drivers/nvme/host/multipath.c

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