On 4/1/22 8:27 PM, mingzhe.zou@xxxxxxxxxxxx wrote:
From: ZouMingzhe <mingzhe.zou@xxxxxxxxxxxx>
We get a kernel crash about "list_add corruption. next->prev should be
prev (ffff9c801bc01210), but was ffff9c77b688237c. (next=ffffae586d8afe68)."
crash> struct list_head 0xffff9c801bc01210
struct list_head {
next = 0xffffae586d8afe68,
prev = 0xffffae586d8afe68
}
crash> struct list_head 0xffff9c77b688237c
struct list_head {
next = 0x0,
prev = 0x0
}
crash> struct list_head 0xffffae586d8afe68
struct list_head struct: invalid kernel virtual address: ffffae586d8afe68 type: "gdb_readmem_callback"
Cannot access memory at address 0xffffae586d8afe68
[230469.019492] Call Trace:
[230469.032041] prepare_to_wait+0x8a/0xb0
[230469.044363] ? bch_btree_keys_free+0x6c/0xc0 [escache]
[230469.056533] mca_cannibalize_lock+0x72/0x90 [escache]
[230469.068788] mca_alloc+0x2ae/0x450 [escache]
[230469.080790] bch_btree_node_get+0x136/0x2d0 [escache]
[230469.092681] bch_btree_check_thread+0x1e1/0x260 [escache]
[230469.104382] ? finish_wait+0x80/0x80
[230469.115884] ? bch_btree_check_recurse+0x1a0/0x1a0 [escache]
[230469.127259] kthread+0x112/0x130
[230469.138448] ? kthread_flush_work_fn+0x10/0x10
[230469.149477] ret_from_fork+0x35/0x40
bch_btree_check_thread() and bch_dirty_init_thread() maybe call
mca_cannibalize() to cannibalize other cached btree nodes. Only
one thread can do it at a time, so the op of other threads will
be added to the btree_cache_wait list.
We must call finish_wait() to remove op from btree_cache_wait
before free it's memory address. Otherwise, the list will be
damaged. Also should call bch_cannibalize_unlock() to release
the btree_cache_alloc_lock and wake_up other waiters.
Signed-off-by: Mingzhe Zou <mingzhe.zou@xxxxxxxxxxxx>
Thank you for this fix, it is really cool to find such defect on
cannibalize lock/unlock. It spent me a little time to understand how it
happens, and reply you late.
I feel the root cause is not from where you patched in bch_btree_check()
and bch_root_node_dirty_init(), something really fishing when using
mca_cannibalize_lock(),
843 static int mca_cannibalize_lock(struct cache_set *c, struct btree_op
*op)
844 {
845 spin_lock(&c->btree_cannibalize_lock);
846 if (likely(c->btree_cache_alloc_lock == NULL)) {
847 c->btree_cache_alloc_lock = current;
848 } else if (c->btree_cache_alloc_lock != current) {
849 if (op)
850 prepare_to_wait(&c->btree_cache_wait, &op->wait,
851 TASK_UNINTERRUPTIBLE);
852 spin_unlock(&c->btree_cannibalize_lock);
853 return -EINTR;
854 }
855 spin_unlock(&c->btree_cannibalize_lock);
856
857 return 0;
858 }
In line 849-851, if the cannibalized locking failed, insert current
op->wait into c->btree_cache_wait. Then at line 852, return -EINTR to
indicate the caller should retry. But it seems no caller checks whether
the return value is -EINTR and handles it properly.
Your patch should work, but I feel the issue of bch_cannibalize_lock()
is not solved yet. Maybe we should work on handling -EINTR returned from
mca_cannibalize_lock() IMHO.
BTW, when you observe the panic, how are the hardware configurations about,
- CPU cores
- Memory size
- Cache size
- Number of keys on the btree root node
Thanks.
Coly Li
---
drivers/md/bcache/btree.c | 10 +++++++++-
drivers/md/bcache/btree.h | 2 ++
drivers/md/bcache/writeback.c | 8 ++++++++
3 files changed, 19 insertions(+), 1 deletion(-)
diff --git a/drivers/md/bcache/btree.c b/drivers/md/bcache/btree.c
index ad9f16689419..f8e6f5c7c736 100644
--- a/drivers/md/bcache/btree.c
+++ b/drivers/md/bcache/btree.c
@@ -885,7 +885,7 @@ static struct btree *mca_cannibalize(struct cache_set *c, struct btree_op *op,
* cannibalize_bucket() will take. This means every time we unlock the root of
* the btree, we need to release this lock if we have it held.
*/
-static void bch_cannibalize_unlock(struct cache_set *c)
+void bch_cannibalize_unlock(struct cache_set *c)
{
spin_lock(&c->btree_cannibalize_lock);
if (c->btree_cache_alloc_lock == current) {
@@ -1968,6 +1968,14 @@ static int bch_btree_check_thread(void *arg)
c->gc_stats.nodes++;
bch_btree_op_init(&op, 0);
ret = bcache_btree(check_recurse, p, c->root, &op);
+ /* The op may be added to cache_set's btree_cache_wait
+ * in mca_cannibalize(), must ensure it is removed from
+ * the list and release btree_cache_alloc_lock before
+ * free op memory.
+ * Otherwise, the btree_cache_wait will be damaged.
+ */
+ bch_cannibalize_unlock(c);
+ finish_wait(&c->btree_cache_wait, &(&op)->wait);
if (ret)
goto out;
}
diff --git a/drivers/md/bcache/btree.h b/drivers/md/bcache/btree.h
index 50482107134f..435e82574ac3 100644
--- a/drivers/md/bcache/btree.h
+++ b/drivers/md/bcache/btree.h
@@ -365,6 +365,8 @@ static inline void force_wake_up_gc(struct cache_set *c)
_r; \
})
+void bch_cannibalize_unlock(struct cache_set *c);
+
#define MAP_DONE 0
#define MAP_CONTINUE 1
diff --git a/drivers/md/bcache/writeback.c b/drivers/md/bcache/writeback.c
index 9ee0005874cd..5b828555bca8 100644
--- a/drivers/md/bcache/writeback.c
+++ b/drivers/md/bcache/writeback.c
@@ -865,6 +865,14 @@ static int bch_root_node_dirty_init(struct cache_set *c,
}
} while (ret == -EAGAIN);
+ /* The op may be added to cache_set's btree_cache_wait
+ * in mca_cannibalize(), must ensure it is removed from
+ * the list and release btree_cache_alloc_lock before
+ * free op memory.
+ * Otherwise, the btree_cache_wait will be damaged.
+ */
+ bch_cannibalize_unlock(c);
+ finish_wait(&c->btree_cache_wait, &(&op.op)->wait);
return ret;
}
