block: force noio scope in blk_mq_freeze_queue

When block drivers or the core block code perform allocations with a frozen queue, this could try to recurse into the block device to reclaim memory and deadlock. Thus all allocations done by a process that froze a queue need to be done without __GFP_IO and __GFP_FS. Instead of tying to track all of them down, force a noio scope as part of freezing the queue. Note that nvme is a bit of a mess here due to the non-owner freezes, and they will be addressed separately. Signed-off-by: Christoph Hellwig <[email protected]> Link: https://lore.kernel.org/r/[email protected] Signed-off-by: Jens Axboe <[email protected]>
multipath-tcp · Jan 31, 2025 · 1e1a9ce · 1e1a9ce
1 parent 14ef496
commit 1e1a9ce
Show file tree

Hide file tree

Showing 26 changed files with 136 additions and 84 deletions.
diff --git a/block/blk-cgroup.c b/block/blk-cgroup.c
@@ -1545,6 +1545,7 @@ int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol)
 	struct request_queue *q = disk->queue;
 	struct blkg_policy_data *pd_prealloc = NULL;
 	struct blkcg_gq *blkg, *pinned_blkg = NULL;
+	unsigned int memflags;
 	int ret;
 
 	if (blkcg_policy_enabled(q, pol))
@@ -1559,7 +1560,7 @@ int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol)
 		return -EINVAL;
 
 	if (queue_is_mq(q))
-		blk_mq_freeze_queue(q);
+		memflags = blk_mq_freeze_queue(q);
 retry:
 	spin_lock_irq(&q->queue_lock);
 
@@ -1623,7 +1624,7 @@ int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol)
 	spin_unlock_irq(&q->queue_lock);
 out:
 	if (queue_is_mq(q))
-		blk_mq_unfreeze_queue(q);
+		blk_mq_unfreeze_queue(q, memflags);
 	if (pinned_blkg)
 		blkg_put(pinned_blkg);
 	if (pd_prealloc)
@@ -1667,12 +1668,13 @@ void blkcg_deactivate_policy(struct gendisk *disk,
 {
 	struct request_queue *q = disk->queue;
 	struct blkcg_gq *blkg;
+	unsigned int memflags;
 
 	if (!blkcg_policy_enabled(q, pol))
 		return;
 
 	if (queue_is_mq(q))
-		blk_mq_freeze_queue(q);
+		memflags = blk_mq_freeze_queue(q);
 
 	mutex_lock(&q->blkcg_mutex);
 	spin_lock_irq(&q->queue_lock);
@@ -1696,7 +1698,7 @@ void blkcg_deactivate_policy(struct gendisk *disk,
 	mutex_unlock(&q->blkcg_mutex);
 
 	if (queue_is_mq(q))
-		blk_mq_unfreeze_queue(q);
+		blk_mq_unfreeze_queue(q, memflags);
 }
 EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
 

diff --git a/block/blk-iocost.c b/block/blk-iocost.c
@@ -3224,6 +3224,7 @@ static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
 	u32 qos[NR_QOS_PARAMS];
 	bool enable, user;
 	char *body, *p;
+	unsigned int memflags;
 	int ret;
 
 	blkg_conf_init(&ctx, input);
@@ -3247,7 +3248,7 @@ static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
 		ioc = q_to_ioc(disk->queue);
 	}
 
-	blk_mq_freeze_queue(disk->queue);
+	memflags = blk_mq_freeze_queue(disk->queue);
 	blk_mq_quiesce_queue(disk->queue);
 
 	spin_lock_irq(&ioc->lock);
@@ -3347,15 +3348,15 @@ static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
 		wbt_enable_default(disk);
 
 	blk_mq_unquiesce_queue(disk->queue);
-	blk_mq_unfreeze_queue(disk->queue);
+	blk_mq_unfreeze_queue(disk->queue, memflags);
 
 	blkg_conf_exit(&ctx);
 	return nbytes;
 einval:
 	spin_unlock_irq(&ioc->lock);
 
 	blk_mq_unquiesce_queue(disk->queue);
-	blk_mq_unfreeze_queue(disk->queue);
+	blk_mq_unfreeze_queue(disk->queue, memflags);
 
 	ret = -EINVAL;
 err:
@@ -3414,6 +3415,7 @@ static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
 {
 	struct blkg_conf_ctx ctx;
 	struct request_queue *q;
+	unsigned int memflags;
 	struct ioc *ioc;
 	u64 u[NR_I_LCOEFS];
 	bool user;
@@ -3441,7 +3443,7 @@ static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
 		ioc = q_to_ioc(q);
 	}
 
-	blk_mq_freeze_queue(q);
+	memflags = blk_mq_freeze_queue(q);
 	blk_mq_quiesce_queue(q);
 
 	spin_lock_irq(&ioc->lock);
@@ -3493,7 +3495,7 @@ static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
 	spin_unlock_irq(&ioc->lock);
 
 	blk_mq_unquiesce_queue(q);
-	blk_mq_unfreeze_queue(q);
+	blk_mq_unfreeze_queue(q, memflags);
 
 	blkg_conf_exit(&ctx);
 	return nbytes;
@@ -3502,7 +3504,7 @@ static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
 	spin_unlock_irq(&ioc->lock);
 
 	blk_mq_unquiesce_queue(q);
-	blk_mq_unfreeze_queue(q);
+	blk_mq_unfreeze_queue(q, memflags);
 
 	ret = -EINVAL;
 err:

diff --git a/block/blk-iolatency.c b/block/blk-iolatency.c
@@ -749,9 +749,11 @@ static void blkiolatency_enable_work_fn(struct work_struct *work)
 	 */
 	enabled = atomic_read(&blkiolat->enable_cnt);
 	if (enabled != blkiolat->enabled) {
-		blk_mq_freeze_queue(blkiolat->rqos.disk->queue);
+		unsigned int memflags;
+
+		memflags = blk_mq_freeze_queue(blkiolat->rqos.disk->queue);
 		blkiolat->enabled = enabled;
-		blk_mq_unfreeze_queue(blkiolat->rqos.disk->queue);
+		blk_mq_unfreeze_queue(blkiolat->rqos.disk->queue, memflags);
 	}
 }
 

diff --git a/block/blk-mq.c b/block/blk-mq.c
@@ -210,12 +210,12 @@ int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
 }
 EXPORT_SYMBOL_GPL(blk_mq_freeze_queue_wait_timeout);
 
-void blk_mq_freeze_queue(struct request_queue *q)
+void blk_mq_freeze_queue_nomemsave(struct request_queue *q)
 {
 	blk_freeze_queue_start(q);
 	blk_mq_freeze_queue_wait(q);
 }
-EXPORT_SYMBOL_GPL(blk_mq_freeze_queue);
+EXPORT_SYMBOL_GPL(blk_mq_freeze_queue_nomemsave);
 
 bool __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic)
 {
@@ -236,12 +236,12 @@ bool __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic)
 	return unfreeze;
 }
 
-void blk_mq_unfreeze_queue(struct request_queue *q)
+void blk_mq_unfreeze_queue_nomemrestore(struct request_queue *q)
 {
 	if (__blk_mq_unfreeze_queue(q, false))
 		blk_unfreeze_release_lock(q);
 }
-EXPORT_SYMBOL_GPL(blk_mq_unfreeze_queue);
+EXPORT_SYMBOL_GPL(blk_mq_unfreeze_queue_nomemrestore);
 
 /*
  * non_owner variant of blk_freeze_queue_start
@@ -4223,13 +4223,14 @@ static void blk_mq_update_tag_set_shared(struct blk_mq_tag_set *set,
 					 bool shared)
 {
 	struct request_queue *q;
+	unsigned int memflags;
 
 	lockdep_assert_held(&set->tag_list_lock);
 
 	list_for_each_entry(q, &set->tag_list, tag_set_list) {
-		blk_mq_freeze_queue(q);
+		memflags = blk_mq_freeze_queue(q);
 		queue_set_hctx_shared(q, shared);
-		blk_mq_unfreeze_queue(q);
+		blk_mq_unfreeze_queue(q, memflags);
 	}
 }
 
@@ -4992,6 +4993,7 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
 	struct request_queue *q;
 	LIST_HEAD(head);
 	int prev_nr_hw_queues = set->nr_hw_queues;
+	unsigned int memflags;
 	int i;
 
 	lockdep_assert_held(&set->tag_list_lock);
@@ -5003,8 +5005,10 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
 	if (set->nr_maps == 1 && nr_hw_queues == set->nr_hw_queues)
 		return;
 
+	memflags = memalloc_noio_save();
 	list_for_each_entry(q, &set->tag_list, tag_set_list)
-		blk_mq_freeze_queue(q);
+		blk_mq_freeze_queue_nomemsave(q);
+
 	/*
 	 * Switch IO scheduler to 'none', cleaning up the data associated
 	 * with the previous scheduler. We will switch back once we are done
@@ -5052,7 +5056,8 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set,
 		blk_mq_elv_switch_back(&head, q);
 
 	list_for_each_entry(q, &set->tag_list, tag_set_list)
-		blk_mq_unfreeze_queue(q);
+		blk_mq_unfreeze_queue_nomemrestore(q);
+	memalloc_noio_restore(memflags);
 
 	/* Free the excess tags when nr_hw_queues shrink. */
 	for (i = set->nr_hw_queues; i < prev_nr_hw_queues; i++)

diff --git a/block/blk-pm.c b/block/blk-pm.c
@@ -89,7 +89,7 @@ int blk_pre_runtime_suspend(struct request_queue *q)
 	if (percpu_ref_is_zero(&q->q_usage_counter))
 		ret = 0;
 	/* Switch q_usage_counter back to per-cpu mode. */
-	blk_mq_unfreeze_queue(q);
+	blk_mq_unfreeze_queue_nomemrestore(q);
 
 	if (ret < 0) {
 		spin_lock_irq(&q->queue_lock);

diff --git a/block/blk-rq-qos.c b/block/blk-rq-qos.c
@@ -299,6 +299,7 @@ int rq_qos_add(struct rq_qos *rqos, struct gendisk *disk, enum rq_qos_id id,
 		const struct rq_qos_ops *ops)
 {
 	struct request_queue *q = disk->queue;
+	unsigned int memflags;
 
 	lockdep_assert_held(&q->rq_qos_mutex);
 
@@ -310,14 +311,14 @@ int rq_qos_add(struct rq_qos *rqos, struct gendisk *disk, enum rq_qos_id id,
 	 * No IO can be in-flight when adding rqos, so freeze queue, which
 	 * is fine since we only support rq_qos for blk-mq queue.
 	 */
-	blk_mq_freeze_queue(q);
+	memflags = blk_mq_freeze_queue(q);
 
 	if (rq_qos_id(q, rqos->id))
 		goto ebusy;
 	rqos->next = q->rq_qos;
 	q->rq_qos = rqos;
 
-	blk_mq_unfreeze_queue(q);
+	blk_mq_unfreeze_queue(q, memflags);
 
 	if (rqos->ops->debugfs_attrs) {
 		mutex_lock(&q->debugfs_mutex);
@@ -327,25 +328,26 @@ int rq_qos_add(struct rq_qos *rqos, struct gendisk *disk, enum rq_qos_id id,
 
 	return 0;
 ebusy:
-	blk_mq_unfreeze_queue(q);
+	blk_mq_unfreeze_queue(q, memflags);
 	return -EBUSY;
 }
 
 void rq_qos_del(struct rq_qos *rqos)
 {
 	struct request_queue *q = rqos->disk->queue;
 	struct rq_qos **cur;
+	unsigned int memflags;
 
 	lockdep_assert_held(&q->rq_qos_mutex);
 
-	blk_mq_freeze_queue(q);
+	memflags = blk_mq_freeze_queue(q);
 	for (cur = &q->rq_qos; *cur; cur = &(*cur)->next) {
 		if (*cur == rqos) {
 			*cur = rqos->next;
 			break;
 		}
 	}
-	blk_mq_unfreeze_queue(q);
+	blk_mq_unfreeze_queue(q, memflags);
 
 	mutex_lock(&q->debugfs_mutex);
 	blk_mq_debugfs_unregister_rqos(rqos);

diff --git a/block/blk-settings.c b/block/blk-settings.c
@@ -461,11 +461,12 @@ EXPORT_SYMBOL_GPL(queue_limits_commit_update);
 int queue_limits_commit_update_frozen(struct request_queue *q,
 		struct queue_limits *lim)
 {
+	unsigned int memflags;
 	int ret;
 
-	blk_mq_freeze_queue(q);
+	memflags = blk_mq_freeze_queue(q);
 	ret = queue_limits_commit_update(q, lim);
-	blk_mq_unfreeze_queue(q);
+	blk_mq_unfreeze_queue(q, memflags);
 
 	return ret;
 }

diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
@@ -681,7 +681,7 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
 	struct queue_sysfs_entry *entry = to_queue(attr);
 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
 	struct request_queue *q = disk->queue;
-	unsigned int noio_flag;
+	unsigned int memflags;
 	ssize_t res;
 
 	if (!entry->store_limit && !entry->store)
@@ -711,11 +711,9 @@ queue_attr_store(struct kobject *kobj, struct attribute *attr,
 	}
 
 	mutex_lock(&q->sysfs_lock);
-	blk_mq_freeze_queue(q);
-	noio_flag = memalloc_noio_save();
+	memflags = blk_mq_freeze_queue(q);
 	res = entry->store(disk, page, length);
-	memalloc_noio_restore(noio_flag);
-	blk_mq_unfreeze_queue(q);
+	blk_mq_unfreeze_queue(q, memflags);
 	mutex_unlock(&q->sysfs_lock);
 	return res;
 }

diff --git a/block/blk-throttle.c b/block/blk-throttle.c
@@ -1202,6 +1202,7 @@ static int blk_throtl_init(struct gendisk *disk)
 {
 	struct request_queue *q = disk->queue;
 	struct throtl_data *td;
+	unsigned int memflags;
 	int ret;
 
 	td = kzalloc_node(sizeof(*td), GFP_KERNEL, q->node);
@@ -1215,7 +1216,7 @@ static int blk_throtl_init(struct gendisk *disk)
 	 * Freeze queue before activating policy, to synchronize with IO path,
 	 * which is protected by 'q_usage_counter'.
 	 */
-	blk_mq_freeze_queue(disk->queue);
+	memflags = blk_mq_freeze_queue(disk->queue);
 	blk_mq_quiesce_queue(disk->queue);
 
 	q->td = td;
@@ -1239,7 +1240,7 @@ static int blk_throtl_init(struct gendisk *disk)
 
 out:
 	blk_mq_unquiesce_queue(disk->queue);
-	blk_mq_unfreeze_queue(disk->queue);
+	blk_mq_unfreeze_queue(disk->queue, memflags);
 
 	return ret;
 }

diff --git a/block/blk-zoned.c b/block/blk-zoned.c
@@ -1717,9 +1717,10 @@ int blk_revalidate_disk_zones(struct gendisk *disk)
 	else
 		pr_warn("%s: failed to revalidate zones\n", disk->disk_name);
 	if (ret) {
-		blk_mq_freeze_queue(q);
+		unsigned int memflags = blk_mq_freeze_queue(q);
+
 		disk_free_zone_resources(disk);
-		blk_mq_unfreeze_queue(q);
+		blk_mq_unfreeze_queue(q, memflags);
 	}
 
 	return ret;