mm: fix __access_remote_vm() GUP failure case
authorLinus Torvalds <torvalds@linux-foundation.org>
Wed, 28 Jun 2023 19:20:24 +0000 (12:20 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 28 Jun 2023 19:20:24 +0000 (12:20 -0700)
Commit ca5e863233e8 ("mm/gup: remove vmas parameter from
get_user_pages_remote()") removed the vma argument from GUP handling,
and instead added a helper function (get_user_page_vma_remote()) that
looks it up separately using 'vma_lookup()'.  And then converted
existing users that needed a vma to use the helper instead.

However, the helper function intentionally acts exactly like the old
get_user_pages_remote() did, and only fills in 'vma' on successful page
lookup.  Fine so far.

However, __access_remote_vm() wants the vma even for the unsuccessful
case, and used to do a

vma = vma_lookup(mm, addr);

explicitly to look it up when the get_user_page() failed.

However, that conversion commit incorrectly removed that vma lookup,
thinking that get_user_page_vma_remote() would have done it.  Not so.

So add the vma_lookup() back in.

Fixes: ca5e863233e8 ("mm/gup: remove vmas parameter from get_user_pages_remote()")
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memory.c

index f758f59..1188218 100644 (file)
@@ -5592,6 +5592,7 @@ int __access_remote_vm(struct mm_struct *mm, unsigned long addr, void *buf,
                         * Check if this is a VM_IO | VM_PFNMAP VMA, which
                         * we can access using slightly different code.
                         */
+                       vma = vma_lookup(mm, addr);
                        if (!vma)
                                break;
                        if (vma->vm_ops && vma->vm_ops->access)