On 2021/7/28 21:23, Michal Hocko wrote:
On Wed 28-07-21 17:13:48, Wang Hai wrote:
When I use kfree_rcu() to free a large memory allocated by
kmalloc_node(), the following dump occurs.
BUG: kernel NULL pointer dereference, address: 0000000000000020
[...]
Oops: 0000 [#1] SMP
[...]
Workqueue: events kfree_rcu_work
RIP: 0010:__obj_to_index include/linux/slub_def.h:182 [inline]
RIP: 0010:obj_to_index include/linux/slub_def.h:191 [inline]
RIP: 0010:memcg_slab_free_hook+0x120/0x260 mm/slab.h:363
[...]
Call Trace:
kmem_cache_free_bulk+0x58/0x630 mm/slub.c:3293
kfree_bulk include/linux/slab.h:413 [inline]
kfree_rcu_work+0x1ab/0x200 kernel/rcu/tree.c:3300
process_one_work+0x207/0x530 kernel/workqueue.c:2276
worker_thread+0x320/0x610 kernel/workqueue.c:2422
kthread+0x13d/0x160 kernel/kthread.c:313
ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:294
When kmalloc_node() a large memory, page is allocated, not slab,
so when freeing memory via kfree_rcu(), this large memory should not
be used by memcg_slab_free_hook(), because memcg_slab_free_hook() is
is used for slab.
So in this case, there is no need to do anything with this large
page in memcg_slab_free_hook(), just skip it.
Fixes: 270c6a71460e ("mm: memcontrol/slab: Use helpers to access slab page's memcg_data")
Are you sure that this commit is really breaking the code. Unless I have
Yes, we confirmed that this commit introduces the bug.
missed something there shouldn't be any real change wrt. large
allocations here. page_has_obj_cgroups is just a different name for what
what page_objcgs is giving us.
maybe we could simply use page_objcgs_check to fix the issue ? we will
check it again.
diff --git a/mm/slab.h b/mm/slab.h
index f997fd5e42c8..58c01a34e5b8 100644
--- a/mm/slab.h
+++ b/mm/slab.h
@@ -346,7 +346,7 @@ static inline void memcg_slab_free_hook(struct
kmem_cache *s_orig,
continue;
page = virt_to_head_page(p[i]);
- objcgs = page_objcgs(page);
+ objcgs = page_objcgs_check(page);
if (!objcgs)
continue;
I haven't studied the kfree_rcu part but isn't the problem its use of
kmem_cache_free_bulk or isn't the problem right there in the bulk free?
Signed-off-by: Wang Hai <wanghai38@xxxxxxxxxx>
---
mm/slab.h | 15 ++++++++++-----
1 file changed, 10 insertions(+), 5 deletions(-)
diff --git a/mm/slab.h b/mm/slab.h
index 67e06637ff2e..247d3f9c21f7 100644
--- a/mm/slab.h
+++ b/mm/slab.h
@@ -339,15 +339,20 @@ static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
continue;
page = virt_to_head_page(p[i]);
+ if (!s_orig) {
+ if (unlikely(!PageSlab(page))) {
+ BUG_ON(!PageCompound(page));
the same logical is in build_detached_freelist()
BUG_ON is not really a good idea here. Why should we crash the kernel
just because of an unexpected page showing up. Leaking it would be more
appropriate (the same would apply to kfree btw). I would just warn
here. Also don't we need any hookd here. Looking at kfree path it does
call kfree_hook. Why is that not needed here?
kmem_cache_free_bulk
-- memcg_slab_free_hook // skip objcgs ops for page
-- build_detached_freelist // kfree_hook and page free
