The fix in commit 0cbb4b4f4c44 ("userfaultfd: clear the
vma->vm_userfaultfd_ctx if UFFD_EVENT_FORK fails") cleared the
vma->vm_userfaultfd_ctx but kept userfaultfd flags in vma->vm_flags that
were copied from the parent process VMA.
As the result, there is an inconsistency between the values of
vma->vm_userfaultfd_ctx.ctx and vma->vm_flags which triggers BUG_ON in
userfaultfd_release().
Clearing the uffd flags from vma->vm_flags in case of UFFD_EVENT_FORK
failure resolves the issue.
Signed-off-by: Mike Rapoport <rppt@xxxxxxxxxxxxxxxxxx>
Reported-by: syzbot+121be635a7a35ddb7dcb@xxxxxxxxxxxxxxxxxxxxxxxxx
Cc: Andrea Arcangeli <aarcange@xxxxxxxxxx>
Cc: Eric Biggers <ebiggers3@xxxxxxxxx>
Cc: stable@xxxxxxxxxxxxxxx
---
fs/userfaultfd.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index 594d192b2331..bad9cea37f12 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -633,8 +633,10 @@ static void userfaultfd_event_wait_completion(struct userfaultfd_ctx *ctx,
/* the various vma->vm_userfaultfd_ctx still points to it */
down_write(&mm->mmap_sem);
for (vma = mm->mmap; vma; vma = vma->vm_next)
- if (vma->vm_userfaultfd_ctx.ctx == release_new_ctx)
+ if (vma->vm_userfaultfd_ctx.ctx == release_new_ctx) {
vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
+ vma->vm_flags &= ~(VM_UFFD_WP | VM_UFFD_MISSING);
+ }
up_write(&mm->mmap_sem);
userfaultfd_ctx_put(release_new_ctx);
--
2.7.4
