2 * Copyright (c) 2017 Christoph Hellwig.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 #include <linux/moduleparam.h>
17 static bool multipath = true;
18 module_param(multipath, bool, 0644);
19 MODULE_PARM_DESC(multipath,
20 "turn on native support for multiple controllers per subsystem");
22 void nvme_failover_req(struct request *req)
24 struct nvme_ns *ns = req->q->queuedata;
27 spin_lock_irqsave(&ns->head->requeue_lock, flags);
28 blk_steal_bios(&ns->head->requeue_list, req);
29 spin_unlock_irqrestore(&ns->head->requeue_lock, flags);
30 blk_mq_end_request(req, 0);
32 nvme_reset_ctrl(ns->ctrl);
33 kblockd_schedule_work(&ns->head->requeue_work);
36 bool nvme_req_needs_failover(struct request *req, blk_status_t error)
38 if (!(req->cmd_flags & REQ_NVME_MPATH))
40 return blk_path_error(error);
43 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
47 down_read(&ctrl->namespaces_rwsem);
48 list_for_each_entry(ns, &ctrl->namespaces, list) {
50 kblockd_schedule_work(&ns->head->requeue_work);
52 up_read(&ctrl->namespaces_rwsem);
55 static struct nvme_ns *__nvme_find_path(struct nvme_ns_head *head)
59 list_for_each_entry_rcu(ns, &head->list, siblings) {
60 if (ns->ctrl->state == NVME_CTRL_LIVE) {
61 rcu_assign_pointer(head->current_path, ns);
69 inline struct nvme_ns *nvme_find_path(struct nvme_ns_head *head)
71 struct nvme_ns *ns = srcu_dereference(head->current_path, &head->srcu);
73 if (unlikely(!ns || ns->ctrl->state != NVME_CTRL_LIVE))
74 ns = __nvme_find_path(head);
78 static blk_qc_t nvme_ns_head_make_request(struct request_queue *q,
81 struct nvme_ns_head *head = q->queuedata;
82 struct device *dev = disk_to_dev(head->disk);
84 blk_qc_t ret = BLK_QC_T_NONE;
87 srcu_idx = srcu_read_lock(&head->srcu);
88 ns = nvme_find_path(head);
90 bio->bi_disk = ns->disk;
91 bio->bi_opf |= REQ_NVME_MPATH;
92 ret = direct_make_request(bio);
93 } else if (!list_empty_careful(&head->list)) {
94 dev_warn_ratelimited(dev, "no path available - requeuing I/O\n");
96 spin_lock_irq(&head->requeue_lock);
97 bio_list_add(&head->requeue_list, bio);
98 spin_unlock_irq(&head->requeue_lock);
100 dev_warn_ratelimited(dev, "no path - failing I/O\n");
102 bio->bi_status = BLK_STS_IOERR;
106 srcu_read_unlock(&head->srcu, srcu_idx);
110 static bool nvme_ns_head_poll(struct request_queue *q, blk_qc_t qc)
112 struct nvme_ns_head *head = q->queuedata;
117 srcu_idx = srcu_read_lock(&head->srcu);
118 ns = srcu_dereference(head->current_path, &head->srcu);
119 if (likely(ns && ns->ctrl->state == NVME_CTRL_LIVE))
120 found = ns->queue->poll_fn(q, qc);
121 srcu_read_unlock(&head->srcu, srcu_idx);
125 static void nvme_requeue_work(struct work_struct *work)
127 struct nvme_ns_head *head =
128 container_of(work, struct nvme_ns_head, requeue_work);
129 struct bio *bio, *next;
131 spin_lock_irq(&head->requeue_lock);
132 next = bio_list_get(&head->requeue_list);
133 spin_unlock_irq(&head->requeue_lock);
135 while ((bio = next) != NULL) {
140 * Reset disk to the mpath node and resubmit to select a new
143 bio->bi_disk = head->disk;
144 generic_make_request(bio);
148 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl, struct nvme_ns_head *head)
150 struct request_queue *q;
153 bio_list_init(&head->requeue_list);
154 spin_lock_init(&head->requeue_lock);
155 INIT_WORK(&head->requeue_work, nvme_requeue_work);
158 * Add a multipath node if the subsystems supports multiple controllers.
159 * We also do this for private namespaces as the namespace sharing data could
160 * change after a rescan.
162 if (!(ctrl->subsys->cmic & (1 << 1)) || !multipath)
165 q = blk_alloc_queue_node(GFP_KERNEL, NUMA_NO_NODE, NULL);
169 blk_queue_make_request(q, nvme_ns_head_make_request);
170 q->poll_fn = nvme_ns_head_poll;
171 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
172 /* set to a default value for 512 until disk is validated */
173 blk_queue_logical_block_size(q, 512);
175 /* we need to propagate up the VMC settings */
176 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
178 blk_queue_write_cache(q, vwc, vwc);
180 head->disk = alloc_disk(0);
182 goto out_cleanup_queue;
183 head->disk->fops = &nvme_ns_head_ops;
184 head->disk->private_data = head;
185 head->disk->queue = q;
186 head->disk->flags = GENHD_FL_EXT_DEVT;
187 sprintf(head->disk->disk_name, "nvme%dn%d",
188 ctrl->subsys->instance, head->instance);
192 blk_cleanup_queue(q);
197 void nvme_mpath_add_disk(struct nvme_ns_head *head)
202 mutex_lock(&head->subsys->lock);
203 if (!(head->disk->flags & GENHD_FL_UP)) {
204 device_add_disk(&head->subsys->dev, head->disk);
205 if (sysfs_create_group(&disk_to_dev(head->disk)->kobj,
206 &nvme_ns_id_attr_group))
207 pr_warn("%s: failed to create sysfs group for identification\n",
208 head->disk->disk_name);
210 mutex_unlock(&head->subsys->lock);
213 void nvme_mpath_remove_disk(struct nvme_ns_head *head)
217 sysfs_remove_group(&disk_to_dev(head->disk)->kobj,
218 &nvme_ns_id_attr_group);
219 del_gendisk(head->disk);
220 blk_set_queue_dying(head->disk->queue);
221 /* make sure all pending bios are cleaned up */
222 kblockd_schedule_work(&head->requeue_work);
223 flush_work(&head->requeue_work);
224 blk_cleanup_queue(head->disk->queue);
225 put_disk(head->disk);