Re: [PATCH] workqueue: Don't record workqueue stack holding raw_spin_lock

From: Shuah Khan
Date: Mon Aug 30 2021 - 12:27:51 EST


On 8/29/21 8:18 PM, Lai Jiangshan wrote:
On Sat, Aug 28, 2021 at 7:49 AM Shuah Khan <skhan@xxxxxxxxxxxxxxxxxxx> wrote:

When CONFIG_PROVE_RAW_LOCK_NESTING and CONFIG_KASAN are enabled,
kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
it tries to allocate memory attempting to acquire spinlock in page
allocation code while holding workqueue pool raw_spinlock.

There are several instances of this problem when block layer tries
to __queue_work(). Call trace from one of these instances is below:

kblockd_mod_delayed_work_on()
mod_delayed_work_on()
__queue_delayed_work()
__queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
insert_work()
kasan_record_aux_stack()
kasan_save_stack()
stack_depot_save()
alloc_pages()
__alloc_pages()
get_page_from_freelist()
rm_queue()
rm_queue_pcplist()
local_lock_irqsave(&pagesets.lock, flags);
[ BUG: Invalid wait context triggered ]

Fix it by calling kasan_record_aux_stack() without holding pool lock.
There is no need to hold pool lock in this path.

=============================
[ BUG: Invalid wait context ]
5.14.0-rc7+ #8 Not tainted
-----------------------------
snap/532 is trying to lock:
ffff888374f32ba0 (lock#2){..-.}-{3:3}, at: get_page_from_freelist (mm/page_alloc.c:3665 mm/page_alloc.c:3703 mm/page_alloc.c:4165)
other info that might help us debug this:
context-{5:5}
3 locks held by snap/532:
#0: ffff888139fa4408 (&type->i_mutex_dir_key#10){.+.+}-{4:4}, at: walk_component (fs/namei.c:1663 fs/namei.c:1959)
#1: ffffffffab870c40 (rcu_read_lock){....}-{1:3}, at: __queue_work (./arch/x86/include/asm/preempt.h:80 ./include/linux/rcupdate.h:68 ./include/linux/rcupdate.h:685 kernel/workqueue.c:1421)
#2: ffff888374f36cd8 (&pool->lock){-.-.}-{2:2}, at: __queue_work (kernel/workqueue.c:1466)
stack backtrace:
CPU: 14 PID: 532 Comm: snap Not tainted 5.14.0-rc7+ #8
Hardware name: LENOVO 90Q30008US/3728, BIOS O4ZKT1CA 09/16/2020
Call Trace:
dump_stack_lvl (lib/dump_stack.c:106 (discriminator 4))
dump_stack (lib/dump_stack.c:113)
__lock_acquire.cold (kernel/locking/lockdep.c:4666 kernel/locking/lockdep.c:4727 kernel/locking/lockdep.c:4965)
? lockdep_hardirqs_on_prepare (kernel/locking/lockdep.c:4873)
? lock_is_held_type (kernel/locking/lockdep.c:5368 kernel/locking/lockdep.c:5668)
lock_acquire (kernel/locking/lockdep.c:438 kernel/locking/lockdep.c:5627 kernel/locking/lockdep.c:5590)
? get_page_from_freelist (mm/page_alloc.c:3665 mm/page_alloc.c:3703 mm/page_alloc.c:4165)
? lock_release (kernel/locking/lockdep.c:5593)
? __kasan_check_read (mm/kasan/shadow.c:32)
? __lock_acquire (kernel/locking/lockdep.c:5019)
? __zone_watermark_ok (./include/linux/list.h:282 ./include/linux/mmzone.h:111 mm/page_alloc.c:3908)
get_page_from_freelist (./include/linux/local_lock_internal.h:43 mm/page_alloc.c:3665 mm/page_alloc.c:3703 mm/page_alloc.c:4165)
? get_page_from_freelist (mm/page_alloc.c:3665 mm/page_alloc.c:3703 mm/page_alloc.c:4165)
? lock_is_held_type (kernel/locking/lockdep.c:5368 kernel/locking/lockdep.c:5668)
? is_bpf_text_address (./arch/x86/include/asm/preempt.h:85 ./include/linux/rcupdate.h:73 ./include/linux/rcupdate.h:719 kernel/bpf/core.c:708)
? lock_downgrade (kernel/locking/lockdep.c:5633)
? __zone_watermark_ok (mm/page_alloc.c:4054)
__alloc_pages (mm/page_alloc.c:5391)
? __alloc_pages_slowpath.constprop.0 (mm/page_alloc.c:5354)
? create_prof_cpu_mask (kernel/stacktrace.c:82)
? _find_first_bit (lib/find_bit.c:83)
alloc_pages (mm/mempolicy.c:2249)
stack_depot_save (lib/stackdepot.c:304)
? lock_is_held_type (kernel/locking/lockdep.c:5368 kernel/locking/lockdep.c:5668)
kasan_save_stack (mm/kasan/common.c:41)
? kasan_save_stack (mm/kasan/common.c:39)
? kasan_record_aux_stack (mm/kasan/generic.c:348)
? insert_work (./include/linux/instrumented.h:71 ./include/asm-generic/bitops/instrumented-non-atomic.h:134 kernel/workqueue.c:616 kernel/workqueue.c:623 kernel/workqueue.c:1335)
? __queue_work (kernel/workqueue.c:1501)
? __queue_delayed_work (kernel/workqueue.c:1657)
? mod_delayed_work_on (kernel/workqueue.c:1720)
? kblockd_mod_delayed_work_on (block/blk-core.c:1633)
? __blk_mq_delay_run_hw_queue (block/blk-mq.c:1567)
? blk_mq_run_hw_queue (block/blk-mq.c:1610)
? blk_mq_sched_insert_request (block/blk-mq-sched.c:480)
? blk_mq_submit_bio (block/blk-mq.c:2276)

Fixes: e89a85d63fb2 ("workqueue: kasan: record workqueue stack")
Signed-off-by: Shuah Khan <skhan@xxxxxxxxxxxxxxxxxxx>
---
kernel/workqueue.c | 18 +++++++++++++++---
1 file changed, 15 insertions(+), 3 deletions(-)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index f148eacda55a..e647b29b9fb0 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -1328,9 +1328,6 @@ static void insert_work(struct pool_workqueue *pwq, struct work_struct *work,
{
struct worker_pool *pool = pwq->pool;

- /* record the work call stack in order to print it in KASAN reports */
- kasan_record_aux_stack(work);
-
/* we own @work, set data and link */
set_work_pwq(work, pwq, extra_flags);
list_add_tail(&work->entry, head);
@@ -1499,6 +1496,14 @@ static void __queue_work(int cpu, struct workqueue_struct *wq,

out:
raw_spin_unlock(&pwq->pool->lock);
+
+ /*
+ * record the work call stack in order to print it in KASAN reports
+ * Do this without pool raw_spinlock hold to avoid nesting raw
+ * spinlock with page allocation spinlock.
+ */
+ kasan_record_aux_stack(work);
+

Hello

After the pool lock is released, the work item can be executed and be freed
before kasan_record_aux_stack() is called. Is it Ok?


Right. That is a concern. As Dmitry pointed out there is a bug in the
code that kasan_record_aux_stack() is in the "out:" path.

And I think queue_work() is allowed to be called with other raw_spin held.
If it is the case, kasan_record_aux_stack() is still in raw_spin held section.


I looked for queue_work() -> queue_work_on() -> __queue_work() and
function header doc. I couldn't find any place it says it is okay to
call queue_work() with raw spinlock held.

I am thinking that is not the case, since __queue_work() acquires pool
lock which is a raw spinlock. The lockdep warning will occur when the
pool lock acquire happens. I haven't seen it. We have 1290 callers of
queue_work() and we would have seen this before.

thanks,
-- Shuah