]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
userfaultfd: release page in error path to avoid BUG_ON
authorAxel Rasmussen <axelrasmussen@google.com>
Sat, 15 May 2021 00:27:19 +0000 (17:27 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 15 May 2021 02:41:32 +0000 (19:41 -0700)
Consider the following sequence of events:

1. Userspace issues a UFFD ioctl, which ends up calling into
   shmem_mfill_atomic_pte(). We successfully account the blocks, we
   shmem_alloc_page(), but then the copy_from_user() fails. We return
   -ENOENT. We don't release the page we allocated.
2. Our caller detects this error code, tries the copy_from_user() after
   dropping the mmap_lock, and retries, calling back into
   shmem_mfill_atomic_pte().
3. Meanwhile, let's say another process filled up the tmpfs being used.
4. So shmem_mfill_atomic_pte() fails to account blocks this time, and
   immediately returns - without releasing the page.

This triggers a BUG_ON in our caller, which asserts that the page
should always be consumed, unless -ENOENT is returned.

To fix this, detect if we have such a "dangling" page when accounting
fails, and if so, release it before returning.

Link: https://lkml.kernel.org/r/20210428230858.348400-1-axelrasmussen@google.com
Fixes: cb658a453b93 ("userfaultfd: shmem: avoid leaking blocks and used blocks in UFFDIO_COPY")
Signed-off-by: Axel Rasmussen <axelrasmussen@google.com>
Reported-by: Hugh Dickins <hughd@google.com>
Acked-by: Hugh Dickins <hughd@google.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/shmem.c

index eb131b9fb1909c1da52b73ff65393f65a8a2caf7..5d46611cba8dca00882a9c2799288f93127fca48 100644 (file)
@@ -2361,8 +2361,18 @@ static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
        pgoff_t offset, max_off;
 
        ret = -ENOMEM;
-       if (!shmem_inode_acct_block(inode, 1))
+       if (!shmem_inode_acct_block(inode, 1)) {
+               /*
+                * We may have got a page, returned -ENOENT triggering a retry,
+                * and now we find ourselves with -ENOMEM. Release the page, to
+                * avoid a BUG_ON in our caller.
+                */
+               if (unlikely(*pagep)) {
+                       put_page(*pagep);
+                       *pagep = NULL;
+               }
                goto out;
+       }
 
        if (!*pagep) {
                page = shmem_alloc_page(gfp, info, pgoff);