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/list_sort.h>
  15 #include <linux/slab.h>
  16 #include <linux/types.h>
  17 #include <linux/pr.h>
  18 #include <linux/ptrace.h>
  19 #include <linux/nvme_ioctl.h>
  20 #include <linux/t10-pi.h>
  21 #include <linux/pm_qos.h>
  22 #include <asm/unaligned.h>
  23
  24 #define CREATE_TRACE_POINTS
  25 #include "trace.h"
  26
  27 #include "nvme.h"
  28 #include "fabrics.h"
  29
  30 #define NVME_MINORS             (1U << MINORBITS)
  31
  32 unsigned int admin_timeout = 60;
  33 module_param(admin_timeout, uint, 0644);
  34 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  35 EXPORT_SYMBOL_GPL(admin_timeout);
  36
  37 unsigned int nvme_io_timeout = 30;
  38 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  39 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  40 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  41
  42 static unsigned char shutdown_timeout = 5;
  43 module_param(shutdown_timeout, byte, 0644);
  44 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  45
  46 static u8 nvme_max_retries = 5;
  47 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  48 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  49
  50 static unsigned long default_ps_max_latency_us = 100000;
  51 module_param(default_ps_max_latency_us, ulong, 0644);
  52 MODULE_PARM_DESC(default_ps_max_latency_us,
  53                  "max power saving latency for new devices; use PM QOS to change per device");
  54
  55 static bool force_apst;
  56 module_param(force_apst, bool, 0644);
  57 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  58
  59 static bool streams;
  60 module_param(streams, bool, 0644);
  61 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  62
  63 /*
  64  * nvme_wq - hosts nvme related works that are not reset or delete
  65  * nvme_reset_wq - hosts nvme reset works
  66  * nvme_delete_wq - hosts nvme delete works
  67  *
  68  * nvme_wq will host works such are scan, aen handling, fw activation,
  69  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
  70  * runs reset works which also flush works hosted on nvme_wq for
  71  * serialization purposes. nvme_delete_wq host controller deletion
  72  * works which flush reset works for serialization.
  73  */
  74 struct workqueue_struct *nvme_wq;
  75 EXPORT_SYMBOL_GPL(nvme_wq);
  76
  77 struct workqueue_struct *nvme_reset_wq;
  78 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  79
  80 struct workqueue_struct *nvme_delete_wq;
  81 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  82
  83 static DEFINE_IDA(nvme_subsystems_ida);
  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         revalidate_disk(ns->disk);
 106         blk_set_queue_dying(ns->queue);
 107         /* Forcibly unquiesce queues to avoid blocking dispatch */
 108         blk_mq_unquiesce_queue(ns->queue);
 109 }
 110
 111 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 112 {
 113         /*
 114          * Only new queue scan work when admin and IO queues are both alive
 115          */
 116         if (ctrl->state == NVME_CTRL_LIVE)
 117                 queue_work(nvme_wq, &ctrl->scan_work);
 118 }
 119
 120 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 121 {
 122         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 123                 return -EBUSY;
 124         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 125                 return -EBUSY;
 126         return 0;
 127 }
 128 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 129
 130 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 131 {
 132         int ret;
 133
 134         ret = nvme_reset_ctrl(ctrl);
 135         if (!ret) {
 136                 flush_work(&ctrl->reset_work);
 137                 if (ctrl->state != NVME_CTRL_LIVE &&
 138                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
 139                         ret = -ENETRESET;
 140         }
 141
 142         return ret;
 143 }
 144 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 145
 146 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 147 {
 148         dev_info(ctrl->device,
 149                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 150
 151         flush_work(&ctrl->reset_work);
 152         nvme_stop_ctrl(ctrl);
 153         nvme_remove_namespaces(ctrl);
 154         ctrl->ops->delete_ctrl(ctrl);
 155         nvme_uninit_ctrl(ctrl);
 156         nvme_put_ctrl(ctrl);
 157 }
 158
 159 static void nvme_delete_ctrl_work(struct work_struct *work)
 160 {
 161         struct nvme_ctrl *ctrl =
 162                 container_of(work, struct nvme_ctrl, delete_work);
 163
 164         nvme_do_delete_ctrl(ctrl);
 165 }
 166
 167 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 168 {
 169         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 170                 return -EBUSY;
 171         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 172                 return -EBUSY;
 173         return 0;
 174 }
 175 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 176
 177 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 178 {
 179         int ret = 0;
 180
 181         /*
 182          * Keep a reference until nvme_do_delete_ctrl() complete,
 183          * since ->delete_ctrl can free the controller.
 184          */
 185         nvme_get_ctrl(ctrl);
 186         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 187                 ret = -EBUSY;
 188         if (!ret)
 189                 nvme_do_delete_ctrl(ctrl);
 190         nvme_put_ctrl(ctrl);
 191         return ret;
 192 }
 193
 194 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 195 {
 196         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 197 }
 198
 199 static blk_status_t nvme_error_status(struct request *req)
 200 {
 201         switch (nvme_req(req)->status & 0x7ff) {
 202         case NVME_SC_SUCCESS:
 203                 return BLK_STS_OK;
 204         case NVME_SC_CAP_EXCEEDED:
 205                 return BLK_STS_NOSPC;
 206         case NVME_SC_LBA_RANGE:
 207                 return BLK_STS_TARGET;
 208         case NVME_SC_BAD_ATTRIBUTES:
 209         case NVME_SC_ONCS_NOT_SUPPORTED:
 210         case NVME_SC_INVALID_OPCODE:
 211         case NVME_SC_INVALID_FIELD:
 212         case NVME_SC_INVALID_NS:
 213                 return BLK_STS_NOTSUPP;
 214         case NVME_SC_WRITE_FAULT:
 215         case NVME_SC_READ_ERROR:
 216         case NVME_SC_UNWRITTEN_BLOCK:
 217         case NVME_SC_ACCESS_DENIED:
 218         case NVME_SC_READ_ONLY:
 219         case NVME_SC_COMPARE_FAILED:
 220                 return BLK_STS_MEDIUM;
 221         case NVME_SC_GUARD_CHECK:
 222         case NVME_SC_APPTAG_CHECK:
 223         case NVME_SC_REFTAG_CHECK:
 224         case NVME_SC_INVALID_PI:
 225                 return BLK_STS_PROTECTION;
 226         case NVME_SC_RESERVATION_CONFLICT:
 227                 return BLK_STS_NEXUS;
 228         default:
 229                 return BLK_STS_IOERR;
 230         }
 231 }
 232
 233 static inline bool nvme_req_needs_retry(struct request *req)
 234 {
 235         if (blk_noretry_request(req))
 236                 return false;
 237         if (nvme_req(req)->status & NVME_SC_DNR)
 238                 return false;
 239         if (nvme_req(req)->retries >= nvme_max_retries)
 240                 return false;
 241         return true;
 242 }
 243
 244 static void nvme_retry_req(struct request *req)
 245 {
 246         struct nvme_ns *ns = req->q->queuedata;
 247         unsigned long delay = 0;
 248         u16 crd;
 249
 250         /* The mask and shift result must be <= 3 */
 251         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 252         if (ns && crd)
 253                 delay = ns->ctrl->crdt[crd - 1] * 100;
 254
 255         nvme_req(req)->retries++;
 256         blk_mq_requeue_request(req, false);
 257         blk_mq_delay_kick_requeue_list(req->q, delay);
 258 }
 259
 260 void nvme_complete_rq(struct request *req)
 261 {
 262         blk_status_t status = nvme_error_status(req);
 263
 264         trace_nvme_complete_rq(req);
 265
 266         if (nvme_req(req)->ctrl->kas)
 267                 nvme_req(req)->ctrl->comp_seen = true;
 268
 269         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 270                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
 271                     blk_path_error(status)) {
 272                         nvme_failover_req(req);
 273                         return;
 274                 }
 275
 276                 if (!blk_queue_dying(req->q)) {
 277                         nvme_retry_req(req);
 278                         return;
 279                 }
 280         }
 281         blk_mq_end_request(req, status);
 282 }
 283 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 284
 285 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 286 {
 287         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 288                                 "Cancelling I/O %d", req->tag);
 289
 290         nvme_req(req)->status = NVME_SC_ABORT_REQ;
 291         blk_mq_complete_request_sync(req);
 292         return true;
 293 }
 294 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 295
 296 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 297                 enum nvme_ctrl_state new_state)
 298 {
 299         enum nvme_ctrl_state old_state;
 300         unsigned long flags;
 301         bool changed = false;
 302
 303         spin_lock_irqsave(&ctrl->lock, flags);
 304
 305         old_state = ctrl->state;
 306         switch (new_state) {
 307         case NVME_CTRL_ADMIN_ONLY:
 308                 switch (old_state) {
 309                 case NVME_CTRL_CONNECTING:
 310                         changed = true;
 311                         /* FALLTHRU */
 312                 default:
 313                         break;
 314                 }
 315                 break;
 316         case NVME_CTRL_LIVE:
 317                 switch (old_state) {
 318                 case NVME_CTRL_NEW:
 319                 case NVME_CTRL_RESETTING:
 320                 case NVME_CTRL_CONNECTING:
 321                         changed = true;
 322                         /* FALLTHRU */
 323                 default:
 324                         break;
 325                 }
 326                 break;
 327         case NVME_CTRL_RESETTING:
 328                 switch (old_state) {
 329                 case NVME_CTRL_NEW:
 330                 case NVME_CTRL_LIVE:
 331                 case NVME_CTRL_ADMIN_ONLY:
 332                         changed = true;
 333                         /* FALLTHRU */
 334                 default:
 335                         break;
 336                 }
 337                 break;
 338         case NVME_CTRL_CONNECTING:
 339                 switch (old_state) {
 340                 case NVME_CTRL_NEW:
 341                 case NVME_CTRL_RESETTING:
 342                         changed = true;
 343                         /* FALLTHRU */
 344                 default:
 345                         break;
 346                 }
 347                 break;
 348         case NVME_CTRL_DELETING:
 349                 switch (old_state) {
 350                 case NVME_CTRL_LIVE:
 351                 case NVME_CTRL_ADMIN_ONLY:
 352                 case NVME_CTRL_RESETTING:
 353                 case NVME_CTRL_CONNECTING:
 354                         changed = true;
 355                         /* FALLTHRU */
 356                 default:
 357                         break;
 358                 }
 359                 break;
 360         case NVME_CTRL_DEAD:
 361                 switch (old_state) {
 362                 case NVME_CTRL_DELETING:
 363                         changed = true;
 364                         /* FALLTHRU */
 365                 default:
 366                         break;
 367                 }
 368                 break;
 369         default:
 370                 break;
 371         }
 372
 373         if (changed)
 374                 ctrl->state = new_state;
 375
 376         spin_unlock_irqrestore(&ctrl->lock, flags);
 377         if (changed && ctrl->state == NVME_CTRL_LIVE)
 378                 nvme_kick_requeue_lists(ctrl);
 379         return changed;
 380 }
 381 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 382
 383 static void nvme_free_ns_head(struct kref *ref)
 384 {
 385         struct nvme_ns_head *head =
 386                 container_of(ref, struct nvme_ns_head, ref);
 387
 388         nvme_mpath_remove_disk(head);
 389         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 390         list_del_init(&head->entry);
 391         cleanup_srcu_struct(&head->srcu);
 392         nvme_put_subsystem(head->subsys);
 393         kfree(head);
 394 }
 395
 396 static void nvme_put_ns_head(struct nvme_ns_head *head)
 397 {
 398         kref_put(&head->ref, nvme_free_ns_head);
 399 }
 400
 401 static void nvme_free_ns(struct kref *kref)
 402 {
 403         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 404
 405         if (ns->ndev)
 406                 nvme_nvm_unregister(ns);
 407
 408         put_disk(ns->disk);
 409         nvme_put_ns_head(ns->head);
 410         nvme_put_ctrl(ns->ctrl);
 411         kfree(ns);
 412 }
 413
 414 static void nvme_put_ns(struct nvme_ns *ns)
 415 {
 416         kref_put(&ns->kref, nvme_free_ns);
 417 }
 418
 419 static inline void nvme_clear_nvme_request(struct request *req)
 420 {
 421         if (!(req->rq_flags & RQF_DONTPREP)) {
 422                 nvme_req(req)->retries = 0;
 423                 nvme_req(req)->flags = 0;
 424                 req->rq_flags |= RQF_DONTPREP;
 425         }
 426 }
 427
 428 struct request *nvme_alloc_request(struct request_queue *q,
 429                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 430 {
 431         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 432         struct request *req;
 433
 434         if (qid == NVME_QID_ANY) {
 435                 req = blk_mq_alloc_request(q, op, flags);
 436         } else {
 437                 req = blk_mq_alloc_request_hctx(q, op, flags,
 438                                 qid ? qid - 1 : 0);
 439         }
 440         if (IS_ERR(req))
 441                 return req;
 442
 443         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 444         nvme_clear_nvme_request(req);
 445         nvme_req(req)->cmd = cmd;
 446
 447         return req;
 448 }
 449 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 450
 451 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 452 {
 453         struct nvme_command c;
 454
 455         memset(&c, 0, sizeof(c));
 456
 457         c.directive.opcode = nvme_admin_directive_send;
 458         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 459         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 460         c.directive.dtype = NVME_DIR_IDENTIFY;
 461         c.directive.tdtype = NVME_DIR_STREAMS;
 462         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 463
 464         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 465 }
 466
 467 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 468 {
 469         return nvme_toggle_streams(ctrl, false);
 470 }
 471
 472 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 473 {
 474         return nvme_toggle_streams(ctrl, true);
 475 }
 476
 477 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 478                                   struct streams_directive_params *s, u32 nsid)
 479 {
 480         struct nvme_command c;
 481
 482         memset(&c, 0, sizeof(c));
 483         memset(s, 0, sizeof(*s));
 484
 485         c.directive.opcode = nvme_admin_directive_recv;
 486         c.directive.nsid = cpu_to_le32(nsid);
 487         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 488         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 489         c.directive.dtype = NVME_DIR_STREAMS;
 490
 491         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 492 }
 493
 494 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 495 {
 496         struct streams_directive_params s;
 497         int ret;
 498
 499         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 500                 return 0;
 501         if (!streams)
 502                 return 0;
 503
 504         ret = nvme_enable_streams(ctrl);
 505         if (ret)
 506                 return ret;
 507
 508         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 509         if (ret)
 510                 return ret;
 511
 512         ctrl->nssa = le16_to_cpu(s.nssa);
 513         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 514                 dev_info(ctrl->device, "too few streams (%u) available\n",
 515                                         ctrl->nssa);
 516                 nvme_disable_streams(ctrl);
 517                 return 0;
 518         }
 519
 520         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 521         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 522         return 0;
 523 }
 524
 525 /*
 526  * Check if 'req' has a write hint associated with it. If it does, assign
 527  * a valid namespace stream to the write.
 528  */
 529 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 530                                      struct request *req, u16 *control,
 531                                      u32 *dsmgmt)
 532 {
 533         enum rw_hint streamid = req->write_hint;
 534
 535         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 536                 streamid = 0;
 537         else {
 538                 streamid--;
 539                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 540                         return;
 541
 542                 *control |= NVME_RW_DTYPE_STREAMS;
 543                 *dsmgmt |= streamid << 16;
 544         }
 545
 546         if (streamid < ARRAY_SIZE(req->q->write_hints))
 547                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 548 }
 549
 550 static inline void nvme_setup_flush(struct nvme_ns *ns,
 551                 struct nvme_command *cmnd)
 552 {
 553         cmnd->common.opcode = nvme_cmd_flush;
 554         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 555 }
 556
 557 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 558                 struct nvme_command *cmnd)
 559 {
 560         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 561         struct nvme_dsm_range *range;
 562         struct bio *bio;
 563
 564         range = kmalloc_array(segments, sizeof(*range),
 565                                 GFP_ATOMIC | __GFP_NOWARN);
 566         if (!range) {
 567                 /*
 568                  * If we fail allocation our range, fallback to the controller
 569                  * discard page. If that's also busy, it's safe to return
 570                  * busy, as we know we can make progress once that's freed.
 571                  */
 572                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 573                         return BLK_STS_RESOURCE;
 574
 575                 range = page_address(ns->ctrl->discard_page);
 576         }
 577
 578         __rq_for_each_bio(bio, req) {
 579                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 580                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 581
 582                 if (n < segments) {
 583                         range[n].cattr = cpu_to_le32(0);
 584                         range[n].nlb = cpu_to_le32(nlb);
 585                         range[n].slba = cpu_to_le64(slba);
 586                 }
 587                 n++;
 588         }
 589
 590         if (WARN_ON_ONCE(n != segments)) {
 591                 if (virt_to_page(range) == ns->ctrl->discard_page)
 592                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 593                 else
 594                         kfree(range);
 595                 return BLK_STS_IOERR;
 596         }
 597
 598         cmnd->dsm.opcode = nvme_cmd_dsm;
 599         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 600         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 601         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 602
 603         req->special_vec.bv_page = virt_to_page(range);
 604         req->special_vec.bv_offset = offset_in_page(range);
 605         req->special_vec.bv_len = sizeof(*range) * segments;
 606         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 607
 608         return BLK_STS_OK;
 609 }
 610
 611 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 612                 struct request *req, struct nvme_command *cmnd)
 613 {
 614         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 615                 return nvme_setup_discard(ns, req, cmnd);
 616
 617         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 618         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 619         cmnd->write_zeroes.slba =
 620                 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 621         cmnd->write_zeroes.length =
 622                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 623         cmnd->write_zeroes.control = 0;
 624         return BLK_STS_OK;
 625 }
 626
 627 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 628                 struct request *req, struct nvme_command *cmnd)
 629 {
 630         struct nvme_ctrl *ctrl = ns->ctrl;
 631         u16 control = 0;
 632         u32 dsmgmt = 0;
 633
 634         if (req->cmd_flags & REQ_FUA)
 635                 control |= NVME_RW_FUA;
 636         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 637                 control |= NVME_RW_LR;
 638
 639         if (req->cmd_flags & REQ_RAHEAD)
 640                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 641
 642         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 643         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 644         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 645         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 646
 647         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 648                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 649
 650         if (ns->ms) {
 651                 /*
 652                  * If formated with metadata, the block layer always provides a
 653                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 654                  * we enable the PRACT bit for protection information or set the
 655                  * namespace capacity to zero to prevent any I/O.
 656                  */
 657                 if (!blk_integrity_rq(req)) {
 658                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 659                                 return BLK_STS_NOTSUPP;
 660                         control |= NVME_RW_PRINFO_PRACT;
 661                 } else if (req_op(req) == REQ_OP_WRITE) {
 662                         t10_pi_prepare(req, ns->pi_type);
 663                 }
 664
 665                 switch (ns->pi_type) {
 666                 case NVME_NS_DPS_PI_TYPE3:
 667                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 668                         break;
 669                 case NVME_NS_DPS_PI_TYPE1:
 670                 case NVME_NS_DPS_PI_TYPE2:
 671                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 672                                         NVME_RW_PRINFO_PRCHK_REF;
 673                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 674                         break;
 675                 }
 676         }
 677
 678         cmnd->rw.control = cpu_to_le16(control);
 679         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 680         return 0;
 681 }
 682
 683 void nvme_cleanup_cmd(struct request *req)
 684 {
 685         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
 686             nvme_req(req)->status == 0) {
 687                 struct nvme_ns *ns = req->rq_disk->private_data;
 688
 689                 t10_pi_complete(req, ns->pi_type,
 690                                 blk_rq_bytes(req) >> ns->lba_shift);
 691         }
 692         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 693                 struct nvme_ns *ns = req->rq_disk->private_data;
 694                 struct page *page = req->special_vec.bv_page;
 695
 696                 if (page == ns->ctrl->discard_page)
 697                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 698                 else
 699                         kfree(page_address(page) + req->special_vec.bv_offset);
 700         }
 701 }
 702 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 703
 704 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 705                 struct nvme_command *cmd)
 706 {
 707         blk_status_t ret = BLK_STS_OK;
 708
 709         nvme_clear_nvme_request(req);
 710
 711         memset(cmd, 0, sizeof(*cmd));
 712         switch (req_op(req)) {
 713         case REQ_OP_DRV_IN:
 714         case REQ_OP_DRV_OUT:
 715                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 716                 break;
 717         case REQ_OP_FLUSH:
 718                 nvme_setup_flush(ns, cmd);
 719                 break;
 720         case REQ_OP_WRITE_ZEROES:
 721                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 722                 break;
 723         case REQ_OP_DISCARD:
 724                 ret = nvme_setup_discard(ns, req, cmd);
 725                 break;
 726         case REQ_OP_READ:
 727         case REQ_OP_WRITE:
 728                 ret = nvme_setup_rw(ns, req, cmd);
 729                 break;
 730         default:
 731                 WARN_ON_ONCE(1);
 732                 return BLK_STS_IOERR;
 733         }
 734
 735         cmd->common.command_id = req->tag;
 736         trace_nvme_setup_cmd(req, cmd);
 737         return ret;
 738 }
 739 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 740
 741 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 742 {
 743         struct completion *waiting = rq->end_io_data;
 744
 745         rq->end_io_data = NULL;
 746         complete(waiting);
 747 }
 748
 749 static void nvme_execute_rq_polled(struct request_queue *q,
 750                 struct gendisk *bd_disk, struct request *rq, int at_head)
 751 {
 752         DECLARE_COMPLETION_ONSTACK(wait);
 753
 754         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 755
 756         rq->cmd_flags |= REQ_HIPRI;
 757         rq->end_io_data = &wait;
 758         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 759
 760         while (!completion_done(&wait)) {
 761                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 762                 cond_resched();
 763         }
 764 }
 765
 766 /*
 767  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 768  * if the result is positive, it's an NVM Express status code
 769  */
 770 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 771                 union nvme_result *result, void *buffer, unsigned bufflen,
 772                 unsigned timeout, int qid, int at_head,
 773                 blk_mq_req_flags_t flags, bool poll)
 774 {
 775         struct request *req;
 776         int ret;
 777
 778         req = nvme_alloc_request(q, cmd, flags, qid);
 779         if (IS_ERR(req))
 780                 return PTR_ERR(req);
 781
 782         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 783
 784         if (buffer && bufflen) {
 785                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 786                 if (ret)
 787                         goto out;
 788         }
 789
 790         if (poll)
 791                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 792         else
 793                 blk_execute_rq(req->q, NULL, req, at_head);
 794         if (result)
 795                 *result = nvme_req(req)->result;
 796         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 797                 ret = -EINTR;
 798         else
 799                 ret = nvme_req(req)->status;
 800  out:
 801         blk_mq_free_request(req);
 802         return ret;
 803 }
 804 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 805
 806 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 807                 void *buffer, unsigned bufflen)
 808 {
 809         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 810                         NVME_QID_ANY, 0, 0, false);
 811 }
 812 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 813
 814 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 815                 unsigned len, u32 seed, bool write)
 816 {
 817         struct bio_integrity_payload *bip;
 818         int ret = -ENOMEM;
 819         void *buf;
 820
 821         buf = kmalloc(len, GFP_KERNEL);
 822         if (!buf)
 823                 goto out;
 824
 825         ret = -EFAULT;
 826         if (write && copy_from_user(buf, ubuf, len))
 827                 goto out_free_meta;
 828
 829         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 830         if (IS_ERR(bip)) {
 831                 ret = PTR_ERR(bip);
 832                 goto out_free_meta;
 833         }
 834
 835         bip->bip_iter.bi_size = len;
 836         bip->bip_iter.bi_sector = seed;
 837         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 838                         offset_in_page(buf));
 839         if (ret == len)
 840                 return buf;
 841         ret = -ENOMEM;
 842 out_free_meta:
 843         kfree(buf);
 844 out:
 845         return ERR_PTR(ret);
 846 }
 847
 848 static int nvme_submit_user_cmd(struct request_queue *q,
 849                 struct nvme_command *cmd, void __user *ubuffer,
 850                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 851                 u32 meta_seed, u32 *result, unsigned timeout)
 852 {
 853         bool write = nvme_is_write(cmd);
 854         struct nvme_ns *ns = q->queuedata;
 855         struct gendisk *disk = ns ? ns->disk : NULL;
 856         struct request *req;
 857         struct bio *bio = NULL;
 858         void *meta = NULL;
 859         int ret;
 860
 861         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 862         if (IS_ERR(req))
 863                 return PTR_ERR(req);
 864
 865         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 866         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 867
 868         if (ubuffer && bufflen) {
 869                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 870                                 GFP_KERNEL);
 871                 if (ret)
 872                         goto out;
 873                 bio = req->bio;
 874                 bio->bi_disk = disk;
 875                 if (disk && meta_buffer && meta_len) {
 876                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 877                                         meta_seed, write);
 878                         if (IS_ERR(meta)) {
 879                                 ret = PTR_ERR(meta);
 880                                 goto out_unmap;
 881                         }
 882                         req->cmd_flags |= REQ_INTEGRITY;
 883                 }
 884         }
 885
 886         blk_execute_rq(req->q, disk, req, 0);
 887         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 888                 ret = -EINTR;
 889         else
 890                 ret = nvme_req(req)->status;
 891         if (result)
 892                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 893         if (meta && !ret && !write) {
 894                 if (copy_to_user(meta_buffer, meta, meta_len))
 895                         ret = -EFAULT;
 896         }
 897         kfree(meta);
 898  out_unmap:
 899         if (bio)
 900                 blk_rq_unmap_user(bio);
 901  out:
 902         blk_mq_free_request(req);
 903         return ret;
 904 }
 905
 906 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 907 {
 908         struct nvme_ctrl *ctrl = rq->end_io_data;
 909         unsigned long flags;
 910         bool startka = false;
 911
 912         blk_mq_free_request(rq);
 913
 914         if (status) {
 915                 dev_err(ctrl->device,
 916                         "failed nvme_keep_alive_end_io error=%d\n",
 917                                 status);
 918                 return;
 919         }
 920
 921         ctrl->comp_seen = false;
 922         spin_lock_irqsave(&ctrl->lock, flags);
 923         if (ctrl->state == NVME_CTRL_LIVE ||
 924             ctrl->state == NVME_CTRL_CONNECTING)
 925                 startka = true;
 926         spin_unlock_irqrestore(&ctrl->lock, flags);
 927         if (startka)
 928                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 929 }
 930
 931 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 932 {
 933         struct request *rq;
 934
 935         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 936                         NVME_QID_ANY);
 937         if (IS_ERR(rq))
 938                 return PTR_ERR(rq);
 939
 940         rq->timeout = ctrl->kato * HZ;
 941         rq->end_io_data = ctrl;
 942
 943         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 944
 945         return 0;
 946 }
 947
 948 static void nvme_keep_alive_work(struct work_struct *work)
 949 {
 950         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 951                         struct nvme_ctrl, ka_work);
 952         bool comp_seen = ctrl->comp_seen;
 953
 954         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
 955                 dev_dbg(ctrl->device,
 956                         "reschedule traffic based keep-alive timer\n");
 957                 ctrl->comp_seen = false;
 958                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 959                 return;
 960         }
 961
 962         if (nvme_keep_alive(ctrl)) {
 963                 /* allocation failure, reset the controller */
 964                 dev_err(ctrl->device, "keep-alive failed\n");
 965                 nvme_reset_ctrl(ctrl);
 966                 return;
 967         }
 968 }
 969
 970 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 971 {
 972         if (unlikely(ctrl->kato == 0))
 973                 return;
 974
 975         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 976 }
 977
 978 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 979 {
 980         if (unlikely(ctrl->kato == 0))
 981                 return;
 982
 983         cancel_delayed_work_sync(&ctrl->ka_work);
 984 }
 985 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 986
 987 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 988 {
 989         struct nvme_command c = { };
 990         int error;
 991
 992         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 993         c.identify.opcode = nvme_admin_identify;
 994         c.identify.cns = NVME_ID_CNS_CTRL;
 995
 996         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 997         if (!*id)
 998                 return -ENOMEM;
 999
1000         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1001                         sizeof(struct nvme_id_ctrl));
1002         if (error)
1003                 kfree(*id);
1004         return error;
1005 }
1006
1007 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1008                 struct nvme_ns_ids *ids)
1009 {
1010         struct nvme_command c = { };
1011         int status;
1012         void *data;
1013         int pos;
1014         int len;
1015
1016         c.identify.opcode = nvme_admin_identify;
1017         c.identify.nsid = cpu_to_le32(nsid);
1018         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1019
1020         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1021         if (!data)
1022                 return -ENOMEM;
1023
1024         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1025                                       NVME_IDENTIFY_DATA_SIZE);
1026         if (status)
1027                 goto free_data;
1028
1029         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1030                 struct nvme_ns_id_desc *cur = data + pos;
1031
1032                 if (cur->nidl == 0)
1033                         break;
1034
1035                 switch (cur->nidt) {
1036                 case NVME_NIDT_EUI64:
1037                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1038                                 dev_warn(ctrl->device,
1039                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1040                                          cur->nidl);
1041                                 goto free_data;
1042                         }
1043                         len = NVME_NIDT_EUI64_LEN;
1044                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1045                         break;
1046                 case NVME_NIDT_NGUID:
1047                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1048                                 dev_warn(ctrl->device,
1049                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1050                                          cur->nidl);
1051                                 goto free_data;
1052                         }
1053                         len = NVME_NIDT_NGUID_LEN;
1054                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1055                         break;
1056                 case NVME_NIDT_UUID:
1057                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
1058                                 dev_warn(ctrl->device,
1059                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1060                                          cur->nidl);
1061                                 goto free_data;
1062                         }
1063                         len = NVME_NIDT_UUID_LEN;
1064                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1065                         break;
1066                 default:
1067                         /* Skip unknown types */
1068                         len = cur->nidl;
1069                         break;
1070                 }
1071
1072                 len += sizeof(*cur);
1073         }
1074 free_data:
1075         kfree(data);
1076         return status;
1077 }
1078
1079 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1080 {
1081         struct nvme_command c = { };
1082
1083         c.identify.opcode = nvme_admin_identify;
1084         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1085         c.identify.nsid = cpu_to_le32(nsid);
1086         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1087                                     NVME_IDENTIFY_DATA_SIZE);
1088 }
1089
1090 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1091                 unsigned nsid)
1092 {
1093         struct nvme_id_ns *id;
1094         struct nvme_command c = { };
1095         int error;
1096
1097         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1098         c.identify.opcode = nvme_admin_identify;
1099         c.identify.nsid = cpu_to_le32(nsid);
1100         c.identify.cns = NVME_ID_CNS_NS;
1101
1102         id = kmalloc(sizeof(*id), GFP_KERNEL);
1103         if (!id)
1104                 return NULL;
1105
1106         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1107         if (error) {
1108                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1109                 kfree(id);
1110                 return NULL;
1111         }
1112
1113         return id;
1114 }
1115
1116 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1117                       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 = nvme_admin_set_features;
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_queue_count(struct nvme_ctrl *ctrl, int *count)
1136 {
1137         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1138         u32 result;
1139         int status, nr_io_queues;
1140
1141         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1142                         &result);
1143         if (status < 0)
1144                 return status;
1145
1146         /*
1147          * Degraded controllers might return an error when setting the queue
1148          * count.  We still want to be able to bring them online and offer
1149          * access to the admin queue, as that might be only way to fix them up.
1150          */
1151         if (status > 0) {
1152                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1153                 *count = 0;
1154         } else {
1155                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1156                 *count = min(*count, nr_io_queues);
1157         }
1158
1159         return 0;
1160 }
1161 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1162
1163 #define NVME_AEN_SUPPORTED \
1164         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1165
1166 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1167 {
1168         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1169         int status;
1170
1171         if (!supported_aens)
1172                 return;
1173
1174         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1175                         NULL, 0, &result);
1176         if (status)
1177                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1178                          supported_aens);
1179 }
1180
1181 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1182 {
1183         struct nvme_user_io io;
1184         struct nvme_command c;
1185         unsigned length, meta_len;
1186         void __user *metadata;
1187
1188         if (copy_from_user(&io, uio, sizeof(io)))
1189                 return -EFAULT;
1190         if (io.flags)
1191                 return -EINVAL;
1192
1193         switch (io.opcode) {
1194         case nvme_cmd_write:
1195         case nvme_cmd_read:
1196         case nvme_cmd_compare:
1197                 break;
1198         default:
1199                 return -EINVAL;
1200         }
1201
1202         length = (io.nblocks + 1) << ns->lba_shift;
1203         meta_len = (io.nblocks + 1) * ns->ms;
1204         metadata = (void __user *)(uintptr_t)io.metadata;
1205
1206         if (ns->ext) {
1207                 length += meta_len;
1208                 meta_len = 0;
1209         } else if (meta_len) {
1210                 if ((io.metadata & 3) || !io.metadata)
1211                         return -EINVAL;
1212         }
1213
1214         memset(&c, 0, sizeof(c));
1215         c.rw.opcode = io.opcode;
1216         c.rw.flags = io.flags;
1217         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1218         c.rw.slba = cpu_to_le64(io.slba);
1219         c.rw.length = cpu_to_le16(io.nblocks);
1220         c.rw.control = cpu_to_le16(io.control);
1221         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1222         c.rw.reftag = cpu_to_le32(io.reftag);
1223         c.rw.apptag = cpu_to_le16(io.apptag);
1224         c.rw.appmask = cpu_to_le16(io.appmask);
1225
1226         return nvme_submit_user_cmd(ns->queue, &c,
1227                         (void __user *)(uintptr_t)io.addr, length,
1228                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1229 }
1230
1231 static u32 nvme_known_admin_effects(u8 opcode)
1232 {
1233         switch (opcode) {
1234         case nvme_admin_format_nvm:
1235                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1236                                         NVME_CMD_EFFECTS_CSE_MASK;
1237         case nvme_admin_sanitize_nvm:
1238                 return NVME_CMD_EFFECTS_CSE_MASK;
1239         default:
1240                 break;
1241         }
1242         return 0;
1243 }
1244
1245 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1246                                                                 u8 opcode)
1247 {
1248         u32 effects = 0;
1249
1250         if (ns) {
1251                 if (ctrl->effects)
1252                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1253                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1254                         dev_warn(ctrl->device,
1255                                  "IO command:%02x has unhandled effects:%08x\n",
1256                                  opcode, effects);
1257                 return 0;
1258         }
1259
1260         if (ctrl->effects)
1261                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1262         effects |= nvme_known_admin_effects(opcode);
1263
1264         /*
1265          * For simplicity, IO to all namespaces is quiesced even if the command
1266          * effects say only one namespace is affected.
1267          */
1268         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1269                 mutex_lock(&ctrl->scan_lock);
1270                 nvme_start_freeze(ctrl);
1271                 nvme_wait_freeze(ctrl);
1272         }
1273         return effects;
1274 }
1275
1276 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1277 {
1278         struct nvme_ns *ns;
1279
1280         down_read(&ctrl->namespaces_rwsem);
1281         list_for_each_entry(ns, &ctrl->namespaces, list)
1282                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1283                         nvme_set_queue_dying(ns);
1284         up_read(&ctrl->namespaces_rwsem);
1285
1286         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1287 }
1288
1289 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1290 {
1291         /*
1292          * Revalidate LBA changes prior to unfreezing. This is necessary to
1293          * prevent memory corruption if a logical block size was changed by
1294          * this command.
1295          */
1296         if (effects & NVME_CMD_EFFECTS_LBCC)
1297                 nvme_update_formats(ctrl);
1298         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1299                 nvme_unfreeze(ctrl);
1300                 mutex_unlock(&ctrl->scan_lock);
1301         }
1302         if (effects & NVME_CMD_EFFECTS_CCC)
1303                 nvme_init_identify(ctrl);
1304         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1305                 nvme_queue_scan(ctrl);
1306 }
1307
1308 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1309                         struct nvme_passthru_cmd __user *ucmd)
1310 {
1311         struct nvme_passthru_cmd cmd;
1312         struct nvme_command c;
1313         unsigned timeout = 0;
1314         u32 effects;
1315         int status;
1316
1317         if (!capable(CAP_SYS_ADMIN))
1318                 return -EACCES;
1319         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1320                 return -EFAULT;
1321         if (cmd.flags)
1322                 return -EINVAL;
1323
1324         memset(&c, 0, sizeof(c));
1325         c.common.opcode = cmd.opcode;
1326         c.common.flags = cmd.flags;
1327         c.common.nsid = cpu_to_le32(cmd.nsid);
1328         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1329         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1330         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1331         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1332         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1333         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1334         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1335         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1336
1337         if (cmd.timeout_ms)
1338                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1339
1340         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1341         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1342                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1343                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1344                         0, &cmd.result, timeout);
1345         nvme_passthru_end(ctrl, effects);
1346
1347         if (status >= 0) {
1348                 if (put_user(cmd.result, &ucmd->result))
1349                         return -EFAULT;
1350         }
1351
1352         return status;
1353 }
1354
1355 /*
1356  * Issue ioctl requests on the first available path.  Note that unlike normal
1357  * block layer requests we will not retry failed request on another controller.
1358  */
1359 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1360                 struct nvme_ns_head **head, int *srcu_idx)
1361 {
1362 #ifdef CONFIG_NVME_MULTIPATH
1363         if (disk->fops == &nvme_ns_head_ops) {
1364                 struct nvme_ns *ns;
1365
1366                 *head = disk->private_data;
1367                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1368                 ns = nvme_find_path(*head);
1369                 if (!ns)
1370                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1371                 return ns;
1372         }
1373 #endif
1374         *head = NULL;
1375         *srcu_idx = -1;
1376         return disk->private_data;
1377 }
1378
1379 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1380 {
1381         if (head)
1382                 srcu_read_unlock(&head->srcu, idx);
1383 }
1384
1385 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1386                 unsigned int cmd, unsigned long arg)
1387 {
1388         struct nvme_ns_head *head = NULL;
1389         void __user *argp = (void __user *)arg;
1390         struct nvme_ns *ns;
1391         int srcu_idx, ret;
1392
1393         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1394         if (unlikely(!ns))
1395                 return -EWOULDBLOCK;
1396
1397         /*
1398          * Handle ioctls that apply to the controller instead of the namespace
1399          * seperately and drop the ns SRCU reference early.  This avoids a
1400          * deadlock when deleting namespaces using the passthrough interface.
1401          */
1402         if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1403                 struct nvme_ctrl *ctrl = ns->ctrl;
1404
1405                 nvme_get_ctrl(ns->ctrl);
1406                 nvme_put_ns_from_disk(head, srcu_idx);
1407
1408                 if (cmd == NVME_IOCTL_ADMIN_CMD)
1409                         ret = nvme_user_cmd(ctrl, NULL, argp);
1410                 else
1411                         ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1412
1413                 nvme_put_ctrl(ctrl);
1414                 return ret;
1415         }
1416
1417         switch (cmd) {
1418         case NVME_IOCTL_ID:
1419                 force_successful_syscall_return();
1420                 ret = ns->head->ns_id;
1421                 break;
1422         case NVME_IOCTL_IO_CMD:
1423                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1424                 break;
1425         case NVME_IOCTL_SUBMIT_IO:
1426                 ret = nvme_submit_io(ns, argp);
1427                 break;
1428         default:
1429                 if (ns->ndev)
1430                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1431                 else
1432                         ret = -ENOTTY;
1433         }
1434
1435         nvme_put_ns_from_disk(head, srcu_idx);
1436         return ret;
1437 }
1438
1439 static int nvme_open(struct block_device *bdev, fmode_t mode)
1440 {
1441         struct nvme_ns *ns = bdev->bd_disk->private_data;
1442
1443 #ifdef CONFIG_NVME_MULTIPATH
1444         /* should never be called due to GENHD_FL_HIDDEN */
1445         if (WARN_ON_ONCE(ns->head->disk))
1446                 goto fail;
1447 #endif
1448         if (!kref_get_unless_zero(&ns->kref))
1449                 goto fail;
1450         if (!try_module_get(ns->ctrl->ops->module))
1451                 goto fail_put_ns;
1452
1453         return 0;
1454
1455 fail_put_ns:
1456         nvme_put_ns(ns);
1457 fail:
1458         return -ENXIO;
1459 }
1460
1461 static void nvme_release(struct gendisk *disk, fmode_t mode)
1462 {
1463         struct nvme_ns *ns = disk->private_data;
1464
1465         module_put(ns->ctrl->ops->module);
1466         nvme_put_ns(ns);
1467 }
1468
1469 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1470 {
1471         /* some standard values */
1472         geo->heads = 1 << 6;
1473         geo->sectors = 1 << 5;
1474         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1475         return 0;
1476 }
1477
1478 #ifdef CONFIG_BLK_DEV_INTEGRITY
1479 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1480 {
1481         struct blk_integrity integrity;
1482
1483         memset(&integrity, 0, sizeof(integrity));
1484         switch (pi_type) {
1485         case NVME_NS_DPS_PI_TYPE3:
1486                 integrity.profile = &t10_pi_type3_crc;
1487                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1488                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1489                 break;
1490         case NVME_NS_DPS_PI_TYPE1:
1491         case NVME_NS_DPS_PI_TYPE2:
1492                 integrity.profile = &t10_pi_type1_crc;
1493                 integrity.tag_size = sizeof(u16);
1494                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1495                 break;
1496         default:
1497                 integrity.profile = NULL;
1498                 break;
1499         }
1500         integrity.tuple_size = ms;
1501         blk_integrity_register(disk, &integrity);
1502         blk_queue_max_integrity_segments(disk->queue, 1);
1503 }
1504 #else
1505 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1506 {
1507 }
1508 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1509
1510 static void nvme_set_chunk_size(struct nvme_ns *ns)
1511 {
1512         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1513         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1514 }
1515
1516 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1517 {
1518         struct nvme_ctrl *ctrl = ns->ctrl;
1519         struct request_queue *queue = disk->queue;
1520         u32 size = queue_logical_block_size(queue);
1521
1522         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1523                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1524                 return;
1525         }
1526
1527         if (ctrl->nr_streams && ns->sws && ns->sgs)
1528                 size *= ns->sws * ns->sgs;
1529
1530         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1531                         NVME_DSM_MAX_RANGES);
1532
1533         queue->limits.discard_alignment = 0;
1534         queue->limits.discard_granularity = size;
1535
1536         /* If discard is already enabled, don't reset queue limits */
1537         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1538                 return;
1539
1540         blk_queue_max_discard_sectors(queue, UINT_MAX);
1541         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1542
1543         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1544                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1545 }
1546
1547 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1548 {
1549         u32 max_sectors;
1550         unsigned short bs = 1 << ns->lba_shift;
1551
1552         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1553             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1554                 return;
1555         /*
1556          * Even though NVMe spec explicitly states that MDTS is not
1557          * applicable to the write-zeroes:- "The restriction does not apply to
1558          * commands that do not transfer data between the host and the
1559          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1560          * In order to be more cautious use controller's max_hw_sectors value
1561          * to configure the maximum sectors for the write-zeroes which is
1562          * configured based on the controller's MDTS field in the
1563          * nvme_init_identify() if available.
1564          */
1565         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1566                 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1567         else
1568                 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1569
1570         blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1571 }
1572
1573 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1574                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1575 {
1576         memset(ids, 0, sizeof(*ids));
1577
1578         if (ctrl->vs >= NVME_VS(1, 1, 0))
1579                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1580         if (ctrl->vs >= NVME_VS(1, 2, 0))
1581                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1582         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1583                  /* Don't treat error as fatal we potentially
1584                   * already have a NGUID or EUI-64
1585                   */
1586                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1587                         dev_warn(ctrl->device,
1588                                  "%s: Identify Descriptors failed\n", __func__);
1589         }
1590 }
1591
1592 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1593 {
1594         return !uuid_is_null(&ids->uuid) ||
1595                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1596                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1597 }
1598
1599 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1600 {
1601         return uuid_equal(&a->uuid, &b->uuid) &&
1602                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1603                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1604 }
1605
1606 static void nvme_update_disk_info(struct gendisk *disk,
1607                 struct nvme_ns *ns, struct nvme_id_ns *id)
1608 {
1609         sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1610         unsigned short bs = 1 << ns->lba_shift;
1611
1612         if (ns->lba_shift > PAGE_SHIFT) {
1613                 /* unsupported block size, set capacity to 0 later */
1614                 bs = (1 << 9);
1615         }
1616         blk_mq_freeze_queue(disk->queue);
1617         blk_integrity_unregister(disk);
1618
1619         blk_queue_logical_block_size(disk->queue, bs);
1620         blk_queue_physical_block_size(disk->queue, bs);
1621         blk_queue_io_min(disk->queue, bs);
1622
1623         if (ns->ms && !ns->ext &&
1624             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1625                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1626         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1627             ns->lba_shift > PAGE_SHIFT)
1628                 capacity = 0;
1629
1630         set_capacity(disk, capacity);
1631
1632         nvme_config_discard(disk, ns);
1633         nvme_config_write_zeroes(disk, ns);
1634
1635         if (id->nsattr & (1 << 0))
1636                 set_disk_ro(disk, true);
1637         else
1638                 set_disk_ro(disk, false);
1639
1640         blk_mq_unfreeze_queue(disk->queue);
1641 }
1642
1643 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1644 {
1645         struct nvme_ns *ns = disk->private_data;
1646
1647         /*
1648          * If identify namespace failed, use default 512 byte block size so
1649          * block layer can use before failing read/write for 0 capacity.
1650          */
1651         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1652         if (ns->lba_shift == 0)
1653                 ns->lba_shift = 9;
1654         ns->noiob = le16_to_cpu(id->noiob);
1655         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1656         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1657         /* the PI implementation requires metadata equal t10 pi tuple size */
1658         if (ns->ms == sizeof(struct t10_pi_tuple))
1659                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1660         else
1661                 ns->pi_type = 0;
1662
1663         if (ns->noiob)
1664                 nvme_set_chunk_size(ns);
1665         nvme_update_disk_info(disk, ns, id);
1666 #ifdef CONFIG_NVME_MULTIPATH
1667         if (ns->head->disk) {
1668                 nvme_update_disk_info(ns->head->disk, ns, id);
1669                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1670         }
1671 #endif
1672 }
1673
1674 static int nvme_revalidate_disk(struct gendisk *disk)
1675 {
1676         struct nvme_ns *ns = disk->private_data;
1677         struct nvme_ctrl *ctrl = ns->ctrl;
1678         struct nvme_id_ns *id;
1679         struct nvme_ns_ids ids;
1680         int ret = 0;
1681
1682         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1683                 set_capacity(disk, 0);
1684                 return -ENODEV;
1685         }
1686
1687         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1688         if (!id)
1689                 return -ENODEV;
1690
1691         if (id->ncap == 0) {
1692                 ret = -ENODEV;
1693                 goto out;
1694         }
1695
1696         __nvme_revalidate_disk(disk, id);
1697         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1698         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1699                 dev_err(ctrl->device,
1700                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1701                 ret = -ENODEV;
1702         }
1703
1704 out:
1705         kfree(id);
1706         return ret;
1707 }
1708
1709 static char nvme_pr_type(enum pr_type type)
1710 {
1711         switch (type) {
1712         case PR_WRITE_EXCLUSIVE:
1713                 return 1;
1714         case PR_EXCLUSIVE_ACCESS:
1715                 return 2;
1716         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1717                 return 3;
1718         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1719                 return 4;
1720         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1721                 return 5;
1722         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1723                 return 6;
1724         default:
1725                 return 0;
1726         }
1727 };
1728
1729 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1730                                 u64 key, u64 sa_key, u8 op)
1731 {
1732         struct nvme_ns_head *head = NULL;
1733         struct nvme_ns *ns;
1734         struct nvme_command c;
1735         int srcu_idx, ret;
1736         u8 data[16] = { 0, };
1737
1738         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1739         if (unlikely(!ns))
1740                 return -EWOULDBLOCK;
1741
1742         put_unaligned_le64(key, &data[0]);
1743         put_unaligned_le64(sa_key, &data[8]);
1744
1745         memset(&c, 0, sizeof(c));
1746         c.common.opcode = op;
1747         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1748         c.common.cdw10 = cpu_to_le32(cdw10);
1749
1750         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1751         nvme_put_ns_from_disk(head, srcu_idx);
1752         return ret;
1753 }
1754
1755 static int nvme_pr_register(struct block_device *bdev, u64 old,
1756                 u64 new, unsigned flags)
1757 {
1758         u32 cdw10;
1759
1760         if (flags & ~PR_FL_IGNORE_KEY)
1761                 return -EOPNOTSUPP;
1762
1763         cdw10 = old ? 2 : 0;
1764         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1765         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1766         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1767 }
1768
1769 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1770                 enum pr_type type, unsigned flags)
1771 {
1772         u32 cdw10;
1773
1774         if (flags & ~PR_FL_IGNORE_KEY)
1775                 return -EOPNOTSUPP;
1776
1777         cdw10 = nvme_pr_type(type) << 8;
1778         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1779         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1780 }
1781
1782 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1783                 enum pr_type type, bool abort)
1784 {
1785         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1786         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1787 }
1788
1789 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1790 {
1791         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1792         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1793 }
1794
1795 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1796 {
1797         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1798         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1799 }
1800
1801 static const struct pr_ops nvme_pr_ops = {
1802         .pr_register    = nvme_pr_register,
1803         .pr_reserve     = nvme_pr_reserve,
1804         .pr_release     = nvme_pr_release,
1805         .pr_preempt     = nvme_pr_preempt,
1806         .pr_clear       = nvme_pr_clear,
1807 };
1808
1809 #ifdef CONFIG_BLK_SED_OPAL
1810 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1811                 bool send)
1812 {
1813         struct nvme_ctrl *ctrl = data;
1814         struct nvme_command cmd;
1815
1816         memset(&cmd, 0, sizeof(cmd));
1817         if (send)
1818                 cmd.common.opcode = nvme_admin_security_send;
1819         else
1820                 cmd.common.opcode = nvme_admin_security_recv;
1821         cmd.common.nsid = 0;
1822         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1823         cmd.common.cdw11 = cpu_to_le32(len);
1824
1825         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1826                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1827 }
1828 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1829 #endif /* CONFIG_BLK_SED_OPAL */
1830
1831 static const struct block_device_operations nvme_fops = {
1832         .owner          = THIS_MODULE,
1833         .ioctl          = nvme_ioctl,
1834         .compat_ioctl   = nvme_ioctl,
1835         .open           = nvme_open,
1836         .release        = nvme_release,
1837         .getgeo         = nvme_getgeo,
1838         .revalidate_disk= nvme_revalidate_disk,
1839         .pr_ops         = &nvme_pr_ops,
1840 };
1841
1842 #ifdef CONFIG_NVME_MULTIPATH
1843 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1844 {
1845         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1846
1847         if (!kref_get_unless_zero(&head->ref))
1848                 return -ENXIO;
1849         return 0;
1850 }
1851
1852 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1853 {
1854         nvme_put_ns_head(disk->private_data);
1855 }
1856
1857 const struct block_device_operations nvme_ns_head_ops = {
1858         .owner          = THIS_MODULE,
1859         .open           = nvme_ns_head_open,
1860         .release        = nvme_ns_head_release,
1861         .ioctl          = nvme_ioctl,
1862         .compat_ioctl   = nvme_ioctl,
1863         .getgeo         = nvme_getgeo,
1864         .pr_ops         = &nvme_pr_ops,
1865 };
1866 #endif /* CONFIG_NVME_MULTIPATH */
1867
1868 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1869 {
1870         unsigned long timeout =
1871                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1872         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1873         int ret;
1874
1875         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1876                 if (csts == ~0)
1877                         return -ENODEV;
1878                 if ((csts & NVME_CSTS_RDY) == bit)
1879                         break;
1880
1881                 msleep(100);
1882                 if (fatal_signal_pending(current))
1883                         return -EINTR;
1884                 if (time_after(jiffies, timeout)) {
1885                         dev_err(ctrl->device,
1886                                 "Device not ready; aborting %s\n", enabled ?
1887                                                 "initialisation" : "reset");
1888                         return -ENODEV;
1889                 }
1890         }
1891
1892         return ret;
1893 }
1894
1895 /*
1896  * If the device has been passed off to us in an enabled state, just clear
1897  * the enabled bit.  The spec says we should set the 'shutdown notification
1898  * bits', but doing so may cause the device to complete commands to the
1899  * admin queue ... and we don't know what memory that might be pointing at!
1900  */
1901 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1902 {
1903         int ret;
1904
1905         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1906         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1907
1908         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1909         if (ret)
1910                 return ret;
1911
1912         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1913                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1914
1915         return nvme_wait_ready(ctrl, cap, false);
1916 }
1917 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1918
1919 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1920 {
1921         /*
1922          * Default to a 4K page size, with the intention to update this
1923          * path in the future to accomodate architectures with differing
1924          * kernel and IO page sizes.
1925          */
1926         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1927         int ret;
1928
1929         if (page_shift < dev_page_min) {
1930                 dev_err(ctrl->device,
1931                         "Minimum device page size %u too large for host (%u)\n",
1932                         1 << dev_page_min, 1 << page_shift);
1933                 return -ENODEV;
1934         }
1935
1936         ctrl->page_size = 1 << page_shift;
1937
1938         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1939         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1940         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1941         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1942         ctrl->ctrl_config |= NVME_CC_ENABLE;
1943
1944         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1945         if (ret)
1946                 return ret;
1947         return nvme_wait_ready(ctrl, cap, true);
1948 }
1949 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1950
1951 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1952 {
1953         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1954         u32 csts;
1955         int ret;
1956
1957         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1958         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1959
1960         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1961         if (ret)
1962                 return ret;
1963
1964         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1965                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1966                         break;
1967
1968                 msleep(100);
1969                 if (fatal_signal_pending(current))
1970                         return -EINTR;
1971                 if (time_after(jiffies, timeout)) {
1972                         dev_err(ctrl->device,
1973                                 "Device shutdown incomplete; abort shutdown\n");
1974                         return -ENODEV;
1975                 }
1976         }
1977
1978         return ret;
1979 }
1980 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1981
1982 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1983                 struct request_queue *q)
1984 {
1985         bool vwc = false;
1986
1987         if (ctrl->max_hw_sectors) {
1988                 u32 max_segments =
1989                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1990
1991                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1992                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1993                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1994         }
1995         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1996             is_power_of_2(ctrl->max_hw_sectors))
1997                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1998         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1999         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2000                 vwc = true;
2001         blk_queue_write_cache(q, vwc, vwc);
2002 }
2003
2004 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2005 {
2006         __le64 ts;
2007         int ret;
2008
2009         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2010                 return 0;
2011
2012         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2013         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2014                         NULL);
2015         if (ret)
2016                 dev_warn_once(ctrl->device,
2017                         "could not set timestamp (%d)\n", ret);
2018         return ret;
2019 }
2020
2021 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2022 {
2023         struct nvme_feat_host_behavior *host;
2024         int ret;
2025
2026         /* Don't bother enabling the feature if retry delay is not reported */
2027         if (!ctrl->crdt[0])
2028                 return 0;
2029
2030         host = kzalloc(sizeof(*host), GFP_KERNEL);
2031         if (!host)
2032                 return 0;
2033
2034         host->acre = NVME_ENABLE_ACRE;
2035         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2036                                 host, sizeof(*host), NULL);
2037         kfree(host);
2038         return ret;
2039 }
2040
2041 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2042 {
2043         /*
2044          * APST (Autonomous Power State Transition) lets us program a
2045          * table of power state transitions that the controller will
2046          * perform automatically.  We configure it with a simple
2047          * heuristic: we are willing to spend at most 2% of the time
2048          * transitioning between power states.  Therefore, when running
2049          * in any given state, we will enter the next lower-power
2050          * non-operational state after waiting 50 * (enlat + exlat)
2051          * microseconds, as long as that state's exit latency is under
2052          * the requested maximum latency.
2053          *
2054          * We will not autonomously enter any non-operational state for
2055          * which the total latency exceeds ps_max_latency_us.  Users
2056          * can set ps_max_latency_us to zero to turn off APST.
2057          */
2058
2059         unsigned apste;
2060         struct nvme_feat_auto_pst *table;
2061         u64 max_lat_us = 0;
2062         int max_ps = -1;
2063         int ret;
2064
2065         /*
2066          * If APST isn't supported or if we haven't been initialized yet,
2067          * then don't do anything.
2068          */
2069         if (!ctrl->apsta)
2070                 return 0;
2071
2072         if (ctrl->npss > 31) {
2073                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2074                 return 0;
2075         }
2076
2077         table = kzalloc(sizeof(*table), GFP_KERNEL);
2078         if (!table)
2079                 return 0;
2080
2081         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2082                 /* Turn off APST. */
2083                 apste = 0;
2084                 dev_dbg(ctrl->device, "APST disabled\n");
2085         } else {
2086                 __le64 target = cpu_to_le64(0);
2087                 int state;
2088
2089                 /*
2090                  * Walk through all states from lowest- to highest-power.
2091                  * According to the spec, lower-numbered states use more
2092                  * power.  NPSS, despite the name, is the index of the
2093                  * lowest-power state, not the number of states.
2094                  */
2095                 for (state = (int)ctrl->npss; state >= 0; state--) {
2096                         u64 total_latency_us, exit_latency_us, transition_ms;
2097
2098                         if (target)
2099                                 table->entries[state] = target;
2100
2101                         /*
2102                          * Don't allow transitions to the deepest state
2103                          * if it's quirked off.
2104                          */
2105                         if (state == ctrl->npss &&
2106                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2107                                 continue;
2108
2109                         /*
2110                          * Is this state a useful non-operational state for
2111                          * higher-power states to autonomously transition to?
2112                          */
2113                         if (!(ctrl->psd[state].flags &
2114                               NVME_PS_FLAGS_NON_OP_STATE))
2115                                 continue;
2116
2117                         exit_latency_us =
2118                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2119                         if (exit_latency_us > ctrl->ps_max_latency_us)
2120                                 continue;
2121
2122                         total_latency_us =
2123                                 exit_latency_us +
2124                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2125
2126                         /*
2127                          * This state is good.  Use it as the APST idle
2128                          * target for higher power states.
2129                          */
2130                         transition_ms = total_latency_us + 19;
2131                         do_div(transition_ms, 20);
2132                         if (transition_ms > (1 << 24) - 1)
2133                                 transition_ms = (1 << 24) - 1;
2134
2135                         target = cpu_to_le64((state << 3) |
2136                                              (transition_ms << 8));
2137
2138                         if (max_ps == -1)
2139                                 max_ps = state;
2140
2141                         if (total_latency_us > max_lat_us)
2142                                 max_lat_us = total_latency_us;
2143                 }
2144
2145                 apste = 1;
2146
2147                 if (max_ps == -1) {
2148                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2149                 } else {
2150                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2151                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2152                 }
2153         }
2154
2155         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2156                                 table, sizeof(*table), NULL);
2157         if (ret)
2158                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2159
2160         kfree(table);
2161         return ret;
2162 }
2163
2164 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2165 {
2166         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2167         u64 latency;
2168
2169         switch (val) {
2170         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2171         case PM_QOS_LATENCY_ANY:
2172                 latency = U64_MAX;
2173                 break;
2174
2175         default:
2176                 latency = val;
2177         }
2178
2179         if (ctrl->ps_max_latency_us != latency) {
2180                 ctrl->ps_max_latency_us = latency;
2181                 nvme_configure_apst(ctrl);
2182         }
2183 }
2184
2185 struct nvme_core_quirk_entry {
2186         /*
2187          * NVMe model and firmware strings are padded with spaces.  For
2188          * simplicity, strings in the quirk table are padded with NULLs
2189          * instead.
2190          */
2191         u16 vid;
2192         const char *mn;
2193         const char *fr;
2194         unsigned long quirks;
2195 };
2196
2197 static const struct nvme_core_quirk_entry core_quirks[] = {
2198         {
2199                 /*
2200                  * This Toshiba device seems to die using any APST states.  See:
2201                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2202                  */
2203                 .vid = 0x1179,
2204                 .mn = "THNSF5256GPUK TOSHIBA",
2205                 .quirks = NVME_QUIRK_NO_APST,
2206         }
2207 };
2208
2209 /* match is null-terminated but idstr is space-padded. */
2210 static bool string_matches(const char *idstr, const char *match, size_t len)
2211 {
2212         size_t matchlen;
2213
2214         if (!match)
2215                 return true;
2216
2217         matchlen = strlen(match);
2218         WARN_ON_ONCE(matchlen > len);
2219
2220         if (memcmp(idstr, match, matchlen))
2221                 return false;
2222
2223         for (; matchlen < len; matchlen++)
2224                 if (idstr[matchlen] != ' ')
2225                         return false;
2226
2227         return true;
2228 }
2229
2230 static bool quirk_matches(const struct nvme_id_ctrl *id,
2231                           const struct nvme_core_quirk_entry *q)
2232 {
2233         return q->vid == le16_to_cpu(id->vid) &&
2234                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2235                 string_matches(id->fr, q->fr, sizeof(id->fr));
2236 }
2237
2238 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2239                 struct nvme_id_ctrl *id)
2240 {
2241         size_t nqnlen;
2242         int off;
2243
2244         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2245                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2246                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2247                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2248                         return;
2249                 }
2250
2251                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2252                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2253         }
2254
2255         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2256         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2257                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2258                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2259         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2260         off += sizeof(id->sn);
2261         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2262         off += sizeof(id->mn);
2263         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2264 }
2265
2266 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2267 {
2268         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2269         kfree(subsys);
2270 }
2271
2272 static void nvme_release_subsystem(struct device *dev)
2273 {
2274         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2275 }
2276
2277 static void nvme_destroy_subsystem(struct kref *ref)
2278 {
2279         struct nvme_subsystem *subsys =
2280                         container_of(ref, struct nvme_subsystem, ref);
2281
2282         mutex_lock(&nvme_subsystems_lock);
2283         list_del(&subsys->entry);
2284         mutex_unlock(&nvme_subsystems_lock);
2285
2286         ida_destroy(&subsys->ns_ida);
2287         device_del(&subsys->dev);
2288         put_device(&subsys->dev);
2289 }
2290
2291 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2292 {
2293         kref_put(&subsys->ref, nvme_destroy_subsystem);
2294 }
2295
2296 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2297 {
2298         struct nvme_subsystem *subsys;
2299
2300         lockdep_assert_held(&nvme_subsystems_lock);
2301
2302         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2303                 if (strcmp(subsys->subnqn, subsysnqn))
2304                         continue;
2305                 if (!kref_get_unless_zero(&subsys->ref))
2306                         continue;
2307                 return subsys;
2308         }
2309
2310         return NULL;
2311 }
2312
2313 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2314         struct device_attribute subsys_attr_##_name = \
2315                 __ATTR(_name, _mode, _show, NULL)
2316
2317 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2318                                     struct device_attribute *attr,
2319                                     char *buf)
2320 {
2321         struct nvme_subsystem *subsys =
2322                 container_of(dev, struct nvme_subsystem, dev);
2323
2324         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2325 }
2326 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2327
2328 #define nvme_subsys_show_str_function(field)                            \
2329 static ssize_t subsys_##field##_show(struct device *dev,                \
2330                             struct device_attribute *attr, char *buf)   \
2331 {                                                                       \
2332         struct nvme_subsystem *subsys =                                 \
2333                 container_of(dev, struct nvme_subsystem, dev);          \
2334         return sprintf(buf, "%.*s\n",                                   \
2335                        (int)sizeof(subsys->field), subsys->field);      \
2336 }                                                                       \
2337 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2338
2339 nvme_subsys_show_str_function(model);
2340 nvme_subsys_show_str_function(serial);
2341 nvme_subsys_show_str_function(firmware_rev);
2342
2343 static struct attribute *nvme_subsys_attrs[] = {
2344         &subsys_attr_model.attr,
2345         &subsys_attr_serial.attr,
2346         &subsys_attr_firmware_rev.attr,
2347         &subsys_attr_subsysnqn.attr,
2348 #ifdef CONFIG_NVME_MULTIPATH
2349         &subsys_attr_iopolicy.attr,
2350 #endif
2351         NULL,
2352 };
2353
2354 static struct attribute_group nvme_subsys_attrs_group = {
2355         .attrs = nvme_subsys_attrs,
2356 };
2357
2358 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2359         &nvme_subsys_attrs_group,
2360         NULL,
2361 };
2362
2363 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2364                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2365 {
2366         struct nvme_ctrl *tmp;
2367
2368         lockdep_assert_held(&nvme_subsystems_lock);
2369
2370         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2371                 if (ctrl->state == NVME_CTRL_DELETING ||
2372                     ctrl->state == NVME_CTRL_DEAD)
2373                         continue;
2374
2375                 if (tmp->cntlid == ctrl->cntlid) {
2376                         dev_err(ctrl->device,
2377                                 "Duplicate cntlid %u with %s, rejecting\n",
2378                                 ctrl->cntlid, dev_name(tmp->device));
2379                         return false;
2380                 }
2381
2382                 if ((id->cmic & (1 << 1)) ||
2383                     (ctrl->opts && ctrl->opts->discovery_nqn))
2384                         continue;
2385
2386                 dev_err(ctrl->device,
2387                         "Subsystem does not support multiple controllers\n");
2388                 return false;
2389         }
2390
2391         return true;
2392 }
2393
2394 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2395 {
2396         struct nvme_subsystem *subsys, *found;
2397         int ret;
2398
2399         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2400         if (!subsys)
2401                 return -ENOMEM;
2402         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2403         if (ret < 0) {
2404                 kfree(subsys);
2405                 return ret;
2406         }
2407         subsys->instance = ret;
2408         mutex_init(&subsys->lock);
2409         kref_init(&subsys->ref);
2410         INIT_LIST_HEAD(&subsys->ctrls);
2411         INIT_LIST_HEAD(&subsys->nsheads);
2412         nvme_init_subnqn(subsys, ctrl, id);
2413         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2414         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2415         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2416         subsys->vendor_id = le16_to_cpu(id->vid);
2417         subsys->cmic = id->cmic;
2418 #ifdef CONFIG_NVME_MULTIPATH
2419         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2420 #endif
2421
2422         subsys->dev.class = nvme_subsys_class;
2423         subsys->dev.release = nvme_release_subsystem;
2424         subsys->dev.groups = nvme_subsys_attrs_groups;
2425         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2426         device_initialize(&subsys->dev);
2427
2428         mutex_lock(&nvme_subsystems_lock);
2429         found = __nvme_find_get_subsystem(subsys->subnqn);
2430         if (found) {
2431                 __nvme_release_subsystem(subsys);
2432                 subsys = found;
2433
2434                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2435                         ret = -EINVAL;
2436                         goto out_put_subsystem;
2437                 }
2438         } else {
2439                 ret = device_add(&subsys->dev);
2440                 if (ret) {
2441                         dev_err(ctrl->device,
2442                                 "failed to register subsystem device.\n");
2443                         goto out_unlock;
2444                 }
2445                 ida_init(&subsys->ns_ida);
2446                 list_add_tail(&subsys->entry, &nvme_subsystems);
2447         }
2448
2449         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2450                         dev_name(ctrl->device))) {
2451                 dev_err(ctrl->device,
2452                         "failed to create sysfs link from subsystem.\n");
2453                 goto out_put_subsystem;
2454         }
2455
2456         ctrl->subsys = subsys;
2457         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2458         mutex_unlock(&nvme_subsystems_lock);
2459         return 0;
2460
2461 out_put_subsystem:
2462         nvme_put_subsystem(subsys);
2463 out_unlock:
2464         mutex_unlock(&nvme_subsystems_lock);
2465         put_device(&subsys->dev);
2466         return ret;
2467 }
2468
2469 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2470                 void *log, size_t size, u64 offset)
2471 {
2472         struct nvme_command c = { };
2473         unsigned long dwlen = size / 4 - 1;
2474
2475         c.get_log_page.opcode = nvme_admin_get_log_page;
2476         c.get_log_page.nsid = cpu_to_le32(nsid);
2477         c.get_log_page.lid = log_page;
2478         c.get_log_page.lsp = lsp;
2479         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2480         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2481         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2482         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2483
2484         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2485 }
2486
2487 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2488 {
2489         int ret;
2490
2491         if (!ctrl->effects)
2492                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2493
2494         if (!ctrl->effects)
2495                 return 0;
2496
2497         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2498                         ctrl->effects, sizeof(*ctrl->effects), 0);
2499         if (ret) {
2500                 kfree(ctrl->effects);
2501                 ctrl->effects = NULL;
2502         }
2503         return ret;
2504 }
2505
2506 /*
2507  * Initialize the cached copies of the Identify data and various controller
2508  * register in our nvme_ctrl structure.  This should be called as soon as
2509  * the admin queue is fully up and running.
2510  */
2511 int nvme_init_identify(struct nvme_ctrl *ctrl)
2512 {
2513         struct nvme_id_ctrl *id;
2514         u64 cap;
2515         int ret, page_shift;
2516         u32 max_hw_sectors;
2517         bool prev_apst_enabled;
2518
2519         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2520         if (ret) {
2521                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2522                 return ret;
2523         }
2524
2525         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2526         if (ret) {
2527                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2528                 return ret;
2529         }
2530         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2531
2532         if (ctrl->vs >= NVME_VS(1, 1, 0))
2533                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2534
2535         ret = nvme_identify_ctrl(ctrl, &id);
2536         if (ret) {
2537                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2538                 return -EIO;
2539         }
2540
2541         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2542                 ret = nvme_get_effects_log(ctrl);
2543                 if (ret < 0)
2544                         goto out_free;
2545         }
2546
2547         if (!ctrl->identified) {
2548                 int i;
2549
2550                 ret = nvme_init_subsystem(ctrl, id);
2551                 if (ret)
2552                         goto out_free;
2553
2554                 /*
2555                  * Check for quirks.  Quirk can depend on firmware version,
2556                  * so, in principle, the set of quirks present can change
2557                  * across a reset.  As a possible future enhancement, we
2558                  * could re-scan for quirks every time we reinitialize
2559                  * the device, but we'd have to make sure that the driver
2560                  * behaves intelligently if the quirks change.
2561                  */
2562                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2563                         if (quirk_matches(id, &core_quirks[i]))
2564                                 ctrl->quirks |= core_quirks[i].quirks;
2565                 }
2566         }
2567
2568         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2569                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2570                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2571         }
2572
2573         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2574         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2575         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2576
2577         ctrl->oacs = le16_to_cpu(id->oacs);
2578         ctrl->oncs = le16_to_cpu(id->oncs);
2579         ctrl->mtfa = le16_to_cpu(id->mtfa);
2580         ctrl->oaes = le32_to_cpu(id->oaes);
2581         atomic_set(&ctrl->abort_limit, id->acl + 1);
2582         ctrl->vwc = id->vwc;
2583         if (id->mdts)
2584                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2585         else
2586                 max_hw_sectors = UINT_MAX;
2587         ctrl->max_hw_sectors =
2588                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2589
2590         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2591         ctrl->sgls = le32_to_cpu(id->sgls);
2592         ctrl->kas = le16_to_cpu(id->kas);
2593         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2594         ctrl->ctratt = le32_to_cpu(id->ctratt);
2595
2596         if (id->rtd3e) {
2597                 /* us -> s */
2598                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2599
2600                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2601                                                  shutdown_timeout, 60);
2602
2603                 if (ctrl->shutdown_timeout != shutdown_timeout)
2604                         dev_info(ctrl->device,
2605                                  "Shutdown timeout set to %u seconds\n",
2606                                  ctrl->shutdown_timeout);
2607         } else
2608                 ctrl->shutdown_timeout = shutdown_timeout;
2609
2610         ctrl->npss = id->npss;
2611         ctrl->apsta = id->apsta;
2612         prev_apst_enabled = ctrl->apst_enabled;
2613         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2614                 if (force_apst && id->apsta) {
2615                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2616                         ctrl->apst_enabled = true;
2617                 } else {
2618                         ctrl->apst_enabled = false;
2619                 }
2620         } else {
2621                 ctrl->apst_enabled = id->apsta;
2622         }
2623         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2624
2625         if (ctrl->ops->flags & NVME_F_FABRICS) {
2626                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2627                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2628                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2629                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2630
2631                 /*
2632                  * In fabrics we need to verify the cntlid matches the
2633                  * admin connect
2634                  */
2635                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2636                         ret = -EINVAL;
2637                         goto out_free;
2638                 }
2639
2640                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2641                         dev_err(ctrl->device,
2642                                 "keep-alive support is mandatory for fabrics\n");
2643                         ret = -EINVAL;
2644                         goto out_free;
2645                 }
2646         } else {
2647                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2648                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2649                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2650                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2651                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2652         }
2653
2654         ret = nvme_mpath_init(ctrl, id);
2655         kfree(id);
2656
2657         if (ret < 0)
2658                 return ret;
2659
2660         if (ctrl->apst_enabled && !prev_apst_enabled)
2661                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2662         else if (!ctrl->apst_enabled && prev_apst_enabled)
2663                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2664
2665         ret = nvme_configure_apst(ctrl);
2666         if (ret < 0)
2667                 return ret;
2668
2669         ret = nvme_configure_timestamp(ctrl);
2670         if (ret < 0)
2671                 return ret;
2672
2673         ret = nvme_configure_directives(ctrl);
2674         if (ret < 0)
2675                 return ret;
2676
2677         ret = nvme_configure_acre(ctrl);
2678         if (ret < 0)
2679                 return ret;
2680
2681         ctrl->identified = true;
2682
2683         return 0;
2684
2685 out_free:
2686         kfree(id);
2687         return ret;
2688 }
2689 EXPORT_SYMBOL_GPL(nvme_init_identify);
2690
2691 static int nvme_dev_open(struct inode *inode, struct file *file)
2692 {
2693         struct nvme_ctrl *ctrl =
2694                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2695
2696         switch (ctrl->state) {
2697         case NVME_CTRL_LIVE:
2698         case NVME_CTRL_ADMIN_ONLY:
2699                 break;
2700         default:
2701                 return -EWOULDBLOCK;
2702         }
2703
2704         file->private_data = ctrl;
2705         return 0;
2706 }
2707
2708 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2709 {
2710         struct nvme_ns *ns;
2711         int ret;
2712
2713         down_read(&ctrl->namespaces_rwsem);
2714         if (list_empty(&ctrl->namespaces)) {
2715                 ret = -ENOTTY;
2716                 goto out_unlock;
2717         }
2718
2719         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2720         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2721                 dev_warn(ctrl->device,
2722                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2723                 ret = -EINVAL;
2724                 goto out_unlock;
2725         }
2726
2727         dev_warn(ctrl->device,
2728                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2729         kref_get(&ns->kref);
2730         up_read(&ctrl->namespaces_rwsem);
2731
2732         ret = nvme_user_cmd(ctrl, ns, argp);
2733         nvme_put_ns(ns);
2734         return ret;
2735
2736 out_unlock:
2737         up_read(&ctrl->namespaces_rwsem);
2738         return ret;
2739 }
2740
2741 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2742                 unsigned long arg)
2743 {
2744         struct nvme_ctrl *ctrl = file->private_data;
2745         void __user *argp = (void __user *)arg;
2746
2747         switch (cmd) {
2748         case NVME_IOCTL_ADMIN_CMD:
2749                 return nvme_user_cmd(ctrl, NULL, argp);
2750         case NVME_IOCTL_IO_CMD:
2751                 return nvme_dev_user_cmd(ctrl, argp);
2752         case NVME_IOCTL_RESET:
2753                 dev_warn(ctrl->device, "resetting controller\n");
2754                 return nvme_reset_ctrl_sync(ctrl);
2755         case NVME_IOCTL_SUBSYS_RESET:
2756                 return nvme_reset_subsystem(ctrl);
2757         case NVME_IOCTL_RESCAN:
2758                 nvme_queue_scan(ctrl);
2759                 return 0;
2760         default:
2761                 return -ENOTTY;
2762         }
2763 }
2764
2765 static const struct file_operations nvme_dev_fops = {
2766         .owner          = THIS_MODULE,
2767         .open           = nvme_dev_open,
2768         .unlocked_ioctl = nvme_dev_ioctl,
2769         .compat_ioctl   = nvme_dev_ioctl,
2770 };
2771
2772 static ssize_t nvme_sysfs_reset(struct device *dev,
2773                                 struct device_attribute *attr, const char *buf,
2774                                 size_t count)
2775 {
2776         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2777         int ret;
2778
2779         ret = nvme_reset_ctrl_sync(ctrl);
2780         if (ret < 0)
2781                 return ret;
2782         return count;
2783 }
2784 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2785
2786 static ssize_t nvme_sysfs_rescan(struct device *dev,
2787                                 struct device_attribute *attr, const char *buf,
2788                                 size_t count)
2789 {
2790         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2791
2792         nvme_queue_scan(ctrl);
2793         return count;
2794 }
2795 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2796
2797 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2798 {
2799         struct gendisk *disk = dev_to_disk(dev);
2800
2801         if (disk->fops == &nvme_fops)
2802                 return nvme_get_ns_from_dev(dev)->head;
2803         else
2804                 return disk->private_data;
2805 }
2806
2807 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2808                 char *buf)
2809 {
2810         struct nvme_ns_head *head = dev_to_ns_head(dev);
2811         struct nvme_ns_ids *ids = &head->ids;
2812         struct nvme_subsystem *subsys = head->subsys;
2813         int serial_len = sizeof(subsys->serial);
2814         int model_len = sizeof(subsys->model);
2815
2816         if (!uuid_is_null(&ids->uuid))
2817                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2818
2819         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2820                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2821
2822         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2823                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2824
2825         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2826                                   subsys->serial[serial_len - 1] == '\0'))
2827                 serial_len--;
2828         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2829                                  subsys->model[model_len - 1] == '\0'))
2830                 model_len--;
2831
2832         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2833                 serial_len, subsys->serial, model_len, subsys->model,
2834                 head->ns_id);
2835 }
2836 static DEVICE_ATTR_RO(wwid);
2837
2838 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2839                 char *buf)
2840 {
2841         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2842 }
2843 static DEVICE_ATTR_RO(nguid);
2844
2845 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2846                 char *buf)
2847 {
2848         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2849
2850         /* For backward compatibility expose the NGUID to userspace if
2851          * we have no UUID set
2852          */
2853         if (uuid_is_null(&ids->uuid)) {
2854                 printk_ratelimited(KERN_WARNING
2855                                    "No UUID available providing old NGUID\n");
2856                 return sprintf(buf, "%pU\n", ids->nguid);
2857         }
2858         return sprintf(buf, "%pU\n", &ids->uuid);
2859 }
2860 static DEVICE_ATTR_RO(uuid);
2861
2862 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2863                 char *buf)
2864 {
2865         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2866 }
2867 static DEVICE_ATTR_RO(eui);
2868
2869 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2870                 char *buf)
2871 {
2872         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2873 }
2874 static DEVICE_ATTR_RO(nsid);
2875
2876 static struct attribute *nvme_ns_id_attrs[] = {
2877         &dev_attr_wwid.attr,
2878         &dev_attr_uuid.attr,
2879         &dev_attr_nguid.attr,
2880         &dev_attr_eui.attr,
2881         &dev_attr_nsid.attr,
2882 #ifdef CONFIG_NVME_MULTIPATH
2883         &dev_attr_ana_grpid.attr,
2884         &dev_attr_ana_state.attr,
2885 #endif
2886         NULL,
2887 };
2888
2889 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2890                 struct attribute *a, int n)
2891 {
2892         struct device *dev = container_of(kobj, struct device, kobj);
2893         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2894
2895         if (a == &dev_attr_uuid.attr) {
2896                 if (uuid_is_null(&ids->uuid) &&
2897                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2898                         return 0;
2899         }
2900         if (a == &dev_attr_nguid.attr) {
2901                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2902                         return 0;
2903         }
2904         if (a == &dev_attr_eui.attr) {
2905                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2906                         return 0;
2907         }
2908 #ifdef CONFIG_NVME_MULTIPATH
2909         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2910                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2911                         return 0;
2912                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2913                         return 0;
2914         }
2915 #endif
2916         return a->mode;
2917 }
2918
2919 static const struct attribute_group nvme_ns_id_attr_group = {
2920         .attrs          = nvme_ns_id_attrs,
2921         .is_visible     = nvme_ns_id_attrs_are_visible,
2922 };
2923
2924 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2925         &nvme_ns_id_attr_group,
2926 #ifdef CONFIG_NVM
2927         &nvme_nvm_attr_group,
2928 #endif
2929         NULL,
2930 };
2931
2932 #define nvme_show_str_function(field)                                           \
2933 static ssize_t  field##_show(struct device *dev,                                \
2934                             struct device_attribute *attr, char *buf)           \
2935 {                                                                               \
2936         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2937         return sprintf(buf, "%.*s\n",                                           \
2938                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2939 }                                                                               \
2940 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2941
2942 nvme_show_str_function(model);
2943 nvme_show_str_function(serial);
2944 nvme_show_str_function(firmware_rev);
2945
2946 #define nvme_show_int_function(field)                                           \
2947 static ssize_t  field##_show(struct device *dev,                                \
2948                             struct device_attribute *attr, char *buf)           \
2949 {                                                                               \
2950         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2951         return sprintf(buf, "%d\n", ctrl->field);       \
2952 }                                                                               \
2953 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2954
2955 nvme_show_int_function(cntlid);
2956 nvme_show_int_function(numa_node);
2957
2958 static ssize_t nvme_sysfs_delete(struct device *dev,
2959                                 struct device_attribute *attr, const char *buf,
2960                                 size_t count)
2961 {
2962         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2963
2964         if (device_remove_file_self(dev, attr))
2965                 nvme_delete_ctrl_sync(ctrl);
2966         return count;
2967 }
2968 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2969
2970 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2971                                          struct device_attribute *attr,
2972                                          char *buf)
2973 {
2974         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2975
2976         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2977 }
2978 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2979
2980 static ssize_t nvme_sysfs_show_state(struct device *dev,
2981                                      struct device_attribute *attr,
2982                                      char *buf)
2983 {
2984         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2985         static const char *const state_name[] = {
2986                 [NVME_CTRL_NEW]         = "new",
2987                 [NVME_CTRL_LIVE]        = "live",
2988                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2989                 [NVME_CTRL_RESETTING]   = "resetting",
2990                 [NVME_CTRL_CONNECTING]  = "connecting",
2991                 [NVME_CTRL_DELETING]    = "deleting",
2992                 [NVME_CTRL_DEAD]        = "dead",
2993         };
2994
2995         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2996             state_name[ctrl->state])
2997                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2998
2999         return sprintf(buf, "unknown state\n");
3000 }
3001
3002 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3003
3004 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3005                                          struct device_attribute *attr,
3006                                          char *buf)
3007 {
3008         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3009
3010         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3011 }
3012 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3013
3014 static ssize_t nvme_sysfs_show_address(struct device *dev,
3015                                          struct device_attribute *attr,
3016                                          char *buf)
3017 {
3018         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3019
3020         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3021 }
3022 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3023
3024 static struct attribute *nvme_dev_attrs[] = {
3025         &dev_attr_reset_controller.attr,
3026         &dev_attr_rescan_controller.attr,
3027         &dev_attr_model.attr,
3028         &dev_attr_serial.attr,
3029         &dev_attr_firmware_rev.attr,
3030         &dev_attr_cntlid.attr,
3031         &dev_attr_delete_controller.attr,
3032         &dev_attr_transport.attr,
3033         &dev_attr_subsysnqn.attr,
3034         &dev_attr_address.attr,
3035         &dev_attr_state.attr,
3036         &dev_attr_numa_node.attr,
3037         NULL
3038 };
3039
3040 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3041                 struct attribute *a, int n)
3042 {
3043         struct device *dev = container_of(kobj, struct device, kobj);
3044         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3045
3046         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3047                 return 0;
3048         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3049                 return 0;
3050
3051         return a->mode;
3052 }
3053
3054 static struct attribute_group nvme_dev_attrs_group = {
3055         .attrs          = nvme_dev_attrs,
3056         .is_visible     = nvme_dev_attrs_are_visible,
3057 };
3058
3059 static const struct attribute_group *nvme_dev_attr_groups[] = {
3060         &nvme_dev_attrs_group,
3061         NULL,
3062 };
3063
3064 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3065                 unsigned nsid)
3066 {
3067         struct nvme_ns_head *h;
3068
3069         lockdep_assert_held(&subsys->lock);
3070
3071         list_for_each_entry(h, &subsys->nsheads, entry) {
3072                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3073                         return h;
3074         }
3075
3076         return NULL;
3077 }
3078
3079 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3080                 struct nvme_ns_head *new)
3081 {
3082         struct nvme_ns_head *h;
3083
3084         lockdep_assert_held(&subsys->lock);
3085
3086         list_for_each_entry(h, &subsys->nsheads, entry) {
3087                 if (nvme_ns_ids_valid(&new->ids) &&
3088                     !list_empty(&h->list) &&
3089                     nvme_ns_ids_equal(&new->ids, &h->ids))
3090                         return -EINVAL;
3091         }
3092
3093         return 0;
3094 }
3095
3096 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3097                 unsigned nsid, struct nvme_id_ns *id)
3098 {
3099         struct nvme_ns_head *head;
3100         size_t size = sizeof(*head);
3101         int ret = -ENOMEM;
3102
3103 #ifdef CONFIG_NVME_MULTIPATH
3104         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3105 #endif
3106
3107         head = kzalloc(size, GFP_KERNEL);
3108         if (!head)
3109                 goto out;
3110         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3111         if (ret < 0)
3112                 goto out_free_head;
3113         head->instance = ret;
3114         INIT_LIST_HEAD(&head->list);
3115         ret = init_srcu_struct(&head->srcu);
3116         if (ret)
3117                 goto out_ida_remove;
3118         head->subsys = ctrl->subsys;
3119         head->ns_id = nsid;
3120         kref_init(&head->ref);
3121
3122         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3123
3124         ret = __nvme_check_ids(ctrl->subsys, head);
3125         if (ret) {
3126                 dev_err(ctrl->device,
3127                         "duplicate IDs for nsid %d\n", nsid);
3128                 goto out_cleanup_srcu;
3129         }
3130
3131         ret = nvme_mpath_alloc_disk(ctrl, head);
3132         if (ret)
3133                 goto out_cleanup_srcu;
3134
3135         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3136
3137         kref_get(&ctrl->subsys->ref);
3138
3139         return head;
3140 out_cleanup_srcu:
3141         cleanup_srcu_struct(&head->srcu);
3142 out_ida_remove:
3143         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3144 out_free_head:
3145         kfree(head);
3146 out:
3147         return ERR_PTR(ret);
3148 }
3149
3150 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3151                 struct nvme_id_ns *id)
3152 {
3153         struct nvme_ctrl *ctrl = ns->ctrl;
3154         bool is_shared = id->nmic & (1 << 0);
3155         struct nvme_ns_head *head = NULL;
3156         int ret = 0;
3157
3158         mutex_lock(&ctrl->subsys->lock);
3159         if (is_shared)
3160                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3161         if (!head) {
3162                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3163                 if (IS_ERR(head)) {
3164                         ret = PTR_ERR(head);
3165                         goto out_unlock;
3166                 }
3167         } else {
3168                 struct nvme_ns_ids ids;
3169
3170                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3171                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3172                         dev_err(ctrl->device,
3173                                 "IDs don't match for shared namespace %d\n",
3174                                         nsid);
3175                         ret = -EINVAL;
3176                         goto out_unlock;
3177                 }
3178         }
3179
3180         list_add_tail(&ns->siblings, &head->list);
3181         ns->head = head;
3182
3183 out_unlock:
3184         mutex_unlock(&ctrl->subsys->lock);
3185         return ret;
3186 }
3187
3188 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3189 {
3190         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3191         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3192
3193         return nsa->head->ns_id - nsb->head->ns_id;
3194 }
3195
3196 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3197 {
3198         struct nvme_ns *ns, *ret = NULL;
3199
3200         down_read(&ctrl->namespaces_rwsem);
3201         list_for_each_entry(ns, &ctrl->namespaces, list) {
3202                 if (ns->head->ns_id == nsid) {
3203                         if (!kref_get_unless_zero(&ns->kref))
3204                                 continue;
3205                         ret = ns;
3206                         break;
3207                 }
3208                 if (ns->head->ns_id > nsid)
3209                         break;
3210         }
3211         up_read(&ctrl->namespaces_rwsem);
3212         return ret;
3213 }
3214
3215 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3216 {
3217         struct streams_directive_params s;
3218         int ret;
3219
3220         if (!ctrl->nr_streams)
3221                 return 0;
3222
3223         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3224         if (ret)
3225                 return ret;
3226
3227         ns->sws = le32_to_cpu(s.sws);
3228         ns->sgs = le16_to_cpu(s.sgs);
3229
3230         if (ns->sws) {
3231                 unsigned int bs = 1 << ns->lba_shift;
3232
3233                 blk_queue_io_min(ns->queue, bs * ns->sws);
3234                 if (ns->sgs)
3235                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3236         }
3237
3238         return 0;
3239 }
3240
3241 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3242 {
3243         struct nvme_ns *ns;
3244         struct gendisk *disk;
3245         struct nvme_id_ns *id;
3246         char disk_name[DISK_NAME_LEN];
3247         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3248
3249         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3250         if (!ns)
3251                 return -ENOMEM;
3252
3253         ns->queue = blk_mq_init_queue(ctrl->tagset);
3254         if (IS_ERR(ns->queue)) {
3255                 ret = PTR_ERR(ns->queue);
3256                 goto out_free_ns;
3257         }
3258
3259         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3260         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3261                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3262
3263         ns->queue->queuedata = ns;
3264         ns->ctrl = ctrl;
3265
3266         kref_init(&ns->kref);
3267         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3268
3269         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3270         nvme_set_queue_limits(ctrl, ns->queue);
3271
3272         id = nvme_identify_ns(ctrl, nsid);
3273         if (!id) {
3274                 ret = -EIO;
3275                 goto out_free_queue;
3276         }
3277
3278         if (id->ncap == 0) {
3279                 ret = -EINVAL;
3280                 goto out_free_id;
3281         }
3282
3283         ret = nvme_init_ns_head(ns, nsid, id);
3284         if (ret)
3285                 goto out_free_id;
3286         nvme_setup_streams_ns(ctrl, ns);
3287         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3288
3289         disk = alloc_disk_node(0, node);
3290         if (!disk) {
3291                 ret = -ENOMEM;
3292                 goto out_unlink_ns;
3293         }
3294
3295         disk->fops = &nvme_fops;
3296         disk->private_data = ns;
3297         disk->queue = ns->queue;
3298         disk->flags = flags;
3299         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3300         ns->disk = disk;
3301
3302         __nvme_revalidate_disk(disk, id);
3303
3304         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3305                 ret = nvme_nvm_register(ns, disk_name, node);
3306                 if (ret) {
3307                         dev_warn(ctrl->device, "LightNVM init failure\n");
3308                         goto out_put_disk;
3309                 }
3310         }
3311
3312         down_write(&ctrl->namespaces_rwsem);
3313         list_add_tail(&ns->list, &ctrl->namespaces);
3314         up_write(&ctrl->namespaces_rwsem);
3315
3316         nvme_get_ctrl(ctrl);
3317
3318         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3319
3320         nvme_mpath_add_disk(ns, id);
3321         nvme_fault_inject_init(ns);
3322         kfree(id);
3323
3324         return 0;
3325  out_put_disk:
3326         put_disk(ns->disk);
3327  out_unlink_ns:
3328         mutex_lock(&ctrl->subsys->lock);
3329         list_del_rcu(&ns->siblings);
3330         mutex_unlock(&ctrl->subsys->lock);
3331         nvme_put_ns_head(ns->head);
3332  out_free_id:
3333         kfree(id);
3334  out_free_queue:
3335         blk_cleanup_queue(ns->queue);
3336  out_free_ns:
3337         kfree(ns);
3338         return ret;
3339 }
3340
3341 static void nvme_ns_remove(struct nvme_ns *ns)
3342 {
3343         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3344                 return;
3345
3346         nvme_fault_inject_fini(ns);
3347         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3348                 del_gendisk(ns->disk);
3349                 blk_cleanup_queue(ns->queue);
3350                 if (blk_get_integrity(ns->disk))
3351                         blk_integrity_unregister(ns->disk);
3352         }
3353
3354         mutex_lock(&ns->ctrl->subsys->lock);
3355         list_del_rcu(&ns->siblings);
3356         nvme_mpath_clear_current_path(ns);
3357         mutex_unlock(&ns->ctrl->subsys->lock);
3358
3359         down_write(&ns->ctrl->namespaces_rwsem);
3360         list_del_init(&ns->list);
3361         up_write(&ns->ctrl->namespaces_rwsem);
3362
3363         synchronize_srcu(&ns->head->srcu);
3364         nvme_mpath_check_last_path(ns);
3365         nvme_put_ns(ns);
3366 }
3367
3368 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3369 {
3370         struct nvme_ns *ns;
3371
3372         ns = nvme_find_get_ns(ctrl, nsid);
3373         if (ns) {
3374                 if (ns->disk && revalidate_disk(ns->disk))
3375                         nvme_ns_remove(ns);
3376                 nvme_put_ns(ns);
3377         } else
3378                 nvme_alloc_ns(ctrl, nsid);
3379 }
3380
3381 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3382                                         unsigned nsid)
3383 {
3384         struct nvme_ns *ns, *next;
3385         LIST_HEAD(rm_list);
3386
3387         down_write(&ctrl->namespaces_rwsem);
3388         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3389                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3390                         list_move_tail(&ns->list, &rm_list);
3391         }
3392         up_write(&ctrl->namespaces_rwsem);
3393
3394         list_for_each_entry_safe(ns, next, &rm_list, list)
3395                 nvme_ns_remove(ns);
3396
3397 }
3398
3399 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3400 {
3401         struct nvme_ns *ns;
3402         __le32 *ns_list;
3403         unsigned i, j, nsid, prev = 0;
3404         unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3405         int ret = 0;
3406
3407         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3408         if (!ns_list)
3409                 return -ENOMEM;
3410
3411         for (i = 0; i < num_lists; i++) {
3412                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3413                 if (ret)
3414                         goto free;
3415
3416                 for (j = 0; j < min(nn, 1024U); j++) {
3417                         nsid = le32_to_cpu(ns_list[j]);
3418                         if (!nsid)
3419                                 goto out;
3420
3421                         nvme_validate_ns(ctrl, nsid);
3422
3423                         while (++prev < nsid) {
3424                                 ns = nvme_find_get_ns(ctrl, prev);
3425                                 if (ns) {
3426                                         nvme_ns_remove(ns);
3427                                         nvme_put_ns(ns);
3428                                 }
3429                         }
3430                 }
3431                 nn -= j;
3432         }
3433  out:
3434         nvme_remove_invalid_namespaces(ctrl, prev);
3435  free:
3436         kfree(ns_list);
3437         return ret;
3438 }
3439
3440 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3441 {
3442         unsigned i;
3443
3444         for (i = 1; i <= nn; i++)
3445                 nvme_validate_ns(ctrl, i);
3446
3447         nvme_remove_invalid_namespaces(ctrl, nn);
3448 }
3449
3450 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3451 {
3452         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3453         __le32 *log;
3454         int error;
3455
3456         log = kzalloc(log_size, GFP_KERNEL);
3457         if (!log)
3458                 return;
3459
3460         /*
3461          * We need to read the log to clear the AEN, but we don't want to rely
3462          * on it for the changed namespace information as userspace could have
3463          * raced with us in reading the log page, which could cause us to miss
3464          * updates.
3465          */
3466         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3467                         log_size, 0);
3468         if (error)
3469                 dev_warn(ctrl->device,
3470                         "reading changed ns log failed: %d\n", error);
3471
3472         kfree(log);
3473 }
3474
3475 static void nvme_scan_work(struct work_struct *work)
3476 {
3477         struct nvme_ctrl *ctrl =
3478                 container_of(work, struct nvme_ctrl, scan_work);
3479         struct nvme_id_ctrl *id;
3480         unsigned nn;
3481
3482         if (ctrl->state != NVME_CTRL_LIVE)
3483                 return;
3484
3485         WARN_ON_ONCE(!ctrl->tagset);
3486
3487         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3488                 dev_info(ctrl->device, "rescanning namespaces.\n");
3489                 nvme_clear_changed_ns_log(ctrl);
3490         }
3491
3492         if (nvme_identify_ctrl(ctrl, &id))
3493                 return;
3494
3495         mutex_lock(&ctrl->scan_lock);
3496         nn = le32_to_cpu(id->nn);
3497         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3498             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3499                 if (!nvme_scan_ns_list(ctrl, nn))
3500                         goto out_free_id;
3501         }
3502         nvme_scan_ns_sequential(ctrl, nn);
3503 out_free_id:
3504         mutex_unlock(&ctrl->scan_lock);
3505         kfree(id);
3506         down_write(&ctrl->namespaces_rwsem);
3507         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3508         up_write(&ctrl->namespaces_rwsem);
3509 }
3510
3511 /*
3512  * This function iterates the namespace list unlocked to allow recovery from
3513  * controller failure. It is up to the caller to ensure the namespace list is
3514  * not modified by scan work while this function is executing.
3515  */
3516 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3517 {
3518         struct nvme_ns *ns, *next;
3519         LIST_HEAD(ns_list);
3520
3521         /* prevent racing with ns scanning */
3522         flush_work(&ctrl->scan_work);
3523
3524         /*
3525          * The dead states indicates the controller was not gracefully
3526          * disconnected. In that case, we won't be able to flush any data while
3527          * removing the namespaces' disks; fail all the queues now to avoid
3528          * potentially having to clean up the failed sync later.
3529          */
3530         if (ctrl->state == NVME_CTRL_DEAD)
3531                 nvme_kill_queues(ctrl);
3532
3533         down_write(&ctrl->namespaces_rwsem);
3534         list_splice_init(&ctrl->namespaces, &ns_list);
3535         up_write(&ctrl->namespaces_rwsem);
3536
3537         list_for_each_entry_safe(ns, next, &ns_list, list)
3538                 nvme_ns_remove(ns);
3539 }
3540 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3541
3542 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3543 {
3544         char *envp[2] = { NULL, NULL };
3545         u32 aen_result = ctrl->aen_result;
3546
3547         ctrl->aen_result = 0;
3548         if (!aen_result)
3549                 return;
3550
3551         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3552         if (!envp[0])
3553                 return;
3554         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3555         kfree(envp[0]);
3556 }
3557
3558 static void nvme_async_event_work(struct work_struct *work)
3559 {
3560         struct nvme_ctrl *ctrl =
3561                 container_of(work, struct nvme_ctrl, async_event_work);
3562
3563         nvme_aen_uevent(ctrl);
3564         ctrl->ops->submit_async_event(ctrl);
3565 }
3566
3567 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3568 {
3569
3570         u32 csts;
3571
3572         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3573                 return false;
3574
3575         if (csts == ~0)
3576                 return false;
3577
3578         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3579 }
3580
3581 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3582 {
3583         struct nvme_fw_slot_info_log *log;
3584
3585         log = kmalloc(sizeof(*log), GFP_KERNEL);
3586         if (!log)
3587                 return;
3588
3589         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3590                         sizeof(*log), 0))
3591                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3592         kfree(log);
3593 }
3594
3595 static void nvme_fw_act_work(struct work_struct *work)
3596 {
3597         struct nvme_ctrl *ctrl = container_of(work,
3598                                 struct nvme_ctrl, fw_act_work);
3599         unsigned long fw_act_timeout;
3600
3601         if (ctrl->mtfa)
3602                 fw_act_timeout = jiffies +
3603                                 msecs_to_jiffies(ctrl->mtfa * 100);
3604         else
3605                 fw_act_timeout = jiffies +
3606                                 msecs_to_jiffies(admin_timeout * 1000);
3607
3608         nvme_stop_queues(ctrl);
3609         while (nvme_ctrl_pp_status(ctrl)) {
3610                 if (time_after(jiffies, fw_act_timeout)) {
3611                         dev_warn(ctrl->device,
3612                                 "Fw activation timeout, reset controller\n");
3613                         nvme_reset_ctrl(ctrl);
3614                         break;
3615                 }
3616                 msleep(100);
3617         }
3618
3619         if (ctrl->state != NVME_CTRL_LIVE)
3620                 return;
3621
3622         nvme_start_queues(ctrl);
3623         /* read FW slot information to clear the AER */
3624         nvme_get_fw_slot_info(ctrl);
3625 }
3626
3627 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3628 {
3629         u32 aer_notice_type = (result & 0xff00) >> 8;
3630
3631         trace_nvme_async_event(ctrl, aer_notice_type);
3632
3633         switch (aer_notice_type) {
3634         case NVME_AER_NOTICE_NS_CHANGED:
3635                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3636                 nvme_queue_scan(ctrl);
3637                 break;
3638         case NVME_AER_NOTICE_FW_ACT_STARTING:
3639                 queue_work(nvme_wq, &ctrl->fw_act_work);
3640                 break;
3641 #ifdef CONFIG_NVME_MULTIPATH
3642         case NVME_AER_NOTICE_ANA:
3643                 if (!ctrl->ana_log_buf)
3644                         break;
3645                 queue_work(nvme_wq, &ctrl->ana_work);
3646                 break;
3647 #endif
3648         default:
3649                 dev_warn(ctrl->device, "async event result %08x\n", result);
3650         }
3651 }
3652
3653 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3654                 volatile union nvme_result *res)
3655 {
3656         u32 result = le32_to_cpu(res->u32);
3657         u32 aer_type = result & 0x07;
3658
3659         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3660                 return;
3661
3662         switch (aer_type) {
3663         case NVME_AER_NOTICE:
3664                 nvme_handle_aen_notice(ctrl, result);
3665                 break;
3666         case NVME_AER_ERROR:
3667         case NVME_AER_SMART:
3668         case NVME_AER_CSS:
3669         case NVME_AER_VS:
3670                 trace_nvme_async_event(ctrl, aer_type);
3671                 ctrl->aen_result = result;
3672                 break;
3673         default:
3674                 break;
3675         }
3676         queue_work(nvme_wq, &ctrl->async_event_work);
3677 }
3678 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3679
3680 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3681 {
3682         nvme_mpath_stop(ctrl);
3683         nvme_stop_keep_alive(ctrl);
3684         flush_work(&ctrl->async_event_work);
3685         cancel_work_sync(&ctrl->fw_act_work);
3686 }
3687 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3688
3689 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3690 {
3691         if (ctrl->kato)
3692                 nvme_start_keep_alive(ctrl);
3693
3694         if (ctrl->queue_count > 1) {
3695                 nvme_queue_scan(ctrl);
3696                 nvme_enable_aen(ctrl);
3697                 queue_work(nvme_wq, &ctrl->async_event_work);
3698                 nvme_start_queues(ctrl);
3699         }
3700 }
3701 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3702
3703 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3704 {
3705         dev_pm_qos_hide_latency_tolerance(ctrl->device);
3706         cdev_device_del(&ctrl->cdev, ctrl->device);
3707 }
3708 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3709
3710 static void nvme_free_ctrl(struct device *dev)
3711 {
3712         struct nvme_ctrl *ctrl =
3713                 container_of(dev, struct nvme_ctrl, ctrl_device);
3714         struct nvme_subsystem *subsys = ctrl->subsys;
3715
3716         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3717         kfree(ctrl->effects);
3718         nvme_mpath_uninit(ctrl);
3719         __free_page(ctrl->discard_page);
3720
3721         if (subsys) {
3722                 mutex_lock(&nvme_subsystems_lock);
3723                 list_del(&ctrl->subsys_entry);
3724                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3725                 mutex_unlock(&nvme_subsystems_lock);
3726         }
3727
3728         ctrl->ops->free_ctrl(ctrl);
3729
3730         if (subsys)
3731                 nvme_put_subsystem(subsys);
3732 }
3733
3734 /*
3735  * Initialize a NVMe controller structures.  This needs to be called during
3736  * earliest initialization so that we have the initialized structured around
3737  * during probing.
3738  */
3739 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3740                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3741 {
3742         int ret;
3743
3744         ctrl->state = NVME_CTRL_NEW;
3745         spin_lock_init(&ctrl->lock);
3746         mutex_init(&ctrl->scan_lock);
3747         INIT_LIST_HEAD(&ctrl->namespaces);
3748         init_rwsem(&ctrl->namespaces_rwsem);
3749         ctrl->dev = dev;
3750         ctrl->ops = ops;
3751         ctrl->quirks = quirks;
3752         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3753         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3754         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3755         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3756
3757         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3758         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3759         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3760
3761         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3762                         PAGE_SIZE);
3763         ctrl->discard_page = alloc_page(GFP_KERNEL);
3764         if (!ctrl->discard_page) {
3765                 ret = -ENOMEM;
3766                 goto out;
3767         }
3768
3769         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3770         if (ret < 0)
3771                 goto out;
3772         ctrl->instance = ret;
3773
3774         device_initialize(&ctrl->ctrl_device);
3775         ctrl->device = &ctrl->ctrl_device;
3776         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3777         ctrl->device->class = nvme_class;
3778         ctrl->device->parent = ctrl->dev;
3779         ctrl->device->groups = nvme_dev_attr_groups;
3780         ctrl->device->release = nvme_free_ctrl;
3781         dev_set_drvdata(ctrl->device, ctrl);
3782         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3783         if (ret)
3784                 goto out_release_instance;
3785
3786         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3787         ctrl->cdev.owner = ops->module;
3788         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3789         if (ret)
3790                 goto out_free_name;
3791
3792         /*
3793          * Initialize latency tolerance controls.  The sysfs files won't
3794          * be visible to userspace unless the device actually supports APST.
3795          */
3796         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3797         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3798                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3799
3800         return 0;
3801 out_free_name:
3802         kfree_const(ctrl->device->kobj.name);
3803 out_release_instance:
3804         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3805 out:
3806         if (ctrl->discard_page)
3807                 __free_page(ctrl->discard_page);
3808         return ret;
3809 }
3810 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3811
3812 /**
3813  * nvme_kill_queues(): Ends all namespace queues
3814  * @ctrl: the dead controller that needs to end
3815  *
3816  * Call this function when the driver determines it is unable to get the
3817  * controller in a state capable of servicing IO.
3818  */
3819 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3820 {
3821         struct nvme_ns *ns;
3822
3823         down_read(&ctrl->namespaces_rwsem);
3824
3825         /* Forcibly unquiesce queues to avoid blocking dispatch */
3826         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3827                 blk_mq_unquiesce_queue(ctrl->admin_q);
3828
3829         list_for_each_entry(ns, &ctrl->namespaces, list)
3830                 nvme_set_queue_dying(ns);
3831
3832         up_read(&ctrl->namespaces_rwsem);
3833 }
3834 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3835
3836 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3837 {
3838         struct nvme_ns *ns;
3839
3840         down_read(&ctrl->namespaces_rwsem);
3841         list_for_each_entry(ns, &ctrl->namespaces, list)
3842                 blk_mq_unfreeze_queue(ns->queue);
3843         up_read(&ctrl->namespaces_rwsem);
3844 }
3845 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3846
3847 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3848 {
3849         struct nvme_ns *ns;
3850
3851         down_read(&ctrl->namespaces_rwsem);
3852         list_for_each_entry(ns, &ctrl->namespaces, list) {
3853                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3854                 if (timeout <= 0)
3855                         break;
3856         }
3857         up_read(&ctrl->namespaces_rwsem);
3858 }
3859 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3860
3861 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3862 {
3863         struct nvme_ns *ns;
3864
3865         down_read(&ctrl->namespaces_rwsem);
3866         list_for_each_entry(ns, &ctrl->namespaces, list)
3867                 blk_mq_freeze_queue_wait(ns->queue);
3868         up_read(&ctrl->namespaces_rwsem);
3869 }
3870 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3871
3872 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3873 {
3874         struct nvme_ns *ns;
3875
3876         down_read(&ctrl->namespaces_rwsem);
3877         list_for_each_entry(ns, &ctrl->namespaces, list)
3878                 blk_freeze_queue_start(ns->queue);
3879         up_read(&ctrl->namespaces_rwsem);
3880 }
3881 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3882
3883 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3884 {
3885         struct nvme_ns *ns;
3886
3887         down_read(&ctrl->namespaces_rwsem);
3888         list_for_each_entry(ns, &ctrl->namespaces, list)
3889                 blk_mq_quiesce_queue(ns->queue);
3890         up_read(&ctrl->namespaces_rwsem);
3891 }
3892 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3893
3894 void nvme_start_queues(struct nvme_ctrl *ctrl)
3895 {
3896         struct nvme_ns *ns;
3897
3898         down_read(&ctrl->namespaces_rwsem);
3899         list_for_each_entry(ns, &ctrl->namespaces, list)
3900                 blk_mq_unquiesce_queue(ns->queue);
3901         up_read(&ctrl->namespaces_rwsem);
3902 }
3903 EXPORT_SYMBOL_GPL(nvme_start_queues);
3904
3905
3906 void nvme_sync_queues(struct nvme_ctrl *ctrl)
3907 {
3908         struct nvme_ns *ns;
3909
3910         down_read(&ctrl->namespaces_rwsem);
3911         list_for_each_entry(ns, &ctrl->namespaces, list)
3912                 blk_sync_queue(ns->queue);
3913         up_read(&ctrl->namespaces_rwsem);
3914 }
3915 EXPORT_SYMBOL_GPL(nvme_sync_queues);
3916
3917 /*
3918  * Check we didn't inadvertently grow the command structure sizes:
3919  */
3920 static inline void _nvme_check_size(void)
3921 {
3922         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3923         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3924         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3925         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3926         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3927         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3928         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3929         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3930         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3931         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3932         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3933         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
3934         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
3935         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
3936         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
3937         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
3938         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
3939 }
3940
3941
3942 static int __init nvme_core_init(void)
3943 {
3944         int result = -ENOMEM;
3945
3946         _nvme_check_size();
3947
3948         nvme_wq = alloc_workqueue("nvme-wq",
3949                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3950         if (!nvme_wq)
3951                 goto out;
3952
3953         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3954                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3955         if (!nvme_reset_wq)
3956                 goto destroy_wq;
3957
3958         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3959                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3960         if (!nvme_delete_wq)
3961                 goto destroy_reset_wq;
3962
3963         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3964         if (result < 0)
3965                 goto destroy_delete_wq;
3966
3967         nvme_class = class_create(THIS_MODULE, "nvme");
3968         if (IS_ERR(nvme_class)) {
3969                 result = PTR_ERR(nvme_class);
3970                 goto unregister_chrdev;
3971         }
3972
3973         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3974         if (IS_ERR(nvme_subsys_class)) {
3975                 result = PTR_ERR(nvme_subsys_class);
3976                 goto destroy_class;
3977         }
3978         return 0;
3979
3980 destroy_class:
3981         class_destroy(nvme_class);
3982 unregister_chrdev:
3983         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3984 destroy_delete_wq:
3985         destroy_workqueue(nvme_delete_wq);
3986 destroy_reset_wq:
3987         destroy_workqueue(nvme_reset_wq);
3988 destroy_wq:
3989         destroy_workqueue(nvme_wq);
3990 out:
3991         return result;
3992 }
3993
3994 static void __exit nvme_core_exit(void)
3995 {
3996         ida_destroy(&nvme_subsystems_ida);
3997         class_destroy(nvme_subsys_class);
3998         class_destroy(nvme_class);
3999         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4000         destroy_workqueue(nvme_delete_wq);
4001         destroy_workqueue(nvme_reset_wq);
4002         destroy_workqueue(nvme_wq);
4003 }
4004
4005 MODULE_LICENSE("GPL");
4006 MODULE_VERSION("1.0");
4007 module_init(nvme_core_init);
4008 module_exit(nvme_core_exit);