folio_is_secretmem() currently relies on secretmem folios being LRU folios, to save some cycles.
However, folios might reside in a folio batch without the LRU flag set, or temporarily have their LRU flag cleared. Consequently, the LRU flag is unreliable for this purpose.
In particular, this is the case when secretmem_fault() allocates a fresh page and calls filemap_add_folio()->folio_add_lru(). The folio might be added to the per-cpu folio batch and won't get the LRU flag set until the batch was drained using e.g., lru_add_drain().
Consequently, folio_is_secretmem() might not detect secretmem folios and GUP-fast can succeed in grabbing a secretmem folio, crashing the kernel when we would later try reading/writing to the folio, because the folio has been unmapped from the directmap.
Fix it by removing that unreliable check.
Reported-by: xingwei lee xrivendell7@gmail.com Reported-by: yue sun samsun1006219@gmail.com Closes: https://lore.kernel.org/lkml/CABOYnLyevJeravW=QrH0JUPYEcDN160aZFb7kwndm-J2rm... Debugged-by: Miklos Szeredi miklos@szeredi.hu Tested-by: Miklos Szeredi mszeredi@redhat.com Fixes: 1507f51255c9 ("mm: introduce memfd_secret system call to create "secret" memory areas") Cc: stable@vger.kernel.org Signed-off-by: David Hildenbrand david@redhat.com --- include/linux/secretmem.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/include/linux/secretmem.h b/include/linux/secretmem.h index 35f3a4a8ceb1..acf7e1a3f3de 100644 --- a/include/linux/secretmem.h +++ b/include/linux/secretmem.h @@ -13,10 +13,10 @@ static inline bool folio_is_secretmem(struct folio *folio) /* * Using folio_mapping() is quite slow because of the actual call * instruction. - * We know that secretmem pages are not compound and LRU so we can + * We know that secretmem pages are not compound, so we can * save a couple of cycles here. */ - if (folio_test_large(folio) || !folio_test_lru(folio)) + if (folio_test_large(folio)) return false;
mapping = (struct address_space *)