix that?
> By adding a second series of file_t
> and multiplexing here.

Sure, if you would prefer it that way. I'll send v3 in a moment.

> Thanks for your work on this,
> Samuel

Sergey
Samuel Thibault Dec. 2, 2022, 1:55 p.m. UTC | #3
Sergey Bugaev, le ven. 02 déc. 2022 16:18:33 +0300, a ecrit:
> On Fri, Dec 2, 2022 at 12:17 PM Samuel Thibault <samuel.thibault@gnu.org> wrote:
> > Sergey Bugaev, le ven. 02 déc. 2022 12:13:51 +0300, a ecrit:
> > > Caveat: this new implementation does not respect the GRND_RANDOM flag
> > > and always uses /dev/urandom to read random data.
> >
> > It should be easy to fix that?
> > By adding a second series of file_t
> > and multiplexing here.
> 
> Sure, if you would prefer it that way.

Yes, please, better not hide an issue :)

Samuel
diff mbox series

Patch

diff --git a/sysdeps/mach/hurd/getrandom.c b/sysdeps/mach/hurd/getrandom.c
index ad2d3ba3..70a0e582 100644
--- a/sysdeps/mach/hurd/getrandom.c
+++ b/sysdeps/mach/hurd/getrandom.c
@@ -16,10 +16,12 @@ 
    License along with the GNU C Library; if not, see
    <https://www.gnu.org/licenses/>.  */
 
+#include <hurd.h>
 #include <sys/random.h>
 #include <fcntl.h>
-#include <unistd.h>
-#include <not-cancel.h>
+
+__libc_rwlock_define_initialized (static, lock);
+static file_t random_server, random_server_nonblock;
 
 extern char *__trivfs_server_name __attribute__((weak));
 
@@ -28,10 +30,14 @@  extern char *__trivfs_server_name __attribute__((weak));
 ssize_t
 __getrandom (void *buffer, size_t length, unsigned int flags)
 {
-  const char *random_source = "/dev/urandom";
-  int open_flags = O_RDONLY | O_CLOEXEC;
-  size_t amount_read;
-  int fd;
+  file_t server, *cached_server;;
+  error_t err;
+  int open_flags;
+  char *data = buffer;
+  mach_msg_type_number_t nread = length;
+
+  if (flags & ~(GRND_RANDOM | GRND_NONBLOCK))
+    return __hurd_fail (EINVAL);
 
   if (&__trivfs_server_name && __trivfs_server_name
       && __trivfs_server_name[0] == 'r'
@@ -44,19 +50,82 @@  __getrandom (void *buffer, size_t length, unsigned int flags)
     /* We are random, don't try to read ourselves!  */
     return length;
 
-  if (flags & GRND_RANDOM)
-    random_source = "/dev/random";
+  cached_server = (flags & GRND_NONBLOCK) ? &random_server_nonblock
+                                          : &random_server;
+
+again:
+  __libc_rwlock_rdlock (lock);
+  server = *cached_server;
+  if (MACH_PORT_VALID (server))
+    /* Attempt to read some random data using this port.  */
+    err = __io_read (server, &data, &nread, -1, length);
+  else
+    err = MACH_SEND_INVALID_DEST;
+  __libc_rwlock_unlock (lock);
+
+  if (err == MACH_SEND_INVALID_DEST || err == MIG_SERVER_DIED)
+    {
+      file_t oldserver = server;
+      mach_port_urefs_t urefs;
+
+      /* Slow path: the cached port didn't work, or there was no
+         cached port in the first place.  */
+
+      __libc_rwlock_wrlock (lock);
+      server = *cached_server;
+      if (server != oldserver)
+        {
+          /* Someone else must have refetched the port while we were
+             waiting for the lock. */
+          __libc_rwlock_unlock (lock);
+          goto again;
+        }
+
+      if (MACH_PORT_VALID (server))
+        {
+          /* It could be that someone else has refetched the port and
+             it got the very same name.  So check whether it is a send
+             right (and not a dead name).  */
+          err = __mach_port_get_refs (__mach_task_self (), server,
+                                      MACH_PORT_RIGHT_SEND, &urefs);
+          if (!err && urefs > 0)
+            {
+              __libc_rwlock_unlock (lock);
+              goto again;
+            }
+
+          /* Now we're sure that it's dead.  */
+          __mach_port_deallocate (__mach_task_self (), server);
+        }
+
+      open_flags = O_RDONLY | O_NOCTTY;
+      if (flags & GRND_NONBLOCK)
+        open_flags |= O_NONBLOCK;
+      server = *cached_server = __file_name_lookup ("/dev/urandom",
+                                                    open_flags, 0);
+      __libc_rwlock_unlock (lock);
+      if (!MACH_PORT_VALID (server))
+        /* No luck.  */
+        return -1;
+
+      goto again;
+    }
 
-  if (flags & GRND_NONBLOCK)
-    open_flags |= O_NONBLOCK;
+  if (err)
+    return __hurd_fail (err);
 
-  fd = __open_nocancel(random_source, open_flags);
-  if (fd == -1)
-    return -1;
+  if (data != buffer)
+    {
+      if (nread > length)
+        {
+          __vm_deallocate (__mach_task_self (), (vm_address_t) data, nread);
+          return __hurd_fail (EGRATUITOUS);
+        }
+      memcpy (buffer, data, nread);
+      __vm_deallocate (__mach_task_self (), (vm_address_t) data, nread);
+    }
 
-  amount_read = __read_nocancel(fd, buffer, length);
-  __close_nocancel_nostatus(fd);
-  return amount_read;
+  return nread;
 }
 
 libc_hidden_def (__getrandom)