userfaultfd: non-cooperative: fix fork use after free
authorAndrea Arcangeli <aarcange@redhat.com>
Tue, 3 Oct 2017 23:15:38 +0000 (16:15 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 4 Oct 2017 00:54:25 +0000 (17:54 -0700)
When reading the event from the uffd, we put it on a temporary
fork_event list to detect if we can still access it after releasing and
retaking the event_wqh.lock.

If fork aborts and removes the event from the fork_event all is fine as
long as we're still in the userfault read context and fork_event head is
still alive.

We've to put the event allocated in the fork kernel stack, back from
fork_event list-head to the event_wqh head, before returning from
userfaultfd_ctx_read, because the fork_event head lifetime is limited to
the userfaultfd_ctx_read stack lifetime.

Forgetting to move the event back to its event_wqh place then results in
__remove_wait_queue(&ctx->event_wqh, &ewq->wq); in
userfaultfd_event_wait_completion to remove it from a head that has been
already freed from the reader stack.

This could only happen if resolve_userfault_fork failed (for example if
there are no file descriptors available to allocate the fork uffd).  If
it succeeded it was put back correctly.

Furthermore, after find_userfault_evt receives a fork event, the forked
userfault context in fork_nctx and uwq->msg.arg.reserved.reserved1 can
be released by the fork thread as soon as the event_wqh.lock is
released.  Taking a reference on the fork_nctx before dropping the lock
prevents an use after free in resolve_userfault_fork().

If the fork side aborted and it already released everything, we still
try to succeed resolve_userfault_fork(), if possible.

Fixes: 893e26e61d04eac9 ("userfaultfd: non-cooperative: Add fork() event")
Link: http://lkml.kernel.org/r/20170920180413.26713-1-aarcange@redhat.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reported-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Mark Rutland <mark.rutland@arm.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/userfaultfd.c

index ef4b48d1ea4270f59f46e29ae898a48cdb0c8aa6..1c713fd5b3e67966c3d998979d2c30eb8e14ba07 100644 (file)
@@ -588,6 +588,12 @@ static void userfaultfd_event_wait_completion(struct userfaultfd_ctx *ctx,
                        break;
                if (ACCESS_ONCE(ctx->released) ||
                    fatal_signal_pending(current)) {
+                       /*
+                        * &ewq->wq may be queued in fork_event, but
+                        * __remove_wait_queue ignores the head
+                        * parameter. It would be a problem if it
+                        * didn't.
+                        */
                        __remove_wait_queue(&ctx->event_wqh, &ewq->wq);
                        if (ewq->msg.event == UFFD_EVENT_FORK) {
                                struct userfaultfd_ctx *new;
@@ -1061,6 +1067,12 @@ static ssize_t userfaultfd_ctx_read(struct userfaultfd_ctx *ctx, int no_wait,
                                        (unsigned long)
                                        uwq->msg.arg.reserved.reserved1;
                                list_move(&uwq->wq.entry, &fork_event);
+                               /*
+                                * fork_nctx can be freed as soon as
+                                * we drop the lock, unless we take a
+                                * reference on it.
+                                */
+                               userfaultfd_ctx_get(fork_nctx);
                                spin_unlock(&ctx->event_wqh.lock);
                                ret = 0;
                                break;
@@ -1091,19 +1103,53 @@ static ssize_t userfaultfd_ctx_read(struct userfaultfd_ctx *ctx, int no_wait,
 
        if (!ret && msg->event == UFFD_EVENT_FORK) {
                ret = resolve_userfault_fork(ctx, fork_nctx, msg);
+               spin_lock(&ctx->event_wqh.lock);
+               if (!list_empty(&fork_event)) {
+                       /*
+                        * The fork thread didn't abort, so we can
+                        * drop the temporary refcount.
+                        */
+                       userfaultfd_ctx_put(fork_nctx);
+
+                       uwq = list_first_entry(&fork_event,
+                                              typeof(*uwq),
+                                              wq.entry);
+                       /*
+                        * If fork_event list wasn't empty and in turn
+                        * the event wasn't already released by fork
+                        * (the event is allocated on fork kernel
+                        * stack), put the event back to its place in
+                        * the event_wq. fork_event head will be freed
+                        * as soon as we return so the event cannot
+                        * stay queued there no matter the current
+                        * "ret" value.
+                        */
+                       list_del(&uwq->wq.entry);
+                       __add_wait_queue(&ctx->event_wqh, &uwq->wq);
 
-               if (!ret) {
-                       spin_lock(&ctx->event_wqh.lock);
-                       if (!list_empty(&fork_event)) {
-                               uwq = list_first_entry(&fork_event,
-                                                      typeof(*uwq),
-                                                      wq.entry);
-                               list_del(&uwq->wq.entry);
-                               __add_wait_queue(&ctx->event_wqh, &uwq->wq);
+                       /*
+                        * Leave the event in the waitqueue and report
+                        * error to userland if we failed to resolve
+                        * the userfault fork.
+                        */
+                       if (likely(!ret))
                                userfaultfd_event_complete(ctx, uwq);
-                       }
-                       spin_unlock(&ctx->event_wqh.lock);
+               } else {
+                       /*
+                        * Here the fork thread aborted and the
+                        * refcount from the fork thread on fork_nctx
+                        * has already been released. We still hold
+                        * the reference we took before releasing the
+                        * lock above. If resolve_userfault_fork
+                        * failed we've to drop it because the
+                        * fork_nctx has to be freed in such case. If
+                        * it succeeded we'll hold it because the new
+                        * uffd references it.
+                        */
+                       if (ret)
+                               userfaultfd_ctx_put(fork_nctx);
                }
+               spin_unlock(&ctx->event_wqh.lock);
        }
 
        return ret;