mm/page-writeback: document folio_mark_dirty() locking more explicitly

We have had bugs where set_page_dirty() was used on a page from GUP
without appropriate locking, leading to UAF, in:

 - KVM, see
   https://lore.kernel.org/r/20260810-x86-kvm-setpagedirty-v1-1-85f180892d4f@google.com
 - i915, see commit 0d4bbe3d407f ("drm/i915/userptr: Try to acquire the
   page lock around set_page_dirty()").
 - VMCI, see commit 5a16c535409f ("VMCI: Use set_page_dirty_lock() when
   unregistering guest memory")
 - kpc2000 staging driver, see commit b6d13bd9f2c1 ("staging: kpc2000:
   kpc_dma: Convert set_page_dirty() --> set_page_dirty_lock()")

I think set_page_dirty() and folio_mark_dirty() need more explicit
documentation on how they should be used with pages from GUP; so add a
comment on top of set_page_dirty() and make the comment above
folio_mark_dirty() more explicit.

Link: https://lore.kernel.org/20260810-set-page-dirty-warnings-v2-1-1bd40fadfacd@google.com
Signed-off-by: Jann Horn <jannh@google.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

diff --git a/mm/folio-compat.c b/mm/folio-compat.c
index a02179a..6212fdd 100644 (file)
--- a/mm/folio-compat.c
+++ b/mm/folio-compat.c
@@ -41,6 +41,7 @@ void set_page_writeback(struct page *page)
 }
 EXPORT_SYMBOL(set_page_writeback);
 
+/* Read the comment above folio_mark_dirty() regarding required locks! */
 bool set_page_dirty(struct page *page)
 {
        return folio_mark_dirty(page_folio(page));

diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index 47495be..3a03e04 100644 (file)
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -2763,6 +2763,11 @@ EXPORT_SYMBOL(folio_redirty_for_writepage);
  * in this folio.  Truncation will block on the page table lock as it
  * unmaps pages before removing the folio from its mapping.
  *
+ * .. DANGER::
+ *    Do not use this on a folio obtained from a function like
+ *    get_user_pages_fast() without holding appropriate locks; you might want to
+ *    use set_page_dirty_lock() or folio_mark_dirty_lock() instead.
+ *
  * Return: True if the folio was newly dirtied, false if it was already dirty.
  */
 bool folio_mark_dirty(struct folio *folio)