Re: [PATCH] mbcache: initialize entry->e_referenced in mb_cache_entry_create()

[Eric Biggers]

+Cc linux-ext4

On Thu, Jul 06, 2017 at 08:27:25PM +0200, Alexander Potapenko wrote:
> KMSAN reported use of uninitialized |entry->e_referenced| in a condition
> in mb_cache_shrink():
> 
> ==================================================================
> BUG: KMSAN: use of uninitialized memory in mb_cache_shrink+0x3b4/0xc50 fs/mbcache.c:287
> CPU: 2 PID: 816 Comm: kswapd1 Not tainted 4.11.0-rc5+ #2877
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs
> 01/01/2011
> Call Trace:
>  __dump_stack lib/dump_stack.c:16 [inline]
>  dump_stack+0x172/0x1c0 lib/dump_stack.c:52
>  kmsan_report+0x12a/0x180 mm/kmsan/kmsan.c:927
>  __msan_warning_32+0x61/0xb0 mm/kmsan/kmsan_instr.c:469
>  mb_cache_shrink+0x3b4/0xc50 fs/mbcache.c:287
>  mb_cache_scan+0x67/0x80 fs/mbcache.c:321
>  do_shrink_slab mm/vmscan.c:397 [inline]
>  shrink_slab+0xc3d/0x12d0 mm/vmscan.c:500
>  shrink_node+0x208f/0x2fd0 mm/vmscan.c:2603
>  kswapd_shrink_node mm/vmscan.c:3172 [inline]
>  balance_pgdat mm/vmscan.c:3289 [inline]
>  kswapd+0x160f/0x2850 mm/vmscan.c:3478
>  kthread+0x46c/0x5f0 kernel/kthread.c:230
>  ret_from_fork+0x29/0x40 arch/x86/entry/entry_64.S:430
> chained origin:
>  save_stack_trace+0x37/0x40 arch/x86/kernel/stacktrace.c:59
>  kmsan_save_stack_with_flags mm/kmsan/kmsan.c:302 [inline]
>  kmsan_save_stack mm/kmsan/kmsan.c:317 [inline]
>  kmsan_internal_chain_origin+0x12a/0x1f0 mm/kmsan/kmsan.c:547
>  __msan_store_shadow_origin_1+0xac/0x110 mm/kmsan/kmsan_instr.c:257
>  mb_cache_entry_create+0x3b3/0xc60 fs/mbcache.c:95
>  ext4_xattr_cache_insert fs/ext4/xattr.c:1647 [inline]
>  ext4_xattr_block_set+0x4c82/0x5530 fs/ext4/xattr.c:1022
>  ext4_xattr_set_handle+0x1332/0x20a0 fs/ext4/xattr.c:1252
>  ext4_xattr_set+0x4d2/0x680 fs/ext4/xattr.c:1306
>  ext4_xattr_trusted_set+0x8d/0xa0 fs/ext4/xattr_trusted.c:36
>  __vfs_setxattr+0x703/0x790 fs/xattr.c:149
>  __vfs_setxattr_noperm+0x27a/0x6f0 fs/xattr.c:180
>  vfs_setxattr fs/xattr.c:223 [inline]
>  setxattr+0x6ae/0x790 fs/xattr.c:449
>  path_setxattr+0x1eb/0x380 fs/xattr.c:468
>  SYSC_lsetxattr+0x8d/0xb0 fs/xattr.c:490
>  SyS_lsetxattr+0x77/0xa0 fs/xattr.c:486
>  entry_SYSCALL_64_fastpath+0x13/0x94
> origin:
>  save_stack_trace+0x37/0x40 arch/x86/kernel/stacktrace.c:59
>  kmsan_save_stack_with_flags mm/kmsan/kmsan.c:302 [inline]
>  kmsan_internal_poison_shadow+0xb1/0x1a0 mm/kmsan/kmsan.c:198
>  kmsan_kmalloc+0x7f/0xe0 mm/kmsan/kmsan.c:337
>  kmem_cache_alloc+0x1c2/0x1e0 mm/slub.c:2766
>  mb_cache_entry_create+0x283/0xc60 fs/mbcache.c:86
>  ext4_xattr_cache_insert fs/ext4/xattr.c:1647 [inline]
>  ext4_xattr_block_set+0x4c82/0x5530 fs/ext4/xattr.c:1022
>  ext4_xattr_set_handle+0x1332/0x20a0 fs/ext4/xattr.c:1252
>  ext4_xattr_set+0x4d2/0x680 fs/ext4/xattr.c:1306
>  ext4_xattr_trusted_set+0x8d/0xa0 fs/ext4/xattr_trusted.c:36
>  __vfs_setxattr+0x703/0x790 fs/xattr.c:149
>  __vfs_setxattr_noperm+0x27a/0x6f0 fs/xattr.c:180
>  vfs_setxattr fs/xattr.c:223 [inline]
>  setxattr+0x6ae/0x790 fs/xattr.c:449
>  path_setxattr+0x1eb/0x380 fs/xattr.c:468
>  SYSC_lsetxattr+0x8d/0xb0 fs/xattr.c:490
>  SyS_lsetxattr+0x77/0xa0 fs/xattr.c:486
>  entry_SYSCALL_64_fastpath+0x13/0x94
> ==================================================================
> 
> Signed-off-by: Alexander Potapenko <glider@xxxxxxxxxx>
> ---
>  fs/mbcache.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/fs/mbcache.c b/fs/mbcache.c
> index b19be429d655..fdfe8933ac6b 100644
> --- a/fs/mbcache.c
> +++ b/fs/mbcache.c
> @@ -93,6 +93,7 @@ int mb_cache_entry_create(struct mb_cache *cache, gfp_t mask, u32 key,
>  	entry->e_key = key;
>  	entry->e_block = block;
>  	entry->e_reusable = reusable;
> +	entry->e_referenced = 0;
>  	head = mb_cache_entry_head(cache, key);
>  	hlist_bl_lock(head);
>  	hlist_bl_for_each_entry(dup, dup_node, head, e_hash_list) {
> -- 

Nice catch!  This looks fine.  I was wondering whether e_referenced should be
initialized to 1 instead, but I'm thinking 0 is better so that only entries that
have actually been deduplicated against get an extra pass through the list.

Reviewed-by: Eric Biggers <ebiggers@xxxxxxxxxx>

- Eric