xref: /openbmc/linux/drivers/nvme/host/multipath.c (revision 884caada)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2017-2018 Christoph Hellwig.
4  */
5 
6 #include <linux/moduleparam.h>
7 #include <trace/events/block.h>
8 #include "nvme.h"
9 
10 static bool multipath = true;
11 module_param(multipath, bool, 0444);
12 MODULE_PARM_DESC(multipath,
13 	"turn on native support for multiple controllers per subsystem");
14 
15 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
16 {
17 	struct nvme_ns_head *h;
18 
19 	lockdep_assert_held(&subsys->lock);
20 	list_for_each_entry(h, &subsys->nsheads, entry)
21 		if (h->disk)
22 			blk_mq_unfreeze_queue(h->disk->queue);
23 }
24 
25 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
26 {
27 	struct nvme_ns_head *h;
28 
29 	lockdep_assert_held(&subsys->lock);
30 	list_for_each_entry(h, &subsys->nsheads, entry)
31 		if (h->disk)
32 			blk_mq_freeze_queue_wait(h->disk->queue);
33 }
34 
35 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
36 {
37 	struct nvme_ns_head *h;
38 
39 	lockdep_assert_held(&subsys->lock);
40 	list_for_each_entry(h, &subsys->nsheads, entry)
41 		if (h->disk)
42 			blk_freeze_queue_start(h->disk->queue);
43 }
44 
45 /*
46  * If multipathing is enabled we need to always use the subsystem instance
47  * number for numbering our devices to avoid conflicts between subsystems that
48  * have multiple controllers and thus use the multipath-aware subsystem node
49  * and those that have a single controller and use the controller node
50  * directly.
51  */
52 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
53 			struct nvme_ctrl *ctrl, int *flags)
54 {
55 	if (!multipath) {
56 		sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
57 	} else if (ns->head->disk) {
58 		sprintf(disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
59 				ctrl->instance, ns->head->instance);
60 		*flags = GENHD_FL_HIDDEN;
61 	} else {
62 		sprintf(disk_name, "nvme%dn%d", ctrl->subsys->instance,
63 				ns->head->instance);
64 	}
65 }
66 
67 void nvme_failover_req(struct request *req)
68 {
69 	struct nvme_ns *ns = req->q->queuedata;
70 	u16 status = nvme_req(req)->status;
71 	unsigned long flags;
72 
73 	spin_lock_irqsave(&ns->head->requeue_lock, flags);
74 	blk_steal_bios(&ns->head->requeue_list, req);
75 	spin_unlock_irqrestore(&ns->head->requeue_lock, flags);
76 	blk_mq_end_request(req, 0);
77 
78 	switch (status & 0x7ff) {
79 	case NVME_SC_ANA_TRANSITION:
80 	case NVME_SC_ANA_INACCESSIBLE:
81 	case NVME_SC_ANA_PERSISTENT_LOSS:
82 		/*
83 		 * If we got back an ANA error we know the controller is alive,
84 		 * but not ready to serve this namespaces.  The spec suggests
85 		 * we should update our general state here, but due to the fact
86 		 * that the admin and I/O queues are not serialized that is
87 		 * fundamentally racy.  So instead just clear the current path,
88 		 * mark the the path as pending and kick of a re-read of the ANA
89 		 * log page ASAP.
90 		 */
91 		nvme_mpath_clear_current_path(ns);
92 		if (ns->ctrl->ana_log_buf) {
93 			set_bit(NVME_NS_ANA_PENDING, &ns->flags);
94 			queue_work(nvme_wq, &ns->ctrl->ana_work);
95 		}
96 		break;
97 	case NVME_SC_HOST_PATH_ERROR:
98 		/*
99 		 * Temporary transport disruption in talking to the controller.
100 		 * Try to send on a new path.
101 		 */
102 		nvme_mpath_clear_current_path(ns);
103 		break;
104 	default:
105 		/*
106 		 * Reset the controller for any non-ANA error as we don't know
107 		 * what caused the error.
108 		 */
109 		nvme_reset_ctrl(ns->ctrl);
110 		break;
111 	}
112 
113 	kblockd_schedule_work(&ns->head->requeue_work);
114 }
115 
116 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
117 {
118 	struct nvme_ns *ns;
119 
120 	down_read(&ctrl->namespaces_rwsem);
121 	list_for_each_entry(ns, &ctrl->namespaces, list) {
122 		if (ns->head->disk)
123 			kblockd_schedule_work(&ns->head->requeue_work);
124 	}
125 	up_read(&ctrl->namespaces_rwsem);
126 }
127 
128 static const char *nvme_ana_state_names[] = {
129 	[0]				= "invalid state",
130 	[NVME_ANA_OPTIMIZED]		= "optimized",
131 	[NVME_ANA_NONOPTIMIZED]		= "non-optimized",
132 	[NVME_ANA_INACCESSIBLE]		= "inaccessible",
133 	[NVME_ANA_PERSISTENT_LOSS]	= "persistent-loss",
134 	[NVME_ANA_CHANGE]		= "change",
135 };
136 
137 bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
138 {
139 	struct nvme_ns_head *head = ns->head;
140 	bool changed = false;
141 	int node;
142 
143 	if (!head)
144 		goto out;
145 
146 	for_each_node(node) {
147 		if (ns == rcu_access_pointer(head->current_path[node])) {
148 			rcu_assign_pointer(head->current_path[node], NULL);
149 			changed = true;
150 		}
151 	}
152 out:
153 	return changed;
154 }
155 
156 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
157 {
158 	struct nvme_ns *ns;
159 
160 	mutex_lock(&ctrl->scan_lock);
161 	list_for_each_entry(ns, &ctrl->namespaces, list)
162 		if (nvme_mpath_clear_current_path(ns))
163 			kblockd_schedule_work(&ns->head->requeue_work);
164 	mutex_unlock(&ctrl->scan_lock);
165 }
166 
167 static bool nvme_path_is_disabled(struct nvme_ns *ns)
168 {
169 	return ns->ctrl->state != NVME_CTRL_LIVE ||
170 		test_bit(NVME_NS_ANA_PENDING, &ns->flags) ||
171 		test_bit(NVME_NS_REMOVING, &ns->flags);
172 }
173 
174 static struct nvme_ns *__nvme_find_path(struct nvme_ns_head *head, int node)
175 {
176 	int found_distance = INT_MAX, fallback_distance = INT_MAX, distance;
177 	struct nvme_ns *found = NULL, *fallback = NULL, *ns;
178 
179 	list_for_each_entry_rcu(ns, &head->list, siblings) {
180 		if (nvme_path_is_disabled(ns))
181 			continue;
182 
183 		if (READ_ONCE(head->subsys->iopolicy) == NVME_IOPOLICY_NUMA)
184 			distance = node_distance(node, ns->ctrl->numa_node);
185 		else
186 			distance = LOCAL_DISTANCE;
187 
188 		switch (ns->ana_state) {
189 		case NVME_ANA_OPTIMIZED:
190 			if (distance < found_distance) {
191 				found_distance = distance;
192 				found = ns;
193 			}
194 			break;
195 		case NVME_ANA_NONOPTIMIZED:
196 			if (distance < fallback_distance) {
197 				fallback_distance = distance;
198 				fallback = ns;
199 			}
200 			break;
201 		default:
202 			break;
203 		}
204 	}
205 
206 	if (!found)
207 		found = fallback;
208 	if (found)
209 		rcu_assign_pointer(head->current_path[node], found);
210 	return found;
211 }
212 
213 static struct nvme_ns *nvme_next_ns(struct nvme_ns_head *head,
214 		struct nvme_ns *ns)
215 {
216 	ns = list_next_or_null_rcu(&head->list, &ns->siblings, struct nvme_ns,
217 			siblings);
218 	if (ns)
219 		return ns;
220 	return list_first_or_null_rcu(&head->list, struct nvme_ns, siblings);
221 }
222 
223 static struct nvme_ns *nvme_round_robin_path(struct nvme_ns_head *head,
224 		int node, struct nvme_ns *old)
225 {
226 	struct nvme_ns *ns, *found, *fallback = NULL;
227 
228 	if (list_is_singular(&head->list)) {
229 		if (nvme_path_is_disabled(old))
230 			return NULL;
231 		return old;
232 	}
233 
234 	for (ns = nvme_next_ns(head, old);
235 	     ns != old;
236 	     ns = nvme_next_ns(head, ns)) {
237 		if (nvme_path_is_disabled(ns))
238 			continue;
239 
240 		if (ns->ana_state == NVME_ANA_OPTIMIZED) {
241 			found = ns;
242 			goto out;
243 		}
244 		if (ns->ana_state == NVME_ANA_NONOPTIMIZED)
245 			fallback = ns;
246 	}
247 
248 	if (!fallback)
249 		return NULL;
250 	found = fallback;
251 out:
252 	rcu_assign_pointer(head->current_path[node], found);
253 	return found;
254 }
255 
256 static inline bool nvme_path_is_optimized(struct nvme_ns *ns)
257 {
258 	return ns->ctrl->state == NVME_CTRL_LIVE &&
259 		ns->ana_state == NVME_ANA_OPTIMIZED;
260 }
261 
262 inline struct nvme_ns *nvme_find_path(struct nvme_ns_head *head)
263 {
264 	int node = numa_node_id();
265 	struct nvme_ns *ns;
266 
267 	ns = srcu_dereference(head->current_path[node], &head->srcu);
268 	if (READ_ONCE(head->subsys->iopolicy) == NVME_IOPOLICY_RR && ns)
269 		ns = nvme_round_robin_path(head, node, ns);
270 	if (unlikely(!ns || !nvme_path_is_optimized(ns)))
271 		ns = __nvme_find_path(head, node);
272 	return ns;
273 }
274 
275 static bool nvme_available_path(struct nvme_ns_head *head)
276 {
277 	struct nvme_ns *ns;
278 
279 	list_for_each_entry_rcu(ns, &head->list, siblings) {
280 		switch (ns->ctrl->state) {
281 		case NVME_CTRL_LIVE:
282 		case NVME_CTRL_RESETTING:
283 		case NVME_CTRL_CONNECTING:
284 			/* fallthru */
285 			return true;
286 		default:
287 			break;
288 		}
289 	}
290 	return false;
291 }
292 
293 static blk_qc_t nvme_ns_head_make_request(struct request_queue *q,
294 		struct bio *bio)
295 {
296 	struct nvme_ns_head *head = q->queuedata;
297 	struct device *dev = disk_to_dev(head->disk);
298 	struct nvme_ns *ns;
299 	blk_qc_t ret = BLK_QC_T_NONE;
300 	int srcu_idx;
301 
302 	/*
303 	 * The namespace might be going away and the bio might
304 	 * be moved to a different queue via blk_steal_bios(),
305 	 * so we need to use the bio_split pool from the original
306 	 * queue to allocate the bvecs from.
307 	 */
308 	blk_queue_split(q, &bio);
309 
310 	srcu_idx = srcu_read_lock(&head->srcu);
311 	ns = nvme_find_path(head);
312 	if (likely(ns)) {
313 		bio->bi_disk = ns->disk;
314 		bio->bi_opf |= REQ_NVME_MPATH;
315 		trace_block_bio_remap(bio->bi_disk->queue, bio,
316 				      disk_devt(ns->head->disk),
317 				      bio->bi_iter.bi_sector);
318 		ret = direct_make_request(bio);
319 	} else if (nvme_available_path(head)) {
320 		dev_warn_ratelimited(dev, "no usable path - requeuing I/O\n");
321 
322 		spin_lock_irq(&head->requeue_lock);
323 		bio_list_add(&head->requeue_list, bio);
324 		spin_unlock_irq(&head->requeue_lock);
325 	} else {
326 		dev_warn_ratelimited(dev, "no available path - failing I/O\n");
327 
328 		bio->bi_status = BLK_STS_IOERR;
329 		bio_endio(bio);
330 	}
331 
332 	srcu_read_unlock(&head->srcu, srcu_idx);
333 	return ret;
334 }
335 
336 static void nvme_requeue_work(struct work_struct *work)
337 {
338 	struct nvme_ns_head *head =
339 		container_of(work, struct nvme_ns_head, requeue_work);
340 	struct bio *bio, *next;
341 
342 	spin_lock_irq(&head->requeue_lock);
343 	next = bio_list_get(&head->requeue_list);
344 	spin_unlock_irq(&head->requeue_lock);
345 
346 	while ((bio = next) != NULL) {
347 		next = bio->bi_next;
348 		bio->bi_next = NULL;
349 
350 		/*
351 		 * Reset disk to the mpath node and resubmit to select a new
352 		 * path.
353 		 */
354 		bio->bi_disk = head->disk;
355 		generic_make_request(bio);
356 	}
357 }
358 
359 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl, struct nvme_ns_head *head)
360 {
361 	struct request_queue *q;
362 	bool vwc = false;
363 
364 	mutex_init(&head->lock);
365 	bio_list_init(&head->requeue_list);
366 	spin_lock_init(&head->requeue_lock);
367 	INIT_WORK(&head->requeue_work, nvme_requeue_work);
368 
369 	/*
370 	 * Add a multipath node if the subsystems supports multiple controllers.
371 	 * We also do this for private namespaces as the namespace sharing data could
372 	 * change after a rescan.
373 	 */
374 	if (!(ctrl->subsys->cmic & (1 << 1)) || !multipath)
375 		return 0;
376 
377 	q = blk_alloc_queue_node(GFP_KERNEL, ctrl->numa_node);
378 	if (!q)
379 		goto out;
380 	q->queuedata = head;
381 	blk_queue_make_request(q, nvme_ns_head_make_request);
382 	blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
383 	/* set to a default value for 512 until disk is validated */
384 	blk_queue_logical_block_size(q, 512);
385 	blk_set_stacking_limits(&q->limits);
386 
387 	/* we need to propagate up the VMC settings */
388 	if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
389 		vwc = true;
390 	blk_queue_write_cache(q, vwc, vwc);
391 
392 	head->disk = alloc_disk(0);
393 	if (!head->disk)
394 		goto out_cleanup_queue;
395 	head->disk->fops = &nvme_ns_head_ops;
396 	head->disk->private_data = head;
397 	head->disk->queue = q;
398 	head->disk->flags = GENHD_FL_EXT_DEVT;
399 	sprintf(head->disk->disk_name, "nvme%dn%d",
400 			ctrl->subsys->instance, head->instance);
401 	return 0;
402 
403 out_cleanup_queue:
404 	blk_cleanup_queue(q);
405 out:
406 	return -ENOMEM;
407 }
408 
409 static void nvme_mpath_set_live(struct nvme_ns *ns)
410 {
411 	struct nvme_ns_head *head = ns->head;
412 
413 	lockdep_assert_held(&ns->head->lock);
414 
415 	if (!head->disk)
416 		return;
417 
418 	if (!(head->disk->flags & GENHD_FL_UP))
419 		device_add_disk(&head->subsys->dev, head->disk,
420 				nvme_ns_id_attr_groups);
421 
422 	if (nvme_path_is_optimized(ns)) {
423 		int node, srcu_idx;
424 
425 		srcu_idx = srcu_read_lock(&head->srcu);
426 		for_each_node(node)
427 			__nvme_find_path(head, node);
428 		srcu_read_unlock(&head->srcu, srcu_idx);
429 	}
430 
431 	synchronize_srcu(&ns->head->srcu);
432 	kblockd_schedule_work(&ns->head->requeue_work);
433 }
434 
435 static int nvme_parse_ana_log(struct nvme_ctrl *ctrl, void *data,
436 		int (*cb)(struct nvme_ctrl *ctrl, struct nvme_ana_group_desc *,
437 			void *))
438 {
439 	void *base = ctrl->ana_log_buf;
440 	size_t offset = sizeof(struct nvme_ana_rsp_hdr);
441 	int error, i;
442 
443 	lockdep_assert_held(&ctrl->ana_lock);
444 
445 	for (i = 0; i < le16_to_cpu(ctrl->ana_log_buf->ngrps); i++) {
446 		struct nvme_ana_group_desc *desc = base + offset;
447 		u32 nr_nsids = le32_to_cpu(desc->nnsids);
448 		size_t nsid_buf_size = nr_nsids * sizeof(__le32);
449 
450 		if (WARN_ON_ONCE(desc->grpid == 0))
451 			return -EINVAL;
452 		if (WARN_ON_ONCE(le32_to_cpu(desc->grpid) > ctrl->anagrpmax))
453 			return -EINVAL;
454 		if (WARN_ON_ONCE(desc->state == 0))
455 			return -EINVAL;
456 		if (WARN_ON_ONCE(desc->state > NVME_ANA_CHANGE))
457 			return -EINVAL;
458 
459 		offset += sizeof(*desc);
460 		if (WARN_ON_ONCE(offset > ctrl->ana_log_size - nsid_buf_size))
461 			return -EINVAL;
462 
463 		error = cb(ctrl, desc, data);
464 		if (error)
465 			return error;
466 
467 		offset += nsid_buf_size;
468 		if (WARN_ON_ONCE(offset > ctrl->ana_log_size - sizeof(*desc)))
469 			return -EINVAL;
470 	}
471 
472 	return 0;
473 }
474 
475 static inline bool nvme_state_is_live(enum nvme_ana_state state)
476 {
477 	return state == NVME_ANA_OPTIMIZED || state == NVME_ANA_NONOPTIMIZED;
478 }
479 
480 static void nvme_update_ns_ana_state(struct nvme_ana_group_desc *desc,
481 		struct nvme_ns *ns)
482 {
483 	mutex_lock(&ns->head->lock);
484 	ns->ana_grpid = le32_to_cpu(desc->grpid);
485 	ns->ana_state = desc->state;
486 	clear_bit(NVME_NS_ANA_PENDING, &ns->flags);
487 
488 	if (nvme_state_is_live(ns->ana_state))
489 		nvme_mpath_set_live(ns);
490 	mutex_unlock(&ns->head->lock);
491 }
492 
493 static int nvme_update_ana_state(struct nvme_ctrl *ctrl,
494 		struct nvme_ana_group_desc *desc, void *data)
495 {
496 	u32 nr_nsids = le32_to_cpu(desc->nnsids), n = 0;
497 	unsigned *nr_change_groups = data;
498 	struct nvme_ns *ns;
499 
500 	dev_dbg(ctrl->device, "ANA group %d: %s.\n",
501 			le32_to_cpu(desc->grpid),
502 			nvme_ana_state_names[desc->state]);
503 
504 	if (desc->state == NVME_ANA_CHANGE)
505 		(*nr_change_groups)++;
506 
507 	if (!nr_nsids)
508 		return 0;
509 
510 	down_write(&ctrl->namespaces_rwsem);
511 	list_for_each_entry(ns, &ctrl->namespaces, list) {
512 		unsigned nsid = le32_to_cpu(desc->nsids[n]);
513 
514 		if (ns->head->ns_id < nsid)
515 			continue;
516 		if (ns->head->ns_id == nsid)
517 			nvme_update_ns_ana_state(desc, ns);
518 		if (++n == nr_nsids)
519 			break;
520 	}
521 	up_write(&ctrl->namespaces_rwsem);
522 	return 0;
523 }
524 
525 static int nvme_read_ana_log(struct nvme_ctrl *ctrl, bool groups_only)
526 {
527 	u32 nr_change_groups = 0;
528 	int error;
529 
530 	mutex_lock(&ctrl->ana_lock);
531 	error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_ANA,
532 			groups_only ? NVME_ANA_LOG_RGO : 0,
533 			ctrl->ana_log_buf, ctrl->ana_log_size, 0);
534 	if (error) {
535 		dev_warn(ctrl->device, "Failed to get ANA log: %d\n", error);
536 		goto out_unlock;
537 	}
538 
539 	error = nvme_parse_ana_log(ctrl, &nr_change_groups,
540 			nvme_update_ana_state);
541 	if (error)
542 		goto out_unlock;
543 
544 	/*
545 	 * In theory we should have an ANATT timer per group as they might enter
546 	 * the change state at different times.  But that is a lot of overhead
547 	 * just to protect against a target that keeps entering new changes
548 	 * states while never finishing previous ones.  But we'll still
549 	 * eventually time out once all groups are in change state, so this
550 	 * isn't a big deal.
551 	 *
552 	 * We also double the ANATT value to provide some slack for transports
553 	 * or AEN processing overhead.
554 	 */
555 	if (nr_change_groups)
556 		mod_timer(&ctrl->anatt_timer, ctrl->anatt * HZ * 2 + jiffies);
557 	else
558 		del_timer_sync(&ctrl->anatt_timer);
559 out_unlock:
560 	mutex_unlock(&ctrl->ana_lock);
561 	return error;
562 }
563 
564 static void nvme_ana_work(struct work_struct *work)
565 {
566 	struct nvme_ctrl *ctrl = container_of(work, struct nvme_ctrl, ana_work);
567 
568 	nvme_read_ana_log(ctrl, false);
569 }
570 
571 static void nvme_anatt_timeout(struct timer_list *t)
572 {
573 	struct nvme_ctrl *ctrl = from_timer(ctrl, t, anatt_timer);
574 
575 	dev_info(ctrl->device, "ANATT timeout, resetting controller.\n");
576 	nvme_reset_ctrl(ctrl);
577 }
578 
579 void nvme_mpath_stop(struct nvme_ctrl *ctrl)
580 {
581 	if (!nvme_ctrl_use_ana(ctrl))
582 		return;
583 	del_timer_sync(&ctrl->anatt_timer);
584 	cancel_work_sync(&ctrl->ana_work);
585 }
586 
587 #define SUBSYS_ATTR_RW(_name, _mode, _show, _store)  \
588 	struct device_attribute subsys_attr_##_name =	\
589 		__ATTR(_name, _mode, _show, _store)
590 
591 static const char *nvme_iopolicy_names[] = {
592 	[NVME_IOPOLICY_NUMA]	= "numa",
593 	[NVME_IOPOLICY_RR]	= "round-robin",
594 };
595 
596 static ssize_t nvme_subsys_iopolicy_show(struct device *dev,
597 		struct device_attribute *attr, char *buf)
598 {
599 	struct nvme_subsystem *subsys =
600 		container_of(dev, struct nvme_subsystem, dev);
601 
602 	return sprintf(buf, "%s\n",
603 			nvme_iopolicy_names[READ_ONCE(subsys->iopolicy)]);
604 }
605 
606 static ssize_t nvme_subsys_iopolicy_store(struct device *dev,
607 		struct device_attribute *attr, const char *buf, size_t count)
608 {
609 	struct nvme_subsystem *subsys =
610 		container_of(dev, struct nvme_subsystem, dev);
611 	int i;
612 
613 	for (i = 0; i < ARRAY_SIZE(nvme_iopolicy_names); i++) {
614 		if (sysfs_streq(buf, nvme_iopolicy_names[i])) {
615 			WRITE_ONCE(subsys->iopolicy, i);
616 			return count;
617 		}
618 	}
619 
620 	return -EINVAL;
621 }
622 SUBSYS_ATTR_RW(iopolicy, S_IRUGO | S_IWUSR,
623 		      nvme_subsys_iopolicy_show, nvme_subsys_iopolicy_store);
624 
625 static ssize_t ana_grpid_show(struct device *dev, struct device_attribute *attr,
626 		char *buf)
627 {
628 	return sprintf(buf, "%d\n", nvme_get_ns_from_dev(dev)->ana_grpid);
629 }
630 DEVICE_ATTR_RO(ana_grpid);
631 
632 static ssize_t ana_state_show(struct device *dev, struct device_attribute *attr,
633 		char *buf)
634 {
635 	struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
636 
637 	return sprintf(buf, "%s\n", nvme_ana_state_names[ns->ana_state]);
638 }
639 DEVICE_ATTR_RO(ana_state);
640 
641 static int nvme_set_ns_ana_state(struct nvme_ctrl *ctrl,
642 		struct nvme_ana_group_desc *desc, void *data)
643 {
644 	struct nvme_ns *ns = data;
645 
646 	if (ns->ana_grpid == le32_to_cpu(desc->grpid)) {
647 		nvme_update_ns_ana_state(desc, ns);
648 		return -ENXIO; /* just break out of the loop */
649 	}
650 
651 	return 0;
652 }
653 
654 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id)
655 {
656 	if (nvme_ctrl_use_ana(ns->ctrl)) {
657 		mutex_lock(&ns->ctrl->ana_lock);
658 		ns->ana_grpid = le32_to_cpu(id->anagrpid);
659 		nvme_parse_ana_log(ns->ctrl, ns, nvme_set_ns_ana_state);
660 		mutex_unlock(&ns->ctrl->ana_lock);
661 	} else {
662 		mutex_lock(&ns->head->lock);
663 		ns->ana_state = NVME_ANA_OPTIMIZED;
664 		nvme_mpath_set_live(ns);
665 		mutex_unlock(&ns->head->lock);
666 	}
667 }
668 
669 void nvme_mpath_remove_disk(struct nvme_ns_head *head)
670 {
671 	if (!head->disk)
672 		return;
673 	if (head->disk->flags & GENHD_FL_UP)
674 		del_gendisk(head->disk);
675 	blk_set_queue_dying(head->disk->queue);
676 	/* make sure all pending bios are cleaned up */
677 	kblockd_schedule_work(&head->requeue_work);
678 	flush_work(&head->requeue_work);
679 	blk_cleanup_queue(head->disk->queue);
680 	put_disk(head->disk);
681 }
682 
683 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
684 {
685 	int error;
686 
687 	/* check if multipath is enabled and we have the capability */
688 	if (!multipath || !ctrl->subsys || !(ctrl->subsys->cmic & (1 << 3)))
689 		return 0;
690 
691 	ctrl->anacap = id->anacap;
692 	ctrl->anatt = id->anatt;
693 	ctrl->nanagrpid = le32_to_cpu(id->nanagrpid);
694 	ctrl->anagrpmax = le32_to_cpu(id->anagrpmax);
695 
696 	mutex_init(&ctrl->ana_lock);
697 	timer_setup(&ctrl->anatt_timer, nvme_anatt_timeout, 0);
698 	ctrl->ana_log_size = sizeof(struct nvme_ana_rsp_hdr) +
699 		ctrl->nanagrpid * sizeof(struct nvme_ana_group_desc);
700 	ctrl->ana_log_size += ctrl->max_namespaces * sizeof(__le32);
701 
702 	if (ctrl->ana_log_size > ctrl->max_hw_sectors << SECTOR_SHIFT) {
703 		dev_err(ctrl->device,
704 			"ANA log page size (%zd) larger than MDTS (%d).\n",
705 			ctrl->ana_log_size,
706 			ctrl->max_hw_sectors << SECTOR_SHIFT);
707 		dev_err(ctrl->device, "disabling ANA support.\n");
708 		return 0;
709 	}
710 
711 	INIT_WORK(&ctrl->ana_work, nvme_ana_work);
712 	ctrl->ana_log_buf = kmalloc(ctrl->ana_log_size, GFP_KERNEL);
713 	if (!ctrl->ana_log_buf) {
714 		error = -ENOMEM;
715 		goto out;
716 	}
717 
718 	error = nvme_read_ana_log(ctrl, true);
719 	if (error)
720 		goto out_free_ana_log_buf;
721 	return 0;
722 out_free_ana_log_buf:
723 	kfree(ctrl->ana_log_buf);
724 	ctrl->ana_log_buf = NULL;
725 out:
726 	return error;
727 }
728 
729 void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
730 {
731 	kfree(ctrl->ana_log_buf);
732 	ctrl->ana_log_buf = NULL;
733 }
734 
735