1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2017-2018 Christoph Hellwig.
6 #include <linux/backing-dev.h>
7 #include <linux/moduleparam.h>
8 #include <trace/events/block.h>
11 static bool multipath = true;
12 module_param(multipath, bool, 0444);
13 MODULE_PARM_DESC(multipath,
14 "turn on native support for multiple controllers per subsystem");
16 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
18 struct nvme_ns_head *h;
20 lockdep_assert_held(&subsys->lock);
21 list_for_each_entry(h, &subsys->nsheads, entry)
23 blk_mq_unfreeze_queue(h->disk->queue);
26 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
28 struct nvme_ns_head *h;
30 lockdep_assert_held(&subsys->lock);
31 list_for_each_entry(h, &subsys->nsheads, entry)
33 blk_mq_freeze_queue_wait(h->disk->queue);
36 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
38 struct nvme_ns_head *h;
40 lockdep_assert_held(&subsys->lock);
41 list_for_each_entry(h, &subsys->nsheads, entry)
43 blk_freeze_queue_start(h->disk->queue);
47 * If multipathing is enabled we need to always use the subsystem instance
48 * number for numbering our devices to avoid conflicts between subsystems that
49 * have multiple controllers and thus use the multipath-aware subsystem node
50 * and those that have a single controller and use the controller node
53 bool nvme_mpath_set_disk_name(struct nvme_ns *ns, char *disk_name, int *flags)
57 if (!ns->head->disk) {
58 sprintf(disk_name, "nvme%dn%d", ns->ctrl->subsys->instance,
62 sprintf(disk_name, "nvme%dc%dn%d", ns->ctrl->subsys->instance,
63 ns->ctrl->instance, ns->head->instance);
64 *flags = GENHD_FL_HIDDEN;
68 void nvme_failover_req(struct request *req)
70 struct nvme_ns *ns = req->q->queuedata;
71 u16 status = nvme_req(req)->status & 0x7ff;
75 nvme_mpath_clear_current_path(ns);
78 * If we got back an ANA error, we know the controller is alive but not
79 * ready to serve this namespace. Kick of a re-read of the ANA
80 * information page, and just try any other available path for now.
82 if (nvme_is_ana_error(status) && ns->ctrl->ana_log_buf) {
83 set_bit(NVME_NS_ANA_PENDING, &ns->flags);
84 queue_work(nvme_wq, &ns->ctrl->ana_work);
87 spin_lock_irqsave(&ns->head->requeue_lock, flags);
88 for (bio = req->bio; bio; bio = bio->bi_next) {
89 bio_set_dev(bio, ns->head->disk->part0);
90 if (bio->bi_opf & REQ_POLLED) {
91 bio->bi_opf &= ~REQ_POLLED;
92 bio->bi_cookie = BLK_QC_T_NONE;
95 blk_steal_bios(&ns->head->requeue_list, req);
96 spin_unlock_irqrestore(&ns->head->requeue_lock, flags);
98 blk_mq_end_request(req, 0);
99 kblockd_schedule_work(&ns->head->requeue_work);
102 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
106 down_read(&ctrl->namespaces_rwsem);
107 list_for_each_entry(ns, &ctrl->namespaces, list) {
110 kblockd_schedule_work(&ns->head->requeue_work);
111 if (ctrl->state == NVME_CTRL_LIVE)
112 disk_uevent(ns->head->disk, KOBJ_CHANGE);
114 up_read(&ctrl->namespaces_rwsem);
117 static const char *nvme_ana_state_names[] = {
118 [0] = "invalid state",
119 [NVME_ANA_OPTIMIZED] = "optimized",
120 [NVME_ANA_NONOPTIMIZED] = "non-optimized",
121 [NVME_ANA_INACCESSIBLE] = "inaccessible",
122 [NVME_ANA_PERSISTENT_LOSS] = "persistent-loss",
123 [NVME_ANA_CHANGE] = "change",
126 bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
128 struct nvme_ns_head *head = ns->head;
129 bool changed = false;
135 for_each_node(node) {
136 if (ns == rcu_access_pointer(head->current_path[node])) {
137 rcu_assign_pointer(head->current_path[node], NULL);
145 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
149 down_read(&ctrl->namespaces_rwsem);
150 list_for_each_entry(ns, &ctrl->namespaces, list) {
151 nvme_mpath_clear_current_path(ns);
152 kblockd_schedule_work(&ns->head->requeue_work);
154 up_read(&ctrl->namespaces_rwsem);
157 void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
159 struct nvme_ns_head *head = ns->head;
160 sector_t capacity = get_capacity(head->disk);
163 list_for_each_entry_rcu(ns, &head->list, siblings) {
164 if (capacity != get_capacity(ns->disk))
165 clear_bit(NVME_NS_READY, &ns->flags);
169 rcu_assign_pointer(head->current_path[node], NULL);
172 static bool nvme_path_is_disabled(struct nvme_ns *ns)
175 * We don't treat NVME_CTRL_DELETING as a disabled path as I/O should
176 * still be able to complete assuming that the controller is connected.
177 * Otherwise it will fail immediately and return to the requeue list.
179 if (ns->ctrl->state != NVME_CTRL_LIVE &&
180 ns->ctrl->state != NVME_CTRL_DELETING)
182 if (test_bit(NVME_NS_ANA_PENDING, &ns->flags) ||
183 !test_bit(NVME_NS_READY, &ns->flags))
188 static struct nvme_ns *__nvme_find_path(struct nvme_ns_head *head, int node)
190 int found_distance = INT_MAX, fallback_distance = INT_MAX, distance;
191 struct nvme_ns *found = NULL, *fallback = NULL, *ns;
193 list_for_each_entry_rcu(ns, &head->list, siblings) {
194 if (nvme_path_is_disabled(ns))
197 if (READ_ONCE(head->subsys->iopolicy) == NVME_IOPOLICY_NUMA)
198 distance = node_distance(node, ns->ctrl->numa_node);
200 distance = LOCAL_DISTANCE;
202 switch (ns->ana_state) {
203 case NVME_ANA_OPTIMIZED:
204 if (distance < found_distance) {
205 found_distance = distance;
209 case NVME_ANA_NONOPTIMIZED:
210 if (distance < fallback_distance) {
211 fallback_distance = distance;
223 rcu_assign_pointer(head->current_path[node], found);
227 static struct nvme_ns *nvme_next_ns(struct nvme_ns_head *head,
230 ns = list_next_or_null_rcu(&head->list, &ns->siblings, struct nvme_ns,
234 return list_first_or_null_rcu(&head->list, struct nvme_ns, siblings);
237 static struct nvme_ns *nvme_round_robin_path(struct nvme_ns_head *head,
238 int node, struct nvme_ns *old)
240 struct nvme_ns *ns, *found = NULL;
242 if (list_is_singular(&head->list)) {
243 if (nvme_path_is_disabled(old))
248 for (ns = nvme_next_ns(head, old);
250 ns = nvme_next_ns(head, ns)) {
251 if (nvme_path_is_disabled(ns))
254 if (ns->ana_state == NVME_ANA_OPTIMIZED) {
258 if (ns->ana_state == NVME_ANA_NONOPTIMIZED)
263 * The loop above skips the current path for round-robin semantics.
264 * Fall back to the current path if either:
265 * - no other optimized path found and current is optimized,
266 * - no other usable path found and current is usable.
268 if (!nvme_path_is_disabled(old) &&
269 (old->ana_state == NVME_ANA_OPTIMIZED ||
270 (!found && old->ana_state == NVME_ANA_NONOPTIMIZED)))
276 rcu_assign_pointer(head->current_path[node], found);
280 static inline bool nvme_path_is_optimized(struct nvme_ns *ns)
282 return ns->ctrl->state == NVME_CTRL_LIVE &&
283 ns->ana_state == NVME_ANA_OPTIMIZED;
286 inline struct nvme_ns *nvme_find_path(struct nvme_ns_head *head)
288 int node = numa_node_id();
291 ns = srcu_dereference(head->current_path[node], &head->srcu);
293 return __nvme_find_path(head, node);
295 if (READ_ONCE(head->subsys->iopolicy) == NVME_IOPOLICY_RR)
296 return nvme_round_robin_path(head, node, ns);
297 if (unlikely(!nvme_path_is_optimized(ns)))
298 return __nvme_find_path(head, node);
302 static bool nvme_available_path(struct nvme_ns_head *head)
306 list_for_each_entry_rcu(ns, &head->list, siblings) {
307 if (test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ns->ctrl->flags))
309 switch (ns->ctrl->state) {
311 case NVME_CTRL_RESETTING:
312 case NVME_CTRL_CONNECTING:
322 static void nvme_ns_head_submit_bio(struct bio *bio)
324 struct nvme_ns_head *head = bio->bi_bdev->bd_disk->private_data;
325 struct device *dev = disk_to_dev(head->disk);
330 * The namespace might be going away and the bio might be moved to a
331 * different queue via blk_steal_bios(), so we need to use the bio_split
332 * pool from the original queue to allocate the bvecs from.
334 blk_queue_split(&bio);
336 srcu_idx = srcu_read_lock(&head->srcu);
337 ns = nvme_find_path(head);
339 bio_set_dev(bio, ns->disk->part0);
340 bio->bi_opf |= REQ_NVME_MPATH;
341 trace_block_bio_remap(bio, disk_devt(ns->head->disk),
342 bio->bi_iter.bi_sector);
343 submit_bio_noacct(bio);
344 } else if (nvme_available_path(head)) {
345 dev_warn_ratelimited(dev, "no usable path - requeuing I/O\n");
347 spin_lock_irq(&head->requeue_lock);
348 bio_list_add(&head->requeue_list, bio);
349 spin_unlock_irq(&head->requeue_lock);
351 dev_warn_ratelimited(dev, "no available path - failing I/O\n");
353 bio->bi_status = BLK_STS_IOERR;
357 srcu_read_unlock(&head->srcu, srcu_idx);
360 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
362 if (!nvme_tryget_ns_head(bdev->bd_disk->private_data))
367 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
369 nvme_put_ns_head(disk->private_data);
372 #ifdef CONFIG_BLK_DEV_ZONED
373 static int nvme_ns_head_report_zones(struct gendisk *disk, sector_t sector,
374 unsigned int nr_zones, report_zones_cb cb, void *data)
376 struct nvme_ns_head *head = disk->private_data;
378 int srcu_idx, ret = -EWOULDBLOCK;
380 srcu_idx = srcu_read_lock(&head->srcu);
381 ns = nvme_find_path(head);
383 ret = nvme_ns_report_zones(ns, sector, nr_zones, cb, data);
384 srcu_read_unlock(&head->srcu, srcu_idx);
388 #define nvme_ns_head_report_zones NULL
389 #endif /* CONFIG_BLK_DEV_ZONED */
391 const struct block_device_operations nvme_ns_head_ops = {
392 .owner = THIS_MODULE,
393 .submit_bio = nvme_ns_head_submit_bio,
394 .open = nvme_ns_head_open,
395 .release = nvme_ns_head_release,
396 .ioctl = nvme_ns_head_ioctl,
397 .getgeo = nvme_getgeo,
398 .report_zones = nvme_ns_head_report_zones,
399 .pr_ops = &nvme_pr_ops,
402 static inline struct nvme_ns_head *cdev_to_ns_head(struct cdev *cdev)
404 return container_of(cdev, struct nvme_ns_head, cdev);
407 static int nvme_ns_head_chr_open(struct inode *inode, struct file *file)
409 if (!nvme_tryget_ns_head(cdev_to_ns_head(inode->i_cdev)))
414 static int nvme_ns_head_chr_release(struct inode *inode, struct file *file)
416 nvme_put_ns_head(cdev_to_ns_head(inode->i_cdev));
420 static const struct file_operations nvme_ns_head_chr_fops = {
421 .owner = THIS_MODULE,
422 .open = nvme_ns_head_chr_open,
423 .release = nvme_ns_head_chr_release,
424 .unlocked_ioctl = nvme_ns_head_chr_ioctl,
425 .compat_ioctl = compat_ptr_ioctl,
428 static int nvme_add_ns_head_cdev(struct nvme_ns_head *head)
432 head->cdev_device.parent = &head->subsys->dev;
433 ret = dev_set_name(&head->cdev_device, "ng%dn%d",
434 head->subsys->instance, head->instance);
437 ret = nvme_cdev_add(&head->cdev, &head->cdev_device,
438 &nvme_ns_head_chr_fops, THIS_MODULE);
442 static void nvme_requeue_work(struct work_struct *work)
444 struct nvme_ns_head *head =
445 container_of(work, struct nvme_ns_head, requeue_work);
446 struct bio *bio, *next;
448 spin_lock_irq(&head->requeue_lock);
449 next = bio_list_get(&head->requeue_list);
450 spin_unlock_irq(&head->requeue_lock);
452 while ((bio = next) != NULL) {
456 submit_bio_noacct(bio);
460 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl, struct nvme_ns_head *head)
464 mutex_init(&head->lock);
465 bio_list_init(&head->requeue_list);
466 spin_lock_init(&head->requeue_lock);
467 INIT_WORK(&head->requeue_work, nvme_requeue_work);
470 * Add a multipath node if the subsystems supports multiple controllers.
471 * We also do this for private namespaces as the namespace sharing data could
472 * change after a rescan.
474 if (!(ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) || !multipath)
477 head->disk = blk_alloc_disk(ctrl->numa_node);
480 head->disk->fops = &nvme_ns_head_ops;
481 head->disk->private_data = head;
482 sprintf(head->disk->disk_name, "nvme%dn%d",
483 ctrl->subsys->instance, head->instance);
485 blk_queue_flag_set(QUEUE_FLAG_NONROT, head->disk->queue);
486 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, head->disk->queue);
488 * This assumes all controllers that refer to a namespace either
489 * support poll queues or not. That is not a strict guarantee,
490 * but if the assumption is wrong the effect is only suboptimal
491 * performance but not correctness problem.
493 if (ctrl->tagset->nr_maps > HCTX_TYPE_POLL &&
494 ctrl->tagset->map[HCTX_TYPE_POLL].nr_queues)
495 blk_queue_flag_set(QUEUE_FLAG_POLL, head->disk->queue);
497 /* set to a default value of 512 until the disk is validated */
498 blk_queue_logical_block_size(head->disk->queue, 512);
499 blk_set_stacking_limits(&head->disk->queue->limits);
501 /* we need to propagate up the VMC settings */
502 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
504 blk_queue_write_cache(head->disk->queue, vwc, vwc);
508 static void nvme_mpath_set_live(struct nvme_ns *ns)
510 struct nvme_ns_head *head = ns->head;
517 * test_and_set_bit() is used because it is protecting against two nvme
518 * paths simultaneously calling device_add_disk() on the same namespace
521 if (!test_and_set_bit(NVME_NSHEAD_DISK_LIVE, &head->flags)) {
522 rc = device_add_disk(&head->subsys->dev, head->disk,
523 nvme_ns_id_attr_groups);
525 clear_bit(NVME_NSHEAD_DISK_LIVE, &ns->flags);
528 nvme_add_ns_head_cdev(head);
531 mutex_lock(&head->lock);
532 if (nvme_path_is_optimized(ns)) {
535 srcu_idx = srcu_read_lock(&head->srcu);
537 __nvme_find_path(head, node);
538 srcu_read_unlock(&head->srcu, srcu_idx);
540 mutex_unlock(&head->lock);
542 synchronize_srcu(&head->srcu);
543 kblockd_schedule_work(&head->requeue_work);
546 static int nvme_parse_ana_log(struct nvme_ctrl *ctrl, void *data,
547 int (*cb)(struct nvme_ctrl *ctrl, struct nvme_ana_group_desc *,
550 void *base = ctrl->ana_log_buf;
551 size_t offset = sizeof(struct nvme_ana_rsp_hdr);
554 lockdep_assert_held(&ctrl->ana_lock);
556 for (i = 0; i < le16_to_cpu(ctrl->ana_log_buf->ngrps); i++) {
557 struct nvme_ana_group_desc *desc = base + offset;
559 size_t nsid_buf_size;
561 if (WARN_ON_ONCE(offset > ctrl->ana_log_size - sizeof(*desc)))
564 nr_nsids = le32_to_cpu(desc->nnsids);
565 nsid_buf_size = flex_array_size(desc, nsids, nr_nsids);
567 if (WARN_ON_ONCE(desc->grpid == 0))
569 if (WARN_ON_ONCE(le32_to_cpu(desc->grpid) > ctrl->anagrpmax))
571 if (WARN_ON_ONCE(desc->state == 0))
573 if (WARN_ON_ONCE(desc->state > NVME_ANA_CHANGE))
576 offset += sizeof(*desc);
577 if (WARN_ON_ONCE(offset > ctrl->ana_log_size - nsid_buf_size))
580 error = cb(ctrl, desc, data);
584 offset += nsid_buf_size;
590 static inline bool nvme_state_is_live(enum nvme_ana_state state)
592 return state == NVME_ANA_OPTIMIZED || state == NVME_ANA_NONOPTIMIZED;
595 static void nvme_update_ns_ana_state(struct nvme_ana_group_desc *desc,
598 ns->ana_grpid = le32_to_cpu(desc->grpid);
599 ns->ana_state = desc->state;
600 clear_bit(NVME_NS_ANA_PENDING, &ns->flags);
602 if (nvme_state_is_live(ns->ana_state))
603 nvme_mpath_set_live(ns);
606 static int nvme_update_ana_state(struct nvme_ctrl *ctrl,
607 struct nvme_ana_group_desc *desc, void *data)
609 u32 nr_nsids = le32_to_cpu(desc->nnsids), n = 0;
610 unsigned *nr_change_groups = data;
613 dev_dbg(ctrl->device, "ANA group %d: %s.\n",
614 le32_to_cpu(desc->grpid),
615 nvme_ana_state_names[desc->state]);
617 if (desc->state == NVME_ANA_CHANGE)
618 (*nr_change_groups)++;
623 down_read(&ctrl->namespaces_rwsem);
624 list_for_each_entry(ns, &ctrl->namespaces, list) {
627 nsid = le32_to_cpu(desc->nsids[n]);
628 if (ns->head->ns_id < nsid)
630 if (ns->head->ns_id == nsid)
631 nvme_update_ns_ana_state(desc, ns);
634 if (ns->head->ns_id > nsid)
637 up_read(&ctrl->namespaces_rwsem);
641 static int nvme_read_ana_log(struct nvme_ctrl *ctrl)
643 u32 nr_change_groups = 0;
646 mutex_lock(&ctrl->ana_lock);
647 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_ANA, 0, NVME_CSI_NVM,
648 ctrl->ana_log_buf, ctrl->ana_log_size, 0);
650 dev_warn(ctrl->device, "Failed to get ANA log: %d\n", error);
654 error = nvme_parse_ana_log(ctrl, &nr_change_groups,
655 nvme_update_ana_state);
660 * In theory we should have an ANATT timer per group as they might enter
661 * the change state at different times. But that is a lot of overhead
662 * just to protect against a target that keeps entering new changes
663 * states while never finishing previous ones. But we'll still
664 * eventually time out once all groups are in change state, so this
667 * We also double the ANATT value to provide some slack for transports
668 * or AEN processing overhead.
670 if (nr_change_groups)
671 mod_timer(&ctrl->anatt_timer, ctrl->anatt * HZ * 2 + jiffies);
673 del_timer_sync(&ctrl->anatt_timer);
675 mutex_unlock(&ctrl->ana_lock);
679 static void nvme_ana_work(struct work_struct *work)
681 struct nvme_ctrl *ctrl = container_of(work, struct nvme_ctrl, ana_work);
683 if (ctrl->state != NVME_CTRL_LIVE)
686 nvme_read_ana_log(ctrl);
689 static void nvme_anatt_timeout(struct timer_list *t)
691 struct nvme_ctrl *ctrl = from_timer(ctrl, t, anatt_timer);
693 dev_info(ctrl->device, "ANATT timeout, resetting controller.\n");
694 nvme_reset_ctrl(ctrl);
697 void nvme_mpath_stop(struct nvme_ctrl *ctrl)
699 if (!nvme_ctrl_use_ana(ctrl))
701 del_timer_sync(&ctrl->anatt_timer);
702 cancel_work_sync(&ctrl->ana_work);
705 #define SUBSYS_ATTR_RW(_name, _mode, _show, _store) \
706 struct device_attribute subsys_attr_##_name = \
707 __ATTR(_name, _mode, _show, _store)
709 static const char *nvme_iopolicy_names[] = {
710 [NVME_IOPOLICY_NUMA] = "numa",
711 [NVME_IOPOLICY_RR] = "round-robin",
714 static ssize_t nvme_subsys_iopolicy_show(struct device *dev,
715 struct device_attribute *attr, char *buf)
717 struct nvme_subsystem *subsys =
718 container_of(dev, struct nvme_subsystem, dev);
720 return sysfs_emit(buf, "%s\n",
721 nvme_iopolicy_names[READ_ONCE(subsys->iopolicy)]);
724 static ssize_t nvme_subsys_iopolicy_store(struct device *dev,
725 struct device_attribute *attr, const char *buf, size_t count)
727 struct nvme_subsystem *subsys =
728 container_of(dev, struct nvme_subsystem, dev);
731 for (i = 0; i < ARRAY_SIZE(nvme_iopolicy_names); i++) {
732 if (sysfs_streq(buf, nvme_iopolicy_names[i])) {
733 WRITE_ONCE(subsys->iopolicy, i);
740 SUBSYS_ATTR_RW(iopolicy, S_IRUGO | S_IWUSR,
741 nvme_subsys_iopolicy_show, nvme_subsys_iopolicy_store);
743 static ssize_t ana_grpid_show(struct device *dev, struct device_attribute *attr,
746 return sysfs_emit(buf, "%d\n", nvme_get_ns_from_dev(dev)->ana_grpid);
748 DEVICE_ATTR_RO(ana_grpid);
750 static ssize_t ana_state_show(struct device *dev, struct device_attribute *attr,
753 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
755 return sysfs_emit(buf, "%s\n", nvme_ana_state_names[ns->ana_state]);
757 DEVICE_ATTR_RO(ana_state);
759 static int nvme_lookup_ana_group_desc(struct nvme_ctrl *ctrl,
760 struct nvme_ana_group_desc *desc, void *data)
762 struct nvme_ana_group_desc *dst = data;
764 if (desc->grpid != dst->grpid)
768 return -ENXIO; /* just break out of the loop */
771 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id)
773 if (nvme_ctrl_use_ana(ns->ctrl)) {
774 struct nvme_ana_group_desc desc = {
775 .grpid = id->anagrpid,
779 mutex_lock(&ns->ctrl->ana_lock);
780 ns->ana_grpid = le32_to_cpu(id->anagrpid);
781 nvme_parse_ana_log(ns->ctrl, &desc, nvme_lookup_ana_group_desc);
782 mutex_unlock(&ns->ctrl->ana_lock);
784 /* found the group desc: update */
785 nvme_update_ns_ana_state(&desc, ns);
787 /* group desc not found: trigger a re-read */
788 set_bit(NVME_NS_ANA_PENDING, &ns->flags);
789 queue_work(nvme_wq, &ns->ctrl->ana_work);
792 ns->ana_state = NVME_ANA_OPTIMIZED;
793 nvme_mpath_set_live(ns);
796 if (blk_queue_stable_writes(ns->queue) && ns->head->disk)
797 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES,
798 ns->head->disk->queue);
799 #ifdef CONFIG_BLK_DEV_ZONED
800 if (blk_queue_is_zoned(ns->queue) && ns->head->disk)
801 ns->head->disk->queue->nr_zones = ns->queue->nr_zones;
805 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
809 kblockd_schedule_work(&head->requeue_work);
810 if (test_bit(NVME_NSHEAD_DISK_LIVE, &head->flags)) {
811 nvme_cdev_del(&head->cdev, &head->cdev_device);
812 del_gendisk(head->disk);
816 void nvme_mpath_remove_disk(struct nvme_ns_head *head)
820 blk_set_queue_dying(head->disk->queue);
821 /* make sure all pending bios are cleaned up */
822 kblockd_schedule_work(&head->requeue_work);
823 flush_work(&head->requeue_work);
824 blk_cleanup_disk(head->disk);
827 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
829 mutex_init(&ctrl->ana_lock);
830 timer_setup(&ctrl->anatt_timer, nvme_anatt_timeout, 0);
831 INIT_WORK(&ctrl->ana_work, nvme_ana_work);
834 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
836 size_t max_transfer_size = ctrl->max_hw_sectors << SECTOR_SHIFT;
840 /* check if multipath is enabled and we have the capability */
841 if (!multipath || !ctrl->subsys ||
842 !(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA))
845 if (!ctrl->max_namespaces ||
846 ctrl->max_namespaces > le32_to_cpu(id->nn)) {
847 dev_err(ctrl->device,
848 "Invalid MNAN value %u\n", ctrl->max_namespaces);
852 ctrl->anacap = id->anacap;
853 ctrl->anatt = id->anatt;
854 ctrl->nanagrpid = le32_to_cpu(id->nanagrpid);
855 ctrl->anagrpmax = le32_to_cpu(id->anagrpmax);
857 ana_log_size = sizeof(struct nvme_ana_rsp_hdr) +
858 ctrl->nanagrpid * sizeof(struct nvme_ana_group_desc) +
859 ctrl->max_namespaces * sizeof(__le32);
860 if (ana_log_size > max_transfer_size) {
861 dev_err(ctrl->device,
862 "ANA log page size (%zd) larger than MDTS (%zd).\n",
863 ana_log_size, max_transfer_size);
864 dev_err(ctrl->device, "disabling ANA support.\n");
867 if (ana_log_size > ctrl->ana_log_size) {
868 nvme_mpath_stop(ctrl);
869 kfree(ctrl->ana_log_buf);
870 ctrl->ana_log_buf = kmalloc(ana_log_size, GFP_KERNEL);
871 if (!ctrl->ana_log_buf)
874 ctrl->ana_log_size = ana_log_size;
875 error = nvme_read_ana_log(ctrl);
881 nvme_mpath_uninit(ctrl);
885 void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
887 kfree(ctrl->ana_log_buf);
888 ctrl->ana_log_buf = NULL;