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 the work is flushed since ->delete_ctrl
 183          * 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(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_quiesced(&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\n");
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)
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         else
1263                 effects = nvme_known_admin_effects(opcode);
1264
1265         /*
1266          * For simplicity, IO to all namespaces is quiesced even if the command
1267          * effects say only one namespace is affected.
1268          */
1269         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1270                 mutex_lock(&ctrl->scan_lock);
1271                 nvme_start_freeze(ctrl);
1272                 nvme_wait_freeze(ctrl);
1273         }
1274         return effects;
1275 }
1276
1277 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1278 {
1279         struct nvme_ns *ns;
1280
1281         down_read(&ctrl->namespaces_rwsem);
1282         list_for_each_entry(ns, &ctrl->namespaces, list)
1283                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1284                         nvme_set_queue_dying(ns);
1285         up_read(&ctrl->namespaces_rwsem);
1286
1287         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1288 }
1289
1290 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1291 {
1292         /*
1293          * Revalidate LBA changes prior to unfreezing. This is necessary to
1294          * prevent memory corruption if a logical block size was changed by
1295          * this command.
1296          */
1297         if (effects & NVME_CMD_EFFECTS_LBCC)
1298                 nvme_update_formats(ctrl);
1299         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1300                 nvme_unfreeze(ctrl);
1301                 mutex_unlock(&ctrl->scan_lock);
1302         }
1303         if (effects & NVME_CMD_EFFECTS_CCC)
1304                 nvme_init_identify(ctrl);
1305         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1306                 nvme_queue_scan(ctrl);
1307 }
1308
1309 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1310                         struct nvme_passthru_cmd __user *ucmd)
1311 {
1312         struct nvme_passthru_cmd cmd;
1313         struct nvme_command c;
1314         unsigned timeout = 0;
1315         u32 effects;
1316         int status;
1317
1318         if (!capable(CAP_SYS_ADMIN))
1319                 return -EACCES;
1320         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1321                 return -EFAULT;
1322         if (cmd.flags)
1323                 return -EINVAL;
1324
1325         memset(&c, 0, sizeof(c));
1326         c.common.opcode = cmd.opcode;
1327         c.common.flags = cmd.flags;
1328         c.common.nsid = cpu_to_le32(cmd.nsid);
1329         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1330         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1331         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1332         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1333         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1334         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1335         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1336         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1337
1338         if (cmd.timeout_ms)
1339                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1340
1341         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1342         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1343                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1344                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1345                         0, &cmd.result, timeout);
1346         nvme_passthru_end(ctrl, effects);
1347
1348         if (status >= 0) {
1349                 if (put_user(cmd.result, &ucmd->result))
1350                         return -EFAULT;
1351         }
1352
1353         return status;
1354 }
1355
1356 /*
1357  * Issue ioctl requests on the first available path.  Note that unlike normal
1358  * block layer requests we will not retry failed request on another controller.
1359  */
1360 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1361                 struct nvme_ns_head **head, int *srcu_idx)
1362 {
1363 #ifdef CONFIG_NVME_MULTIPATH
1364         if (disk->fops == &nvme_ns_head_ops) {
1365                 *head = disk->private_data;
1366                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1367                 return nvme_find_path(*head);
1368         }
1369 #endif
1370         *head = NULL;
1371         *srcu_idx = -1;
1372         return disk->private_data;
1373 }
1374
1375 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1376 {
1377         if (head)
1378                 srcu_read_unlock(&head->srcu, idx);
1379 }
1380
1381 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1382 {
1383         switch (cmd) {
1384         case NVME_IOCTL_ID:
1385                 force_successful_syscall_return();
1386                 return ns->head->ns_id;
1387         case NVME_IOCTL_ADMIN_CMD:
1388                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1389         case NVME_IOCTL_IO_CMD:
1390                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1391         case NVME_IOCTL_SUBMIT_IO:
1392                 return nvme_submit_io(ns, (void __user *)arg);
1393         default:
1394 #ifdef CONFIG_NVM
1395                 if (ns->ndev)
1396                         return nvme_nvm_ioctl(ns, cmd, arg);
1397 #endif
1398                 if (is_sed_ioctl(cmd))
1399                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1400                                          (void __user *) arg);
1401                 return -ENOTTY;
1402         }
1403 }
1404
1405 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1406                 unsigned int cmd, unsigned long arg)
1407 {
1408         struct nvme_ns_head *head = NULL;
1409         struct nvme_ns *ns;
1410         int srcu_idx, ret;
1411
1412         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1413         if (unlikely(!ns))
1414                 ret = -EWOULDBLOCK;
1415         else
1416                 ret = nvme_ns_ioctl(ns, cmd, arg);
1417         nvme_put_ns_from_disk(head, srcu_idx);
1418         return ret;
1419 }
1420
1421 static int nvme_open(struct block_device *bdev, fmode_t mode)
1422 {
1423         struct nvme_ns *ns = bdev->bd_disk->private_data;
1424
1425 #ifdef CONFIG_NVME_MULTIPATH
1426         /* should never be called due to GENHD_FL_HIDDEN */
1427         if (WARN_ON_ONCE(ns->head->disk))
1428                 goto fail;
1429 #endif
1430         if (!kref_get_unless_zero(&ns->kref))
1431                 goto fail;
1432         if (!try_module_get(ns->ctrl->ops->module))
1433                 goto fail_put_ns;
1434
1435         return 0;
1436
1437 fail_put_ns:
1438         nvme_put_ns(ns);
1439 fail:
1440         return -ENXIO;
1441 }
1442
1443 static void nvme_release(struct gendisk *disk, fmode_t mode)
1444 {
1445         struct nvme_ns *ns = disk->private_data;
1446
1447         module_put(ns->ctrl->ops->module);
1448         nvme_put_ns(ns);
1449 }
1450
1451 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1452 {
1453         /* some standard values */
1454         geo->heads = 1 << 6;
1455         geo->sectors = 1 << 5;
1456         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1457         return 0;
1458 }
1459
1460 #ifdef CONFIG_BLK_DEV_INTEGRITY
1461 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1462 {
1463         struct blk_integrity integrity;
1464
1465         memset(&integrity, 0, sizeof(integrity));
1466         switch (pi_type) {
1467         case NVME_NS_DPS_PI_TYPE3:
1468                 integrity.profile = &t10_pi_type3_crc;
1469                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1470                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1471                 break;
1472         case NVME_NS_DPS_PI_TYPE1:
1473         case NVME_NS_DPS_PI_TYPE2:
1474                 integrity.profile = &t10_pi_type1_crc;
1475                 integrity.tag_size = sizeof(u16);
1476                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1477                 break;
1478         default:
1479                 integrity.profile = NULL;
1480                 break;
1481         }
1482         integrity.tuple_size = ms;
1483         blk_integrity_register(disk, &integrity);
1484         blk_queue_max_integrity_segments(disk->queue, 1);
1485 }
1486 #else
1487 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1488 {
1489 }
1490 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1491
1492 static void nvme_set_chunk_size(struct nvme_ns *ns)
1493 {
1494         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1495         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1496 }
1497
1498 static void nvme_config_discard(struct nvme_ns *ns)
1499 {
1500         struct nvme_ctrl *ctrl = ns->ctrl;
1501         struct request_queue *queue = ns->queue;
1502         u32 size = queue_logical_block_size(queue);
1503
1504         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1505                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1506                 return;
1507         }
1508
1509         if (ctrl->nr_streams && ns->sws && ns->sgs)
1510                 size *= ns->sws * ns->sgs;
1511
1512         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1513                         NVME_DSM_MAX_RANGES);
1514
1515         queue->limits.discard_alignment = 0;
1516         queue->limits.discard_granularity = size;
1517
1518         /* If discard is already enabled, don't reset queue limits */
1519         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1520                 return;
1521
1522         blk_queue_max_discard_sectors(queue, UINT_MAX);
1523         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1524
1525         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1526                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1527 }
1528
1529 static inline void nvme_config_write_zeroes(struct nvme_ns *ns)
1530 {
1531         u32 max_sectors;
1532         unsigned short bs = 1 << ns->lba_shift;
1533
1534         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES))
1535                 return;
1536         /*
1537          * Even though NVMe spec explicitly states that MDTS is not
1538          * applicable to the write-zeroes:- "The restriction does not apply to
1539          * commands that do not transfer data between the host and the
1540          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1541          * In order to be more cautious use controller's max_hw_sectors value
1542          * to configure the maximum sectors for the write-zeroes which is
1543          * configured based on the controller's MDTS field in the
1544          * nvme_init_identify() if available.
1545          */
1546         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1547                 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1548         else
1549                 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1550
1551         blk_queue_max_write_zeroes_sectors(ns->queue, max_sectors);
1552 }
1553
1554 static inline void nvme_ns_config_oncs(struct nvme_ns *ns)
1555 {
1556         nvme_config_discard(ns);
1557         nvme_config_write_zeroes(ns);
1558 }
1559
1560 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1561                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1562 {
1563         memset(ids, 0, sizeof(*ids));
1564
1565         if (ctrl->vs >= NVME_VS(1, 1, 0))
1566                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1567         if (ctrl->vs >= NVME_VS(1, 2, 0))
1568                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1569         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1570                  /* Don't treat error as fatal we potentially
1571                   * already have a NGUID or EUI-64
1572                   */
1573                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1574                         dev_warn(ctrl->device,
1575                                  "%s: Identify Descriptors failed\n", __func__);
1576         }
1577 }
1578
1579 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1580 {
1581         return !uuid_is_null(&ids->uuid) ||
1582                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1583                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1584 }
1585
1586 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1587 {
1588         return uuid_equal(&a->uuid, &b->uuid) &&
1589                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1590                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1591 }
1592
1593 static void nvme_update_disk_info(struct gendisk *disk,
1594                 struct nvme_ns *ns, struct nvme_id_ns *id)
1595 {
1596         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1597         unsigned short bs = 1 << ns->lba_shift;
1598
1599         blk_mq_freeze_queue(disk->queue);
1600         blk_integrity_unregister(disk);
1601
1602         blk_queue_logical_block_size(disk->queue, bs);
1603         blk_queue_physical_block_size(disk->queue, bs);
1604         blk_queue_io_min(disk->queue, bs);
1605
1606         if (ns->ms && !ns->ext &&
1607             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1608                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1609         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1610                 capacity = 0;
1611
1612         set_capacity(disk, capacity);
1613         nvme_ns_config_oncs(ns);
1614
1615         if (id->nsattr & (1 << 0))
1616                 set_disk_ro(disk, true);
1617         else
1618                 set_disk_ro(disk, false);
1619
1620         blk_mq_unfreeze_queue(disk->queue);
1621 }
1622
1623 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1624 {
1625         struct nvme_ns *ns = disk->private_data;
1626
1627         /*
1628          * If identify namespace failed, use default 512 byte block size so
1629          * block layer can use before failing read/write for 0 capacity.
1630          */
1631         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1632         if (ns->lba_shift == 0)
1633                 ns->lba_shift = 9;
1634         ns->noiob = le16_to_cpu(id->noiob);
1635         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1636         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1637         /* the PI implementation requires metadata equal t10 pi tuple size */
1638         if (ns->ms == sizeof(struct t10_pi_tuple))
1639                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1640         else
1641                 ns->pi_type = 0;
1642
1643         if (ns->noiob)
1644                 nvme_set_chunk_size(ns);
1645         nvme_update_disk_info(disk, ns, id);
1646 #ifdef CONFIG_NVME_MULTIPATH
1647         if (ns->head->disk) {
1648                 nvme_update_disk_info(ns->head->disk, ns, id);
1649                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1650         }
1651 #endif
1652 }
1653
1654 static int nvme_revalidate_disk(struct gendisk *disk)
1655 {
1656         struct nvme_ns *ns = disk->private_data;
1657         struct nvme_ctrl *ctrl = ns->ctrl;
1658         struct nvme_id_ns *id;
1659         struct nvme_ns_ids ids;
1660         int ret = 0;
1661
1662         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1663                 set_capacity(disk, 0);
1664                 return -ENODEV;
1665         }
1666
1667         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1668         if (!id)
1669                 return -ENODEV;
1670
1671         if (id->ncap == 0) {
1672                 ret = -ENODEV;
1673                 goto out;
1674         }
1675
1676         __nvme_revalidate_disk(disk, id);
1677         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1678         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1679                 dev_err(ctrl->device,
1680                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1681                 ret = -ENODEV;
1682         }
1683
1684 out:
1685         kfree(id);
1686         return ret;
1687 }
1688
1689 static char nvme_pr_type(enum pr_type type)
1690 {
1691         switch (type) {
1692         case PR_WRITE_EXCLUSIVE:
1693                 return 1;
1694         case PR_EXCLUSIVE_ACCESS:
1695                 return 2;
1696         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1697                 return 3;
1698         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1699                 return 4;
1700         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1701                 return 5;
1702         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1703                 return 6;
1704         default:
1705                 return 0;
1706         }
1707 };
1708
1709 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1710                                 u64 key, u64 sa_key, u8 op)
1711 {
1712         struct nvme_ns_head *head = NULL;
1713         struct nvme_ns *ns;
1714         struct nvme_command c;
1715         int srcu_idx, ret;
1716         u8 data[16] = { 0, };
1717
1718         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1719         if (unlikely(!ns))
1720                 return -EWOULDBLOCK;
1721
1722         put_unaligned_le64(key, &data[0]);
1723         put_unaligned_le64(sa_key, &data[8]);
1724
1725         memset(&c, 0, sizeof(c));
1726         c.common.opcode = op;
1727         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1728         c.common.cdw10 = cpu_to_le32(cdw10);
1729
1730         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1731         nvme_put_ns_from_disk(head, srcu_idx);
1732         return ret;
1733 }
1734
1735 static int nvme_pr_register(struct block_device *bdev, u64 old,
1736                 u64 new, unsigned flags)
1737 {
1738         u32 cdw10;
1739
1740         if (flags & ~PR_FL_IGNORE_KEY)
1741                 return -EOPNOTSUPP;
1742
1743         cdw10 = old ? 2 : 0;
1744         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1745         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1746         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1747 }
1748
1749 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1750                 enum pr_type type, unsigned flags)
1751 {
1752         u32 cdw10;
1753
1754         if (flags & ~PR_FL_IGNORE_KEY)
1755                 return -EOPNOTSUPP;
1756
1757         cdw10 = nvme_pr_type(type) << 8;
1758         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1759         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1760 }
1761
1762 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1763                 enum pr_type type, bool abort)
1764 {
1765         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1766         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1767 }
1768
1769 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1770 {
1771         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1772         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1773 }
1774
1775 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1776 {
1777         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1778         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1779 }
1780
1781 static const struct pr_ops nvme_pr_ops = {
1782         .pr_register    = nvme_pr_register,
1783         .pr_reserve     = nvme_pr_reserve,
1784         .pr_release     = nvme_pr_release,
1785         .pr_preempt     = nvme_pr_preempt,
1786         .pr_clear       = nvme_pr_clear,
1787 };
1788
1789 #ifdef CONFIG_BLK_SED_OPAL
1790 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1791                 bool send)
1792 {
1793         struct nvme_ctrl *ctrl = data;
1794         struct nvme_command cmd;
1795
1796         memset(&cmd, 0, sizeof(cmd));
1797         if (send)
1798                 cmd.common.opcode = nvme_admin_security_send;
1799         else
1800                 cmd.common.opcode = nvme_admin_security_recv;
1801         cmd.common.nsid = 0;
1802         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1803         cmd.common.cdw11 = cpu_to_le32(len);
1804
1805         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1806                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1807 }
1808 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1809 #endif /* CONFIG_BLK_SED_OPAL */
1810
1811 static const struct block_device_operations nvme_fops = {
1812         .owner          = THIS_MODULE,
1813         .ioctl          = nvme_ioctl,
1814         .compat_ioctl   = nvme_ioctl,
1815         .open           = nvme_open,
1816         .release        = nvme_release,
1817         .getgeo         = nvme_getgeo,
1818         .revalidate_disk= nvme_revalidate_disk,
1819         .pr_ops         = &nvme_pr_ops,
1820 };
1821
1822 #ifdef CONFIG_NVME_MULTIPATH
1823 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1824 {
1825         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1826
1827         if (!kref_get_unless_zero(&head->ref))
1828                 return -ENXIO;
1829         return 0;
1830 }
1831
1832 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1833 {
1834         nvme_put_ns_head(disk->private_data);
1835 }
1836
1837 const struct block_device_operations nvme_ns_head_ops = {
1838         .owner          = THIS_MODULE,
1839         .open           = nvme_ns_head_open,
1840         .release        = nvme_ns_head_release,
1841         .ioctl          = nvme_ioctl,
1842         .compat_ioctl   = nvme_ioctl,
1843         .getgeo         = nvme_getgeo,
1844         .pr_ops         = &nvme_pr_ops,
1845 };
1846 #endif /* CONFIG_NVME_MULTIPATH */
1847
1848 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1849 {
1850         unsigned long timeout =
1851                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1852         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1853         int ret;
1854
1855         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1856                 if (csts == ~0)
1857                         return -ENODEV;
1858                 if ((csts & NVME_CSTS_RDY) == bit)
1859                         break;
1860
1861                 msleep(100);
1862                 if (fatal_signal_pending(current))
1863                         return -EINTR;
1864                 if (time_after(jiffies, timeout)) {
1865                         dev_err(ctrl->device,
1866                                 "Device not ready; aborting %s\n", enabled ?
1867                                                 "initialisation" : "reset");
1868                         return -ENODEV;
1869                 }
1870         }
1871
1872         return ret;
1873 }
1874
1875 /*
1876  * If the device has been passed off to us in an enabled state, just clear
1877  * the enabled bit.  The spec says we should set the 'shutdown notification
1878  * bits', but doing so may cause the device to complete commands to the
1879  * admin queue ... and we don't know what memory that might be pointing at!
1880  */
1881 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1882 {
1883         int ret;
1884
1885         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1886         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1887
1888         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1889         if (ret)
1890                 return ret;
1891
1892         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1893                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1894
1895         return nvme_wait_ready(ctrl, cap, false);
1896 }
1897 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1898
1899 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1900 {
1901         /*
1902          * Default to a 4K page size, with the intention to update this
1903          * path in the future to accomodate architectures with differing
1904          * kernel and IO page sizes.
1905          */
1906         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1907         int ret;
1908
1909         if (page_shift < dev_page_min) {
1910                 dev_err(ctrl->device,
1911                         "Minimum device page size %u too large for host (%u)\n",
1912                         1 << dev_page_min, 1 << page_shift);
1913                 return -ENODEV;
1914         }
1915
1916         ctrl->page_size = 1 << page_shift;
1917
1918         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1919         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1920         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1921         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1922         ctrl->ctrl_config |= NVME_CC_ENABLE;
1923
1924         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1925         if (ret)
1926                 return ret;
1927         return nvme_wait_ready(ctrl, cap, true);
1928 }
1929 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1930
1931 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1932 {
1933         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1934         u32 csts;
1935         int ret;
1936
1937         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1938         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1939
1940         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1941         if (ret)
1942                 return ret;
1943
1944         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1945                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1946                         break;
1947
1948                 msleep(100);
1949                 if (fatal_signal_pending(current))
1950                         return -EINTR;
1951                 if (time_after(jiffies, timeout)) {
1952                         dev_err(ctrl->device,
1953                                 "Device shutdown incomplete; abort shutdown\n");
1954                         return -ENODEV;
1955                 }
1956         }
1957
1958         return ret;
1959 }
1960 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1961
1962 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1963                 struct request_queue *q)
1964 {
1965         bool vwc = false;
1966
1967         if (ctrl->max_hw_sectors) {
1968                 u32 max_segments =
1969                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1970
1971                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1972                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1973                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1974         }
1975         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1976             is_power_of_2(ctrl->max_hw_sectors))
1977                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1978         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1979         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1980                 vwc = true;
1981         blk_queue_write_cache(q, vwc, vwc);
1982 }
1983
1984 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1985 {
1986         __le64 ts;
1987         int ret;
1988
1989         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1990                 return 0;
1991
1992         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1993         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1994                         NULL);
1995         if (ret)
1996                 dev_warn_once(ctrl->device,
1997                         "could not set timestamp (%d)\n", ret);
1998         return ret;
1999 }
2000
2001 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2002 {
2003         struct nvme_feat_host_behavior *host;
2004         int ret;
2005
2006         /* Don't bother enabling the feature if retry delay is not reported */
2007         if (!ctrl->crdt[0])
2008                 return 0;
2009
2010         host = kzalloc(sizeof(*host), GFP_KERNEL);
2011         if (!host)
2012                 return 0;
2013
2014         host->acre = NVME_ENABLE_ACRE;
2015         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2016                                 host, sizeof(*host), NULL);
2017         kfree(host);
2018         return ret;
2019 }
2020
2021 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2022 {
2023         /*
2024          * APST (Autonomous Power State Transition) lets us program a
2025          * table of power state transitions that the controller will
2026          * perform automatically.  We configure it with a simple
2027          * heuristic: we are willing to spend at most 2% of the time
2028          * transitioning between power states.  Therefore, when running
2029          * in any given state, we will enter the next lower-power
2030          * non-operational state after waiting 50 * (enlat + exlat)
2031          * microseconds, as long as that state's exit latency is under
2032          * the requested maximum latency.
2033          *
2034          * We will not autonomously enter any non-operational state for
2035          * which the total latency exceeds ps_max_latency_us.  Users
2036          * can set ps_max_latency_us to zero to turn off APST.
2037          */
2038
2039         unsigned apste;
2040         struct nvme_feat_auto_pst *table;
2041         u64 max_lat_us = 0;
2042         int max_ps = -1;
2043         int ret;
2044
2045         /*
2046          * If APST isn't supported or if we haven't been initialized yet,
2047          * then don't do anything.
2048          */
2049         if (!ctrl->apsta)
2050                 return 0;
2051
2052         if (ctrl->npss > 31) {
2053                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2054                 return 0;
2055         }
2056
2057         table = kzalloc(sizeof(*table), GFP_KERNEL);
2058         if (!table)
2059                 return 0;
2060
2061         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2062                 /* Turn off APST. */
2063                 apste = 0;
2064                 dev_dbg(ctrl->device, "APST disabled\n");
2065         } else {
2066                 __le64 target = cpu_to_le64(0);
2067                 int state;
2068
2069                 /*
2070                  * Walk through all states from lowest- to highest-power.
2071                  * According to the spec, lower-numbered states use more
2072                  * power.  NPSS, despite the name, is the index of the
2073                  * lowest-power state, not the number of states.
2074                  */
2075                 for (state = (int)ctrl->npss; state >= 0; state--) {
2076                         u64 total_latency_us, exit_latency_us, transition_ms;
2077
2078                         if (target)
2079                                 table->entries[state] = target;
2080
2081                         /*
2082                          * Don't allow transitions to the deepest state
2083                          * if it's quirked off.
2084                          */
2085                         if (state == ctrl->npss &&
2086                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2087                                 continue;
2088
2089                         /*
2090                          * Is this state a useful non-operational state for
2091                          * higher-power states to autonomously transition to?
2092                          */
2093                         if (!(ctrl->psd[state].flags &
2094                               NVME_PS_FLAGS_NON_OP_STATE))
2095                                 continue;
2096
2097                         exit_latency_us =
2098                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2099                         if (exit_latency_us > ctrl->ps_max_latency_us)
2100                                 continue;
2101
2102                         total_latency_us =
2103                                 exit_latency_us +
2104                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2105
2106                         /*
2107                          * This state is good.  Use it as the APST idle
2108                          * target for higher power states.
2109                          */
2110                         transition_ms = total_latency_us + 19;
2111                         do_div(transition_ms, 20);
2112                         if (transition_ms > (1 << 24) - 1)
2113                                 transition_ms = (1 << 24) - 1;
2114
2115                         target = cpu_to_le64((state << 3) |
2116                                              (transition_ms << 8));
2117
2118                         if (max_ps == -1)
2119                                 max_ps = state;
2120
2121                         if (total_latency_us > max_lat_us)
2122                                 max_lat_us = total_latency_us;
2123                 }
2124
2125                 apste = 1;
2126
2127                 if (max_ps == -1) {
2128                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2129                 } else {
2130                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2131                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2132                 }
2133         }
2134
2135         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2136                                 table, sizeof(*table), NULL);
2137         if (ret)
2138                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2139
2140         kfree(table);
2141         return ret;
2142 }
2143
2144 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2145 {
2146         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2147         u64 latency;
2148
2149         switch (val) {
2150         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2151         case PM_QOS_LATENCY_ANY:
2152                 latency = U64_MAX;
2153                 break;
2154
2155         default:
2156                 latency = val;
2157         }
2158
2159         if (ctrl->ps_max_latency_us != latency) {
2160                 ctrl->ps_max_latency_us = latency;
2161                 nvme_configure_apst(ctrl);
2162         }
2163 }
2164
2165 struct nvme_core_quirk_entry {
2166         /*
2167          * NVMe model and firmware strings are padded with spaces.  For
2168          * simplicity, strings in the quirk table are padded with NULLs
2169          * instead.
2170          */
2171         u16 vid;
2172         const char *mn;
2173         const char *fr;
2174         unsigned long quirks;
2175 };
2176
2177 static const struct nvme_core_quirk_entry core_quirks[] = {
2178         {
2179                 /*
2180                  * This Toshiba device seems to die using any APST states.  See:
2181                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2182                  */
2183                 .vid = 0x1179,
2184                 .mn = "THNSF5256GPUK TOSHIBA",
2185                 .quirks = NVME_QUIRK_NO_APST,
2186         }
2187 };
2188
2189 /* match is null-terminated but idstr is space-padded. */
2190 static bool string_matches(const char *idstr, const char *match, size_t len)
2191 {
2192         size_t matchlen;
2193
2194         if (!match)
2195                 return true;
2196
2197         matchlen = strlen(match);
2198         WARN_ON_ONCE(matchlen > len);
2199
2200         if (memcmp(idstr, match, matchlen))
2201                 return false;
2202
2203         for (; matchlen < len; matchlen++)
2204                 if (idstr[matchlen] != ' ')
2205                         return false;
2206
2207         return true;
2208 }
2209
2210 static bool quirk_matches(const struct nvme_id_ctrl *id,
2211                           const struct nvme_core_quirk_entry *q)
2212 {
2213         return q->vid == le16_to_cpu(id->vid) &&
2214                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2215                 string_matches(id->fr, q->fr, sizeof(id->fr));
2216 }
2217
2218 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2219                 struct nvme_id_ctrl *id)
2220 {
2221         size_t nqnlen;
2222         int off;
2223
2224         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2225                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2226                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2227                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2228                         return;
2229                 }
2230
2231                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2232                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2233         }
2234
2235         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2236         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2237                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2238                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2239         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2240         off += sizeof(id->sn);
2241         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2242         off += sizeof(id->mn);
2243         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2244 }
2245
2246 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2247 {
2248         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2249         kfree(subsys);
2250 }
2251
2252 static void nvme_release_subsystem(struct device *dev)
2253 {
2254         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2255 }
2256
2257 static void nvme_destroy_subsystem(struct kref *ref)
2258 {
2259         struct nvme_subsystem *subsys =
2260                         container_of(ref, struct nvme_subsystem, ref);
2261
2262         mutex_lock(&nvme_subsystems_lock);
2263         list_del(&subsys->entry);
2264         mutex_unlock(&nvme_subsystems_lock);
2265
2266         ida_destroy(&subsys->ns_ida);
2267         device_del(&subsys->dev);
2268         put_device(&subsys->dev);
2269 }
2270
2271 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2272 {
2273         kref_put(&subsys->ref, nvme_destroy_subsystem);
2274 }
2275
2276 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2277 {
2278         struct nvme_subsystem *subsys;
2279
2280         lockdep_assert_held(&nvme_subsystems_lock);
2281
2282         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2283                 if (strcmp(subsys->subnqn, subsysnqn))
2284                         continue;
2285                 if (!kref_get_unless_zero(&subsys->ref))
2286                         continue;
2287                 return subsys;
2288         }
2289
2290         return NULL;
2291 }
2292
2293 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2294         struct device_attribute subsys_attr_##_name = \
2295                 __ATTR(_name, _mode, _show, NULL)
2296
2297 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2298                                     struct device_attribute *attr,
2299                                     char *buf)
2300 {
2301         struct nvme_subsystem *subsys =
2302                 container_of(dev, struct nvme_subsystem, dev);
2303
2304         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2305 }
2306 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2307
2308 #define nvme_subsys_show_str_function(field)                            \
2309 static ssize_t subsys_##field##_show(struct device *dev,                \
2310                             struct device_attribute *attr, char *buf)   \
2311 {                                                                       \
2312         struct nvme_subsystem *subsys =                                 \
2313                 container_of(dev, struct nvme_subsystem, dev);          \
2314         return sprintf(buf, "%.*s\n",                                   \
2315                        (int)sizeof(subsys->field), subsys->field);      \
2316 }                                                                       \
2317 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2318
2319 nvme_subsys_show_str_function(model);
2320 nvme_subsys_show_str_function(serial);
2321 nvme_subsys_show_str_function(firmware_rev);
2322
2323 static struct attribute *nvme_subsys_attrs[] = {
2324         &subsys_attr_model.attr,
2325         &subsys_attr_serial.attr,
2326         &subsys_attr_firmware_rev.attr,
2327         &subsys_attr_subsysnqn.attr,
2328 #ifdef CONFIG_NVME_MULTIPATH
2329         &subsys_attr_iopolicy.attr,
2330 #endif
2331         NULL,
2332 };
2333
2334 static struct attribute_group nvme_subsys_attrs_group = {
2335         .attrs = nvme_subsys_attrs,
2336 };
2337
2338 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2339         &nvme_subsys_attrs_group,
2340         NULL,
2341 };
2342
2343 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2344 {
2345         int count = 0;
2346         struct nvme_ctrl *ctrl;
2347
2348         mutex_lock(&subsys->lock);
2349         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2350                 if (ctrl->state != NVME_CTRL_DELETING &&
2351                     ctrl->state != NVME_CTRL_DEAD)
2352                         count++;
2353         }
2354         mutex_unlock(&subsys->lock);
2355
2356         return count;
2357 }
2358
2359 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2360 {
2361         struct nvme_subsystem *subsys, *found;
2362         int ret;
2363
2364         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2365         if (!subsys)
2366                 return -ENOMEM;
2367         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2368         if (ret < 0) {
2369                 kfree(subsys);
2370                 return ret;
2371         }
2372         subsys->instance = ret;
2373         mutex_init(&subsys->lock);
2374         kref_init(&subsys->ref);
2375         INIT_LIST_HEAD(&subsys->ctrls);
2376         INIT_LIST_HEAD(&subsys->nsheads);
2377         nvme_init_subnqn(subsys, ctrl, id);
2378         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2379         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2380         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2381         subsys->vendor_id = le16_to_cpu(id->vid);
2382         subsys->cmic = id->cmic;
2383 #ifdef CONFIG_NVME_MULTIPATH
2384         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2385 #endif
2386
2387         subsys->dev.class = nvme_subsys_class;
2388         subsys->dev.release = nvme_release_subsystem;
2389         subsys->dev.groups = nvme_subsys_attrs_groups;
2390         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2391         device_initialize(&subsys->dev);
2392
2393         mutex_lock(&nvme_subsystems_lock);
2394         found = __nvme_find_get_subsystem(subsys->subnqn);
2395         if (found) {
2396                 /*
2397                  * Verify that the subsystem actually supports multiple
2398                  * controllers, else bail out.
2399                  */
2400                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2401                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2402                         dev_err(ctrl->device,
2403                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2404                                 found->subnqn);
2405                         nvme_put_subsystem(found);
2406                         ret = -EINVAL;
2407                         goto out_unlock;
2408                 }
2409
2410                 __nvme_release_subsystem(subsys);
2411                 subsys = found;
2412         } else {
2413                 ret = device_add(&subsys->dev);
2414                 if (ret) {
2415                         dev_err(ctrl->device,
2416                                 "failed to register subsystem device.\n");
2417                         goto out_unlock;
2418                 }
2419                 ida_init(&subsys->ns_ida);
2420                 list_add_tail(&subsys->entry, &nvme_subsystems);
2421         }
2422
2423         ctrl->subsys = subsys;
2424         mutex_unlock(&nvme_subsystems_lock);
2425
2426         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2427                         dev_name(ctrl->device))) {
2428                 dev_err(ctrl->device,
2429                         "failed to create sysfs link from subsystem.\n");
2430                 /* the transport driver will eventually put the subsystem */
2431                 return -EINVAL;
2432         }
2433
2434         mutex_lock(&subsys->lock);
2435         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2436         mutex_unlock(&subsys->lock);
2437
2438         return 0;
2439
2440 out_unlock:
2441         mutex_unlock(&nvme_subsystems_lock);
2442         put_device(&subsys->dev);
2443         return ret;
2444 }
2445
2446 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2447                 void *log, size_t size, u64 offset)
2448 {
2449         struct nvme_command c = { };
2450         unsigned long dwlen = size / 4 - 1;
2451
2452         c.get_log_page.opcode = nvme_admin_get_log_page;
2453         c.get_log_page.nsid = cpu_to_le32(nsid);
2454         c.get_log_page.lid = log_page;
2455         c.get_log_page.lsp = lsp;
2456         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2457         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2458         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2459         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2460
2461         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2462 }
2463
2464 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2465 {
2466         int ret;
2467
2468         if (!ctrl->effects)
2469                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2470
2471         if (!ctrl->effects)
2472                 return 0;
2473
2474         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2475                         ctrl->effects, sizeof(*ctrl->effects), 0);
2476         if (ret) {
2477                 kfree(ctrl->effects);
2478                 ctrl->effects = NULL;
2479         }
2480         return ret;
2481 }
2482
2483 /*
2484  * Initialize the cached copies of the Identify data and various controller
2485  * register in our nvme_ctrl structure.  This should be called as soon as
2486  * the admin queue is fully up and running.
2487  */
2488 int nvme_init_identify(struct nvme_ctrl *ctrl)
2489 {
2490         struct nvme_id_ctrl *id;
2491         u64 cap;
2492         int ret, page_shift;
2493         u32 max_hw_sectors;
2494         bool prev_apst_enabled;
2495
2496         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2497         if (ret) {
2498                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2499                 return ret;
2500         }
2501
2502         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2503         if (ret) {
2504                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2505                 return ret;
2506         }
2507         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2508
2509         if (ctrl->vs >= NVME_VS(1, 1, 0))
2510                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2511
2512         ret = nvme_identify_ctrl(ctrl, &id);
2513         if (ret) {
2514                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2515                 return -EIO;
2516         }
2517
2518         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2519                 ret = nvme_get_effects_log(ctrl);
2520                 if (ret < 0)
2521                         goto out_free;
2522         }
2523
2524         if (!ctrl->identified) {
2525                 int i;
2526
2527                 ret = nvme_init_subsystem(ctrl, id);
2528                 if (ret)
2529                         goto out_free;
2530
2531                 /*
2532                  * Check for quirks.  Quirk can depend on firmware version,
2533                  * so, in principle, the set of quirks present can change
2534                  * across a reset.  As a possible future enhancement, we
2535                  * could re-scan for quirks every time we reinitialize
2536                  * the device, but we'd have to make sure that the driver
2537                  * behaves intelligently if the quirks change.
2538                  */
2539                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2540                         if (quirk_matches(id, &core_quirks[i]))
2541                                 ctrl->quirks |= core_quirks[i].quirks;
2542                 }
2543         }
2544
2545         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2546                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2547                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2548         }
2549
2550         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2551         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2552         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2553
2554         ctrl->oacs = le16_to_cpu(id->oacs);
2555         ctrl->oncs = le16_to_cpup(&id->oncs);
2556         ctrl->oaes = le32_to_cpu(id->oaes);
2557         atomic_set(&ctrl->abort_limit, id->acl + 1);
2558         ctrl->vwc = id->vwc;
2559         if (id->mdts)
2560                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2561         else
2562                 max_hw_sectors = UINT_MAX;
2563         ctrl->max_hw_sectors =
2564                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2565
2566         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2567         ctrl->sgls = le32_to_cpu(id->sgls);
2568         ctrl->kas = le16_to_cpu(id->kas);
2569         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2570         ctrl->ctratt = le32_to_cpu(id->ctratt);
2571
2572         if (id->rtd3e) {
2573                 /* us -> s */
2574                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2575
2576                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2577                                                  shutdown_timeout, 60);
2578
2579                 if (ctrl->shutdown_timeout != shutdown_timeout)
2580                         dev_info(ctrl->device,
2581                                  "Shutdown timeout set to %u seconds\n",
2582                                  ctrl->shutdown_timeout);
2583         } else
2584                 ctrl->shutdown_timeout = shutdown_timeout;
2585
2586         ctrl->npss = id->npss;
2587         ctrl->apsta = id->apsta;
2588         prev_apst_enabled = ctrl->apst_enabled;
2589         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2590                 if (force_apst && id->apsta) {
2591                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2592                         ctrl->apst_enabled = true;
2593                 } else {
2594                         ctrl->apst_enabled = false;
2595                 }
2596         } else {
2597                 ctrl->apst_enabled = id->apsta;
2598         }
2599         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2600
2601         if (ctrl->ops->flags & NVME_F_FABRICS) {
2602                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2603                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2604                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2605                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2606
2607                 /*
2608                  * In fabrics we need to verify the cntlid matches the
2609                  * admin connect
2610                  */
2611                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2612                         ret = -EINVAL;
2613                         goto out_free;
2614                 }
2615
2616                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2617                         dev_err(ctrl->device,
2618                                 "keep-alive support is mandatory for fabrics\n");
2619                         ret = -EINVAL;
2620                         goto out_free;
2621                 }
2622         } else {
2623                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2624                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2625                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2626                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2627                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2628         }
2629
2630         ret = nvme_mpath_init(ctrl, id);
2631         kfree(id);
2632
2633         if (ret < 0)
2634                 return ret;
2635
2636         if (ctrl->apst_enabled && !prev_apst_enabled)
2637                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2638         else if (!ctrl->apst_enabled && prev_apst_enabled)
2639                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2640
2641         ret = nvme_configure_apst(ctrl);
2642         if (ret < 0)
2643                 return ret;
2644
2645         ret = nvme_configure_timestamp(ctrl);
2646         if (ret < 0)
2647                 return ret;
2648
2649         ret = nvme_configure_directives(ctrl);
2650         if (ret < 0)
2651                 return ret;
2652
2653         ret = nvme_configure_acre(ctrl);
2654         if (ret < 0)
2655                 return ret;
2656
2657         ctrl->identified = true;
2658
2659         return 0;
2660
2661 out_free:
2662         kfree(id);
2663         return ret;
2664 }
2665 EXPORT_SYMBOL_GPL(nvme_init_identify);
2666
2667 static int nvme_dev_open(struct inode *inode, struct file *file)
2668 {
2669         struct nvme_ctrl *ctrl =
2670                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2671
2672         switch (ctrl->state) {
2673         case NVME_CTRL_LIVE:
2674         case NVME_CTRL_ADMIN_ONLY:
2675                 break;
2676         default:
2677                 return -EWOULDBLOCK;
2678         }
2679
2680         file->private_data = ctrl;
2681         return 0;
2682 }
2683
2684 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2685 {
2686         struct nvme_ns *ns;
2687         int ret;
2688
2689         down_read(&ctrl->namespaces_rwsem);
2690         if (list_empty(&ctrl->namespaces)) {
2691                 ret = -ENOTTY;
2692                 goto out_unlock;
2693         }
2694
2695         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2696         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2697                 dev_warn(ctrl->device,
2698                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2699                 ret = -EINVAL;
2700                 goto out_unlock;
2701         }
2702
2703         dev_warn(ctrl->device,
2704                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2705         kref_get(&ns->kref);
2706         up_read(&ctrl->namespaces_rwsem);
2707
2708         ret = nvme_user_cmd(ctrl, ns, argp);
2709         nvme_put_ns(ns);
2710         return ret;
2711
2712 out_unlock:
2713         up_read(&ctrl->namespaces_rwsem);
2714         return ret;
2715 }
2716
2717 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2718                 unsigned long arg)
2719 {
2720         struct nvme_ctrl *ctrl = file->private_data;
2721         void __user *argp = (void __user *)arg;
2722
2723         switch (cmd) {
2724         case NVME_IOCTL_ADMIN_CMD:
2725                 return nvme_user_cmd(ctrl, NULL, argp);
2726         case NVME_IOCTL_IO_CMD:
2727                 return nvme_dev_user_cmd(ctrl, argp);
2728         case NVME_IOCTL_RESET:
2729                 dev_warn(ctrl->device, "resetting controller\n");
2730                 return nvme_reset_ctrl_sync(ctrl);
2731         case NVME_IOCTL_SUBSYS_RESET:
2732                 return nvme_reset_subsystem(ctrl);
2733         case NVME_IOCTL_RESCAN:
2734                 nvme_queue_scan(ctrl);
2735                 return 0;
2736         default:
2737                 return -ENOTTY;
2738         }
2739 }
2740
2741 static const struct file_operations nvme_dev_fops = {
2742         .owner          = THIS_MODULE,
2743         .open           = nvme_dev_open,
2744         .unlocked_ioctl = nvme_dev_ioctl,
2745         .compat_ioctl   = nvme_dev_ioctl,
2746 };
2747
2748 static ssize_t nvme_sysfs_reset(struct device *dev,
2749                                 struct device_attribute *attr, const char *buf,
2750                                 size_t count)
2751 {
2752         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2753         int ret;
2754
2755         ret = nvme_reset_ctrl_sync(ctrl);
2756         if (ret < 0)
2757                 return ret;
2758         return count;
2759 }
2760 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2761
2762 static ssize_t nvme_sysfs_rescan(struct device *dev,
2763                                 struct device_attribute *attr, const char *buf,
2764                                 size_t count)
2765 {
2766         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2767
2768         nvme_queue_scan(ctrl);
2769         return count;
2770 }
2771 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2772
2773 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2774 {
2775         struct gendisk *disk = dev_to_disk(dev);
2776
2777         if (disk->fops == &nvme_fops)
2778                 return nvme_get_ns_from_dev(dev)->head;
2779         else
2780                 return disk->private_data;
2781 }
2782
2783 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2784                 char *buf)
2785 {
2786         struct nvme_ns_head *head = dev_to_ns_head(dev);
2787         struct nvme_ns_ids *ids = &head->ids;
2788         struct nvme_subsystem *subsys = head->subsys;
2789         int serial_len = sizeof(subsys->serial);
2790         int model_len = sizeof(subsys->model);
2791
2792         if (!uuid_is_null(&ids->uuid))
2793                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2794
2795         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2796                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2797
2798         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2799                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2800
2801         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2802                                   subsys->serial[serial_len - 1] == '\0'))
2803                 serial_len--;
2804         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2805                                  subsys->model[model_len - 1] == '\0'))
2806                 model_len--;
2807
2808         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2809                 serial_len, subsys->serial, model_len, subsys->model,
2810                 head->ns_id);
2811 }
2812 static DEVICE_ATTR_RO(wwid);
2813
2814 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2815                 char *buf)
2816 {
2817         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2818 }
2819 static DEVICE_ATTR_RO(nguid);
2820
2821 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2822                 char *buf)
2823 {
2824         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2825
2826         /* For backward compatibility expose the NGUID to userspace if
2827          * we have no UUID set
2828          */
2829         if (uuid_is_null(&ids->uuid)) {
2830                 printk_ratelimited(KERN_WARNING
2831                                    "No UUID available providing old NGUID\n");
2832                 return sprintf(buf, "%pU\n", ids->nguid);
2833         }
2834         return sprintf(buf, "%pU\n", &ids->uuid);
2835 }
2836 static DEVICE_ATTR_RO(uuid);
2837
2838 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2839                 char *buf)
2840 {
2841         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2842 }
2843 static DEVICE_ATTR_RO(eui);
2844
2845 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2846                 char *buf)
2847 {
2848         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2849 }
2850 static DEVICE_ATTR_RO(nsid);
2851
2852 static struct attribute *nvme_ns_id_attrs[] = {
2853         &dev_attr_wwid.attr,
2854         &dev_attr_uuid.attr,
2855         &dev_attr_nguid.attr,
2856         &dev_attr_eui.attr,
2857         &dev_attr_nsid.attr,
2858 #ifdef CONFIG_NVME_MULTIPATH
2859         &dev_attr_ana_grpid.attr,
2860         &dev_attr_ana_state.attr,
2861 #endif
2862         NULL,
2863 };
2864
2865 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2866                 struct attribute *a, int n)
2867 {
2868         struct device *dev = container_of(kobj, struct device, kobj);
2869         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2870
2871         if (a == &dev_attr_uuid.attr) {
2872                 if (uuid_is_null(&ids->uuid) &&
2873                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2874                         return 0;
2875         }
2876         if (a == &dev_attr_nguid.attr) {
2877                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2878                         return 0;
2879         }
2880         if (a == &dev_attr_eui.attr) {
2881                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2882                         return 0;
2883         }
2884 #ifdef CONFIG_NVME_MULTIPATH
2885         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2886                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2887                         return 0;
2888                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2889                         return 0;
2890         }
2891 #endif
2892         return a->mode;
2893 }
2894
2895 static const struct attribute_group nvme_ns_id_attr_group = {
2896         .attrs          = nvme_ns_id_attrs,
2897         .is_visible     = nvme_ns_id_attrs_are_visible,
2898 };
2899
2900 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2901         &nvme_ns_id_attr_group,
2902 #ifdef CONFIG_NVM
2903         &nvme_nvm_attr_group,
2904 #endif
2905         NULL,
2906 };
2907
2908 #define nvme_show_str_function(field)                                           \
2909 static ssize_t  field##_show(struct device *dev,                                \
2910                             struct device_attribute *attr, char *buf)           \
2911 {                                                                               \
2912         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2913         return sprintf(buf, "%.*s\n",                                           \
2914                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2915 }                                                                               \
2916 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2917
2918 nvme_show_str_function(model);
2919 nvme_show_str_function(serial);
2920 nvme_show_str_function(firmware_rev);
2921
2922 #define nvme_show_int_function(field)                                           \
2923 static ssize_t  field##_show(struct device *dev,                                \
2924                             struct device_attribute *attr, char *buf)           \
2925 {                                                                               \
2926         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2927         return sprintf(buf, "%d\n", ctrl->field);       \
2928 }                                                                               \
2929 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2930
2931 nvme_show_int_function(cntlid);
2932 nvme_show_int_function(numa_node);
2933
2934 static ssize_t nvme_sysfs_delete(struct device *dev,
2935                                 struct device_attribute *attr, const char *buf,
2936                                 size_t count)
2937 {
2938         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2939
2940         if (device_remove_file_self(dev, attr))
2941                 nvme_delete_ctrl_sync(ctrl);
2942         return count;
2943 }
2944 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2945
2946 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2947                                          struct device_attribute *attr,
2948                                          char *buf)
2949 {
2950         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2951
2952         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2953 }
2954 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2955
2956 static ssize_t nvme_sysfs_show_state(struct device *dev,
2957                                      struct device_attribute *attr,
2958                                      char *buf)
2959 {
2960         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2961         static const char *const state_name[] = {
2962                 [NVME_CTRL_NEW]         = "new",
2963                 [NVME_CTRL_LIVE]        = "live",
2964                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2965                 [NVME_CTRL_RESETTING]   = "resetting",
2966                 [NVME_CTRL_CONNECTING]  = "connecting",
2967                 [NVME_CTRL_DELETING]    = "deleting",
2968                 [NVME_CTRL_DEAD]        = "dead",
2969         };
2970
2971         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2972             state_name[ctrl->state])
2973                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2974
2975         return sprintf(buf, "unknown state\n");
2976 }
2977
2978 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2979
2980 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2981                                          struct device_attribute *attr,
2982                                          char *buf)
2983 {
2984         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2985
2986         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2987 }
2988 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2989
2990 static ssize_t nvme_sysfs_show_address(struct device *dev,
2991                                          struct device_attribute *attr,
2992                                          char *buf)
2993 {
2994         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2995
2996         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2997 }
2998 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2999
3000 static struct attribute *nvme_dev_attrs[] = {
3001         &dev_attr_reset_controller.attr,
3002         &dev_attr_rescan_controller.attr,
3003         &dev_attr_model.attr,
3004         &dev_attr_serial.attr,
3005         &dev_attr_firmware_rev.attr,
3006         &dev_attr_cntlid.attr,
3007         &dev_attr_delete_controller.attr,
3008         &dev_attr_transport.attr,
3009         &dev_attr_subsysnqn.attr,
3010         &dev_attr_address.attr,
3011         &dev_attr_state.attr,
3012         &dev_attr_numa_node.attr,
3013         NULL
3014 };
3015
3016 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3017                 struct attribute *a, int n)
3018 {
3019         struct device *dev = container_of(kobj, struct device, kobj);
3020         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3021
3022         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3023                 return 0;
3024         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3025                 return 0;
3026
3027         return a->mode;
3028 }
3029
3030 static struct attribute_group nvme_dev_attrs_group = {
3031         .attrs          = nvme_dev_attrs,
3032         .is_visible     = nvme_dev_attrs_are_visible,
3033 };
3034
3035 static const struct attribute_group *nvme_dev_attr_groups[] = {
3036         &nvme_dev_attrs_group,
3037         NULL,
3038 };
3039
3040 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3041                 unsigned nsid)
3042 {
3043         struct nvme_ns_head *h;
3044
3045         lockdep_assert_held(&subsys->lock);
3046
3047         list_for_each_entry(h, &subsys->nsheads, entry) {
3048                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3049                         return h;
3050         }
3051
3052         return NULL;
3053 }
3054
3055 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3056                 struct nvme_ns_head *new)
3057 {
3058         struct nvme_ns_head *h;
3059
3060         lockdep_assert_held(&subsys->lock);
3061
3062         list_for_each_entry(h, &subsys->nsheads, entry) {
3063                 if (nvme_ns_ids_valid(&new->ids) &&
3064                     !list_empty(&h->list) &&
3065                     nvme_ns_ids_equal(&new->ids, &h->ids))
3066                         return -EINVAL;
3067         }
3068
3069         return 0;
3070 }
3071
3072 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3073                 unsigned nsid, struct nvme_id_ns *id)
3074 {
3075         struct nvme_ns_head *head;
3076         size_t size = sizeof(*head);
3077         int ret = -ENOMEM;
3078
3079 #ifdef CONFIG_NVME_MULTIPATH
3080         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3081 #endif
3082
3083         head = kzalloc(size, GFP_KERNEL);
3084         if (!head)
3085                 goto out;
3086         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3087         if (ret < 0)
3088                 goto out_free_head;
3089         head->instance = ret;
3090         INIT_LIST_HEAD(&head->list);
3091         ret = init_srcu_struct(&head->srcu);
3092         if (ret)
3093                 goto out_ida_remove;
3094         head->subsys = ctrl->subsys;
3095         head->ns_id = nsid;
3096         kref_init(&head->ref);
3097
3098         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3099
3100         ret = __nvme_check_ids(ctrl->subsys, head);
3101         if (ret) {
3102                 dev_err(ctrl->device,
3103                         "duplicate IDs for nsid %d\n", nsid);
3104                 goto out_cleanup_srcu;
3105         }
3106
3107         ret = nvme_mpath_alloc_disk(ctrl, head);
3108         if (ret)
3109                 goto out_cleanup_srcu;
3110
3111         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3112
3113         kref_get(&ctrl->subsys->ref);
3114
3115         return head;
3116 out_cleanup_srcu:
3117         cleanup_srcu_struct(&head->srcu);
3118 out_ida_remove:
3119         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3120 out_free_head:
3121         kfree(head);
3122 out:
3123         return ERR_PTR(ret);
3124 }
3125
3126 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3127                 struct nvme_id_ns *id)
3128 {
3129         struct nvme_ctrl *ctrl = ns->ctrl;
3130         bool is_shared = id->nmic & (1 << 0);
3131         struct nvme_ns_head *head = NULL;
3132         int ret = 0;
3133
3134         mutex_lock(&ctrl->subsys->lock);
3135         if (is_shared)
3136                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3137         if (!head) {
3138                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3139                 if (IS_ERR(head)) {
3140                         ret = PTR_ERR(head);
3141                         goto out_unlock;
3142                 }
3143         } else {
3144                 struct nvme_ns_ids ids;
3145
3146                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3147                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3148                         dev_err(ctrl->device,
3149                                 "IDs don't match for shared namespace %d\n",
3150                                         nsid);
3151                         ret = -EINVAL;
3152                         goto out_unlock;
3153                 }
3154         }
3155
3156         list_add_tail(&ns->siblings, &head->list);
3157         ns->head = head;
3158
3159 out_unlock:
3160         mutex_unlock(&ctrl->subsys->lock);
3161         return ret;
3162 }
3163
3164 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3165 {
3166         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3167         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3168
3169         return nsa->head->ns_id - nsb->head->ns_id;
3170 }
3171
3172 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3173 {
3174         struct nvme_ns *ns, *ret = NULL;
3175
3176         down_read(&ctrl->namespaces_rwsem);
3177         list_for_each_entry(ns, &ctrl->namespaces, list) {
3178                 if (ns->head->ns_id == nsid) {
3179                         if (!kref_get_unless_zero(&ns->kref))
3180                                 continue;
3181                         ret = ns;
3182                         break;
3183                 }
3184                 if (ns->head->ns_id > nsid)
3185                         break;
3186         }
3187         up_read(&ctrl->namespaces_rwsem);
3188         return ret;
3189 }
3190
3191 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3192 {
3193         struct streams_directive_params s;
3194         int ret;
3195
3196         if (!ctrl->nr_streams)
3197                 return 0;
3198
3199         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3200         if (ret)
3201                 return ret;
3202
3203         ns->sws = le32_to_cpu(s.sws);
3204         ns->sgs = le16_to_cpu(s.sgs);
3205
3206         if (ns->sws) {
3207                 unsigned int bs = 1 << ns->lba_shift;
3208
3209                 blk_queue_io_min(ns->queue, bs * ns->sws);
3210                 if (ns->sgs)
3211                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3212         }
3213
3214         return 0;
3215 }
3216
3217 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3218 {
3219         struct nvme_ns *ns;
3220         struct gendisk *disk;
3221         struct nvme_id_ns *id;
3222         char disk_name[DISK_NAME_LEN];
3223         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3224
3225         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3226         if (!ns)
3227                 return -ENOMEM;
3228
3229         ns->queue = blk_mq_init_queue(ctrl->tagset);
3230         if (IS_ERR(ns->queue)) {
3231                 ret = PTR_ERR(ns->queue);
3232                 goto out_free_ns;
3233         }
3234
3235         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3236         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3237                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3238
3239         ns->queue->queuedata = ns;
3240         ns->ctrl = ctrl;
3241
3242         kref_init(&ns->kref);
3243         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3244
3245         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3246         nvme_set_queue_limits(ctrl, ns->queue);
3247
3248         id = nvme_identify_ns(ctrl, nsid);
3249         if (!id) {
3250                 ret = -EIO;
3251                 goto out_free_queue;
3252         }
3253
3254         if (id->ncap == 0) {
3255                 ret = -EINVAL;
3256                 goto out_free_id;
3257         }
3258
3259         ret = nvme_init_ns_head(ns, nsid, id);
3260         if (ret)
3261                 goto out_free_id;
3262         nvme_setup_streams_ns(ctrl, ns);
3263         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3264
3265         disk = alloc_disk_node(0, node);
3266         if (!disk) {
3267                 ret = -ENOMEM;
3268                 goto out_unlink_ns;
3269         }
3270
3271         disk->fops = &nvme_fops;
3272         disk->private_data = ns;
3273         disk->queue = ns->queue;
3274         disk->flags = flags;
3275         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3276         ns->disk = disk;
3277
3278         __nvme_revalidate_disk(disk, id);
3279
3280         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3281                 ret = nvme_nvm_register(ns, disk_name, node);
3282                 if (ret) {
3283                         dev_warn(ctrl->device, "LightNVM init failure\n");
3284                         goto out_put_disk;
3285                 }
3286         }
3287
3288         down_write(&ctrl->namespaces_rwsem);
3289         list_add_tail(&ns->list, &ctrl->namespaces);
3290         up_write(&ctrl->namespaces_rwsem);
3291
3292         nvme_get_ctrl(ctrl);
3293
3294         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3295
3296         nvme_mpath_add_disk(ns, id);
3297         nvme_fault_inject_init(ns);
3298         kfree(id);
3299
3300         return 0;
3301  out_put_disk:
3302         put_disk(ns->disk);
3303  out_unlink_ns:
3304         mutex_lock(&ctrl->subsys->lock);
3305         list_del_rcu(&ns->siblings);
3306         mutex_unlock(&ctrl->subsys->lock);
3307  out_free_id:
3308         kfree(id);
3309  out_free_queue:
3310         blk_cleanup_queue(ns->queue);
3311  out_free_ns:
3312         kfree(ns);
3313         return ret;
3314 }
3315
3316 static void nvme_ns_remove(struct nvme_ns *ns)
3317 {
3318         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3319                 return;
3320
3321         nvme_fault_inject_fini(ns);
3322         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3323                 del_gendisk(ns->disk);
3324                 blk_cleanup_queue(ns->queue);
3325                 if (blk_get_integrity(ns->disk))
3326                         blk_integrity_unregister(ns->disk);
3327         }
3328
3329         mutex_lock(&ns->ctrl->subsys->lock);
3330         list_del_rcu(&ns->siblings);
3331         nvme_mpath_clear_current_path(ns);
3332         mutex_unlock(&ns->ctrl->subsys->lock);
3333
3334         down_write(&ns->ctrl->namespaces_rwsem);
3335         list_del_init(&ns->list);
3336         up_write(&ns->ctrl->namespaces_rwsem);
3337
3338         synchronize_srcu(&ns->head->srcu);
3339         nvme_mpath_check_last_path(ns);
3340         nvme_put_ns(ns);
3341 }
3342
3343 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3344 {
3345         struct nvme_ns *ns;
3346
3347         ns = nvme_find_get_ns(ctrl, nsid);
3348         if (ns) {
3349                 if (ns->disk && revalidate_disk(ns->disk))
3350                         nvme_ns_remove(ns);
3351                 nvme_put_ns(ns);
3352         } else
3353                 nvme_alloc_ns(ctrl, nsid);
3354 }
3355
3356 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3357                                         unsigned nsid)
3358 {
3359         struct nvme_ns *ns, *next;
3360         LIST_HEAD(rm_list);
3361
3362         down_write(&ctrl->namespaces_rwsem);
3363         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3364                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3365                         list_move_tail(&ns->list, &rm_list);
3366         }
3367         up_write(&ctrl->namespaces_rwsem);
3368
3369         list_for_each_entry_safe(ns, next, &rm_list, list)
3370                 nvme_ns_remove(ns);
3371
3372 }
3373
3374 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3375 {
3376         struct nvme_ns *ns;
3377         __le32 *ns_list;
3378         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3379         int ret = 0;
3380
3381         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3382         if (!ns_list)
3383                 return -ENOMEM;
3384
3385         for (i = 0; i < num_lists; i++) {
3386                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3387                 if (ret)
3388                         goto free;
3389
3390                 for (j = 0; j < min(nn, 1024U); j++) {
3391                         nsid = le32_to_cpu(ns_list[j]);
3392                         if (!nsid)
3393                                 goto out;
3394
3395                         nvme_validate_ns(ctrl, nsid);
3396
3397                         while (++prev < nsid) {
3398                                 ns = nvme_find_get_ns(ctrl, prev);
3399                                 if (ns) {
3400                                         nvme_ns_remove(ns);
3401                                         nvme_put_ns(ns);
3402                                 }
3403                         }
3404                 }
3405                 nn -= j;
3406         }
3407  out:
3408         nvme_remove_invalid_namespaces(ctrl, prev);
3409  free:
3410         kfree(ns_list);
3411         return ret;
3412 }
3413
3414 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3415 {
3416         unsigned i;
3417
3418         for (i = 1; i <= nn; i++)
3419                 nvme_validate_ns(ctrl, i);
3420
3421         nvme_remove_invalid_namespaces(ctrl, nn);
3422 }
3423
3424 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3425 {
3426         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3427         __le32 *log;
3428         int error;
3429
3430         log = kzalloc(log_size, GFP_KERNEL);
3431         if (!log)
3432                 return;
3433
3434         /*
3435          * We need to read the log to clear the AEN, but we don't want to rely
3436          * on it for the changed namespace information as userspace could have
3437          * raced with us in reading the log page, which could cause us to miss
3438          * updates.
3439          */
3440         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3441                         log_size, 0);
3442         if (error)
3443                 dev_warn(ctrl->device,
3444                         "reading changed ns log failed: %d\n", error);
3445
3446         kfree(log);
3447 }
3448
3449 static void nvme_scan_work(struct work_struct *work)
3450 {
3451         struct nvme_ctrl *ctrl =
3452                 container_of(work, struct nvme_ctrl, scan_work);
3453         struct nvme_id_ctrl *id;
3454         unsigned nn;
3455
3456         if (ctrl->state != NVME_CTRL_LIVE)
3457                 return;
3458
3459         WARN_ON_ONCE(!ctrl->tagset);
3460
3461         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3462                 dev_info(ctrl->device, "rescanning namespaces.\n");
3463                 nvme_clear_changed_ns_log(ctrl);
3464         }
3465
3466         if (nvme_identify_ctrl(ctrl, &id))
3467                 return;
3468
3469         mutex_lock(&ctrl->scan_lock);
3470         nn = le32_to_cpu(id->nn);
3471         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3472             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3473                 if (!nvme_scan_ns_list(ctrl, nn))
3474                         goto out_free_id;
3475         }
3476         nvme_scan_ns_sequential(ctrl, nn);
3477 out_free_id:
3478         mutex_unlock(&ctrl->scan_lock);
3479         kfree(id);
3480         down_write(&ctrl->namespaces_rwsem);
3481         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3482         up_write(&ctrl->namespaces_rwsem);
3483 }
3484
3485 /*
3486  * This function iterates the namespace list unlocked to allow recovery from
3487  * controller failure. It is up to the caller to ensure the namespace list is
3488  * not modified by scan work while this function is executing.
3489  */
3490 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3491 {
3492         struct nvme_ns *ns, *next;
3493         LIST_HEAD(ns_list);
3494
3495         /* prevent racing with ns scanning */
3496         flush_work(&ctrl->scan_work);
3497
3498         /*
3499          * The dead states indicates the controller was not gracefully
3500          * disconnected. In that case, we won't be able to flush any data while
3501          * removing the namespaces' disks; fail all the queues now to avoid
3502          * potentially having to clean up the failed sync later.
3503          */
3504         if (ctrl->state == NVME_CTRL_DEAD)
3505                 nvme_kill_queues(ctrl);
3506
3507         down_write(&ctrl->namespaces_rwsem);
3508         list_splice_init(&ctrl->namespaces, &ns_list);
3509         up_write(&ctrl->namespaces_rwsem);
3510
3511         list_for_each_entry_safe(ns, next, &ns_list, list)
3512                 nvme_ns_remove(ns);
3513 }
3514 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3515
3516 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3517 {
3518         char *envp[2] = { NULL, NULL };
3519         u32 aen_result = ctrl->aen_result;
3520
3521         ctrl->aen_result = 0;
3522         if (!aen_result)
3523                 return;
3524
3525         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3526         if (!envp[0])
3527                 return;
3528         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3529         kfree(envp[0]);
3530 }
3531
3532 static void nvme_async_event_work(struct work_struct *work)
3533 {
3534         struct nvme_ctrl *ctrl =
3535                 container_of(work, struct nvme_ctrl, async_event_work);
3536
3537         nvme_aen_uevent(ctrl);
3538         ctrl->ops->submit_async_event(ctrl);
3539 }
3540
3541 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3542 {
3543
3544         u32 csts;
3545
3546         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3547                 return false;
3548
3549         if (csts == ~0)
3550                 return false;
3551
3552         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3553 }
3554
3555 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3556 {
3557         struct nvme_fw_slot_info_log *log;
3558
3559         log = kmalloc(sizeof(*log), GFP_KERNEL);
3560         if (!log)
3561                 return;
3562
3563         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3564                         sizeof(*log), 0))
3565                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3566         kfree(log);
3567 }
3568
3569 static void nvme_fw_act_work(struct work_struct *work)
3570 {
3571         struct nvme_ctrl *ctrl = container_of(work,
3572                                 struct nvme_ctrl, fw_act_work);
3573         unsigned long fw_act_timeout;
3574
3575         if (ctrl->mtfa)
3576                 fw_act_timeout = jiffies +
3577                                 msecs_to_jiffies(ctrl->mtfa * 100);
3578         else
3579                 fw_act_timeout = jiffies +
3580                                 msecs_to_jiffies(admin_timeout * 1000);
3581
3582         nvme_stop_queues(ctrl);
3583         while (nvme_ctrl_pp_status(ctrl)) {
3584                 if (time_after(jiffies, fw_act_timeout)) {
3585                         dev_warn(ctrl->device,
3586                                 "Fw activation timeout, reset controller\n");
3587                         nvme_reset_ctrl(ctrl);
3588                         break;
3589                 }
3590                 msleep(100);
3591         }
3592
3593         if (ctrl->state != NVME_CTRL_LIVE)
3594                 return;
3595
3596         nvme_start_queues(ctrl);
3597         /* read FW slot information to clear the AER */
3598         nvme_get_fw_slot_info(ctrl);
3599 }
3600
3601 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3602 {
3603         u32 aer_notice_type = (result & 0xff00) >> 8;
3604
3605         switch (aer_notice_type) {
3606         case NVME_AER_NOTICE_NS_CHANGED:
3607                 trace_nvme_async_event(ctrl, aer_notice_type);
3608                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3609                 nvme_queue_scan(ctrl);
3610                 break;
3611         case NVME_AER_NOTICE_FW_ACT_STARTING:
3612                 trace_nvme_async_event(ctrl, aer_notice_type);
3613                 queue_work(nvme_wq, &ctrl->fw_act_work);
3614                 break;
3615 #ifdef CONFIG_NVME_MULTIPATH
3616         case NVME_AER_NOTICE_ANA:
3617                 trace_nvme_async_event(ctrl, aer_notice_type);
3618                 if (!ctrl->ana_log_buf)
3619                         break;
3620                 queue_work(nvme_wq, &ctrl->ana_work);
3621                 break;
3622 #endif
3623         default:
3624                 dev_warn(ctrl->device, "async event result %08x\n", result);
3625         }
3626 }
3627
3628 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3629                 volatile union nvme_result *res)
3630 {
3631         u32 result = le32_to_cpu(res->u32);
3632         u32 aer_type = result & 0x07;
3633
3634         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3635                 return;
3636
3637         switch (aer_type) {
3638         case NVME_AER_NOTICE:
3639                 nvme_handle_aen_notice(ctrl, result);
3640                 break;
3641         case NVME_AER_ERROR:
3642         case NVME_AER_SMART:
3643         case NVME_AER_CSS:
3644         case NVME_AER_VS:
3645                 trace_nvme_async_event(ctrl, aer_type);
3646                 ctrl->aen_result = result;
3647                 break;
3648         default:
3649                 break;
3650         }
3651         queue_work(nvme_wq, &ctrl->async_event_work);
3652 }
3653 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3654
3655 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3656 {
3657         nvme_mpath_stop(ctrl);
3658         nvme_stop_keep_alive(ctrl);
3659         flush_work(&ctrl->async_event_work);
3660         cancel_work_sync(&ctrl->fw_act_work);
3661 }
3662 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3663
3664 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3665 {
3666         if (ctrl->kato)
3667                 nvme_start_keep_alive(ctrl);
3668
3669         if (ctrl->queue_count > 1) {
3670                 nvme_queue_scan(ctrl);
3671                 nvme_enable_aen(ctrl);
3672                 queue_work(nvme_wq, &ctrl->async_event_work);
3673                 nvme_start_queues(ctrl);
3674         }
3675 }
3676 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3677
3678 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3679 {
3680         cdev_device_del(&ctrl->cdev, ctrl->device);
3681 }
3682 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3683
3684 static void nvme_free_ctrl(struct device *dev)
3685 {
3686         struct nvme_ctrl *ctrl =
3687                 container_of(dev, struct nvme_ctrl, ctrl_device);
3688         struct nvme_subsystem *subsys = ctrl->subsys;
3689
3690         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3691         kfree(ctrl->effects);
3692         nvme_mpath_uninit(ctrl);
3693         __free_page(ctrl->discard_page);
3694
3695         if (subsys) {
3696                 mutex_lock(&subsys->lock);
3697                 list_del(&ctrl->subsys_entry);
3698                 mutex_unlock(&subsys->lock);
3699                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3700         }
3701
3702         ctrl->ops->free_ctrl(ctrl);
3703
3704         if (subsys)
3705                 nvme_put_subsystem(subsys);
3706 }
3707
3708 /*
3709  * Initialize a NVMe controller structures.  This needs to be called during
3710  * earliest initialization so that we have the initialized structured around
3711  * during probing.
3712  */
3713 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3714                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3715 {
3716         int ret;
3717
3718         ctrl->state = NVME_CTRL_NEW;
3719         spin_lock_init(&ctrl->lock);
3720         mutex_init(&ctrl->scan_lock);
3721         INIT_LIST_HEAD(&ctrl->namespaces);
3722         init_rwsem(&ctrl->namespaces_rwsem);
3723         ctrl->dev = dev;
3724         ctrl->ops = ops;
3725         ctrl->quirks = quirks;
3726         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3727         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3728         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3729         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3730
3731         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3732         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3733         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3734
3735         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3736                         PAGE_SIZE);
3737         ctrl->discard_page = alloc_page(GFP_KERNEL);
3738         if (!ctrl->discard_page) {
3739                 ret = -ENOMEM;
3740                 goto out;
3741         }
3742
3743         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3744         if (ret < 0)
3745                 goto out;
3746         ctrl->instance = ret;
3747
3748         device_initialize(&ctrl->ctrl_device);
3749         ctrl->device = &ctrl->ctrl_device;
3750         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3751         ctrl->device->class = nvme_class;
3752         ctrl->device->parent = ctrl->dev;
3753         ctrl->device->groups = nvme_dev_attr_groups;
3754         ctrl->device->release = nvme_free_ctrl;
3755         dev_set_drvdata(ctrl->device, ctrl);
3756         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3757         if (ret)
3758                 goto out_release_instance;
3759
3760         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3761         ctrl->cdev.owner = ops->module;
3762         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3763         if (ret)
3764                 goto out_free_name;
3765
3766         /*
3767          * Initialize latency tolerance controls.  The sysfs files won't
3768          * be visible to userspace unless the device actually supports APST.
3769          */
3770         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3771         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3772                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3773
3774         return 0;
3775 out_free_name:
3776         kfree_const(ctrl->device->kobj.name);
3777 out_release_instance:
3778         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3779 out:
3780         if (ctrl->discard_page)
3781                 __free_page(ctrl->discard_page);
3782         return ret;
3783 }
3784 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3785
3786 /**
3787  * nvme_kill_queues(): Ends all namespace queues
3788  * @ctrl: the dead controller that needs to end
3789  *
3790  * Call this function when the driver determines it is unable to get the
3791  * controller in a state capable of servicing IO.
3792  */
3793 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3794 {
3795         struct nvme_ns *ns;
3796
3797         down_read(&ctrl->namespaces_rwsem);
3798
3799         /* Forcibly unquiesce queues to avoid blocking dispatch */
3800         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3801                 blk_mq_unquiesce_queue(ctrl->admin_q);
3802
3803         list_for_each_entry(ns, &ctrl->namespaces, list)
3804                 nvme_set_queue_dying(ns);
3805
3806         up_read(&ctrl->namespaces_rwsem);
3807 }
3808 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3809
3810 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3811 {
3812         struct nvme_ns *ns;
3813
3814         down_read(&ctrl->namespaces_rwsem);
3815         list_for_each_entry(ns, &ctrl->namespaces, list)
3816                 blk_mq_unfreeze_queue(ns->queue);
3817         up_read(&ctrl->namespaces_rwsem);
3818 }
3819 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3820
3821 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3822 {
3823         struct nvme_ns *ns;
3824
3825         down_read(&ctrl->namespaces_rwsem);
3826         list_for_each_entry(ns, &ctrl->namespaces, list) {
3827                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3828                 if (timeout <= 0)
3829                         break;
3830         }
3831         up_read(&ctrl->namespaces_rwsem);
3832 }
3833 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3834
3835 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3836 {
3837         struct nvme_ns *ns;
3838
3839         down_read(&ctrl->namespaces_rwsem);
3840         list_for_each_entry(ns, &ctrl->namespaces, list)
3841                 blk_mq_freeze_queue_wait(ns->queue);
3842         up_read(&ctrl->namespaces_rwsem);
3843 }
3844 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3845
3846 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3847 {
3848         struct nvme_ns *ns;
3849
3850         down_read(&ctrl->namespaces_rwsem);
3851         list_for_each_entry(ns, &ctrl->namespaces, list)
3852                 blk_freeze_queue_start(ns->queue);
3853         up_read(&ctrl->namespaces_rwsem);
3854 }
3855 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3856
3857 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3858 {
3859         struct nvme_ns *ns;
3860
3861         down_read(&ctrl->namespaces_rwsem);
3862         list_for_each_entry(ns, &ctrl->namespaces, list)
3863                 blk_mq_quiesce_queue(ns->queue);
3864         up_read(&ctrl->namespaces_rwsem);
3865 }
3866 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3867
3868 void nvme_start_queues(struct nvme_ctrl *ctrl)
3869 {
3870         struct nvme_ns *ns;
3871
3872         down_read(&ctrl->namespaces_rwsem);
3873         list_for_each_entry(ns, &ctrl->namespaces, list)
3874                 blk_mq_unquiesce_queue(ns->queue);
3875         up_read(&ctrl->namespaces_rwsem);
3876 }
3877 EXPORT_SYMBOL_GPL(nvme_start_queues);
3878
3879 int __init nvme_core_init(void)
3880 {
3881         int result = -ENOMEM;
3882
3883         nvme_wq = alloc_workqueue("nvme-wq",
3884                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3885         if (!nvme_wq)
3886                 goto out;
3887
3888         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3889                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3890         if (!nvme_reset_wq)
3891                 goto destroy_wq;
3892
3893         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3894                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3895         if (!nvme_delete_wq)
3896                 goto destroy_reset_wq;
3897
3898         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3899         if (result < 0)
3900                 goto destroy_delete_wq;
3901
3902         nvme_class = class_create(THIS_MODULE, "nvme");
3903         if (IS_ERR(nvme_class)) {
3904                 result = PTR_ERR(nvme_class);
3905                 goto unregister_chrdev;
3906         }
3907
3908         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3909         if (IS_ERR(nvme_subsys_class)) {
3910                 result = PTR_ERR(nvme_subsys_class);
3911                 goto destroy_class;
3912         }
3913         return 0;
3914
3915 destroy_class:
3916         class_destroy(nvme_class);
3917 unregister_chrdev:
3918         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3919 destroy_delete_wq:
3920         destroy_workqueue(nvme_delete_wq);
3921 destroy_reset_wq:
3922         destroy_workqueue(nvme_reset_wq);
3923 destroy_wq:
3924         destroy_workqueue(nvme_wq);
3925 out:
3926         return result;
3927 }
3928
3929 void __exit nvme_core_exit(void)
3930 {
3931         ida_destroy(&nvme_subsystems_ida);
3932         class_destroy(nvme_subsys_class);
3933         class_destroy(nvme_class);
3934         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3935         destroy_workqueue(nvme_delete_wq);
3936         destroy_workqueue(nvme_reset_wq);
3937         destroy_workqueue(nvme_wq);
3938 }
3939
3940 MODULE_LICENSE("GPL");
3941 MODULE_VERSION("1.0");
3942 module_init(nvme_core_init);
3943 module_exit(nvme_core_exit);