drivers/nvme/host/core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * NVM Express device driver
   4  * Copyright (c) 2011-2014, Intel Corporation.
   5  */
   6
   7 #include <linux/blkdev.h>
   8 #include <linux/blk-mq.h>
   9 #include <linux/delay.h>
  10 #include <linux/errno.h>
  11 #include <linux/hdreg.h>
  12 #include <linux/kernel.h>
  13 #include <linux/module.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/list_sort.h>
  16 #include <linux/slab.h>
  17 #include <linux/types.h>
  18 #include <linux/pr.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/nvme_ioctl.h>
  21 #include <linux/t10-pi.h>
  22 #include <linux/pm_qos.h>
  23 #include <asm/unaligned.h>
  24
  25 #include "nvme.h"
  26 #include "fabrics.h"
  27
  28 #define CREATE_TRACE_POINTS
  29 #include "trace.h"
  30
  31 #define NVME_MINORS             (1U << MINORBITS)
  32
  33 unsigned int admin_timeout = 60;
  34 module_param(admin_timeout, uint, 0644);
  35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  36 EXPORT_SYMBOL_GPL(admin_timeout);
  37
  38 unsigned int nvme_io_timeout = 30;
  39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  42
  43 static unsigned char shutdown_timeout = 5;
  44 module_param(shutdown_timeout, byte, 0644);
  45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  46
  47 static u8 nvme_max_retries = 5;
  48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  50
  51 static unsigned long default_ps_max_latency_us = 100000;
  52 module_param(default_ps_max_latency_us, ulong, 0644);
  53 MODULE_PARM_DESC(default_ps_max_latency_us,
  54                  "max power saving latency for new devices; use PM QOS to change per device");
  55
  56 static bool force_apst;
  57 module_param(force_apst, bool, 0644);
  58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  59
  60 static bool streams;
  61 module_param(streams, bool, 0644);
  62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  63
  64 /*
  65  * nvme_wq - hosts nvme related works that are not reset or delete
  66  * nvme_reset_wq - hosts nvme reset works
  67  * nvme_delete_wq - hosts nvme delete works
  68  *
  69  * nvme_wq will host works such are scan, aen handling, fw activation,
  70  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
  71  * runs reset works which also flush works hosted on nvme_wq for
  72  * serialization purposes. nvme_delete_wq host controller deletion
  73  * works which flush reset works for serialization.
  74  */
  75 struct workqueue_struct *nvme_wq;
  76 EXPORT_SYMBOL_GPL(nvme_wq);
  77
  78 struct workqueue_struct *nvme_reset_wq;
  79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  80
  81 struct workqueue_struct *nvme_delete_wq;
  82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  83
  84 static LIST_HEAD(nvme_subsystems);
  85 static DEFINE_MUTEX(nvme_subsystems_lock);
  86
  87 static DEFINE_IDA(nvme_instance_ida);
  88 static dev_t nvme_chr_devt;
  89 static struct class *nvme_class;
  90 static struct class *nvme_subsys_class;
  91
  92 static int nvme_revalidate_disk(struct gendisk *disk);
  93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  95                                            unsigned nsid);
  96
  97 static void nvme_set_queue_dying(struct nvme_ns *ns)
  98 {
  99         /*
 100          * Revalidating a dead namespace sets capacity to 0. This will end
 101          * buffered writers dirtying pages that can't be synced.
 102          */
 103         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 104                 return;
 105         blk_set_queue_dying(ns->queue);
 106         /* Forcibly unquiesce queues to avoid blocking dispatch */
 107         blk_mq_unquiesce_queue(ns->queue);
 108         /*
 109          * Revalidate after unblocking dispatchers that may be holding bd_butex
 110          */
 111         revalidate_disk(ns->disk);
 112 }
 113
 114 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 115 {
 116         /*
 117          * Only new queue scan work when admin and IO queues are both alive
 118          */
 119         if (ctrl->state == NVME_CTRL_LIVE)
 120                 queue_work(nvme_wq, &ctrl->scan_work);
 121 }
 122
 123 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 124 {
 125         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 126                 return -EBUSY;
 127         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 128                 return -EBUSY;
 129         return 0;
 130 }
 131 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 132
 133 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 134 {
 135         int ret;
 136
 137         ret = nvme_reset_ctrl(ctrl);
 138         if (!ret) {
 139                 flush_work(&ctrl->reset_work);
 140                 if (ctrl->state != NVME_CTRL_LIVE &&
 141                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
 142                         ret = -ENETRESET;
 143         }
 144
 145         return ret;
 146 }
 147 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 148
 149 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 150 {
 151         dev_info(ctrl->device,
 152                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 153
 154         flush_work(&ctrl->reset_work);
 155         nvme_stop_ctrl(ctrl);
 156         nvme_remove_namespaces(ctrl);
 157         ctrl->ops->delete_ctrl(ctrl);
 158         nvme_uninit_ctrl(ctrl);
 159         nvme_put_ctrl(ctrl);
 160 }
 161
 162 static void nvme_delete_ctrl_work(struct work_struct *work)
 163 {
 164         struct nvme_ctrl *ctrl =
 165                 container_of(work, struct nvme_ctrl, delete_work);
 166
 167         nvme_do_delete_ctrl(ctrl);
 168 }
 169
 170 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 171 {
 172         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 173                 return -EBUSY;
 174         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 175                 return -EBUSY;
 176         return 0;
 177 }
 178 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 179
 180 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 181 {
 182         int ret = 0;
 183
 184         /*
 185          * Keep a reference until nvme_do_delete_ctrl() complete,
 186          * since ->delete_ctrl can free the controller.
 187          */
 188         nvme_get_ctrl(ctrl);
 189         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 190                 ret = -EBUSY;
 191         if (!ret)
 192                 nvme_do_delete_ctrl(ctrl);
 193         nvme_put_ctrl(ctrl);
 194         return ret;
 195 }
 196
 197 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 198 {
 199         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 200 }
 201
 202 static blk_status_t nvme_error_status(u16 status)
 203 {
 204         switch (status & 0x7ff) {
 205         case NVME_SC_SUCCESS:
 206                 return BLK_STS_OK;
 207         case NVME_SC_CAP_EXCEEDED:
 208                 return BLK_STS_NOSPC;
 209         case NVME_SC_LBA_RANGE:
 210                 return BLK_STS_TARGET;
 211         case NVME_SC_BAD_ATTRIBUTES:
 212         case NVME_SC_ONCS_NOT_SUPPORTED:
 213         case NVME_SC_INVALID_OPCODE:
 214         case NVME_SC_INVALID_FIELD:
 215         case NVME_SC_INVALID_NS:
 216                 return BLK_STS_NOTSUPP;
 217         case NVME_SC_WRITE_FAULT:
 218         case NVME_SC_READ_ERROR:
 219         case NVME_SC_UNWRITTEN_BLOCK:
 220         case NVME_SC_ACCESS_DENIED:
 221         case NVME_SC_READ_ONLY:
 222         case NVME_SC_COMPARE_FAILED:
 223                 return BLK_STS_MEDIUM;
 224         case NVME_SC_GUARD_CHECK:
 225         case NVME_SC_APPTAG_CHECK:
 226         case NVME_SC_REFTAG_CHECK:
 227         case NVME_SC_INVALID_PI:
 228                 return BLK_STS_PROTECTION;
 229         case NVME_SC_RESERVATION_CONFLICT:
 230                 return BLK_STS_NEXUS;
 231         case NVME_SC_HOST_PATH_ERROR:
 232                 return BLK_STS_TRANSPORT;
 233         default:
 234                 return BLK_STS_IOERR;
 235         }
 236 }
 237
 238 static inline bool nvme_req_needs_retry(struct request *req)
 239 {
 240         if (blk_noretry_request(req))
 241                 return false;
 242         if (nvme_req(req)->status & NVME_SC_DNR)
 243                 return false;
 244         if (nvme_req(req)->retries >= nvme_max_retries)
 245                 return false;
 246         return true;
 247 }
 248
 249 static void nvme_retry_req(struct request *req)
 250 {
 251         struct nvme_ns *ns = req->q->queuedata;
 252         unsigned long delay = 0;
 253         u16 crd;
 254
 255         /* The mask and shift result must be <= 3 */
 256         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 257         if (ns && crd)
 258                 delay = ns->ctrl->crdt[crd - 1] * 100;
 259
 260         nvme_req(req)->retries++;
 261         blk_mq_requeue_request(req, false);
 262         blk_mq_delay_kick_requeue_list(req->q, delay);
 263 }
 264
 265 void nvme_complete_rq(struct request *req)
 266 {
 267         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 268
 269         trace_nvme_complete_rq(req);
 270
 271         if (nvme_req(req)->ctrl->kas)
 272                 nvme_req(req)->ctrl->comp_seen = true;
 273
 274         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 275                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
 276                     blk_path_error(status)) {
 277                         nvme_failover_req(req);
 278                         return;
 279                 }
 280
 281                 if (!blk_queue_dying(req->q)) {
 282                         nvme_retry_req(req);
 283                         return;
 284                 }
 285         }
 286
 287         nvme_trace_bio_complete(req, status);
 288         blk_mq_end_request(req, status);
 289 }
 290 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 291
 292 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 293 {
 294         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 295                                 "Cancelling I/O %d", req->tag);
 296
 297         /* don't abort one completed request */
 298         if (blk_mq_request_completed(req))
 299                 return true;
 300
 301         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 302         blk_mq_complete_request(req);
 303         return true;
 304 }
 305 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 306
 307 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 308                 enum nvme_ctrl_state new_state)
 309 {
 310         enum nvme_ctrl_state old_state;
 311         unsigned long flags;
 312         bool changed = false;
 313
 314         spin_lock_irqsave(&ctrl->lock, flags);
 315
 316         old_state = ctrl->state;
 317         switch (new_state) {
 318         case NVME_CTRL_ADMIN_ONLY:
 319                 switch (old_state) {
 320                 case NVME_CTRL_CONNECTING:
 321                         changed = true;
 322                         /* FALLTHRU */
 323                 default:
 324                         break;
 325                 }
 326                 break;
 327         case NVME_CTRL_LIVE:
 328                 switch (old_state) {
 329                 case NVME_CTRL_NEW:
 330                 case NVME_CTRL_RESETTING:
 331                 case NVME_CTRL_CONNECTING:
 332                         changed = true;
 333                         /* FALLTHRU */
 334                 default:
 335                         break;
 336                 }
 337                 break;
 338         case NVME_CTRL_RESETTING:
 339                 switch (old_state) {
 340                 case NVME_CTRL_NEW:
 341                 case NVME_CTRL_LIVE:
 342                 case NVME_CTRL_ADMIN_ONLY:
 343                         changed = true;
 344                         /* FALLTHRU */
 345                 default:
 346                         break;
 347                 }
 348                 break;
 349         case NVME_CTRL_CONNECTING:
 350                 switch (old_state) {
 351                 case NVME_CTRL_NEW:
 352                 case NVME_CTRL_RESETTING:
 353                         changed = true;
 354                         /* FALLTHRU */
 355                 default:
 356                         break;
 357                 }
 358                 break;
 359         case NVME_CTRL_DELETING:
 360                 switch (old_state) {
 361                 case NVME_CTRL_LIVE:
 362                 case NVME_CTRL_ADMIN_ONLY:
 363                 case NVME_CTRL_RESETTING:
 364                 case NVME_CTRL_CONNECTING:
 365                         changed = true;
 366                         /* FALLTHRU */
 367                 default:
 368                         break;
 369                 }
 370                 break;
 371         case NVME_CTRL_DEAD:
 372                 switch (old_state) {
 373                 case NVME_CTRL_DELETING:
 374                         changed = true;
 375                         /* FALLTHRU */
 376                 default:
 377                         break;
 378                 }
 379                 break;
 380         default:
 381                 break;
 382         }
 383
 384         if (changed)
 385                 ctrl->state = new_state;
 386
 387         spin_unlock_irqrestore(&ctrl->lock, flags);
 388         if (changed && ctrl->state == NVME_CTRL_LIVE)
 389                 nvme_kick_requeue_lists(ctrl);
 390         return changed;
 391 }
 392 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 393
 394 static void nvme_free_ns_head(struct kref *ref)
 395 {
 396         struct nvme_ns_head *head =
 397                 container_of(ref, struct nvme_ns_head, ref);
 398
 399         nvme_mpath_remove_disk(head);
 400         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 401         list_del_init(&head->entry);
 402         cleanup_srcu_struct(&head->srcu);
 403         nvme_put_subsystem(head->subsys);
 404         kfree(head);
 405 }
 406
 407 static void nvme_put_ns_head(struct nvme_ns_head *head)
 408 {
 409         kref_put(&head->ref, nvme_free_ns_head);
 410 }
 411
 412 static void nvme_free_ns(struct kref *kref)
 413 {
 414         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 415
 416         if (ns->ndev)
 417                 nvme_nvm_unregister(ns);
 418
 419         put_disk(ns->disk);
 420         nvme_put_ns_head(ns->head);
 421         nvme_put_ctrl(ns->ctrl);
 422         kfree(ns);
 423 }
 424
 425 static void nvme_put_ns(struct nvme_ns *ns)
 426 {
 427         kref_put(&ns->kref, nvme_free_ns);
 428 }
 429
 430 static inline void nvme_clear_nvme_request(struct request *req)
 431 {
 432         if (!(req->rq_flags & RQF_DONTPREP)) {
 433                 nvme_req(req)->retries = 0;
 434                 nvme_req(req)->flags = 0;
 435                 req->rq_flags |= RQF_DONTPREP;
 436         }
 437 }
 438
 439 struct request *nvme_alloc_request(struct request_queue *q,
 440                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 441 {
 442         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 443         struct request *req;
 444
 445         if (qid == NVME_QID_ANY) {
 446                 req = blk_mq_alloc_request(q, op, flags);
 447         } else {
 448                 req = blk_mq_alloc_request_hctx(q, op, flags,
 449                                 qid ? qid - 1 : 0);
 450         }
 451         if (IS_ERR(req))
 452                 return req;
 453
 454         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 455         nvme_clear_nvme_request(req);
 456         nvme_req(req)->cmd = cmd;
 457
 458         return req;
 459 }
 460 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 461
 462 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 463 {
 464         struct nvme_command c;
 465
 466         memset(&c, 0, sizeof(c));
 467
 468         c.directive.opcode = nvme_admin_directive_send;
 469         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 470         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 471         c.directive.dtype = NVME_DIR_IDENTIFY;
 472         c.directive.tdtype = NVME_DIR_STREAMS;
 473         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 474
 475         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 476 }
 477
 478 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 479 {
 480         return nvme_toggle_streams(ctrl, false);
 481 }
 482
 483 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 484 {
 485         return nvme_toggle_streams(ctrl, true);
 486 }
 487
 488 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 489                                   struct streams_directive_params *s, u32 nsid)
 490 {
 491         struct nvme_command c;
 492
 493         memset(&c, 0, sizeof(c));
 494         memset(s, 0, sizeof(*s));
 495
 496         c.directive.opcode = nvme_admin_directive_recv;
 497         c.directive.nsid = cpu_to_le32(nsid);
 498         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 499         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 500         c.directive.dtype = NVME_DIR_STREAMS;
 501
 502         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 503 }
 504
 505 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 506 {
 507         struct streams_directive_params s;
 508         int ret;
 509
 510         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 511                 return 0;
 512         if (!streams)
 513                 return 0;
 514
 515         ret = nvme_enable_streams(ctrl);
 516         if (ret)
 517                 return ret;
 518
 519         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 520         if (ret)
 521                 return ret;
 522
 523         ctrl->nssa = le16_to_cpu(s.nssa);
 524         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 525                 dev_info(ctrl->device, "too few streams (%u) available\n",
 526                                         ctrl->nssa);
 527                 nvme_disable_streams(ctrl);
 528                 return 0;
 529         }
 530
 531         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 532         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 533         return 0;
 534 }
 535
 536 /*
 537  * Check if 'req' has a write hint associated with it. If it does, assign
 538  * a valid namespace stream to the write.
 539  */
 540 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 541                                      struct request *req, u16 *control,
 542                                      u32 *dsmgmt)
 543 {
 544         enum rw_hint streamid = req->write_hint;
 545
 546         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 547                 streamid = 0;
 548         else {
 549                 streamid--;
 550                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 551                         return;
 552
 553                 *control |= NVME_RW_DTYPE_STREAMS;
 554                 *dsmgmt |= streamid << 16;
 555         }
 556
 557         if (streamid < ARRAY_SIZE(req->q->write_hints))
 558                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 559 }
 560
 561 static inline void nvme_setup_flush(struct nvme_ns *ns,
 562                 struct nvme_command *cmnd)
 563 {
 564         cmnd->common.opcode = nvme_cmd_flush;
 565         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 566 }
 567
 568 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 569                 struct nvme_command *cmnd)
 570 {
 571         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 572         struct nvme_dsm_range *range;
 573         struct bio *bio;
 574
 575         range = kmalloc_array(segments, sizeof(*range),
 576                                 GFP_ATOMIC | __GFP_NOWARN);
 577         if (!range) {
 578                 /*
 579                  * If we fail allocation our range, fallback to the controller
 580                  * discard page. If that's also busy, it's safe to return
 581                  * busy, as we know we can make progress once that's freed.
 582                  */
 583                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 584                         return BLK_STS_RESOURCE;
 585
 586                 range = page_address(ns->ctrl->discard_page);
 587         }
 588
 589         __rq_for_each_bio(bio, req) {
 590                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 591                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 592
 593                 if (n < segments) {
 594                         range[n].cattr = cpu_to_le32(0);
 595                         range[n].nlb = cpu_to_le32(nlb);
 596                         range[n].slba = cpu_to_le64(slba);
 597                 }
 598                 n++;
 599         }
 600
 601         if (WARN_ON_ONCE(n != segments)) {
 602                 if (virt_to_page(range) == ns->ctrl->discard_page)
 603                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 604                 else
 605                         kfree(range);
 606                 return BLK_STS_IOERR;
 607         }
 608
 609         cmnd->dsm.opcode = nvme_cmd_dsm;
 610         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 611         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 612         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 613
 614         req->special_vec.bv_page = virt_to_page(range);
 615         req->special_vec.bv_offset = offset_in_page(range);
 616         req->special_vec.bv_len = sizeof(*range) * segments;
 617         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 618
 619         return BLK_STS_OK;
 620 }
 621
 622 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 623                 struct request *req, struct nvme_command *cmnd)
 624 {
 625         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 626                 return nvme_setup_discard(ns, req, cmnd);
 627
 628         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 629         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 630         cmnd->write_zeroes.slba =
 631                 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 632         cmnd->write_zeroes.length =
 633                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 634         cmnd->write_zeroes.control = 0;
 635         return BLK_STS_OK;
 636 }
 637
 638 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 639                 struct request *req, struct nvme_command *cmnd)
 640 {
 641         struct nvme_ctrl *ctrl = ns->ctrl;
 642         u16 control = 0;
 643         u32 dsmgmt = 0;
 644
 645         if (req->cmd_flags & REQ_FUA)
 646                 control |= NVME_RW_FUA;
 647         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 648                 control |= NVME_RW_LR;
 649
 650         if (req->cmd_flags & REQ_RAHEAD)
 651                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 652
 653         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 654         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 655         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 656         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 657
 658         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 659                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 660
 661         if (ns->ms) {
 662                 /*
 663                  * If formated with metadata, the block layer always provides a
 664                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 665                  * we enable the PRACT bit for protection information or set the
 666                  * namespace capacity to zero to prevent any I/O.
 667                  */
 668                 if (!blk_integrity_rq(req)) {
 669                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 670                                 return BLK_STS_NOTSUPP;
 671                         control |= NVME_RW_PRINFO_PRACT;
 672                 }
 673
 674                 switch (ns->pi_type) {
 675                 case NVME_NS_DPS_PI_TYPE3:
 676                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 677                         break;
 678                 case NVME_NS_DPS_PI_TYPE1:
 679                 case NVME_NS_DPS_PI_TYPE2:
 680                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 681                                         NVME_RW_PRINFO_PRCHK_REF;
 682                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 683                         break;
 684                 }
 685         }
 686
 687         cmnd->rw.control = cpu_to_le16(control);
 688         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 689         return 0;
 690 }
 691
 692 void nvme_cleanup_cmd(struct request *req)
 693 {
 694         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 695                 struct nvme_ns *ns = req->rq_disk->private_data;
 696                 struct page *page = req->special_vec.bv_page;
 697
 698                 if (page == ns->ctrl->discard_page)
 699                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 700                 else
 701                         kfree(page_address(page) + req->special_vec.bv_offset);
 702         }
 703 }
 704 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 705
 706 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 707                 struct nvme_command *cmd)
 708 {
 709         blk_status_t ret = BLK_STS_OK;
 710
 711         nvme_clear_nvme_request(req);
 712
 713         memset(cmd, 0, sizeof(*cmd));
 714         switch (req_op(req)) {
 715         case REQ_OP_DRV_IN:
 716         case REQ_OP_DRV_OUT:
 717                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 718                 break;
 719         case REQ_OP_FLUSH:
 720                 nvme_setup_flush(ns, cmd);
 721                 break;
 722         case REQ_OP_WRITE_ZEROES:
 723                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 724                 break;
 725         case REQ_OP_DISCARD:
 726                 ret = nvme_setup_discard(ns, req, cmd);
 727                 break;
 728         case REQ_OP_READ:
 729         case REQ_OP_WRITE:
 730                 ret = nvme_setup_rw(ns, req, cmd);
 731                 break;
 732         default:
 733                 WARN_ON_ONCE(1);
 734                 return BLK_STS_IOERR;
 735         }
 736
 737         cmd->common.command_id = req->tag;
 738         trace_nvme_setup_cmd(req, cmd);
 739         return ret;
 740 }
 741 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 742
 743 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 744 {
 745         struct completion *waiting = rq->end_io_data;
 746
 747         rq->end_io_data = NULL;
 748         complete(waiting);
 749 }
 750
 751 static void nvme_execute_rq_polled(struct request_queue *q,
 752                 struct gendisk *bd_disk, struct request *rq, int at_head)
 753 {
 754         DECLARE_COMPLETION_ONSTACK(wait);
 755
 756         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 757
 758         rq->cmd_flags |= REQ_HIPRI;
 759         rq->end_io_data = &wait;
 760         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 761
 762         while (!completion_done(&wait)) {
 763                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 764                 cond_resched();
 765         }
 766 }
 767
 768 /*
 769  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 770  * if the result is positive, it's an NVM Express status code
 771  */
 772 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 773                 union nvme_result *result, void *buffer, unsigned bufflen,
 774                 unsigned timeout, int qid, int at_head,
 775                 blk_mq_req_flags_t flags, bool poll)
 776 {
 777         struct request *req;
 778         int ret;
 779
 780         req = nvme_alloc_request(q, cmd, flags, qid);
 781         if (IS_ERR(req))
 782                 return PTR_ERR(req);
 783
 784         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 785
 786         if (buffer && bufflen) {
 787                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 788                 if (ret)
 789                         goto out;
 790         }
 791
 792         if (poll)
 793                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 794         else
 795                 blk_execute_rq(req->q, NULL, req, at_head);
 796         if (result)
 797                 *result = nvme_req(req)->result;
 798         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 799                 ret = -EINTR;
 800         else
 801                 ret = nvme_req(req)->status;
 802  out:
 803         blk_mq_free_request(req);
 804         return ret;
 805 }
 806 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 807
 808 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 809                 void *buffer, unsigned bufflen)
 810 {
 811         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 812                         NVME_QID_ANY, 0, 0, false);
 813 }
 814 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 815
 816 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 817                 unsigned len, u32 seed, bool write)
 818 {
 819         struct bio_integrity_payload *bip;
 820         int ret = -ENOMEM;
 821         void *buf;
 822
 823         buf = kmalloc(len, GFP_KERNEL);
 824         if (!buf)
 825                 goto out;
 826
 827         ret = -EFAULT;
 828         if (write && copy_from_user(buf, ubuf, len))
 829                 goto out_free_meta;
 830
 831         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 832         if (IS_ERR(bip)) {
 833                 ret = PTR_ERR(bip);
 834                 goto out_free_meta;
 835         }
 836
 837         bip->bip_iter.bi_size = len;
 838         bip->bip_iter.bi_sector = seed;
 839         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 840                         offset_in_page(buf));
 841         if (ret == len)
 842                 return buf;
 843         ret = -ENOMEM;
 844 out_free_meta:
 845         kfree(buf);
 846 out:
 847         return ERR_PTR(ret);
 848 }
 849
 850 static int nvme_submit_user_cmd(struct request_queue *q,
 851                 struct nvme_command *cmd, void __user *ubuffer,
 852                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 853                 u32 meta_seed, u64 *result, unsigned timeout)
 854 {
 855         bool write = nvme_is_write(cmd);
 856         struct nvme_ns *ns = q->queuedata;
 857         struct gendisk *disk = ns ? ns->disk : NULL;
 858         struct request *req;
 859         struct bio *bio = NULL;
 860         void *meta = NULL;
 861         int ret;
 862
 863         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 864         if (IS_ERR(req))
 865                 return PTR_ERR(req);
 866
 867         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 868         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 869
 870         if (ubuffer && bufflen) {
 871                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 872                                 GFP_KERNEL);
 873                 if (ret)
 874                         goto out;
 875                 bio = req->bio;
 876                 bio->bi_disk = disk;
 877                 if (disk && meta_buffer && meta_len) {
 878                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 879                                         meta_seed, write);
 880                         if (IS_ERR(meta)) {
 881                                 ret = PTR_ERR(meta);
 882                                 goto out_unmap;
 883                         }
 884                         req->cmd_flags |= REQ_INTEGRITY;
 885                 }
 886         }
 887
 888         blk_execute_rq(req->q, disk, req, 0);
 889         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 890                 ret = -EINTR;
 891         else
 892                 ret = nvme_req(req)->status;
 893         if (result)
 894                 *result = le64_to_cpu(nvme_req(req)->result.u64);
 895         if (meta && !ret && !write) {
 896                 if (copy_to_user(meta_buffer, meta, meta_len))
 897                         ret = -EFAULT;
 898         }
 899         kfree(meta);
 900  out_unmap:
 901         if (bio)
 902                 blk_rq_unmap_user(bio);
 903  out:
 904         blk_mq_free_request(req);
 905         return ret;
 906 }
 907
 908 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 909 {
 910         struct nvme_ctrl *ctrl = rq->end_io_data;
 911         unsigned long flags;
 912         bool startka = false;
 913
 914         blk_mq_free_request(rq);
 915
 916         if (status) {
 917                 dev_err(ctrl->device,
 918                         "failed nvme_keep_alive_end_io error=%d\n",
 919                                 status);
 920                 return;
 921         }
 922
 923         ctrl->comp_seen = false;
 924         spin_lock_irqsave(&ctrl->lock, flags);
 925         if (ctrl->state == NVME_CTRL_LIVE ||
 926             ctrl->state == NVME_CTRL_CONNECTING)
 927                 startka = true;
 928         spin_unlock_irqrestore(&ctrl->lock, flags);
 929         if (startka)
 930                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 931 }
 932
 933 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 934 {
 935         struct request *rq;
 936
 937         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 938                         NVME_QID_ANY);
 939         if (IS_ERR(rq))
 940                 return PTR_ERR(rq);
 941
 942         rq->timeout = ctrl->kato * HZ;
 943         rq->end_io_data = ctrl;
 944
 945         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 946
 947         return 0;
 948 }
 949
 950 static void nvme_keep_alive_work(struct work_struct *work)
 951 {
 952         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 953                         struct nvme_ctrl, ka_work);
 954         bool comp_seen = ctrl->comp_seen;
 955
 956         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
 957                 dev_dbg(ctrl->device,
 958                         "reschedule traffic based keep-alive timer\n");
 959                 ctrl->comp_seen = false;
 960                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 961                 return;
 962         }
 963
 964         if (nvme_keep_alive(ctrl)) {
 965                 /* allocation failure, reset the controller */
 966                 dev_err(ctrl->device, "keep-alive failed\n");
 967                 nvme_reset_ctrl(ctrl);
 968                 return;
 969         }
 970 }
 971
 972 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 973 {
 974         if (unlikely(ctrl->kato == 0))
 975                 return;
 976
 977         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 978 }
 979
 980 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 981 {
 982         if (unlikely(ctrl->kato == 0))
 983                 return;
 984
 985         cancel_delayed_work_sync(&ctrl->ka_work);
 986 }
 987 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 988
 989 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 990 {
 991         struct nvme_command c = { };
 992         int error;
 993
 994         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 995         c.identify.opcode = nvme_admin_identify;
 996         c.identify.cns = NVME_ID_CNS_CTRL;
 997
 998         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 999         if (!*id)
1000                 return -ENOMEM;
1001
1002         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1003                         sizeof(struct nvme_id_ctrl));
1004         if (error)
1005                 kfree(*id);
1006         return error;
1007 }
1008
1009 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1010                 struct nvme_ns_ids *ids)
1011 {
1012         struct nvme_command c = { };
1013         int status;
1014         void *data;
1015         int pos;
1016         int len;
1017
1018         c.identify.opcode = nvme_admin_identify;
1019         c.identify.nsid = cpu_to_le32(nsid);
1020         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1021
1022         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1023         if (!data)
1024                 return -ENOMEM;
1025
1026         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1027                                       NVME_IDENTIFY_DATA_SIZE);
1028         if (status)
1029                 goto free_data;
1030
1031         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1032                 struct nvme_ns_id_desc *cur = data + pos;
1033
1034                 if (cur->nidl == 0)
1035                         break;
1036
1037                 switch (cur->nidt) {
1038                 case NVME_NIDT_EUI64:
1039                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1040                                 dev_warn(ctrl->device,
1041                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1042                                          cur->nidl);
1043                                 goto free_data;
1044                         }
1045                         len = NVME_NIDT_EUI64_LEN;
1046                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1047                         break;
1048                 case NVME_NIDT_NGUID:
1049                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1050                                 dev_warn(ctrl->device,
1051                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1052                                          cur->nidl);
1053                                 goto free_data;
1054                         }
1055                         len = NVME_NIDT_NGUID_LEN;
1056                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1057                         break;
1058                 case NVME_NIDT_UUID:
1059                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
1060                                 dev_warn(ctrl->device,
1061                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1062                                          cur->nidl);
1063                                 goto free_data;
1064                         }
1065                         len = NVME_NIDT_UUID_LEN;
1066                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1067                         break;
1068                 default:
1069                         /* Skip unknown types */
1070                         len = cur->nidl;
1071                         break;
1072                 }
1073
1074                 len += sizeof(*cur);
1075         }
1076 free_data:
1077         kfree(data);
1078         return status;
1079 }
1080
1081 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1082 {
1083         struct nvme_command c = { };
1084
1085         c.identify.opcode = nvme_admin_identify;
1086         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1087         c.identify.nsid = cpu_to_le32(nsid);
1088         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1089                                     NVME_IDENTIFY_DATA_SIZE);
1090 }
1091
1092 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1093                 unsigned nsid, struct nvme_id_ns **id)
1094 {
1095         struct nvme_command c = { };
1096         int error;
1097
1098         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1099         c.identify.opcode = nvme_admin_identify;
1100         c.identify.nsid = cpu_to_le32(nsid);
1101         c.identify.cns = NVME_ID_CNS_NS;
1102
1103         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1104         if (!*id)
1105                 return -ENOMEM;
1106
1107         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1108         if (error) {
1109                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1110                 kfree(*id);
1111         }
1112
1113         return error;
1114 }
1115
1116 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1117                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1118 {
1119         struct nvme_command c;
1120         union nvme_result res;
1121         int ret;
1122
1123         memset(&c, 0, sizeof(c));
1124         c.features.opcode = op;
1125         c.features.fid = cpu_to_le32(fid);
1126         c.features.dword11 = cpu_to_le32(dword11);
1127
1128         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1129                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1130         if (ret >= 0 && result)
1131                 *result = le32_to_cpu(res.u32);
1132         return ret;
1133 }
1134
1135 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1136                       unsigned int dword11, void *buffer, size_t buflen,
1137                       u32 *result)
1138 {
1139         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1140                              buflen, result);
1141 }
1142 EXPORT_SYMBOL_GPL(nvme_set_features);
1143
1144 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1145                       unsigned int dword11, void *buffer, size_t buflen,
1146                       u32 *result)
1147 {
1148         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1149                              buflen, result);
1150 }
1151 EXPORT_SYMBOL_GPL(nvme_get_features);
1152
1153 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1154 {
1155         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1156         u32 result;
1157         int status, nr_io_queues;
1158
1159         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1160                         &result);
1161         if (status < 0)
1162                 return status;
1163
1164         /*
1165          * Degraded controllers might return an error when setting the queue
1166          * count.  We still want to be able to bring them online and offer
1167          * access to the admin queue, as that might be only way to fix them up.
1168          */
1169         if (status > 0) {
1170                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1171                 *count = 0;
1172         } else {
1173                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1174                 *count = min(*count, nr_io_queues);
1175         }
1176
1177         return 0;
1178 }
1179 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1180
1181 #define NVME_AEN_SUPPORTED \
1182         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1183          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1184
1185 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1186 {
1187         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1188         int status;
1189
1190         if (!supported_aens)
1191                 return;
1192
1193         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1194                         NULL, 0, &result);
1195         if (status)
1196                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1197                          supported_aens);
1198
1199         queue_work(nvme_wq, &ctrl->async_event_work);
1200 }
1201
1202 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1203 {
1204         struct nvme_user_io io;
1205         struct nvme_command c;
1206         unsigned length, meta_len;
1207         void __user *metadata;
1208
1209         if (copy_from_user(&io, uio, sizeof(io)))
1210                 return -EFAULT;
1211         if (io.flags)
1212                 return -EINVAL;
1213
1214         switch (io.opcode) {
1215         case nvme_cmd_write:
1216         case nvme_cmd_read:
1217         case nvme_cmd_compare:
1218                 break;
1219         default:
1220                 return -EINVAL;
1221         }
1222
1223         length = (io.nblocks + 1) << ns->lba_shift;
1224         meta_len = (io.nblocks + 1) * ns->ms;
1225         metadata = (void __user *)(uintptr_t)io.metadata;
1226
1227         if (ns->ext) {
1228                 length += meta_len;
1229                 meta_len = 0;
1230         } else if (meta_len) {
1231                 if ((io.metadata & 3) || !io.metadata)
1232                         return -EINVAL;
1233         }
1234
1235         memset(&c, 0, sizeof(c));
1236         c.rw.opcode = io.opcode;
1237         c.rw.flags = io.flags;
1238         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1239         c.rw.slba = cpu_to_le64(io.slba);
1240         c.rw.length = cpu_to_le16(io.nblocks);
1241         c.rw.control = cpu_to_le16(io.control);
1242         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1243         c.rw.reftag = cpu_to_le32(io.reftag);
1244         c.rw.apptag = cpu_to_le16(io.apptag);
1245         c.rw.appmask = cpu_to_le16(io.appmask);
1246
1247         return nvme_submit_user_cmd(ns->queue, &c,
1248                         (void __user *)(uintptr_t)io.addr, length,
1249                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1250 }
1251
1252 static u32 nvme_known_admin_effects(u8 opcode)
1253 {
1254         switch (opcode) {
1255         case nvme_admin_format_nvm:
1256                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1257                                         NVME_CMD_EFFECTS_CSE_MASK;
1258         case nvme_admin_sanitize_nvm:
1259                 return NVME_CMD_EFFECTS_CSE_MASK;
1260         default:
1261                 break;
1262         }
1263         return 0;
1264 }
1265
1266 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1267                                                                 u8 opcode)
1268 {
1269         u32 effects = 0;
1270
1271         if (ns) {
1272                 if (ctrl->effects)
1273                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1274                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1275                         dev_warn(ctrl->device,
1276                                  "IO command:%02x has unhandled effects:%08x\n",
1277                                  opcode, effects);
1278                 return 0;
1279         }
1280
1281         if (ctrl->effects)
1282                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1283         effects |= nvme_known_admin_effects(opcode);
1284
1285         /*
1286          * For simplicity, IO to all namespaces is quiesced even if the command
1287          * effects say only one namespace is affected.
1288          */
1289         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1290                 mutex_lock(&ctrl->scan_lock);
1291                 mutex_lock(&ctrl->subsys->lock);
1292                 nvme_mpath_start_freeze(ctrl->subsys);
1293                 nvme_mpath_wait_freeze(ctrl->subsys);
1294                 nvme_start_freeze(ctrl);
1295                 nvme_wait_freeze(ctrl);
1296         }
1297         return effects;
1298 }
1299
1300 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1301 {
1302         struct nvme_ns *ns;
1303
1304         down_read(&ctrl->namespaces_rwsem);
1305         list_for_each_entry(ns, &ctrl->namespaces, list)
1306                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1307                         nvme_set_queue_dying(ns);
1308         up_read(&ctrl->namespaces_rwsem);
1309 }
1310
1311 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1312 {
1313         /*
1314          * Revalidate LBA changes prior to unfreezing. This is necessary to
1315          * prevent memory corruption if a logical block size was changed by
1316          * this command.
1317          */
1318         if (effects & NVME_CMD_EFFECTS_LBCC)
1319                 nvme_update_formats(ctrl);
1320         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1321                 nvme_unfreeze(ctrl);
1322                 nvme_mpath_unfreeze(ctrl->subsys);
1323                 mutex_unlock(&ctrl->subsys->lock);
1324                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1325                 mutex_unlock(&ctrl->scan_lock);
1326         }
1327         if (effects & NVME_CMD_EFFECTS_CCC)
1328                 nvme_init_identify(ctrl);
1329         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1330                 nvme_queue_scan(ctrl);
1331 }
1332
1333 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1334                         struct nvme_passthru_cmd __user *ucmd)
1335 {
1336         struct nvme_passthru_cmd cmd;
1337         struct nvme_command c;
1338         unsigned timeout = 0;
1339         u32 effects;
1340         u64 result;
1341         int status;
1342
1343         if (!capable(CAP_SYS_ADMIN))
1344                 return -EACCES;
1345         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1346                 return -EFAULT;
1347         if (cmd.flags)
1348                 return -EINVAL;
1349
1350         memset(&c, 0, sizeof(c));
1351         c.common.opcode = cmd.opcode;
1352         c.common.flags = cmd.flags;
1353         c.common.nsid = cpu_to_le32(cmd.nsid);
1354         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1355         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1356         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1357         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1358         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1359         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1360         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1361         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1362
1363         if (cmd.timeout_ms)
1364                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1365
1366         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1367         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1368                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1369                         (void __user *)(uintptr_t)cmd.metadata,
1370                         cmd.metadata_len, 0, &result, timeout);
1371         nvme_passthru_end(ctrl, effects);
1372
1373         if (status >= 0) {
1374                 if (put_user(result, &ucmd->result))
1375                         return -EFAULT;
1376         }
1377
1378         return status;
1379 }
1380
1381 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1382                         struct nvme_passthru_cmd64 __user *ucmd)
1383 {
1384         struct nvme_passthru_cmd64 cmd;
1385         struct nvme_command c;
1386         unsigned timeout = 0;
1387         u32 effects;
1388         int status;
1389
1390         if (!capable(CAP_SYS_ADMIN))
1391                 return -EACCES;
1392         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1393                 return -EFAULT;
1394         if (cmd.flags)
1395                 return -EINVAL;
1396
1397         memset(&c, 0, sizeof(c));
1398         c.common.opcode = cmd.opcode;
1399         c.common.flags = cmd.flags;
1400         c.common.nsid = cpu_to_le32(cmd.nsid);
1401         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1402         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1403         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1404         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1405         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1406         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1407         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1408         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1409
1410         if (cmd.timeout_ms)
1411                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1412
1413         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1414         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1415                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1416                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1417                         0, &cmd.result, timeout);
1418         nvme_passthru_end(ctrl, effects);
1419
1420         if (status >= 0) {
1421                 if (put_user(cmd.result, &ucmd->result))
1422                         return -EFAULT;
1423         }
1424
1425         return status;
1426 }
1427
1428 /*
1429  * Issue ioctl requests on the first available path.  Note that unlike normal
1430  * block layer requests we will not retry failed request on another controller.
1431  */
1432 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1433                 struct nvme_ns_head **head, int *srcu_idx)
1434 {
1435 #ifdef CONFIG_NVME_MULTIPATH
1436         if (disk->fops == &nvme_ns_head_ops) {
1437                 struct nvme_ns *ns;
1438
1439                 *head = disk->private_data;
1440                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1441                 ns = nvme_find_path(*head);
1442                 if (!ns)
1443                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1444                 return ns;
1445         }
1446 #endif
1447         *head = NULL;
1448         *srcu_idx = -1;
1449         return disk->private_data;
1450 }
1451
1452 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1453 {
1454         if (head)
1455                 srcu_read_unlock(&head->srcu, idx);
1456 }
1457
1458 static bool is_ctrl_ioctl(unsigned int cmd)
1459 {
1460         if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1461                 return true;
1462         if (is_sed_ioctl(cmd))
1463                 return true;
1464         return false;
1465 }
1466
1467 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1468                                   void __user *argp,
1469                                   struct nvme_ns_head *head,
1470                                   int srcu_idx)
1471 {
1472         struct nvme_ctrl *ctrl = ns->ctrl;
1473         int ret;
1474
1475         nvme_get_ctrl(ns->ctrl);
1476         nvme_put_ns_from_disk(head, srcu_idx);
1477
1478         switch (cmd) {
1479         case NVME_IOCTL_ADMIN_CMD:
1480                 ret = nvme_user_cmd(ctrl, NULL, argp);
1481                 break;
1482         case NVME_IOCTL_ADMIN64_CMD:
1483                 ret = nvme_user_cmd64(ctrl, NULL, argp);
1484                 break;
1485         default:
1486                 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1487                 break;
1488         }
1489         nvme_put_ctrl(ctrl);
1490         return ret;
1491 }
1492
1493 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1494                 unsigned int cmd, unsigned long arg)
1495 {
1496         struct nvme_ns_head *head = NULL;
1497         void __user *argp = (void __user *)arg;
1498         struct nvme_ns *ns;
1499         int srcu_idx, ret;
1500
1501         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1502         if (unlikely(!ns))
1503                 return -EWOULDBLOCK;
1504
1505         /*
1506          * Handle ioctls that apply to the controller instead of the namespace
1507          * seperately and drop the ns SRCU reference early.  This avoids a
1508          * deadlock when deleting namespaces using the passthrough interface.
1509          */
1510         if (is_ctrl_ioctl(cmd))
1511                 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1512
1513         switch (cmd) {
1514         case NVME_IOCTL_ID:
1515                 force_successful_syscall_return();
1516                 ret = ns->head->ns_id;
1517                 break;
1518         case NVME_IOCTL_IO_CMD:
1519                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1520                 break;
1521         case NVME_IOCTL_SUBMIT_IO:
1522                 ret = nvme_submit_io(ns, argp);
1523                 break;
1524         case NVME_IOCTL_IO64_CMD:
1525                 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1526                 break;
1527         default:
1528                 if (ns->ndev)
1529                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1530                 else
1531                         ret = -ENOTTY;
1532         }
1533
1534         nvme_put_ns_from_disk(head, srcu_idx);
1535         return ret;
1536 }
1537
1538 static int nvme_open(struct block_device *bdev, fmode_t mode)
1539 {
1540         struct nvme_ns *ns = bdev->bd_disk->private_data;
1541
1542 #ifdef CONFIG_NVME_MULTIPATH
1543         /* should never be called due to GENHD_FL_HIDDEN */
1544         if (WARN_ON_ONCE(ns->head->disk))
1545                 goto fail;
1546 #endif
1547         if (!kref_get_unless_zero(&ns->kref))
1548                 goto fail;
1549         if (!try_module_get(ns->ctrl->ops->module))
1550                 goto fail_put_ns;
1551
1552         return 0;
1553
1554 fail_put_ns:
1555         nvme_put_ns(ns);
1556 fail:
1557         return -ENXIO;
1558 }
1559
1560 static void nvme_release(struct gendisk *disk, fmode_t mode)
1561 {
1562         struct nvme_ns *ns = disk->private_data;
1563
1564         module_put(ns->ctrl->ops->module);
1565         nvme_put_ns(ns);
1566 }
1567
1568 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1569 {
1570         /* some standard values */
1571         geo->heads = 1 << 6;
1572         geo->sectors = 1 << 5;
1573         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1574         return 0;
1575 }
1576
1577 #ifdef CONFIG_BLK_DEV_INTEGRITY
1578 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1579 {
1580         struct blk_integrity integrity;
1581
1582         memset(&integrity, 0, sizeof(integrity));
1583         switch (pi_type) {
1584         case NVME_NS_DPS_PI_TYPE3:
1585                 integrity.profile = &t10_pi_type3_crc;
1586                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1587                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1588                 break;
1589         case NVME_NS_DPS_PI_TYPE1:
1590         case NVME_NS_DPS_PI_TYPE2:
1591                 integrity.profile = &t10_pi_type1_crc;
1592                 integrity.tag_size = sizeof(u16);
1593                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1594                 break;
1595         default:
1596                 integrity.profile = NULL;
1597                 break;
1598         }
1599         integrity.tuple_size = ms;
1600         blk_integrity_register(disk, &integrity);
1601         blk_queue_max_integrity_segments(disk->queue, 1);
1602 }
1603 #else
1604 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1605 {
1606 }
1607 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1608
1609 static void nvme_set_chunk_size(struct nvme_ns *ns)
1610 {
1611         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1612         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1613 }
1614
1615 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1616 {
1617         struct nvme_ctrl *ctrl = ns->ctrl;
1618         struct request_queue *queue = disk->queue;
1619         u32 size = queue_logical_block_size(queue);
1620
1621         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1622                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1623                 return;
1624         }
1625
1626         if (ctrl->nr_streams && ns->sws && ns->sgs)
1627                 size *= ns->sws * ns->sgs;
1628
1629         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1630                         NVME_DSM_MAX_RANGES);
1631
1632         queue->limits.discard_alignment = 0;
1633         queue->limits.discard_granularity = size;
1634
1635         /* If discard is already enabled, don't reset queue limits */
1636         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1637                 return;
1638
1639         blk_queue_max_discard_sectors(queue, UINT_MAX);
1640         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1641
1642         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1643                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1644 }
1645
1646 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1647 {
1648         u32 max_sectors;
1649         unsigned short bs = 1 << ns->lba_shift;
1650
1651         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1652             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1653                 return;
1654         /*
1655          * Even though NVMe spec explicitly states that MDTS is not
1656          * applicable to the write-zeroes:- "The restriction does not apply to
1657          * commands that do not transfer data between the host and the
1658          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1659          * In order to be more cautious use controller's max_hw_sectors value
1660          * to configure the maximum sectors for the write-zeroes which is
1661          * configured based on the controller's MDTS field in the
1662          * nvme_init_identify() if available.
1663          */
1664         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1665                 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1666         else
1667                 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1668
1669         blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1670 }
1671
1672 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1673                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1674 {
1675         int ret = 0;
1676
1677         memset(ids, 0, sizeof(*ids));
1678
1679         if (ctrl->vs >= NVME_VS(1, 1, 0))
1680                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1681         if (ctrl->vs >= NVME_VS(1, 2, 0))
1682                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1683         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1684                  /* Don't treat error as fatal we potentially
1685                   * already have a NGUID or EUI-64
1686                   */
1687                 ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1688                 if (ret)
1689                         dev_warn(ctrl->device,
1690                                  "Identify Descriptors failed (%d)\n", ret);
1691         }
1692         return ret;
1693 }
1694
1695 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1696 {
1697         return !uuid_is_null(&ids->uuid) ||
1698                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1699                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1700 }
1701
1702 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1703 {
1704         return uuid_equal(&a->uuid, &b->uuid) &&
1705                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1706                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1707 }
1708
1709 static void nvme_update_disk_info(struct gendisk *disk,
1710                 struct nvme_ns *ns, struct nvme_id_ns *id)
1711 {
1712         sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1713         unsigned short bs = 1 << ns->lba_shift;
1714         u32 atomic_bs, phys_bs, io_opt;
1715
1716         if (ns->lba_shift > PAGE_SHIFT) {
1717                 /* unsupported block size, set capacity to 0 later */
1718                 bs = (1 << 9);
1719         }
1720         blk_mq_freeze_queue(disk->queue);
1721         blk_integrity_unregister(disk);
1722
1723         if (id->nabo == 0) {
1724                 /*
1725                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1726                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1727                  * 0 then AWUPF must be used instead.
1728                  */
1729                 if (id->nsfeat & (1 << 1) && id->nawupf)
1730                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1731                 else
1732                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1733         } else {
1734                 atomic_bs = bs;
1735         }
1736         phys_bs = bs;
1737         io_opt = bs;
1738         if (id->nsfeat & (1 << 4)) {
1739                 /* NPWG = Namespace Preferred Write Granularity */
1740                 phys_bs *= 1 + le16_to_cpu(id->npwg);
1741                 /* NOWS = Namespace Optimal Write Size */
1742                 io_opt *= 1 + le16_to_cpu(id->nows);
1743         }
1744
1745         blk_queue_logical_block_size(disk->queue, bs);
1746         /*
1747          * Linux filesystems assume writing a single physical block is
1748          * an atomic operation. Hence limit the physical block size to the
1749          * value of the Atomic Write Unit Power Fail parameter.
1750          */
1751         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1752         blk_queue_io_min(disk->queue, phys_bs);
1753         blk_queue_io_opt(disk->queue, io_opt);
1754
1755         if (ns->ms && !ns->ext &&
1756             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1757                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1758         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1759             ns->lba_shift > PAGE_SHIFT)
1760                 capacity = 0;
1761
1762         set_capacity(disk, capacity);
1763
1764         nvme_config_discard(disk, ns);
1765         nvme_config_write_zeroes(disk, ns);
1766
1767         if (id->nsattr & (1 << 0))
1768                 set_disk_ro(disk, true);
1769         else
1770                 set_disk_ro(disk, false);
1771
1772         blk_mq_unfreeze_queue(disk->queue);
1773 }
1774
1775 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1776 {
1777         struct nvme_ns *ns = disk->private_data;
1778
1779         /*
1780          * If identify namespace failed, use default 512 byte block size so
1781          * block layer can use before failing read/write for 0 capacity.
1782          */
1783         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1784         if (ns->lba_shift == 0)
1785                 ns->lba_shift = 9;
1786         ns->noiob = le16_to_cpu(id->noiob);
1787         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1788         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1789         /* the PI implementation requires metadata equal t10 pi tuple size */
1790         if (ns->ms == sizeof(struct t10_pi_tuple))
1791                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1792         else
1793                 ns->pi_type = 0;
1794
1795         if (ns->noiob)
1796                 nvme_set_chunk_size(ns);
1797         nvme_update_disk_info(disk, ns, id);
1798 #ifdef CONFIG_NVME_MULTIPATH
1799         if (ns->head->disk) {
1800                 nvme_update_disk_info(ns->head->disk, ns, id);
1801                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1802                 revalidate_disk(ns->head->disk);
1803         }
1804 #endif
1805 }
1806
1807 static int nvme_revalidate_disk(struct gendisk *disk)
1808 {
1809         struct nvme_ns *ns = disk->private_data;
1810         struct nvme_ctrl *ctrl = ns->ctrl;
1811         struct nvme_id_ns *id;
1812         struct nvme_ns_ids ids;
1813         int ret = 0;
1814
1815         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1816                 set_capacity(disk, 0);
1817                 return -ENODEV;
1818         }
1819
1820         ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1821         if (ret)
1822                 goto out;
1823
1824         if (id->ncap == 0) {
1825                 ret = -ENODEV;
1826                 goto free_id;
1827         }
1828
1829         __nvme_revalidate_disk(disk, id);
1830         ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1831         if (ret)
1832                 goto free_id;
1833
1834         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1835                 dev_err(ctrl->device,
1836                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1837                 ret = -ENODEV;
1838         }
1839
1840 free_id:
1841         kfree(id);
1842 out:
1843         /*
1844          * Only fail the function if we got a fatal error back from the
1845          * device, otherwise ignore the error and just move on.
1846          */
1847         if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1848                 ret = 0;
1849         else if (ret > 0)
1850                 ret = blk_status_to_errno(nvme_error_status(ret));
1851         return ret;
1852 }
1853
1854 static char nvme_pr_type(enum pr_type type)
1855 {
1856         switch (type) {
1857         case PR_WRITE_EXCLUSIVE:
1858                 return 1;
1859         case PR_EXCLUSIVE_ACCESS:
1860                 return 2;
1861         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1862                 return 3;
1863         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1864                 return 4;
1865         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1866                 return 5;
1867         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1868                 return 6;
1869         default:
1870                 return 0;
1871         }
1872 };
1873
1874 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1875                                 u64 key, u64 sa_key, u8 op)
1876 {
1877         struct nvme_ns_head *head = NULL;
1878         struct nvme_ns *ns;
1879         struct nvme_command c;
1880         int srcu_idx, ret;
1881         u8 data[16] = { 0, };
1882
1883         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1884         if (unlikely(!ns))
1885                 return -EWOULDBLOCK;
1886
1887         put_unaligned_le64(key, &data[0]);
1888         put_unaligned_le64(sa_key, &data[8]);
1889
1890         memset(&c, 0, sizeof(c));
1891         c.common.opcode = op;
1892         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1893         c.common.cdw10 = cpu_to_le32(cdw10);
1894
1895         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1896         nvme_put_ns_from_disk(head, srcu_idx);
1897         return ret;
1898 }
1899
1900 static int nvme_pr_register(struct block_device *bdev, u64 old,
1901                 u64 new, unsigned flags)
1902 {
1903         u32 cdw10;
1904
1905         if (flags & ~PR_FL_IGNORE_KEY)
1906                 return -EOPNOTSUPP;
1907
1908         cdw10 = old ? 2 : 0;
1909         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1910         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1911         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1912 }
1913
1914 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1915                 enum pr_type type, unsigned flags)
1916 {
1917         u32 cdw10;
1918
1919         if (flags & ~PR_FL_IGNORE_KEY)
1920                 return -EOPNOTSUPP;
1921
1922         cdw10 = nvme_pr_type(type) << 8;
1923         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1924         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1925 }
1926
1927 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1928                 enum pr_type type, bool abort)
1929 {
1930         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1931         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1932 }
1933
1934 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1935 {
1936         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1937         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1938 }
1939
1940 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1941 {
1942         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1943         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1944 }
1945
1946 static const struct pr_ops nvme_pr_ops = {
1947         .pr_register    = nvme_pr_register,
1948         .pr_reserve     = nvme_pr_reserve,
1949         .pr_release     = nvme_pr_release,
1950         .pr_preempt     = nvme_pr_preempt,
1951         .pr_clear       = nvme_pr_clear,
1952 };
1953
1954 #ifdef CONFIG_BLK_SED_OPAL
1955 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1956                 bool send)
1957 {
1958         struct nvme_ctrl *ctrl = data;
1959         struct nvme_command cmd;
1960
1961         memset(&cmd, 0, sizeof(cmd));
1962         if (send)
1963                 cmd.common.opcode = nvme_admin_security_send;
1964         else
1965                 cmd.common.opcode = nvme_admin_security_recv;
1966         cmd.common.nsid = 0;
1967         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1968         cmd.common.cdw11 = cpu_to_le32(len);
1969
1970         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1971                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1972 }
1973 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1974 #endif /* CONFIG_BLK_SED_OPAL */
1975
1976 static const struct block_device_operations nvme_fops = {
1977         .owner          = THIS_MODULE,
1978         .ioctl          = nvme_ioctl,
1979         .compat_ioctl   = nvme_ioctl,
1980         .open           = nvme_open,
1981         .release        = nvme_release,
1982         .getgeo         = nvme_getgeo,
1983         .revalidate_disk= nvme_revalidate_disk,
1984         .pr_ops         = &nvme_pr_ops,
1985 };
1986
1987 #ifdef CONFIG_NVME_MULTIPATH
1988 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1989 {
1990         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1991
1992         if (!kref_get_unless_zero(&head->ref))
1993                 return -ENXIO;
1994         return 0;
1995 }
1996
1997 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1998 {
1999         nvme_put_ns_head(disk->private_data);
2000 }
2001
2002 const struct block_device_operations nvme_ns_head_ops = {
2003         .owner          = THIS_MODULE,
2004         .open           = nvme_ns_head_open,
2005         .release        = nvme_ns_head_release,
2006         .ioctl          = nvme_ioctl,
2007         .compat_ioctl   = nvme_ioctl,
2008         .getgeo         = nvme_getgeo,
2009         .pr_ops         = &nvme_pr_ops,
2010 };
2011 #endif /* CONFIG_NVME_MULTIPATH */
2012
2013 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2014 {
2015         unsigned long timeout =
2016                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2017         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2018         int ret;
2019
2020         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2021                 if (csts == ~0)
2022                         return -ENODEV;
2023                 if ((csts & NVME_CSTS_RDY) == bit)
2024                         break;
2025
2026                 msleep(100);
2027                 if (fatal_signal_pending(current))
2028                         return -EINTR;
2029                 if (time_after(jiffies, timeout)) {
2030                         dev_err(ctrl->device,
2031                                 "Device not ready; aborting %s\n", enabled ?
2032                                                 "initialisation" : "reset");
2033                         return -ENODEV;
2034                 }
2035         }
2036
2037         return ret;
2038 }
2039
2040 /*
2041  * If the device has been passed off to us in an enabled state, just clear
2042  * the enabled bit.  The spec says we should set the 'shutdown notification
2043  * bits', but doing so may cause the device to complete commands to the
2044  * admin queue ... and we don't know what memory that might be pointing at!
2045  */
2046 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2047 {
2048         int ret;
2049
2050         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2051         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2052
2053         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2054         if (ret)
2055                 return ret;
2056
2057         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2058                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2059
2060         return nvme_wait_ready(ctrl, ctrl->cap, false);
2061 }
2062 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2063
2064 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2065 {
2066         /*
2067          * Default to a 4K page size, with the intention to update this
2068          * path in the future to accomodate architectures with differing
2069          * kernel and IO page sizes.
2070          */
2071         unsigned dev_page_min, page_shift = 12;
2072         int ret;
2073
2074         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2075         if (ret) {
2076                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2077                 return ret;
2078         }
2079         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2080
2081         if (page_shift < dev_page_min) {
2082                 dev_err(ctrl->device,
2083                         "Minimum device page size %u too large for host (%u)\n",
2084                         1 << dev_page_min, 1 << page_shift);
2085                 return -ENODEV;
2086         }
2087
2088         ctrl->page_size = 1 << page_shift;
2089
2090         ctrl->ctrl_config = NVME_CC_CSS_NVM;
2091         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2092         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2093         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2094         ctrl->ctrl_config |= NVME_CC_ENABLE;
2095
2096         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2097         if (ret)
2098                 return ret;
2099         return nvme_wait_ready(ctrl, ctrl->cap, true);
2100 }
2101 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2102
2103 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2104 {
2105         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2106         u32 csts;
2107         int ret;
2108
2109         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2110         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2111
2112         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2113         if (ret)
2114                 return ret;
2115
2116         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2117                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2118                         break;
2119
2120                 msleep(100);
2121                 if (fatal_signal_pending(current))
2122                         return -EINTR;
2123                 if (time_after(jiffies, timeout)) {
2124                         dev_err(ctrl->device,
2125                                 "Device shutdown incomplete; abort shutdown\n");
2126                         return -ENODEV;
2127                 }
2128         }
2129
2130         return ret;
2131 }
2132 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2133
2134 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2135                 struct request_queue *q)
2136 {
2137         bool vwc = false;
2138
2139         if (ctrl->max_hw_sectors) {
2140                 u32 max_segments =
2141                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2142
2143                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2144                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2145                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2146         }
2147         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2148             is_power_of_2(ctrl->max_hw_sectors))
2149                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2150         blk_queue_virt_boundary(q, ctrl->page_size - 1);
2151         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2152                 vwc = true;
2153         blk_queue_write_cache(q, vwc, vwc);
2154 }
2155
2156 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2157 {
2158         __le64 ts;
2159         int ret;
2160
2161         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2162                 return 0;
2163
2164         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2165         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2166                         NULL);
2167         if (ret)
2168                 dev_warn_once(ctrl->device,
2169                         "could not set timestamp (%d)\n", ret);
2170         return ret;
2171 }
2172
2173 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2174 {
2175         struct nvme_feat_host_behavior *host;
2176         int ret;
2177
2178         /* Don't bother enabling the feature if retry delay is not reported */
2179         if (!ctrl->crdt[0])
2180                 return 0;
2181
2182         host = kzalloc(sizeof(*host), GFP_KERNEL);
2183         if (!host)
2184                 return 0;
2185
2186         host->acre = NVME_ENABLE_ACRE;
2187         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2188                                 host, sizeof(*host), NULL);
2189         kfree(host);
2190         return ret;
2191 }
2192
2193 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2194 {
2195         /*
2196          * APST (Autonomous Power State Transition) lets us program a
2197          * table of power state transitions that the controller will
2198          * perform automatically.  We configure it with a simple
2199          * heuristic: we are willing to spend at most 2% of the time
2200          * transitioning between power states.  Therefore, when running
2201          * in any given state, we will enter the next lower-power
2202          * non-operational state after waiting 50 * (enlat + exlat)
2203          * microseconds, as long as that state's exit latency is under
2204          * the requested maximum latency.
2205          *
2206          * We will not autonomously enter any non-operational state for
2207          * which the total latency exceeds ps_max_latency_us.  Users
2208          * can set ps_max_latency_us to zero to turn off APST.
2209          */
2210
2211         unsigned apste;
2212         struct nvme_feat_auto_pst *table;
2213         u64 max_lat_us = 0;
2214         int max_ps = -1;
2215         int ret;
2216
2217         /*
2218          * If APST isn't supported or if we haven't been initialized yet,
2219          * then don't do anything.
2220          */
2221         if (!ctrl->apsta)
2222                 return 0;
2223
2224         if (ctrl->npss > 31) {
2225                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2226                 return 0;
2227         }
2228
2229         table = kzalloc(sizeof(*table), GFP_KERNEL);
2230         if (!table)
2231                 return 0;
2232
2233         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2234                 /* Turn off APST. */
2235                 apste = 0;
2236                 dev_dbg(ctrl->device, "APST disabled\n");
2237         } else {
2238                 __le64 target = cpu_to_le64(0);
2239                 int state;
2240
2241                 /*
2242                  * Walk through all states from lowest- to highest-power.
2243                  * According to the spec, lower-numbered states use more
2244                  * power.  NPSS, despite the name, is the index of the
2245                  * lowest-power state, not the number of states.
2246                  */
2247                 for (state = (int)ctrl->npss; state >= 0; state--) {
2248                         u64 total_latency_us, exit_latency_us, transition_ms;
2249
2250                         if (target)
2251                                 table->entries[state] = target;
2252
2253                         /*
2254                          * Don't allow transitions to the deepest state
2255                          * if it's quirked off.
2256                          */
2257                         if (state == ctrl->npss &&
2258                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2259                                 continue;
2260
2261                         /*
2262                          * Is this state a useful non-operational state for
2263                          * higher-power states to autonomously transition to?
2264                          */
2265                         if (!(ctrl->psd[state].flags &
2266                               NVME_PS_FLAGS_NON_OP_STATE))
2267                                 continue;
2268
2269                         exit_latency_us =
2270                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2271                         if (exit_latency_us > ctrl->ps_max_latency_us)
2272                                 continue;
2273
2274                         total_latency_us =
2275                                 exit_latency_us +
2276                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2277
2278                         /*
2279                          * This state is good.  Use it as the APST idle
2280                          * target for higher power states.
2281                          */
2282                         transition_ms = total_latency_us + 19;
2283                         do_div(transition_ms, 20);
2284                         if (transition_ms > (1 << 24) - 1)
2285                                 transition_ms = (1 << 24) - 1;
2286
2287                         target = cpu_to_le64((state << 3) |
2288                                              (transition_ms << 8));
2289
2290                         if (max_ps == -1)
2291                                 max_ps = state;
2292
2293                         if (total_latency_us > max_lat_us)
2294                                 max_lat_us = total_latency_us;
2295                 }
2296
2297                 apste = 1;
2298
2299                 if (max_ps == -1) {
2300                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2301                 } else {
2302                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2303                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2304                 }
2305         }
2306
2307         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2308                                 table, sizeof(*table), NULL);
2309         if (ret)
2310                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2311
2312         kfree(table);
2313         return ret;
2314 }
2315
2316 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2317 {
2318         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2319         u64 latency;
2320
2321         switch (val) {
2322         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2323         case PM_QOS_LATENCY_ANY:
2324                 latency = U64_MAX;
2325                 break;
2326
2327         default:
2328                 latency = val;
2329         }
2330
2331         if (ctrl->ps_max_latency_us != latency) {
2332                 ctrl->ps_max_latency_us = latency;
2333                 nvme_configure_apst(ctrl);
2334         }
2335 }
2336
2337 struct nvme_core_quirk_entry {
2338         /*
2339          * NVMe model and firmware strings are padded with spaces.  For
2340          * simplicity, strings in the quirk table are padded with NULLs
2341          * instead.
2342          */
2343         u16 vid;
2344         const char *mn;
2345         const char *fr;
2346         unsigned long quirks;
2347 };
2348
2349 static const struct nvme_core_quirk_entry core_quirks[] = {
2350         {
2351                 /*
2352                  * This Toshiba device seems to die using any APST states.  See:
2353                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2354                  */
2355                 .vid = 0x1179,
2356                 .mn = "THNSF5256GPUK TOSHIBA",
2357                 .quirks = NVME_QUIRK_NO_APST,
2358         },
2359         {
2360                 /*
2361                  * This LiteON CL1-3D*-Q11 firmware version has a race
2362                  * condition associated with actions related to suspend to idle
2363                  * LiteON has resolved the problem in future firmware
2364                  */
2365                 .vid = 0x14a4,
2366                 .fr = "22301111",
2367                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2368         },
2369         {
2370                 /*
2371                  * This Kingston E8FK11.T firmware version has no interrupt
2372                  * after resume with actions related to suspend to idle
2373                  * https://bugzilla.kernel.org/show_bug.cgi?id=204887
2374                  */
2375                 .vid = 0x2646,
2376                 .fr = "E8FK11.T",
2377                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2378         }
2379 };
2380
2381 /* match is null-terminated but idstr is space-padded. */
2382 static bool string_matches(const char *idstr, const char *match, size_t len)
2383 {
2384         size_t matchlen;
2385
2386         if (!match)
2387                 return true;
2388
2389         matchlen = strlen(match);
2390         WARN_ON_ONCE(matchlen > len);
2391
2392         if (memcmp(idstr, match, matchlen))
2393                 return false;
2394
2395         for (; matchlen < len; matchlen++)
2396                 if (idstr[matchlen] != ' ')
2397                         return false;
2398
2399         return true;
2400 }
2401
2402 static bool quirk_matches(const struct nvme_id_ctrl *id,
2403                           const struct nvme_core_quirk_entry *q)
2404 {
2405         return q->vid == le16_to_cpu(id->vid) &&
2406                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2407                 string_matches(id->fr, q->fr, sizeof(id->fr));
2408 }
2409
2410 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2411                 struct nvme_id_ctrl *id)
2412 {
2413         size_t nqnlen;
2414         int off;
2415
2416         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2417                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2418                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2419                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2420                         return;
2421                 }
2422
2423                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2424                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2425         }
2426
2427         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2428         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2429                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2430                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2431         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2432         off += sizeof(id->sn);
2433         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2434         off += sizeof(id->mn);
2435         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2436 }
2437
2438 static void nvme_release_subsystem(struct device *dev)
2439 {
2440         struct nvme_subsystem *subsys =
2441                 container_of(dev, struct nvme_subsystem, dev);
2442
2443         if (subsys->instance >= 0)
2444                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2445         kfree(subsys);
2446 }
2447
2448 static void nvme_destroy_subsystem(struct kref *ref)
2449 {
2450         struct nvme_subsystem *subsys =
2451                         container_of(ref, struct nvme_subsystem, ref);
2452
2453         mutex_lock(&nvme_subsystems_lock);
2454         list_del(&subsys->entry);
2455         mutex_unlock(&nvme_subsystems_lock);
2456
2457         ida_destroy(&subsys->ns_ida);
2458         device_del(&subsys->dev);
2459         put_device(&subsys->dev);
2460 }
2461
2462 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2463 {
2464         kref_put(&subsys->ref, nvme_destroy_subsystem);
2465 }
2466
2467 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2468 {
2469         struct nvme_subsystem *subsys;
2470
2471         lockdep_assert_held(&nvme_subsystems_lock);
2472
2473         /*
2474          * Fail matches for discovery subsystems. This results
2475          * in each discovery controller bound to a unique subsystem.
2476          * This avoids issues with validating controller values
2477          * that can only be true when there is a single unique subsystem.
2478          * There may be multiple and completely independent entities
2479          * that provide discovery controllers.
2480          */
2481         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2482                 return NULL;
2483
2484         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2485                 if (strcmp(subsys->subnqn, subsysnqn))
2486                         continue;
2487                 if (!kref_get_unless_zero(&subsys->ref))
2488                         continue;
2489                 return subsys;
2490         }
2491
2492         return NULL;
2493 }
2494
2495 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2496         struct device_attribute subsys_attr_##_name = \
2497                 __ATTR(_name, _mode, _show, NULL)
2498
2499 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2500                                     struct device_attribute *attr,
2501                                     char *buf)
2502 {
2503         struct nvme_subsystem *subsys =
2504                 container_of(dev, struct nvme_subsystem, dev);
2505
2506         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2507 }
2508 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2509
2510 #define nvme_subsys_show_str_function(field)                            \
2511 static ssize_t subsys_##field##_show(struct device *dev,                \
2512                             struct device_attribute *attr, char *buf)   \
2513 {                                                                       \
2514         struct nvme_subsystem *subsys =                                 \
2515                 container_of(dev, struct nvme_subsystem, dev);          \
2516         return sprintf(buf, "%.*s\n",                                   \
2517                        (int)sizeof(subsys->field), subsys->field);      \
2518 }                                                                       \
2519 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2520
2521 nvme_subsys_show_str_function(model);
2522 nvme_subsys_show_str_function(serial);
2523 nvme_subsys_show_str_function(firmware_rev);
2524
2525 static struct attribute *nvme_subsys_attrs[] = {
2526         &subsys_attr_model.attr,
2527         &subsys_attr_serial.attr,
2528         &subsys_attr_firmware_rev.attr,
2529         &subsys_attr_subsysnqn.attr,
2530 #ifdef CONFIG_NVME_MULTIPATH
2531         &subsys_attr_iopolicy.attr,
2532 #endif
2533         NULL,
2534 };
2535
2536 static struct attribute_group nvme_subsys_attrs_group = {
2537         .attrs = nvme_subsys_attrs,
2538 };
2539
2540 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2541         &nvme_subsys_attrs_group,
2542         NULL,
2543 };
2544
2545 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2546                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2547 {
2548         struct nvme_ctrl *tmp;
2549
2550         lockdep_assert_held(&nvme_subsystems_lock);
2551
2552         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2553                 if (tmp->state == NVME_CTRL_DELETING ||
2554                     tmp->state == NVME_CTRL_DEAD)
2555                         continue;
2556
2557                 if (tmp->cntlid == ctrl->cntlid) {
2558                         dev_err(ctrl->device,
2559                                 "Duplicate cntlid %u with %s, rejecting\n",
2560                                 ctrl->cntlid, dev_name(tmp->device));
2561                         return false;
2562                 }
2563
2564                 if ((id->cmic & (1 << 1)) ||
2565                     (ctrl->opts && ctrl->opts->discovery_nqn))
2566                         continue;
2567
2568                 dev_err(ctrl->device,
2569                         "Subsystem does not support multiple controllers\n");
2570                 return false;
2571         }
2572
2573         return true;
2574 }
2575
2576 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2577 {
2578         struct nvme_subsystem *subsys, *found;
2579         int ret;
2580
2581         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2582         if (!subsys)
2583                 return -ENOMEM;
2584
2585         subsys->instance = -1;
2586         mutex_init(&subsys->lock);
2587         kref_init(&subsys->ref);
2588         INIT_LIST_HEAD(&subsys->ctrls);
2589         INIT_LIST_HEAD(&subsys->nsheads);
2590         nvme_init_subnqn(subsys, ctrl, id);
2591         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2592         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2593         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2594         subsys->vendor_id = le16_to_cpu(id->vid);
2595         subsys->cmic = id->cmic;
2596         subsys->awupf = le16_to_cpu(id->awupf);
2597 #ifdef CONFIG_NVME_MULTIPATH
2598         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2599 #endif
2600
2601         subsys->dev.class = nvme_subsys_class;
2602         subsys->dev.release = nvme_release_subsystem;
2603         subsys->dev.groups = nvme_subsys_attrs_groups;
2604         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2605         device_initialize(&subsys->dev);
2606
2607         mutex_lock(&nvme_subsystems_lock);
2608         found = __nvme_find_get_subsystem(subsys->subnqn);
2609         if (found) {
2610                 put_device(&subsys->dev);
2611                 subsys = found;
2612
2613                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2614                         ret = -EINVAL;
2615                         goto out_put_subsystem;
2616                 }
2617         } else {
2618                 ret = device_add(&subsys->dev);
2619                 if (ret) {
2620                         dev_err(ctrl->device,
2621                                 "failed to register subsystem device.\n");
2622                         put_device(&subsys->dev);
2623                         goto out_unlock;
2624                 }
2625                 ida_init(&subsys->ns_ida);
2626                 list_add_tail(&subsys->entry, &nvme_subsystems);
2627         }
2628
2629         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2630                                 dev_name(ctrl->device));
2631         if (ret) {
2632                 dev_err(ctrl->device,
2633                         "failed to create sysfs link from subsystem.\n");
2634                 goto out_put_subsystem;
2635         }
2636
2637         if (!found)
2638                 subsys->instance = ctrl->instance;
2639         ctrl->subsys = subsys;
2640         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2641         mutex_unlock(&nvme_subsystems_lock);
2642         return 0;
2643
2644 out_put_subsystem:
2645         nvme_put_subsystem(subsys);
2646 out_unlock:
2647         mutex_unlock(&nvme_subsystems_lock);
2648         return ret;
2649 }
2650
2651 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2652                 void *log, size_t size, u64 offset)
2653 {
2654         struct nvme_command c = { };
2655         unsigned long dwlen = size / 4 - 1;
2656
2657         c.get_log_page.opcode = nvme_admin_get_log_page;
2658         c.get_log_page.nsid = cpu_to_le32(nsid);
2659         c.get_log_page.lid = log_page;
2660         c.get_log_page.lsp = lsp;
2661         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2662         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2663         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2664         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2665
2666         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2667 }
2668
2669 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2670 {
2671         int ret;
2672
2673         if (!ctrl->effects)
2674                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2675
2676         if (!ctrl->effects)
2677                 return 0;
2678
2679         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2680                         ctrl->effects, sizeof(*ctrl->effects), 0);
2681         if (ret) {
2682                 kfree(ctrl->effects);
2683                 ctrl->effects = NULL;
2684         }
2685         return ret;
2686 }
2687
2688 /*
2689  * Initialize the cached copies of the Identify data and various controller
2690  * register in our nvme_ctrl structure.  This should be called as soon as
2691  * the admin queue is fully up and running.
2692  */
2693 int nvme_init_identify(struct nvme_ctrl *ctrl)
2694 {
2695         struct nvme_id_ctrl *id;
2696         int ret, page_shift;
2697         u32 max_hw_sectors;
2698         bool prev_apst_enabled;
2699
2700         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2701         if (ret) {
2702                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2703                 return ret;
2704         }
2705         page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2706         ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2707
2708         if (ctrl->vs >= NVME_VS(1, 1, 0))
2709                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2710
2711         ret = nvme_identify_ctrl(ctrl, &id);
2712         if (ret) {
2713                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2714                 return -EIO;
2715         }
2716
2717         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2718                 ret = nvme_get_effects_log(ctrl);
2719                 if (ret < 0)
2720                         goto out_free;
2721         }
2722
2723         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2724                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2725
2726         if (!ctrl->identified) {
2727                 int i;
2728
2729                 ret = nvme_init_subsystem(ctrl, id);
2730                 if (ret)
2731                         goto out_free;
2732
2733                 /*
2734                  * Check for quirks.  Quirk can depend on firmware version,
2735                  * so, in principle, the set of quirks present can change
2736                  * across a reset.  As a possible future enhancement, we
2737                  * could re-scan for quirks every time we reinitialize
2738                  * the device, but we'd have to make sure that the driver
2739                  * behaves intelligently if the quirks change.
2740                  */
2741                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2742                         if (quirk_matches(id, &core_quirks[i]))
2743                                 ctrl->quirks |= core_quirks[i].quirks;
2744                 }
2745         }
2746
2747         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2748                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2749                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2750         }
2751
2752         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2753         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2754         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2755
2756         ctrl->oacs = le16_to_cpu(id->oacs);
2757         ctrl->oncs = le16_to_cpu(id->oncs);
2758         ctrl->mtfa = le16_to_cpu(id->mtfa);
2759         ctrl->oaes = le32_to_cpu(id->oaes);
2760         atomic_set(&ctrl->abort_limit, id->acl + 1);
2761         ctrl->vwc = id->vwc;
2762         if (id->mdts)
2763                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2764         else
2765                 max_hw_sectors = UINT_MAX;
2766         ctrl->max_hw_sectors =
2767                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2768
2769         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2770         ctrl->sgls = le32_to_cpu(id->sgls);
2771         ctrl->kas = le16_to_cpu(id->kas);
2772         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2773         ctrl->ctratt = le32_to_cpu(id->ctratt);
2774
2775         if (id->rtd3e) {
2776                 /* us -> s */
2777                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2778
2779                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2780                                                  shutdown_timeout, 60);
2781
2782                 if (ctrl->shutdown_timeout != shutdown_timeout)
2783                         dev_info(ctrl->device,
2784                                  "Shutdown timeout set to %u seconds\n",
2785                                  ctrl->shutdown_timeout);
2786         } else
2787                 ctrl->shutdown_timeout = shutdown_timeout;
2788
2789         ctrl->npss = id->npss;
2790         ctrl->apsta = id->apsta;
2791         prev_apst_enabled = ctrl->apst_enabled;
2792         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2793                 if (force_apst && id->apsta) {
2794                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2795                         ctrl->apst_enabled = true;
2796                 } else {
2797                         ctrl->apst_enabled = false;
2798                 }
2799         } else {
2800                 ctrl->apst_enabled = id->apsta;
2801         }
2802         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2803
2804         if (ctrl->ops->flags & NVME_F_FABRICS) {
2805                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2806                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2807                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2808                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2809
2810                 /*
2811                  * In fabrics we need to verify the cntlid matches the
2812                  * admin connect
2813                  */
2814                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2815                         ret = -EINVAL;
2816                         goto out_free;
2817                 }
2818
2819                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2820                         dev_err(ctrl->device,
2821                                 "keep-alive support is mandatory for fabrics\n");
2822                         ret = -EINVAL;
2823                         goto out_free;
2824                 }
2825         } else {
2826                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2827                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2828                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2829                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2830         }
2831
2832         ret = nvme_mpath_init(ctrl, id);
2833         kfree(id);
2834
2835         if (ret < 0)
2836                 return ret;
2837
2838         if (ctrl->apst_enabled && !prev_apst_enabled)
2839                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2840         else if (!ctrl->apst_enabled && prev_apst_enabled)
2841                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2842
2843         ret = nvme_configure_apst(ctrl);
2844         if (ret < 0)
2845                 return ret;
2846
2847         ret = nvme_configure_timestamp(ctrl);
2848         if (ret < 0)
2849                 return ret;
2850
2851         ret = nvme_configure_directives(ctrl);
2852         if (ret < 0)
2853                 return ret;
2854
2855         ret = nvme_configure_acre(ctrl);
2856         if (ret < 0)
2857                 return ret;
2858
2859         ctrl->identified = true;
2860
2861         return 0;
2862
2863 out_free:
2864         kfree(id);
2865         return ret;
2866 }
2867 EXPORT_SYMBOL_GPL(nvme_init_identify);
2868
2869 static int nvme_dev_open(struct inode *inode, struct file *file)
2870 {
2871         struct nvme_ctrl *ctrl =
2872                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2873
2874         switch (ctrl->state) {
2875         case NVME_CTRL_LIVE:
2876         case NVME_CTRL_ADMIN_ONLY:
2877                 break;
2878         default:
2879                 return -EWOULDBLOCK;
2880         }
2881
2882         file->private_data = ctrl;
2883         return 0;
2884 }
2885
2886 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2887 {
2888         struct nvme_ns *ns;
2889         int ret;
2890
2891         down_read(&ctrl->namespaces_rwsem);
2892         if (list_empty(&ctrl->namespaces)) {
2893                 ret = -ENOTTY;
2894                 goto out_unlock;
2895         }
2896
2897         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2898         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2899                 dev_warn(ctrl->device,
2900                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2901                 ret = -EINVAL;
2902                 goto out_unlock;
2903         }
2904
2905         dev_warn(ctrl->device,
2906                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2907         kref_get(&ns->kref);
2908         up_read(&ctrl->namespaces_rwsem);
2909
2910         ret = nvme_user_cmd(ctrl, ns, argp);
2911         nvme_put_ns(ns);
2912         return ret;
2913
2914 out_unlock:
2915         up_read(&ctrl->namespaces_rwsem);
2916         return ret;
2917 }
2918
2919 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2920                 unsigned long arg)
2921 {
2922         struct nvme_ctrl *ctrl = file->private_data;
2923         void __user *argp = (void __user *)arg;
2924
2925         switch (cmd) {
2926         case NVME_IOCTL_ADMIN_CMD:
2927                 return nvme_user_cmd(ctrl, NULL, argp);
2928         case NVME_IOCTL_ADMIN64_CMD:
2929                 return nvme_user_cmd64(ctrl, NULL, argp);
2930         case NVME_IOCTL_IO_CMD:
2931                 return nvme_dev_user_cmd(ctrl, argp);
2932         case NVME_IOCTL_RESET:
2933                 dev_warn(ctrl->device, "resetting controller\n");
2934                 return nvme_reset_ctrl_sync(ctrl);
2935         case NVME_IOCTL_SUBSYS_RESET:
2936                 return nvme_reset_subsystem(ctrl);
2937         case NVME_IOCTL_RESCAN:
2938                 nvme_queue_scan(ctrl);
2939                 return 0;
2940         default:
2941                 return -ENOTTY;
2942         }
2943 }
2944
2945 static const struct file_operations nvme_dev_fops = {
2946         .owner          = THIS_MODULE,
2947         .open           = nvme_dev_open,
2948         .unlocked_ioctl = nvme_dev_ioctl,
2949         .compat_ioctl   = nvme_dev_ioctl,
2950 };
2951
2952 static ssize_t nvme_sysfs_reset(struct device *dev,
2953                                 struct device_attribute *attr, const char *buf,
2954                                 size_t count)
2955 {
2956         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2957         int ret;
2958
2959         ret = nvme_reset_ctrl_sync(ctrl);
2960         if (ret < 0)
2961                 return ret;
2962         return count;
2963 }
2964 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2965
2966 static ssize_t nvme_sysfs_rescan(struct device *dev,
2967                                 struct device_attribute *attr, const char *buf,
2968                                 size_t count)
2969 {
2970         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2971
2972         nvme_queue_scan(ctrl);
2973         return count;
2974 }
2975 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2976
2977 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2978 {
2979         struct gendisk *disk = dev_to_disk(dev);
2980
2981         if (disk->fops == &nvme_fops)
2982                 return nvme_get_ns_from_dev(dev)->head;
2983         else
2984                 return disk->private_data;
2985 }
2986
2987 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2988                 char *buf)
2989 {
2990         struct nvme_ns_head *head = dev_to_ns_head(dev);
2991         struct nvme_ns_ids *ids = &head->ids;
2992         struct nvme_subsystem *subsys = head->subsys;
2993         int serial_len = sizeof(subsys->serial);
2994         int model_len = sizeof(subsys->model);
2995
2996         if (!uuid_is_null(&ids->uuid))
2997                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2998
2999         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3000                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3001
3002         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3003                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3004
3005         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3006                                   subsys->serial[serial_len - 1] == '\0'))
3007                 serial_len--;
3008         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3009                                  subsys->model[model_len - 1] == '\0'))
3010                 model_len--;
3011
3012         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3013                 serial_len, subsys->serial, model_len, subsys->model,
3014                 head->ns_id);
3015 }
3016 static DEVICE_ATTR_RO(wwid);
3017
3018 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3019                 char *buf)
3020 {
3021         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3022 }
3023 static DEVICE_ATTR_RO(nguid);
3024
3025 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3026                 char *buf)
3027 {
3028         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3029
3030         /* For backward compatibility expose the NGUID to userspace if
3031          * we have no UUID set
3032          */
3033         if (uuid_is_null(&ids->uuid)) {
3034                 printk_ratelimited(KERN_WARNING
3035                                    "No UUID available providing old NGUID\n");
3036                 return sprintf(buf, "%pU\n", ids->nguid);
3037         }
3038         return sprintf(buf, "%pU\n", &ids->uuid);
3039 }
3040 static DEVICE_ATTR_RO(uuid);
3041
3042 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3043                 char *buf)
3044 {
3045         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3046 }
3047 static DEVICE_ATTR_RO(eui);
3048
3049 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3050                 char *buf)
3051 {
3052         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3053 }
3054 static DEVICE_ATTR_RO(nsid);
3055
3056 static struct attribute *nvme_ns_id_attrs[] = {
3057         &dev_attr_wwid.attr,
3058         &dev_attr_uuid.attr,
3059         &dev_attr_nguid.attr,
3060         &dev_attr_eui.attr,
3061         &dev_attr_nsid.attr,
3062 #ifdef CONFIG_NVME_MULTIPATH
3063         &dev_attr_ana_grpid.attr,
3064         &dev_attr_ana_state.attr,
3065 #endif
3066         NULL,
3067 };
3068
3069 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3070                 struct attribute *a, int n)
3071 {
3072         struct device *dev = container_of(kobj, struct device, kobj);
3073         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3074
3075         if (a == &dev_attr_uuid.attr) {
3076                 if (uuid_is_null(&ids->uuid) &&
3077                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3078                         return 0;
3079         }
3080         if (a == &dev_attr_nguid.attr) {
3081                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3082                         return 0;
3083         }
3084         if (a == &dev_attr_eui.attr) {
3085                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3086                         return 0;
3087         }
3088 #ifdef CONFIG_NVME_MULTIPATH
3089         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3090                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3091                         return 0;
3092                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3093                         return 0;
3094         }
3095 #endif
3096         return a->mode;
3097 }
3098
3099 static const struct attribute_group nvme_ns_id_attr_group = {
3100         .attrs          = nvme_ns_id_attrs,
3101         .is_visible     = nvme_ns_id_attrs_are_visible,
3102 };
3103
3104 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3105         &nvme_ns_id_attr_group,
3106 #ifdef CONFIG_NVM
3107         &nvme_nvm_attr_group,
3108 #endif
3109         NULL,
3110 };
3111
3112 #define nvme_show_str_function(field)                                           \
3113 static ssize_t  field##_show(struct device *dev,                                \
3114                             struct device_attribute *attr, char *buf)           \
3115 {                                                                               \
3116         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3117         return sprintf(buf, "%.*s\n",                                           \
3118                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3119 }                                                                               \
3120 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3121
3122 nvme_show_str_function(model);
3123 nvme_show_str_function(serial);
3124 nvme_show_str_function(firmware_rev);
3125
3126 #define nvme_show_int_function(field)                                           \
3127 static ssize_t  field##_show(struct device *dev,                                \
3128                             struct device_attribute *attr, char *buf)           \
3129 {                                                                               \
3130         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3131         return sprintf(buf, "%d\n", ctrl->field);       \
3132 }                                                                               \
3133 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3134
3135 nvme_show_int_function(cntlid);
3136 nvme_show_int_function(numa_node);
3137 nvme_show_int_function(queue_count);
3138 nvme_show_int_function(sqsize);
3139
3140 static ssize_t nvme_sysfs_delete(struct device *dev,
3141                                 struct device_attribute *attr, const char *buf,
3142                                 size_t count)
3143 {
3144         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3145
3146         if (device_remove_file_self(dev, attr))
3147                 nvme_delete_ctrl_sync(ctrl);
3148         return count;
3149 }
3150 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3151
3152 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3153                                          struct device_attribute *attr,
3154                                          char *buf)
3155 {
3156         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3157
3158         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3159 }
3160 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3161
3162 static ssize_t nvme_sysfs_show_state(struct device *dev,
3163                                      struct device_attribute *attr,
3164                                      char *buf)
3165 {
3166         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3167         static const char *const state_name[] = {
3168                 [NVME_CTRL_NEW]         = "new",
3169                 [NVME_CTRL_LIVE]        = "live",
3170                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
3171                 [NVME_CTRL_RESETTING]   = "resetting",
3172                 [NVME_CTRL_CONNECTING]  = "connecting",
3173                 [NVME_CTRL_DELETING]    = "deleting",
3174                 [NVME_CTRL_DEAD]        = "dead",
3175         };
3176
3177         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3178             state_name[ctrl->state])
3179                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3180
3181         return sprintf(buf, "unknown state\n");
3182 }
3183
3184 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3185
3186 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3187                                          struct device_attribute *attr,
3188                                          char *buf)
3189 {
3190         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3191
3192         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3193 }
3194 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3195
3196 static ssize_t nvme_sysfs_show_address(struct device *dev,
3197                                          struct device_attribute *attr,
3198                                          char *buf)
3199 {
3200         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3201
3202         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3203 }
3204 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3205
3206 static struct attribute *nvme_dev_attrs[] = {
3207         &dev_attr_reset_controller.attr,
3208         &dev_attr_rescan_controller.attr,
3209         &dev_attr_model.attr,
3210         &dev_attr_serial.attr,
3211         &dev_attr_firmware_rev.attr,
3212         &dev_attr_cntlid.attr,
3213         &dev_attr_delete_controller.attr,
3214         &dev_attr_transport.attr,
3215         &dev_attr_subsysnqn.attr,
3216         &dev_attr_address.attr,
3217         &dev_attr_state.attr,
3218         &dev_attr_numa_node.attr,
3219         &dev_attr_queue_count.attr,
3220         &dev_attr_sqsize.attr,
3221         NULL
3222 };
3223
3224 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3225                 struct attribute *a, int n)
3226 {
3227         struct device *dev = container_of(kobj, struct device, kobj);
3228         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3229
3230         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3231                 return 0;
3232         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3233                 return 0;
3234
3235         return a->mode;
3236 }
3237
3238 static struct attribute_group nvme_dev_attrs_group = {
3239         .attrs          = nvme_dev_attrs,
3240         .is_visible     = nvme_dev_attrs_are_visible,
3241 };
3242
3243 static const struct attribute_group *nvme_dev_attr_groups[] = {
3244         &nvme_dev_attrs_group,
3245         NULL,
3246 };
3247
3248 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3249                 unsigned nsid)
3250 {
3251         struct nvme_ns_head *h;
3252
3253         lockdep_assert_held(&subsys->lock);
3254
3255         list_for_each_entry(h, &subsys->nsheads, entry) {
3256                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3257                         return h;
3258         }
3259
3260         return NULL;
3261 }
3262
3263 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3264                 struct nvme_ns_head *new)
3265 {
3266         struct nvme_ns_head *h;
3267
3268         lockdep_assert_held(&subsys->lock);
3269
3270         list_for_each_entry(h, &subsys->nsheads, entry) {
3271                 if (nvme_ns_ids_valid(&new->ids) &&
3272                     !list_empty(&h->list) &&
3273                     nvme_ns_ids_equal(&new->ids, &h->ids))
3274                         return -EINVAL;
3275         }
3276
3277         return 0;
3278 }
3279
3280 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3281                 unsigned nsid, struct nvme_id_ns *id)
3282 {
3283         struct nvme_ns_head *head;
3284         size_t size = sizeof(*head);
3285         int ret = -ENOMEM;
3286
3287 #ifdef CONFIG_NVME_MULTIPATH
3288         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3289 #endif
3290
3291         head = kzalloc(size, GFP_KERNEL);
3292         if (!head)
3293                 goto out;
3294         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3295         if (ret < 0)
3296                 goto out_free_head;
3297         head->instance = ret;
3298         INIT_LIST_HEAD(&head->list);
3299         ret = init_srcu_struct(&head->srcu);
3300         if (ret)
3301                 goto out_ida_remove;
3302         head->subsys = ctrl->subsys;
3303         head->ns_id = nsid;
3304         kref_init(&head->ref);
3305
3306         ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3307         if (ret)
3308                 goto out_cleanup_srcu;
3309
3310         ret = __nvme_check_ids(ctrl->subsys, head);
3311         if (ret) {
3312                 dev_err(ctrl->device,
3313                         "duplicate IDs for nsid %d\n", nsid);
3314                 goto out_cleanup_srcu;
3315         }
3316
3317         ret = nvme_mpath_alloc_disk(ctrl, head);
3318         if (ret)
3319                 goto out_cleanup_srcu;
3320
3321         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3322
3323         kref_get(&ctrl->subsys->ref);
3324
3325         return head;
3326 out_cleanup_srcu:
3327         cleanup_srcu_struct(&head->srcu);
3328 out_ida_remove:
3329         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3330 out_free_head:
3331         kfree(head);
3332 out:
3333         if (ret > 0)
3334                 ret = blk_status_to_errno(nvme_error_status(ret));
3335         return ERR_PTR(ret);
3336 }
3337
3338 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3339                 struct nvme_id_ns *id)
3340 {
3341         struct nvme_ctrl *ctrl = ns->ctrl;
3342         bool is_shared = id->nmic & (1 << 0);
3343         struct nvme_ns_head *head = NULL;
3344         int ret = 0;
3345
3346         mutex_lock(&ctrl->subsys->lock);
3347         if (is_shared)
3348                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3349         if (!head) {
3350                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3351                 if (IS_ERR(head)) {
3352                         ret = PTR_ERR(head);
3353                         goto out_unlock;
3354                 }
3355         } else {
3356                 struct nvme_ns_ids ids;
3357
3358                 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3359                 if (ret)
3360                         goto out_unlock;
3361
3362                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3363                         dev_err(ctrl->device,
3364                                 "IDs don't match for shared namespace %d\n",
3365                                         nsid);
3366                         ret = -EINVAL;
3367                         goto out_unlock;
3368                 }
3369         }
3370
3371         list_add_tail(&ns->siblings, &head->list);
3372         ns->head = head;
3373
3374 out_unlock:
3375         mutex_unlock(&ctrl->subsys->lock);
3376         if (ret > 0)
3377                 ret = blk_status_to_errno(nvme_error_status(ret));
3378         return ret;
3379 }
3380
3381 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3382 {
3383         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3384         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3385
3386         return nsa->head->ns_id - nsb->head->ns_id;
3387 }
3388
3389 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3390 {
3391         struct nvme_ns *ns, *ret = NULL;
3392
3393         down_read(&ctrl->namespaces_rwsem);
3394         list_for_each_entry(ns, &ctrl->namespaces, list) {
3395                 if (ns->head->ns_id == nsid) {
3396                         if (!kref_get_unless_zero(&ns->kref))
3397                                 continue;
3398                         ret = ns;
3399                         break;
3400                 }
3401                 if (ns->head->ns_id > nsid)
3402                         break;
3403         }
3404         up_read(&ctrl->namespaces_rwsem);
3405         return ret;
3406 }
3407
3408 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3409 {
3410         struct streams_directive_params s;
3411         int ret;
3412
3413         if (!ctrl->nr_streams)
3414                 return 0;
3415
3416         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3417         if (ret)
3418                 return ret;
3419
3420         ns->sws = le32_to_cpu(s.sws);
3421         ns->sgs = le16_to_cpu(s.sgs);
3422
3423         if (ns->sws) {
3424                 unsigned int bs = 1 << ns->lba_shift;
3425
3426                 blk_queue_io_min(ns->queue, bs * ns->sws);
3427                 if (ns->sgs)
3428                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3429         }
3430
3431         return 0;
3432 }
3433
3434 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3435 {
3436         struct nvme_ns *ns;
3437         struct gendisk *disk;
3438         struct nvme_id_ns *id;
3439         char disk_name[DISK_NAME_LEN];
3440         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3441
3442         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3443         if (!ns)
3444                 return -ENOMEM;
3445
3446         ns->queue = blk_mq_init_queue(ctrl->tagset);
3447         if (IS_ERR(ns->queue)) {
3448                 ret = PTR_ERR(ns->queue);
3449                 goto out_free_ns;
3450         }
3451
3452         if (ctrl->opts && ctrl->opts->data_digest)
3453                 ns->queue->backing_dev_info->capabilities
3454                         |= BDI_CAP_STABLE_WRITES;
3455
3456         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3457         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3458                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3459
3460         ns->queue->queuedata = ns;
3461         ns->ctrl = ctrl;
3462
3463         kref_init(&ns->kref);
3464         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3465
3466         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3467         nvme_set_queue_limits(ctrl, ns->queue);
3468
3469         ret = nvme_identify_ns(ctrl, nsid, &id);
3470         if (ret)
3471                 goto out_free_queue;
3472
3473         if (id->ncap == 0) {
3474                 ret = -EINVAL;
3475                 goto out_free_id;
3476         }
3477
3478         ret = nvme_init_ns_head(ns, nsid, id);
3479         if (ret)
3480                 goto out_free_id;
3481         nvme_setup_streams_ns(ctrl, ns);
3482         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3483
3484         disk = alloc_disk_node(0, node);
3485         if (!disk) {
3486                 ret = -ENOMEM;
3487                 goto out_unlink_ns;
3488         }
3489
3490         disk->fops = &nvme_fops;
3491         disk->private_data = ns;
3492         disk->queue = ns->queue;
3493         disk->flags = flags;
3494         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3495         ns->disk = disk;
3496
3497         __nvme_revalidate_disk(disk, id);
3498
3499         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3500                 ret = nvme_nvm_register(ns, disk_name, node);
3501                 if (ret) {
3502                         dev_warn(ctrl->device, "LightNVM init failure\n");
3503                         goto out_put_disk;
3504                 }
3505         }
3506
3507         down_write(&ctrl->namespaces_rwsem);
3508         list_add_tail(&ns->list, &ctrl->namespaces);
3509         up_write(&ctrl->namespaces_rwsem);
3510
3511         nvme_get_ctrl(ctrl);
3512
3513         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3514
3515         nvme_mpath_add_disk(ns, id);
3516         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3517         kfree(id);
3518
3519         return 0;
3520  out_put_disk:
3521         put_disk(ns->disk);
3522  out_unlink_ns:
3523         mutex_lock(&ctrl->subsys->lock);
3524         list_del_rcu(&ns->siblings);
3525         mutex_unlock(&ctrl->subsys->lock);
3526         nvme_put_ns_head(ns->head);
3527  out_free_id:
3528         kfree(id);
3529  out_free_queue:
3530         blk_cleanup_queue(ns->queue);
3531  out_free_ns:
3532         kfree(ns);
3533         if (ret > 0)
3534                 ret = blk_status_to_errno(nvme_error_status(ret));
3535         return ret;
3536 }
3537
3538 static void nvme_ns_remove(struct nvme_ns *ns)
3539 {
3540         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3541                 return;
3542
3543         nvme_fault_inject_fini(&ns->fault_inject);
3544
3545         mutex_lock(&ns->ctrl->subsys->lock);
3546         list_del_rcu(&ns->siblings);
3547         mutex_unlock(&ns->ctrl->subsys->lock);
3548         synchronize_rcu(); /* guarantee not available in head->list */
3549         nvme_mpath_clear_current_path(ns);
3550         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3551
3552         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3553                 del_gendisk(ns->disk);
3554                 blk_cleanup_queue(ns->queue);
3555                 if (blk_get_integrity(ns->disk))
3556                         blk_integrity_unregister(ns->disk);
3557         }
3558
3559         down_write(&ns->ctrl->namespaces_rwsem);
3560         list_del_init(&ns->list);
3561         up_write(&ns->ctrl->namespaces_rwsem);
3562
3563         nvme_mpath_check_last_path(ns);
3564         nvme_put_ns(ns);
3565 }
3566
3567 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3568 {
3569         struct nvme_ns *ns;
3570
3571         ns = nvme_find_get_ns(ctrl, nsid);
3572         if (ns) {
3573                 if (ns->disk && revalidate_disk(ns->disk))
3574                         nvme_ns_remove(ns);
3575                 nvme_put_ns(ns);
3576         } else
3577                 nvme_alloc_ns(ctrl, nsid);
3578 }
3579
3580 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3581                                         unsigned nsid)
3582 {
3583         struct nvme_ns *ns, *next;
3584         LIST_HEAD(rm_list);
3585
3586         down_write(&ctrl->namespaces_rwsem);
3587         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3588                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3589                         list_move_tail(&ns->list, &rm_list);
3590         }
3591         up_write(&ctrl->namespaces_rwsem);
3592
3593         list_for_each_entry_safe(ns, next, &rm_list, list)
3594                 nvme_ns_remove(ns);
3595
3596 }
3597
3598 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3599 {
3600         struct nvme_ns *ns;
3601         __le32 *ns_list;
3602         unsigned i, j, nsid, prev = 0;
3603         unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3604         int ret = 0;
3605
3606         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3607         if (!ns_list)
3608                 return -ENOMEM;
3609
3610         for (i = 0; i < num_lists; i++) {
3611                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3612                 if (ret)
3613                         goto free;
3614
3615                 for (j = 0; j < min(nn, 1024U); j++) {
3616                         nsid = le32_to_cpu(ns_list[j]);
3617                         if (!nsid)
3618                                 goto out;
3619
3620                         nvme_validate_ns(ctrl, nsid);
3621
3622                         while (++prev < nsid) {
3623                                 ns = nvme_find_get_ns(ctrl, prev);
3624                                 if (ns) {
3625                                         nvme_ns_remove(ns);
3626                                         nvme_put_ns(ns);
3627                                 }
3628                         }
3629                 }
3630                 nn -= j;
3631         }
3632  out:
3633         nvme_remove_invalid_namespaces(ctrl, prev);
3634  free:
3635         kfree(ns_list);
3636         return ret;
3637 }
3638
3639 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3640 {
3641         unsigned i;
3642
3643         for (i = 1; i <= nn; i++)
3644                 nvme_validate_ns(ctrl, i);
3645
3646         nvme_remove_invalid_namespaces(ctrl, nn);
3647 }
3648
3649 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3650 {
3651         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3652         __le32 *log;
3653         int error;
3654
3655         log = kzalloc(log_size, GFP_KERNEL);
3656         if (!log)
3657                 return;
3658
3659         /*
3660          * We need to read the log to clear the AEN, but we don't want to rely
3661          * on it for the changed namespace information as userspace could have
3662          * raced with us in reading the log page, which could cause us to miss
3663          * updates.
3664          */
3665         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3666                         log_size, 0);
3667         if (error)
3668                 dev_warn(ctrl->device,
3669                         "reading changed ns log failed: %d\n", error);
3670
3671         kfree(log);
3672 }
3673
3674 static void nvme_scan_work(struct work_struct *work)
3675 {
3676         struct nvme_ctrl *ctrl =
3677                 container_of(work, struct nvme_ctrl, scan_work);
3678         struct nvme_id_ctrl *id;
3679         unsigned nn;
3680
3681         if (ctrl->state != NVME_CTRL_LIVE)
3682                 return;
3683
3684         WARN_ON_ONCE(!ctrl->tagset);
3685
3686         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3687                 dev_info(ctrl->device, "rescanning namespaces.\n");
3688                 nvme_clear_changed_ns_log(ctrl);
3689         }
3690
3691         if (nvme_identify_ctrl(ctrl, &id))
3692                 return;
3693
3694         mutex_lock(&ctrl->scan_lock);
3695         nn = le32_to_cpu(id->nn);
3696         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3697             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3698                 if (!nvme_scan_ns_list(ctrl, nn))
3699                         goto out_free_id;
3700         }
3701         nvme_scan_ns_sequential(ctrl, nn);
3702 out_free_id:
3703         mutex_unlock(&ctrl->scan_lock);
3704         kfree(id);
3705         down_write(&ctrl->namespaces_rwsem);
3706         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3707         up_write(&ctrl->namespaces_rwsem);
3708 }
3709
3710 /*
3711  * This function iterates the namespace list unlocked to allow recovery from
3712  * controller failure. It is up to the caller to ensure the namespace list is
3713  * not modified by scan work while this function is executing.
3714  */
3715 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3716 {
3717         struct nvme_ns *ns, *next;
3718         LIST_HEAD(ns_list);
3719
3720         /*
3721          * make sure to requeue I/O to all namespaces as these
3722          * might result from the scan itself and must complete
3723          * for the scan_work to make progress
3724          */
3725         nvme_mpath_clear_ctrl_paths(ctrl);
3726
3727         /* prevent racing with ns scanning */
3728         flush_work(&ctrl->scan_work);
3729
3730         /*
3731          * The dead states indicates the controller was not gracefully
3732          * disconnected. In that case, we won't be able to flush any data while
3733          * removing the namespaces' disks; fail all the queues now to avoid
3734          * potentially having to clean up the failed sync later.
3735          */
3736         if (ctrl->state == NVME_CTRL_DEAD)
3737                 nvme_kill_queues(ctrl);
3738
3739         down_write(&ctrl->namespaces_rwsem);
3740         list_splice_init(&ctrl->namespaces, &ns_list);
3741         up_write(&ctrl->namespaces_rwsem);
3742
3743         list_for_each_entry_safe(ns, next, &ns_list, list)
3744                 nvme_ns_remove(ns);
3745 }
3746 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3747
3748 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3749 {
3750         struct nvme_ctrl *ctrl =
3751                 container_of(dev, struct nvme_ctrl, ctrl_device);
3752         struct nvmf_ctrl_options *opts = ctrl->opts;
3753         int ret;
3754
3755         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3756         if (ret)
3757                 return ret;
3758
3759         if (opts) {
3760                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3761                 if (ret)
3762                         return ret;
3763
3764                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3765                                 opts->trsvcid ?: "none");
3766                 if (ret)
3767                         return ret;
3768
3769                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3770                                 opts->host_traddr ?: "none");
3771         }
3772         return ret;
3773 }
3774
3775 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3776 {
3777         char *envp[2] = { NULL, NULL };
3778         u32 aen_result = ctrl->aen_result;
3779
3780         ctrl->aen_result = 0;
3781         if (!aen_result)
3782                 return;
3783
3784         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3785         if (!envp[0])
3786                 return;
3787         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3788         kfree(envp[0]);
3789 }
3790
3791 static void nvme_async_event_work(struct work_struct *work)
3792 {
3793         struct nvme_ctrl *ctrl =
3794                 container_of(work, struct nvme_ctrl, async_event_work);
3795
3796         nvme_aen_uevent(ctrl);
3797         ctrl->ops->submit_async_event(ctrl);
3798 }
3799
3800 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3801 {
3802
3803         u32 csts;
3804
3805         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3806                 return false;
3807
3808         if (csts == ~0)
3809                 return false;
3810
3811         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3812 }
3813
3814 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3815 {
3816         struct nvme_fw_slot_info_log *log;
3817
3818         log = kmalloc(sizeof(*log), GFP_KERNEL);
3819         if (!log)
3820                 return;
3821
3822         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3823                         sizeof(*log), 0))
3824                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3825         kfree(log);
3826 }
3827
3828 static void nvme_fw_act_work(struct work_struct *work)
3829 {
3830         struct nvme_ctrl *ctrl = container_of(work,
3831                                 struct nvme_ctrl, fw_act_work);
3832         unsigned long fw_act_timeout;
3833
3834         if (ctrl->mtfa)
3835                 fw_act_timeout = jiffies +
3836                                 msecs_to_jiffies(ctrl->mtfa * 100);
3837         else
3838                 fw_act_timeout = jiffies +
3839                                 msecs_to_jiffies(admin_timeout * 1000);
3840
3841         nvme_stop_queues(ctrl);
3842         while (nvme_ctrl_pp_status(ctrl)) {
3843                 if (time_after(jiffies, fw_act_timeout)) {
3844                         dev_warn(ctrl->device,
3845                                 "Fw activation timeout, reset controller\n");
3846                         nvme_reset_ctrl(ctrl);
3847                         break;
3848                 }
3849                 msleep(100);
3850         }
3851
3852         if (ctrl->state != NVME_CTRL_LIVE)
3853                 return;
3854
3855         nvme_start_queues(ctrl);
3856         /* read FW slot information to clear the AER */
3857         nvme_get_fw_slot_info(ctrl);
3858 }
3859
3860 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3861 {
3862         u32 aer_notice_type = (result & 0xff00) >> 8;
3863
3864         trace_nvme_async_event(ctrl, aer_notice_type);
3865
3866         switch (aer_notice_type) {
3867         case NVME_AER_NOTICE_NS_CHANGED:
3868                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3869                 nvme_queue_scan(ctrl);
3870                 break;
3871         case NVME_AER_NOTICE_FW_ACT_STARTING:
3872                 queue_work(nvme_wq, &ctrl->fw_act_work);
3873                 break;
3874 #ifdef CONFIG_NVME_MULTIPATH
3875         case NVME_AER_NOTICE_ANA:
3876                 if (!ctrl->ana_log_buf)
3877                         break;
3878                 queue_work(nvme_wq, &ctrl->ana_work);
3879                 break;
3880 #endif
3881         case NVME_AER_NOTICE_DISC_CHANGED:
3882                 ctrl->aen_result = result;
3883                 break;
3884         default:
3885                 dev_warn(ctrl->device, "async event result %08x\n", result);
3886         }
3887 }
3888
3889 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3890                 volatile union nvme_result *res)
3891 {
3892         u32 result = le32_to_cpu(res->u32);
3893         u32 aer_type = result & 0x07;
3894
3895         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3896                 return;
3897
3898         switch (aer_type) {
3899         case NVME_AER_NOTICE:
3900                 nvme_handle_aen_notice(ctrl, result);
3901                 break;
3902         case NVME_AER_ERROR:
3903         case NVME_AER_SMART:
3904         case NVME_AER_CSS:
3905         case NVME_AER_VS:
3906                 trace_nvme_async_event(ctrl, aer_type);
3907                 ctrl->aen_result = result;
3908                 break;
3909         default:
3910                 break;
3911         }
3912         queue_work(nvme_wq, &ctrl->async_event_work);
3913 }
3914 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3915
3916 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3917 {
3918         nvme_mpath_stop(ctrl);
3919         nvme_stop_keep_alive(ctrl);
3920         flush_work(&ctrl->async_event_work);
3921         cancel_work_sync(&ctrl->fw_act_work);
3922 }
3923 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3924
3925 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3926 {
3927         if (ctrl->kato)
3928                 nvme_start_keep_alive(ctrl);
3929
3930         nvme_enable_aen(ctrl);
3931
3932         if (ctrl->queue_count > 1) {
3933                 nvme_queue_scan(ctrl);
3934                 nvme_start_queues(ctrl);
3935         }
3936 }
3937 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3938
3939 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3940 {
3941         nvme_fault_inject_fini(&ctrl->fault_inject);
3942         dev_pm_qos_hide_latency_tolerance(ctrl->device);
3943         cdev_device_del(&ctrl->cdev, ctrl->device);
3944 }
3945 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3946
3947 static void nvme_free_ctrl(struct device *dev)
3948 {
3949         struct nvme_ctrl *ctrl =
3950                 container_of(dev, struct nvme_ctrl, ctrl_device);
3951         struct nvme_subsystem *subsys = ctrl->subsys;
3952
3953         if (subsys && ctrl->instance != subsys->instance)
3954                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3955
3956         kfree(ctrl->effects);
3957         nvme_mpath_uninit(ctrl);
3958         __free_page(ctrl->discard_page);
3959
3960         if (subsys) {
3961                 mutex_lock(&nvme_subsystems_lock);
3962                 list_del(&ctrl->subsys_entry);
3963                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3964                 mutex_unlock(&nvme_subsystems_lock);
3965         }
3966
3967         ctrl->ops->free_ctrl(ctrl);
3968
3969         if (subsys)
3970                 nvme_put_subsystem(subsys);
3971 }
3972
3973 /*
3974  * Initialize a NVMe controller structures.  This needs to be called during
3975  * earliest initialization so that we have the initialized structured around
3976  * during probing.
3977  */
3978 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3979                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3980 {
3981         int ret;
3982
3983         ctrl->state = NVME_CTRL_NEW;
3984         spin_lock_init(&ctrl->lock);
3985         mutex_init(&ctrl->scan_lock);
3986         INIT_LIST_HEAD(&ctrl->namespaces);
3987         init_rwsem(&ctrl->namespaces_rwsem);
3988         ctrl->dev = dev;
3989         ctrl->ops = ops;
3990         ctrl->quirks = quirks;
3991         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3992         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3993         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3994         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3995
3996         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3997         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3998         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3999
4000         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4001                         PAGE_SIZE);
4002         ctrl->discard_page = alloc_page(GFP_KERNEL);
4003         if (!ctrl->discard_page) {
4004                 ret = -ENOMEM;
4005                 goto out;
4006         }
4007
4008         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4009         if (ret < 0)
4010                 goto out;
4011         ctrl->instance = ret;
4012
4013         device_initialize(&ctrl->ctrl_device);
4014         ctrl->device = &ctrl->ctrl_device;
4015         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4016         ctrl->device->class = nvme_class;
4017         ctrl->device->parent = ctrl->dev;
4018         ctrl->device->groups = nvme_dev_attr_groups;
4019         ctrl->device->release = nvme_free_ctrl;
4020         dev_set_drvdata(ctrl->device, ctrl);
4021         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4022         if (ret)
4023                 goto out_release_instance;
4024
4025         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4026         ctrl->cdev.owner = ops->module;
4027         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4028         if (ret)
4029                 goto out_free_name;
4030
4031         /*
4032          * Initialize latency tolerance controls.  The sysfs files won't
4033          * be visible to userspace unless the device actually supports APST.
4034          */
4035         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4036         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4037                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4038
4039         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4040
4041         return 0;
4042 out_free_name:
4043         kfree_const(ctrl->device->kobj.name);
4044 out_release_instance:
4045         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4046 out:
4047         if (ctrl->discard_page)
4048                 __free_page(ctrl->discard_page);
4049         return ret;
4050 }
4051 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4052
4053 /**
4054  * nvme_kill_queues(): Ends all namespace queues
4055  * @ctrl: the dead controller that needs to end
4056  *
4057  * Call this function when the driver determines it is unable to get the
4058  * controller in a state capable of servicing IO.
4059  */
4060 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4061 {
4062         struct nvme_ns *ns;
4063
4064         down_read(&ctrl->namespaces_rwsem);
4065
4066         /* Forcibly unquiesce queues to avoid blocking dispatch */
4067         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4068                 blk_mq_unquiesce_queue(ctrl->admin_q);
4069
4070         list_for_each_entry(ns, &ctrl->namespaces, list)
4071                 nvme_set_queue_dying(ns);
4072
4073         up_read(&ctrl->namespaces_rwsem);
4074 }
4075 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4076
4077 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4078 {
4079         struct nvme_ns *ns;
4080
4081         down_read(&ctrl->namespaces_rwsem);
4082         list_for_each_entry(ns, &ctrl->namespaces, list)
4083                 blk_mq_unfreeze_queue(ns->queue);
4084         up_read(&ctrl->namespaces_rwsem);
4085 }
4086 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4087
4088 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4089 {
4090         struct nvme_ns *ns;
4091
4092         down_read(&ctrl->namespaces_rwsem);
4093         list_for_each_entry(ns, &ctrl->namespaces, list) {
4094                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4095                 if (timeout <= 0)
4096                         break;
4097         }
4098         up_read(&ctrl->namespaces_rwsem);
4099 }
4100 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4101
4102 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4103 {
4104         struct nvme_ns *ns;
4105
4106         down_read(&ctrl->namespaces_rwsem);
4107         list_for_each_entry(ns, &ctrl->namespaces, list)
4108                 blk_mq_freeze_queue_wait(ns->queue);
4109         up_read(&ctrl->namespaces_rwsem);
4110 }
4111 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4112
4113 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4114 {
4115         struct nvme_ns *ns;
4116
4117         down_read(&ctrl->namespaces_rwsem);
4118         list_for_each_entry(ns, &ctrl->namespaces, list)
4119                 blk_freeze_queue_start(ns->queue);
4120         up_read(&ctrl->namespaces_rwsem);
4121 }
4122 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4123
4124 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4125 {
4126         struct nvme_ns *ns;
4127
4128         down_read(&ctrl->namespaces_rwsem);
4129         list_for_each_entry(ns, &ctrl->namespaces, list)
4130                 blk_mq_quiesce_queue(ns->queue);
4131         up_read(&ctrl->namespaces_rwsem);
4132 }
4133 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4134
4135 void nvme_start_queues(struct nvme_ctrl *ctrl)
4136 {
4137         struct nvme_ns *ns;
4138
4139         down_read(&ctrl->namespaces_rwsem);
4140         list_for_each_entry(ns, &ctrl->namespaces, list)
4141                 blk_mq_unquiesce_queue(ns->queue);
4142         up_read(&ctrl->namespaces_rwsem);
4143 }
4144 EXPORT_SYMBOL_GPL(nvme_start_queues);
4145
4146
4147 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4148 {
4149         struct nvme_ns *ns;
4150
4151         down_read(&ctrl->namespaces_rwsem);
4152         list_for_each_entry(ns, &ctrl->namespaces, list)
4153                 blk_sync_queue(ns->queue);
4154         up_read(&ctrl->namespaces_rwsem);
4155
4156         if (ctrl->admin_q)
4157                 blk_sync_queue(ctrl->admin_q);
4158 }
4159 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4160
4161 /*
4162  * Check we didn't inadvertently grow the command structure sizes:
4163  */
4164 static inline void _nvme_check_size(void)
4165 {
4166         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4167         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4168         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4169         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4170         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4171         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4172         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4173         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4174         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4175         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4176         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4177         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4178         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4179         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4180         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4181         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4182         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4183 }
4184
4185
4186 static int __init nvme_core_init(void)
4187 {
4188         int result = -ENOMEM;
4189
4190         _nvme_check_size();
4191
4192         nvme_wq = alloc_workqueue("nvme-wq",
4193                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4194         if (!nvme_wq)
4195                 goto out;
4196
4197         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4198                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4199         if (!nvme_reset_wq)
4200                 goto destroy_wq;
4201
4202         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4203                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4204         if (!nvme_delete_wq)
4205                 goto destroy_reset_wq;
4206
4207         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4208         if (result < 0)
4209                 goto destroy_delete_wq;
4210
4211         nvme_class = class_create(THIS_MODULE, "nvme");
4212         if (IS_ERR(nvme_class)) {
4213                 result = PTR_ERR(nvme_class);
4214                 goto unregister_chrdev;
4215         }
4216         nvme_class->dev_uevent = nvme_class_uevent;
4217
4218         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4219         if (IS_ERR(nvme_subsys_class)) {
4220                 result = PTR_ERR(nvme_subsys_class);
4221                 goto destroy_class;
4222         }
4223         return 0;
4224
4225 destroy_class:
4226         class_destroy(nvme_class);
4227 unregister_chrdev:
4228         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4229 destroy_delete_wq:
4230         destroy_workqueue(nvme_delete_wq);
4231 destroy_reset_wq:
4232         destroy_workqueue(nvme_reset_wq);
4233 destroy_wq:
4234         destroy_workqueue(nvme_wq);
4235 out:
4236         return result;
4237 }
4238
4239 static void __exit nvme_core_exit(void)
4240 {
4241         class_destroy(nvme_subsys_class);
4242         class_destroy(nvme_class);
4243         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4244         destroy_workqueue(nvme_delete_wq);
4245         destroy_workqueue(nvme_reset_wq);
4246         destroy_workqueue(nvme_wq);
4247 }
4248
4249 MODULE_LICENSE("GPL");
4250 MODULE_VERSION("1.0");
4251 module_init(nvme_core_init);
4252 module_exit(nvme_core_exit);