@@ -395,32 +395,6 @@ vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason)
if (!(vmf->flags & FAULT_FLAG_USER) && (ctx->flags & UFFD_USER_MODE_ONLY))
goto out;
- /*
- * If it's already released don't get it. This avoids to loop
- * in __get_user_pages if userfaultfd_release waits on the
- * caller of handle_userfault to release the mmap_lock.
- */
- if (unlikely(READ_ONCE(ctx->released))) {
- /*
- * Don't return VM_FAULT_SIGBUS in this case, so a non
- * cooperative manager can close the uffd after the
- * last UFFDIO_COPY, without risking to trigger an
- * involuntary SIGBUS if the process was starting the
- * userfaultfd while the userfaultfd was still armed
- * (but after the last UFFDIO_COPY). If the uffd
- * wasn't already closed when the userfault reached
- * this point, that would normally be solved by
- * userfaultfd_must_wait returning 'false'.
- *
- * If we were to return VM_FAULT_SIGBUS here, the non
- * cooperative manager would be instead forced to
- * always call UFFDIO_UNREGISTER before it can safely
- * close the uffd.
- */
- ret = VM_FAULT_NOPAGE;
- goto out;
- }
-
/*
* Check that we can return VM_FAULT_RETRY.
*
@@ -457,6 +431,31 @@ vm_fault_t handle_userfault(struct vm_fault *vmf, unsigned long reason)
if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
goto out;
+ if (unlikely(READ_ONCE(ctx->released))) {
+ /*
+ * If a concurrent release is detected, do not return
+ * VM_FAULT_SIGBUS or VM_FAULT_NOPAGE, but instead always
+ * return VM_FAULT_RETRY with lock released proactively.
+ *
+ * If we were to return VM_FAULT_SIGBUS here, the non
+ * cooperative manager would be instead forced to
+ * always call UFFDIO_UNREGISTER before it can safely
+ * close the uffd, to avoid involuntary SIGBUS triggered.
+ *
+ * If we were to return VM_FAULT_NOPAGE, it would work for
+ * the fault path, in which the lock will be released
+ * later. However for GUP, faultin_page() does nothing
+ * special on NOPAGE, so GUP would spin retrying without
+ * releasing the mmap read lock, causing possible livelock.
+ *
+ * Here only VM_FAULT_RETRY would make sure the mmap lock
+ * be released immediately, so that the thread concurrently
+ * releasing the userfault would always make progress.
+ */
+ release_fault_lock(vmf);
+ goto out;
+ }
+
/* take the reference before dropping the mmap_lock */
userfaultfd_ctx_get(ctx);
This patch should fix a possible userfaultfd release() hang during concurrent GUP. This problem was initially reported by Dimitris Siakavaras in July 2023 [1] in a firecracker use case. Firecracker has a separate process handling page faults remotely, and when the process releases the userfaultfd it can race with a concurrent GUP from KVM trying to fault in a guest page during the secondary MMU page fault process. A similar problem was reported recently again by Jinjiang Tu in March 2025 [2], even though the race happened this time with a mlockall() operation, which does GUP in a similar fashion. In 2017, commit 656710a60e36 ("userfaultfd: non-cooperative: closing the uffd without triggering SIGBUS") was trying to fix this issue. AFAIU, that fixes well the fault paths but may not work yet for GUP. In GUP, the issue is NOPAGE will be almost treated the same as "page fault resolved" in faultin_page(), then the GUP will follow page again, seeing page missing, and it'll keep going into a live lock situation as reported. This change makes core mm return RETRY instead of NOPAGE for both the GUP and fault paths, proactively releasing the mmap read lock. This should guarantee the other release thread make progress on taking the write lock and avoid the live lock even for GUP. When at it, rearrange the comments to make sure it's uptodate. [1] https://lore.kernel.org/r/79375b71-db2e-3e66-346b-254c90d915e2@cslab.ece.ntua.gr [2] https://lore.kernel.org/r/20250307072133.3522652-1-tujinjiang@huawei.com Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Mike Rapoport (IBM) <rppt@kernel.org> Cc: Axel Rasmussen <axelrasmussen@google.com> Cc: Jinjiang Tu <tujinjiang@huawei.com> Cc: Dimitris Siakavaras <jimsiak@cslab.ece.ntua.gr> Signed-off-by: Peter Xu <peterx@redhat.com> --- fs/userfaultfd.c | 51 ++++++++++++++++++++++++------------------------ 1 file changed, 25 insertions(+), 26 deletions(-)