mm/filemap: Fix handling of THPs in generic_file_buffered_read()
authorMatthew Wilcox (Oracle) <willy@infradead.org>
Wed, 23 Feb 2022 15:59:18 +0000 (15:59 +0000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 2 Mar 2022 10:47:47 +0000 (11:47 +0100)
When a THP is present in the page cache, we can return it several times,
leading to userspace seeing the same data repeatedly if doing a read()
that crosses a 64-page boundary.  This is probably not a security issue
(since the data all comes from the same file), but it can be interpreted
as a transient data corruption issue.  Fortunately, it is very rare as
it can only occur when CONFIG_READ_ONLY_THP_FOR_FS is enabled, and it can
only happen to executables.  We don't often call read() on executables.

This bug is fixed differently in v5.17 by commit 6b24ca4a1a8d
("mm: Use multi-index entries in the page cache").  That commit is
unsuitable for backporting, so fix this in the clearest way.  It
sacrifices a little performance for clarity, but this should never
be a performance path in these kernel versions.

Fixes: cbd59c48ae2b ("mm/filemap: use head pages in generic_file_buffered_read")
Cc: stable@vger.kernel.org # v5.15, v5.16
Link: https://lore.kernel.org/r/df3b5d1c-a36b-2c73-3e27-99e74983de3a@suse.cz/
Analyzed-by: Adam Majer <amajer@suse.com>
Analyzed-by: Dirk Mueller <dmueller@suse.com>
Bisected-by: Takashi Iwai <tiwai@suse.de>
Reported-by: Vlastimil Babka <vbabka@suse.cz>
Tested-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mm/filemap.c

index 82a17c35eb96bcd4155d783f544a52a70269b1a7..1293c3409e429225294d117eb5c608aa0441be80 100644 (file)
@@ -2354,8 +2354,12 @@ static void filemap_get_read_batch(struct address_space *mapping,
                        break;
                if (PageReadahead(head))
                        break;
-               xas.xa_index = head->index + thp_nr_pages(head) - 1;
-               xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK;
+               if (PageHead(head)) {
+                       xas_set(&xas, head->index + thp_nr_pages(head));
+                       /* Handle wrap correctly */
+                       if (xas.xa_index - 1 >= max)
+                               break;
+               }
                continue;
 put_page:
                put_page(head);