The previous fix 384632e67e0829deb8015ee6ad916b180049d252 corrected
the refcounting in case of UFFD_EVENT_FORK failure for the fork
userfault paths. That still didn't clear the vma->vm_userfaultfd_ctx
of the vmas that were set to point to the aborted new uffd ctx earlier
in dup_userfaultfd.
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Andrea Arcangeli <aarcange@xxxxxxxxxx>
---
fs/userfaultfd.c | 20 ++++++++++++++++++--
1 file changed, 18 insertions(+), 2 deletions(-)
diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index 896f810b6a06..1a88916455bd 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -591,11 +591,14 @@ int handle_userfault(struct vm_fault *vmf, unsigned long reason)
static void userfaultfd_event_wait_completion(struct userfaultfd_ctx *ctx,
struct userfaultfd_wait_queue *ewq)
{
+ struct userfaultfd_ctx *release_new_ctx;
+
if (WARN_ON_ONCE(current->flags & PF_EXITING))
goto out;
ewq->ctx = ctx;
init_waitqueue_entry(&ewq->wq, current);
+ release_new_ctx = NULL;
spin_lock(&ctx->event_wqh.lock);
/*
@@ -622,8 +625,7 @@ static void userfaultfd_event_wait_completion(struct userfaultfd_ctx *ctx,
new = (struct userfaultfd_ctx *)
(unsigned long)
ewq->msg.arg.reserved.reserved1;
-
- userfaultfd_ctx_put(new);
+ release_new_ctx = new;
}
break;
}
@@ -638,6 +640,20 @@ static void userfaultfd_event_wait_completion(struct userfaultfd_ctx *ctx,
__set_current_state(TASK_RUNNING);
spin_unlock(&ctx->event_wqh.lock);
+ if (release_new_ctx) {
+ struct vm_area_struct *vma;
+ struct mm_struct *mm = release_new_ctx->mm;
+
+ /* 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)
+ vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
+ up_write(&mm->mmap_sem);
+
+ userfaultfd_ctx_put(release_new_ctx);
+ }
+
/*
* ctx may go away after this if the userfault pseudo fd is
* already released.
