From a7cb2e709f2927cc3c76781df3e45de2381b3b9d Mon Sep 17 00:00:00 2001 From: Chengming Zhou Date: Fri, 1 Sep 2023 20:03:06 +0800 Subject: [PATCH] null_blk: fix poll request timeout handling commit 5a26e45edb4690d58406178b5a9ea4c6dcf2c105 upstream. When doing io_uring benchmark on /dev/nullb0, it's easy to crash the kernel if poll requests timeout triggered, as reported by David. [1] BUG: kernel NULL pointer dereference, address: 0000000000000008 Workqueue: kblockd blk_mq_timeout_work RIP: 0010:null_timeout_rq+0x4e/0x91 Call Trace: ? null_timeout_rq+0x4e/0x91 blk_mq_handle_expired+0x31/0x4b bt_iter+0x68/0x84 ? bt_tags_iter+0x81/0x81 __sbitmap_for_each_set.constprop.0+0xb0/0xf2 ? __blk_mq_complete_request_remote+0xf/0xf bt_for_each+0x46/0x64 ? __blk_mq_complete_request_remote+0xf/0xf ? percpu_ref_get_many+0xc/0x2a blk_mq_queue_tag_busy_iter+0x14d/0x18e blk_mq_timeout_work+0x95/0x127 process_one_work+0x185/0x263 worker_thread+0x1b5/0x227 This is indeed a race problem between null_timeout_rq() and null_poll(). null_poll() null_timeout_rq() spin_lock(&nq->poll_lock) list_splice_init(&nq->poll_list, &list) spin_unlock(&nq->poll_lock) while (!list_empty(&list)) req = list_first_entry() list_del_init() ... blk_mq_add_to_batch() // req->rq_next = NULL spin_lock(&nq->poll_lock) // rq->queuelist->next == NULL list_del_init(&rq->queuelist) spin_unlock(&nq->poll_lock) Fix these problems by setting requests state to MQ_RQ_COMPLETE under nq->poll_lock protection, in which null_timeout_rq() can safely detect this race and early return. Note this patch just fix the kernel panic when request timeout happen. [1] https://lore.kernel.org/all/3893581.1691785261@warthog.procyon.org.uk/ Fixes: 0a593fbbc245 ("null_blk: poll queue support") Reported-by: David Howells Tested-by: David Howells Reviewed-by: Ming Lei Signed-off-by: Chengming Zhou Link: https://lore.kernel.org/r/20230901120306.170520-2-chengming.zhou@linux.dev Signed-off-by: Jens Axboe Signed-off-by: Greg Kroah-Hartman --- drivers/block/null_blk/main.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) diff --git a/drivers/block/null_blk/main.c b/drivers/block/null_blk/main.c index 864013019d6b6..968090935eb23 100644 --- a/drivers/block/null_blk/main.c +++ b/drivers/block/null_blk/main.c @@ -1643,9 +1643,12 @@ static int null_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob) struct nullb_queue *nq = hctx->driver_data; LIST_HEAD(list); int nr = 0; + struct request *rq; spin_lock(&nq->poll_lock); list_splice_init(&nq->poll_list, &list); + list_for_each_entry(rq, &list, queuelist) + blk_mq_set_request_complete(rq); spin_unlock(&nq->poll_lock); while (!list_empty(&list)) { @@ -1671,16 +1674,21 @@ static enum blk_eh_timer_return null_timeout_rq(struct request *rq) struct blk_mq_hw_ctx *hctx = rq->mq_hctx; struct nullb_cmd *cmd = blk_mq_rq_to_pdu(rq); - pr_info("rq %p timed out\n", rq); - if (hctx->type == HCTX_TYPE_POLL) { struct nullb_queue *nq = hctx->driver_data; spin_lock(&nq->poll_lock); + /* The request may have completed meanwhile. */ + if (blk_mq_request_completed(rq)) { + spin_unlock(&nq->poll_lock); + return BLK_EH_DONE; + } list_del_init(&rq->queuelist); spin_unlock(&nq->poll_lock); } + pr_info("rq %p timed out\n", rq); + /* * If the device is marked as blocking (i.e. memory backed or zoned * device), the submission path may be blocked waiting for resources