On 22/11/2019 11:26 am, glider@xxxxxxxxxx wrote:
KMSAN doesn't allow treating adjacent memory pages as such, if they were
allocated by different alloc_pages() calls.
ext4_mpage_readpages() however does so: adjacent pages end up being passed
together to dma_direct_map_sg().
Urgh, there are definitely more places where physically-contiguous pages
are coalesced into single scatterlist entries - see
sg_alloc_table_from_pages() for instance. I wouldn't be surprised if
there are further open-coded versions hiding out in various other
drivers/subsystems too. Unless I've misunderstood, this seems like quite
an invasive limitation :(
Robin.
To prevent this, jump directly to the buffer_head-based read function in
KMSAN builds.
Signed-off-by: Alexander Potapenko <glider@xxxxxxxxxx>
Cc: "Theodore Ts'o" <tytso@xxxxxxx>
Cc: Andreas Dilger <adilger.kernel@xxxxxxxxx>
Cc: Vegard Nossum <vegard.nossum@xxxxxxxxxx>
Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
Cc: linux-mm@xxxxxxxxx
---
Change-Id: I54ae8af536626a988e6398ff18a06c179b0ce034
---
fs/ext4/readpage.c | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/fs/ext4/readpage.c b/fs/ext4/readpage.c
index a30b203fa461..a3bb9e3ce5de 100644
--- a/fs/ext4/readpage.c
+++ b/fs/ext4/readpage.c
@@ -252,6 +252,17 @@ int ext4_mpage_readpages(struct address_space *mapping,
if (page_has_buffers(page))
goto confused;
+#if defined(CONFIG_KMSAN)
+ /*
+ * The following code may treat adjacent pages allocated
+ * separately as a bigger contiguous allocation.
+ * KMSAN doesn't allow this, as the corresponding metadata
+ * pages may be separated.
+ * Skip this complex logic for KMSAN builds.
+ */
+ goto confused;
+#endif
+
block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
last_block = block_in_file + nr_pages * blocks_per_page;
last_block_in_file = (ext4_readpage_limit(inode) +
