drivers/nvme/host/core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * NVM Express device driver
   4  * Copyright (c) 2011-2014, Intel Corporation.
   5  */
   6
   7 #include <linux/blkdev.h>
   8 #include <linux/blk-mq.h>
   9 #include <linux/delay.h>
  10 #include <linux/errno.h>
  11 #include <linux/hdreg.h>
  12 #include <linux/kernel.h>
  13 #include <linux/module.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/list_sort.h>
  16 #include <linux/slab.h>
  17 #include <linux/types.h>
  18 #include <linux/pr.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/nvme_ioctl.h>
  21 #include <linux/t10-pi.h>
  22 #include <linux/pm_qos.h>
  23 #include <asm/unaligned.h>
  24
  25 #include "nvme.h"
  26 #include "fabrics.h"
  27
  28 #define CREATE_TRACE_POINTS
  29 #include "trace.h"
  30
  31 #define NVME_MINORS             (1U << MINORBITS)
  32
  33 unsigned int admin_timeout = 60;
  34 module_param(admin_timeout, uint, 0644);
  35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  36 EXPORT_SYMBOL_GPL(admin_timeout);
  37
  38 unsigned int nvme_io_timeout = 30;
  39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  42
  43 static unsigned char shutdown_timeout = 5;
  44 module_param(shutdown_timeout, byte, 0644);
  45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  46
  47 static u8 nvme_max_retries = 5;
  48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  50
  51 static unsigned long default_ps_max_latency_us = 100000;
  52 module_param(default_ps_max_latency_us, ulong, 0644);
  53 MODULE_PARM_DESC(default_ps_max_latency_us,
  54                  "max power saving latency for new devices; use PM QOS to change per device");
  55
  56 static bool force_apst;
  57 module_param(force_apst, bool, 0644);
  58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  59
  60 static bool streams;
  61 module_param(streams, bool, 0644);
  62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  63
  64 /*
  65  * nvme_wq - hosts nvme related works that are not reset or delete
  66  * nvme_reset_wq - hosts nvme reset works
  67  * nvme_delete_wq - hosts nvme delete works
  68  *
  69  * nvme_wq will host works such are scan, aen handling, fw activation,
  70  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
  71  * runs reset works which also flush works hosted on nvme_wq for
  72  * serialization purposes. nvme_delete_wq host controller deletion
  73  * works which flush reset works for serialization.
  74  */
  75 struct workqueue_struct *nvme_wq;
  76 EXPORT_SYMBOL_GPL(nvme_wq);
  77
  78 struct workqueue_struct *nvme_reset_wq;
  79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  80
  81 struct workqueue_struct *nvme_delete_wq;
  82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  83
  84 static LIST_HEAD(nvme_subsystems);
  85 static DEFINE_MUTEX(nvme_subsystems_lock);
  86
  87 static DEFINE_IDA(nvme_instance_ida);
  88 static dev_t nvme_chr_devt;
  89 static struct class *nvme_class;
  90 static struct class *nvme_subsys_class;
  91
  92 static int nvme_revalidate_disk(struct gendisk *disk);
  93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  95                                            unsigned nsid);
  96
  97 static void nvme_set_queue_dying(struct nvme_ns *ns)
  98 {
  99         /*
 100          * Revalidating a dead namespace sets capacity to 0. This will end
 101          * buffered writers dirtying pages that can't be synced.
 102          */
 103         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 104                 return;
 105         blk_set_queue_dying(ns->queue);
 106         /* Forcibly unquiesce queues to avoid blocking dispatch */
 107         blk_mq_unquiesce_queue(ns->queue);
 108         /*
 109          * Revalidate after unblocking dispatchers that may be holding bd_butex
 110          */
 111         revalidate_disk(ns->disk);
 112 }
 113
 114 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 115 {
 116         /*
 117          * Only new queue scan work when admin and IO queues are both alive
 118          */
 119         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 120                 queue_work(nvme_wq, &ctrl->scan_work);
 121 }
 122
 123 /*
 124  * Use this function to proceed with scheduling reset_work for a controller
 125  * that had previously been set to the resetting state. This is intended for
 126  * code paths that can't be interrupted by other reset attempts. A hot removal
 127  * may prevent this from succeeding.
 128  */
 129 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 130 {
 131         if (ctrl->state != NVME_CTRL_RESETTING)
 132                 return -EBUSY;
 133         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 134                 return -EBUSY;
 135         return 0;
 136 }
 137 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 138
 139 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 140 {
 141         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 142                 return -EBUSY;
 143         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 144                 return -EBUSY;
 145         return 0;
 146 }
 147 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 148
 149 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 150 {
 151         int ret;
 152
 153         ret = nvme_reset_ctrl(ctrl);
 154         if (!ret) {
 155                 flush_work(&ctrl->reset_work);
 156                 if (ctrl->state != NVME_CTRL_LIVE)
 157                         ret = -ENETRESET;
 158         }
 159
 160         return ret;
 161 }
 162 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 163
 164 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 165 {
 166         dev_info(ctrl->device,
 167                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 168
 169         flush_work(&ctrl->reset_work);
 170         nvme_stop_ctrl(ctrl);
 171         nvme_remove_namespaces(ctrl);
 172         ctrl->ops->delete_ctrl(ctrl);
 173         nvme_uninit_ctrl(ctrl);
 174         nvme_put_ctrl(ctrl);
 175 }
 176
 177 static void nvme_delete_ctrl_work(struct work_struct *work)
 178 {
 179         struct nvme_ctrl *ctrl =
 180                 container_of(work, struct nvme_ctrl, delete_work);
 181
 182         nvme_do_delete_ctrl(ctrl);
 183 }
 184
 185 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 186 {
 187         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 188                 return -EBUSY;
 189         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 190                 return -EBUSY;
 191         return 0;
 192 }
 193 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 194
 195 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 196 {
 197         int ret = 0;
 198
 199         /*
 200          * Keep a reference until nvme_do_delete_ctrl() complete,
 201          * since ->delete_ctrl can free the controller.
 202          */
 203         nvme_get_ctrl(ctrl);
 204         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 205                 ret = -EBUSY;
 206         if (!ret)
 207                 nvme_do_delete_ctrl(ctrl);
 208         nvme_put_ctrl(ctrl);
 209         return ret;
 210 }
 211
 212 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 213 {
 214         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 215 }
 216
 217 static blk_status_t nvme_error_status(u16 status)
 218 {
 219         switch (status & 0x7ff) {
 220         case NVME_SC_SUCCESS:
 221                 return BLK_STS_OK;
 222         case NVME_SC_CAP_EXCEEDED:
 223                 return BLK_STS_NOSPC;
 224         case NVME_SC_LBA_RANGE:
 225         case NVME_SC_CMD_INTERRUPTED:
 226         case NVME_SC_NS_NOT_READY:
 227                 return BLK_STS_TARGET;
 228         case NVME_SC_BAD_ATTRIBUTES:
 229         case NVME_SC_ONCS_NOT_SUPPORTED:
 230         case NVME_SC_INVALID_OPCODE:
 231         case NVME_SC_INVALID_FIELD:
 232         case NVME_SC_INVALID_NS:
 233                 return BLK_STS_NOTSUPP;
 234         case NVME_SC_WRITE_FAULT:
 235         case NVME_SC_READ_ERROR:
 236         case NVME_SC_UNWRITTEN_BLOCK:
 237         case NVME_SC_ACCESS_DENIED:
 238         case NVME_SC_READ_ONLY:
 239         case NVME_SC_COMPARE_FAILED:
 240                 return BLK_STS_MEDIUM;
 241         case NVME_SC_GUARD_CHECK:
 242         case NVME_SC_APPTAG_CHECK:
 243         case NVME_SC_REFTAG_CHECK:
 244         case NVME_SC_INVALID_PI:
 245                 return BLK_STS_PROTECTION;
 246         case NVME_SC_RESERVATION_CONFLICT:
 247                 return BLK_STS_NEXUS;
 248         case NVME_SC_HOST_PATH_ERROR:
 249                 return BLK_STS_TRANSPORT;
 250         default:
 251                 return BLK_STS_IOERR;
 252         }
 253 }
 254
 255 static inline bool nvme_req_needs_retry(struct request *req)
 256 {
 257         if (blk_noretry_request(req))
 258                 return false;
 259         if (nvme_req(req)->status & NVME_SC_DNR)
 260                 return false;
 261         if (nvme_req(req)->retries >= nvme_max_retries)
 262                 return false;
 263         return true;
 264 }
 265
 266 static void nvme_retry_req(struct request *req)
 267 {
 268         struct nvme_ns *ns = req->q->queuedata;
 269         unsigned long delay = 0;
 270         u16 crd;
 271
 272         /* The mask and shift result must be <= 3 */
 273         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 274         if (ns && crd)
 275                 delay = ns->ctrl->crdt[crd - 1] * 100;
 276
 277         nvme_req(req)->retries++;
 278         blk_mq_requeue_request(req, false);
 279         blk_mq_delay_kick_requeue_list(req->q, delay);
 280 }
 281
 282 void nvme_complete_rq(struct request *req)
 283 {
 284         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 285
 286         trace_nvme_complete_rq(req);
 287
 288         nvme_cleanup_cmd(req);
 289
 290         if (nvme_req(req)->ctrl->kas)
 291                 nvme_req(req)->ctrl->comp_seen = true;
 292
 293         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 294                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
 295                     blk_path_error(status)) {
 296                         nvme_failover_req(req);
 297                         return;
 298                 }
 299
 300                 if (!blk_queue_dying(req->q)) {
 301                         nvme_retry_req(req);
 302                         return;
 303                 }
 304         }
 305
 306         nvme_trace_bio_complete(req, status);
 307         blk_mq_end_request(req, status);
 308 }
 309 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 310
 311 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 312 {
 313         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 314                                 "Cancelling I/O %d", req->tag);
 315
 316         /* don't abort one completed request */
 317         if (blk_mq_request_completed(req))
 318                 return true;
 319
 320         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 321         blk_mq_complete_request(req);
 322         return true;
 323 }
 324 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 325
 326 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 327                 enum nvme_ctrl_state new_state)
 328 {
 329         enum nvme_ctrl_state old_state;
 330         unsigned long flags;
 331         bool changed = false;
 332
 333         spin_lock_irqsave(&ctrl->lock, flags);
 334
 335         old_state = ctrl->state;
 336         switch (new_state) {
 337         case NVME_CTRL_LIVE:
 338                 switch (old_state) {
 339                 case NVME_CTRL_NEW:
 340                 case NVME_CTRL_RESETTING:
 341                 case NVME_CTRL_CONNECTING:
 342                         changed = true;
 343                         /* FALLTHRU */
 344                 default:
 345                         break;
 346                 }
 347                 break;
 348         case NVME_CTRL_RESETTING:
 349                 switch (old_state) {
 350                 case NVME_CTRL_NEW:
 351                 case NVME_CTRL_LIVE:
 352                         changed = true;
 353                         /* FALLTHRU */
 354                 default:
 355                         break;
 356                 }
 357                 break;
 358         case NVME_CTRL_CONNECTING:
 359                 switch (old_state) {
 360                 case NVME_CTRL_NEW:
 361                 case NVME_CTRL_RESETTING:
 362                         changed = true;
 363                         /* FALLTHRU */
 364                 default:
 365                         break;
 366                 }
 367                 break;
 368         case NVME_CTRL_DELETING:
 369                 switch (old_state) {
 370                 case NVME_CTRL_LIVE:
 371                 case NVME_CTRL_RESETTING:
 372                 case NVME_CTRL_CONNECTING:
 373                         changed = true;
 374                         /* FALLTHRU */
 375                 default:
 376                         break;
 377                 }
 378                 break;
 379         case NVME_CTRL_DEAD:
 380                 switch (old_state) {
 381                 case NVME_CTRL_DELETING:
 382                         changed = true;
 383                         /* FALLTHRU */
 384                 default:
 385                         break;
 386                 }
 387                 break;
 388         default:
 389                 break;
 390         }
 391
 392         if (changed) {
 393                 ctrl->state = new_state;
 394                 wake_up_all(&ctrl->state_wq);
 395         }
 396
 397         spin_unlock_irqrestore(&ctrl->lock, flags);
 398         if (changed && ctrl->state == NVME_CTRL_LIVE)
 399                 nvme_kick_requeue_lists(ctrl);
 400         return changed;
 401 }
 402 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 403
 404 /*
 405  * Returns true for sink states that can't ever transition back to live.
 406  */
 407 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 408 {
 409         switch (ctrl->state) {
 410         case NVME_CTRL_NEW:
 411         case NVME_CTRL_LIVE:
 412         case NVME_CTRL_RESETTING:
 413         case NVME_CTRL_CONNECTING:
 414                 return false;
 415         case NVME_CTRL_DELETING:
 416         case NVME_CTRL_DEAD:
 417                 return true;
 418         default:
 419                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 420                 return true;
 421         }
 422 }
 423
 424 /*
 425  * Waits for the controller state to be resetting, or returns false if it is
 426  * not possible to ever transition to that state.
 427  */
 428 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 429 {
 430         wait_event(ctrl->state_wq,
 431                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 432                    nvme_state_terminal(ctrl));
 433         return ctrl->state == NVME_CTRL_RESETTING;
 434 }
 435 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 436
 437 static void nvme_free_ns_head(struct kref *ref)
 438 {
 439         struct nvme_ns_head *head =
 440                 container_of(ref, struct nvme_ns_head, ref);
 441
 442         nvme_mpath_remove_disk(head);
 443         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 444         list_del_init(&head->entry);
 445         cleanup_srcu_struct(&head->srcu);
 446         nvme_put_subsystem(head->subsys);
 447         kfree(head);
 448 }
 449
 450 static void nvme_put_ns_head(struct nvme_ns_head *head)
 451 {
 452         kref_put(&head->ref, nvme_free_ns_head);
 453 }
 454
 455 static void nvme_free_ns(struct kref *kref)
 456 {
 457         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 458
 459         if (ns->ndev)
 460                 nvme_nvm_unregister(ns);
 461
 462         put_disk(ns->disk);
 463         nvme_put_ns_head(ns->head);
 464         nvme_put_ctrl(ns->ctrl);
 465         kfree(ns);
 466 }
 467
 468 static void nvme_put_ns(struct nvme_ns *ns)
 469 {
 470         kref_put(&ns->kref, nvme_free_ns);
 471 }
 472
 473 static inline void nvme_clear_nvme_request(struct request *req)
 474 {
 475         if (!(req->rq_flags & RQF_DONTPREP)) {
 476                 nvme_req(req)->retries = 0;
 477                 nvme_req(req)->flags = 0;
 478                 req->rq_flags |= RQF_DONTPREP;
 479         }
 480 }
 481
 482 struct request *nvme_alloc_request(struct request_queue *q,
 483                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 484 {
 485         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 486         struct request *req;
 487
 488         if (qid == NVME_QID_ANY) {
 489                 req = blk_mq_alloc_request(q, op, flags);
 490         } else {
 491                 req = blk_mq_alloc_request_hctx(q, op, flags,
 492                                 qid ? qid - 1 : 0);
 493         }
 494         if (IS_ERR(req))
 495                 return req;
 496
 497         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 498         nvme_clear_nvme_request(req);
 499         nvme_req(req)->cmd = cmd;
 500
 501         return req;
 502 }
 503 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 504
 505 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 506 {
 507         struct nvme_command c;
 508
 509         memset(&c, 0, sizeof(c));
 510
 511         c.directive.opcode = nvme_admin_directive_send;
 512         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 513         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 514         c.directive.dtype = NVME_DIR_IDENTIFY;
 515         c.directive.tdtype = NVME_DIR_STREAMS;
 516         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 517
 518         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 519 }
 520
 521 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 522 {
 523         return nvme_toggle_streams(ctrl, false);
 524 }
 525
 526 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 527 {
 528         return nvme_toggle_streams(ctrl, true);
 529 }
 530
 531 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 532                                   struct streams_directive_params *s, u32 nsid)
 533 {
 534         struct nvme_command c;
 535
 536         memset(&c, 0, sizeof(c));
 537         memset(s, 0, sizeof(*s));
 538
 539         c.directive.opcode = nvme_admin_directive_recv;
 540         c.directive.nsid = cpu_to_le32(nsid);
 541         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 542         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 543         c.directive.dtype = NVME_DIR_STREAMS;
 544
 545         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 546 }
 547
 548 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 549 {
 550         struct streams_directive_params s;
 551         int ret;
 552
 553         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 554                 return 0;
 555         if (!streams)
 556                 return 0;
 557
 558         ret = nvme_enable_streams(ctrl);
 559         if (ret)
 560                 return ret;
 561
 562         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 563         if (ret)
 564                 return ret;
 565
 566         ctrl->nssa = le16_to_cpu(s.nssa);
 567         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 568                 dev_info(ctrl->device, "too few streams (%u) available\n",
 569                                         ctrl->nssa);
 570                 nvme_disable_streams(ctrl);
 571                 return 0;
 572         }
 573
 574         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 575         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 576         return 0;
 577 }
 578
 579 /*
 580  * Check if 'req' has a write hint associated with it. If it does, assign
 581  * a valid namespace stream to the write.
 582  */
 583 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 584                                      struct request *req, u16 *control,
 585                                      u32 *dsmgmt)
 586 {
 587         enum rw_hint streamid = req->write_hint;
 588
 589         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 590                 streamid = 0;
 591         else {
 592                 streamid--;
 593                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 594                         return;
 595
 596                 *control |= NVME_RW_DTYPE_STREAMS;
 597                 *dsmgmt |= streamid << 16;
 598         }
 599
 600         if (streamid < ARRAY_SIZE(req->q->write_hints))
 601                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 602 }
 603
 604 static inline void nvme_setup_flush(struct nvme_ns *ns,
 605                 struct nvme_command *cmnd)
 606 {
 607         cmnd->common.opcode = nvme_cmd_flush;
 608         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 609 }
 610
 611 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 612                 struct nvme_command *cmnd)
 613 {
 614         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 615         struct nvme_dsm_range *range;
 616         struct bio *bio;
 617
 618         /*
 619          * Some devices do not consider the DSM 'Number of Ranges' field when
 620          * determining how much data to DMA. Always allocate memory for maximum
 621          * number of segments to prevent device reading beyond end of buffer.
 622          */
 623         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 624
 625         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 626         if (!range) {
 627                 /*
 628                  * If we fail allocation our range, fallback to the controller
 629                  * discard page. If that's also busy, it's safe to return
 630                  * busy, as we know we can make progress once that's freed.
 631                  */
 632                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 633                         return BLK_STS_RESOURCE;
 634
 635                 range = page_address(ns->ctrl->discard_page);
 636         }
 637
 638         __rq_for_each_bio(bio, req) {
 639                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 640                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 641
 642                 if (n < segments) {
 643                         range[n].cattr = cpu_to_le32(0);
 644                         range[n].nlb = cpu_to_le32(nlb);
 645                         range[n].slba = cpu_to_le64(slba);
 646                 }
 647                 n++;
 648         }
 649
 650         if (WARN_ON_ONCE(n != segments)) {
 651                 if (virt_to_page(range) == ns->ctrl->discard_page)
 652                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 653                 else
 654                         kfree(range);
 655                 return BLK_STS_IOERR;
 656         }
 657
 658         cmnd->dsm.opcode = nvme_cmd_dsm;
 659         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 660         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 661         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 662
 663         req->special_vec.bv_page = virt_to_page(range);
 664         req->special_vec.bv_offset = offset_in_page(range);
 665         req->special_vec.bv_len = alloc_size;
 666         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 667
 668         return BLK_STS_OK;
 669 }
 670
 671 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 672                 struct request *req, struct nvme_command *cmnd)
 673 {
 674         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 675                 return nvme_setup_discard(ns, req, cmnd);
 676
 677         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 678         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 679         cmnd->write_zeroes.slba =
 680                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 681         cmnd->write_zeroes.length =
 682                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 683         cmnd->write_zeroes.control = 0;
 684         return BLK_STS_OK;
 685 }
 686
 687 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 688                 struct request *req, struct nvme_command *cmnd)
 689 {
 690         struct nvme_ctrl *ctrl = ns->ctrl;
 691         u16 control = 0;
 692         u32 dsmgmt = 0;
 693
 694         if (req->cmd_flags & REQ_FUA)
 695                 control |= NVME_RW_FUA;
 696         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 697                 control |= NVME_RW_LR;
 698
 699         if (req->cmd_flags & REQ_RAHEAD)
 700                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 701
 702         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 703         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 704         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 705         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 706
 707         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 708                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 709
 710         if (ns->ms) {
 711                 /*
 712                  * If formated with metadata, the block layer always provides a
 713                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 714                  * we enable the PRACT bit for protection information or set the
 715                  * namespace capacity to zero to prevent any I/O.
 716                  */
 717                 if (!blk_integrity_rq(req)) {
 718                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 719                                 return BLK_STS_NOTSUPP;
 720                         control |= NVME_RW_PRINFO_PRACT;
 721                 }
 722
 723                 switch (ns->pi_type) {
 724                 case NVME_NS_DPS_PI_TYPE3:
 725                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 726                         break;
 727                 case NVME_NS_DPS_PI_TYPE1:
 728                 case NVME_NS_DPS_PI_TYPE2:
 729                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 730                                         NVME_RW_PRINFO_PRCHK_REF;
 731                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 732                         break;
 733                 }
 734         }
 735
 736         cmnd->rw.control = cpu_to_le16(control);
 737         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 738         return 0;
 739 }
 740
 741 void nvme_cleanup_cmd(struct request *req)
 742 {
 743         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 744                 struct nvme_ns *ns = req->rq_disk->private_data;
 745                 struct page *page = req->special_vec.bv_page;
 746
 747                 if (page == ns->ctrl->discard_page)
 748                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 749                 else
 750                         kfree(page_address(page) + req->special_vec.bv_offset);
 751         }
 752 }
 753 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 754
 755 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 756                 struct nvme_command *cmd)
 757 {
 758         blk_status_t ret = BLK_STS_OK;
 759
 760         nvme_clear_nvme_request(req);
 761
 762         memset(cmd, 0, sizeof(*cmd));
 763         switch (req_op(req)) {
 764         case REQ_OP_DRV_IN:
 765         case REQ_OP_DRV_OUT:
 766                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 767                 break;
 768         case REQ_OP_FLUSH:
 769                 nvme_setup_flush(ns, cmd);
 770                 break;
 771         case REQ_OP_WRITE_ZEROES:
 772                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 773                 break;
 774         case REQ_OP_DISCARD:
 775                 ret = nvme_setup_discard(ns, req, cmd);
 776                 break;
 777         case REQ_OP_READ:
 778         case REQ_OP_WRITE:
 779                 ret = nvme_setup_rw(ns, req, cmd);
 780                 break;
 781         default:
 782                 WARN_ON_ONCE(1);
 783                 return BLK_STS_IOERR;
 784         }
 785
 786         cmd->common.command_id = req->tag;
 787         trace_nvme_setup_cmd(req, cmd);
 788         return ret;
 789 }
 790 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 791
 792 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 793 {
 794         struct completion *waiting = rq->end_io_data;
 795
 796         rq->end_io_data = NULL;
 797         complete(waiting);
 798 }
 799
 800 static void nvme_execute_rq_polled(struct request_queue *q,
 801                 struct gendisk *bd_disk, struct request *rq, int at_head)
 802 {
 803         DECLARE_COMPLETION_ONSTACK(wait);
 804
 805         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 806
 807         rq->cmd_flags |= REQ_HIPRI;
 808         rq->end_io_data = &wait;
 809         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 810
 811         while (!completion_done(&wait)) {
 812                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 813                 cond_resched();
 814         }
 815 }
 816
 817 /*
 818  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 819  * if the result is positive, it's an NVM Express status code
 820  */
 821 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 822                 union nvme_result *result, void *buffer, unsigned bufflen,
 823                 unsigned timeout, int qid, int at_head,
 824                 blk_mq_req_flags_t flags, bool poll)
 825 {
 826         struct request *req;
 827         int ret;
 828
 829         req = nvme_alloc_request(q, cmd, flags, qid);
 830         if (IS_ERR(req))
 831                 return PTR_ERR(req);
 832
 833         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 834
 835         if (buffer && bufflen) {
 836                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 837                 if (ret)
 838                         goto out;
 839         }
 840
 841         if (poll)
 842                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 843         else
 844                 blk_execute_rq(req->q, NULL, req, at_head);
 845         if (result)
 846                 *result = nvme_req(req)->result;
 847         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 848                 ret = -EINTR;
 849         else
 850                 ret = nvme_req(req)->status;
 851  out:
 852         blk_mq_free_request(req);
 853         return ret;
 854 }
 855 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 856
 857 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 858                 void *buffer, unsigned bufflen)
 859 {
 860         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 861                         NVME_QID_ANY, 0, 0, false);
 862 }
 863 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 864
 865 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 866                 unsigned len, u32 seed, bool write)
 867 {
 868         struct bio_integrity_payload *bip;
 869         int ret = -ENOMEM;
 870         void *buf;
 871
 872         buf = kmalloc(len, GFP_KERNEL);
 873         if (!buf)
 874                 goto out;
 875
 876         ret = -EFAULT;
 877         if (write && copy_from_user(buf, ubuf, len))
 878                 goto out_free_meta;
 879
 880         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 881         if (IS_ERR(bip)) {
 882                 ret = PTR_ERR(bip);
 883                 goto out_free_meta;
 884         }
 885
 886         bip->bip_iter.bi_size = len;
 887         bip->bip_iter.bi_sector = seed;
 888         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 889                         offset_in_page(buf));
 890         if (ret == len)
 891                 return buf;
 892         ret = -ENOMEM;
 893 out_free_meta:
 894         kfree(buf);
 895 out:
 896         return ERR_PTR(ret);
 897 }
 898
 899 static int nvme_submit_user_cmd(struct request_queue *q,
 900                 struct nvme_command *cmd, void __user *ubuffer,
 901                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 902                 u32 meta_seed, u64 *result, unsigned timeout)
 903 {
 904         bool write = nvme_is_write(cmd);
 905         struct nvme_ns *ns = q->queuedata;
 906         struct gendisk *disk = ns ? ns->disk : NULL;
 907         struct request *req;
 908         struct bio *bio = NULL;
 909         void *meta = NULL;
 910         int ret;
 911
 912         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 913         if (IS_ERR(req))
 914                 return PTR_ERR(req);
 915
 916         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 917         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 918
 919         if (ubuffer && bufflen) {
 920                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 921                                 GFP_KERNEL);
 922                 if (ret)
 923                         goto out;
 924                 bio = req->bio;
 925                 bio->bi_disk = disk;
 926                 if (disk && meta_buffer && meta_len) {
 927                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 928                                         meta_seed, write);
 929                         if (IS_ERR(meta)) {
 930                                 ret = PTR_ERR(meta);
 931                                 goto out_unmap;
 932                         }
 933                         req->cmd_flags |= REQ_INTEGRITY;
 934                 }
 935         }
 936
 937         blk_execute_rq(req->q, disk, req, 0);
 938         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 939                 ret = -EINTR;
 940         else
 941                 ret = nvme_req(req)->status;
 942         if (result)
 943                 *result = le64_to_cpu(nvme_req(req)->result.u64);
 944         if (meta && !ret && !write) {
 945                 if (copy_to_user(meta_buffer, meta, meta_len))
 946                         ret = -EFAULT;
 947         }
 948         kfree(meta);
 949  out_unmap:
 950         if (bio)
 951                 blk_rq_unmap_user(bio);
 952  out:
 953         blk_mq_free_request(req);
 954         return ret;
 955 }
 956
 957 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 958 {
 959         struct nvme_ctrl *ctrl = rq->end_io_data;
 960         unsigned long flags;
 961         bool startka = false;
 962
 963         blk_mq_free_request(rq);
 964
 965         if (status) {
 966                 dev_err(ctrl->device,
 967                         "failed nvme_keep_alive_end_io error=%d\n",
 968                                 status);
 969                 return;
 970         }
 971
 972         ctrl->comp_seen = false;
 973         spin_lock_irqsave(&ctrl->lock, flags);
 974         if (ctrl->state == NVME_CTRL_LIVE ||
 975             ctrl->state == NVME_CTRL_CONNECTING)
 976                 startka = true;
 977         spin_unlock_irqrestore(&ctrl->lock, flags);
 978         if (startka)
 979                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 980 }
 981
 982 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 983 {
 984         struct request *rq;
 985
 986         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 987                         NVME_QID_ANY);
 988         if (IS_ERR(rq))
 989                 return PTR_ERR(rq);
 990
 991         rq->timeout = ctrl->kato * HZ;
 992         rq->end_io_data = ctrl;
 993
 994         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 995
 996         return 0;
 997 }
 998
 999 static void nvme_keep_alive_work(struct work_struct *work)
1000 {
1001         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1002                         struct nvme_ctrl, ka_work);
1003         bool comp_seen = ctrl->comp_seen;
1004
1005         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1006                 dev_dbg(ctrl->device,
1007                         "reschedule traffic based keep-alive timer\n");
1008                 ctrl->comp_seen = false;
1009                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1010                 return;
1011         }
1012
1013         if (nvme_keep_alive(ctrl)) {
1014                 /* allocation failure, reset the controller */
1015                 dev_err(ctrl->device, "keep-alive failed\n");
1016                 nvme_reset_ctrl(ctrl);
1017                 return;
1018         }
1019 }
1020
1021 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1022 {
1023         if (unlikely(ctrl->kato == 0))
1024                 return;
1025
1026         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1027 }
1028
1029 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1030 {
1031         if (unlikely(ctrl->kato == 0))
1032                 return;
1033
1034         cancel_delayed_work_sync(&ctrl->ka_work);
1035 }
1036 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1037
1038 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1039 {
1040         struct nvme_command c = { };
1041         int error;
1042
1043         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1044         c.identify.opcode = nvme_admin_identify;
1045         c.identify.cns = NVME_ID_CNS_CTRL;
1046
1047         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1048         if (!*id)
1049                 return -ENOMEM;
1050
1051         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1052                         sizeof(struct nvme_id_ctrl));
1053         if (error)
1054                 kfree(*id);
1055         return error;
1056 }
1057
1058 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1059                 struct nvme_ns_ids *ids)
1060 {
1061         struct nvme_command c = { };
1062         int status;
1063         void *data;
1064         int pos;
1065         int len;
1066
1067         c.identify.opcode = nvme_admin_identify;
1068         c.identify.nsid = cpu_to_le32(nsid);
1069         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1070
1071         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1072         if (!data)
1073                 return -ENOMEM;
1074
1075         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1076                                       NVME_IDENTIFY_DATA_SIZE);
1077         if (status)
1078                 goto free_data;
1079
1080         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1081                 struct nvme_ns_id_desc *cur = data + pos;
1082
1083                 if (cur->nidl == 0)
1084                         break;
1085
1086                 switch (cur->nidt) {
1087                 case NVME_NIDT_EUI64:
1088                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1089                                 dev_warn(ctrl->device,
1090                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1091                                          cur->nidl);
1092                                 goto free_data;
1093                         }
1094                         len = NVME_NIDT_EUI64_LEN;
1095                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1096                         break;
1097                 case NVME_NIDT_NGUID:
1098                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1099                                 dev_warn(ctrl->device,
1100                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1101                                          cur->nidl);
1102                                 goto free_data;
1103                         }
1104                         len = NVME_NIDT_NGUID_LEN;
1105                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1106                         break;
1107                 case NVME_NIDT_UUID:
1108                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
1109                                 dev_warn(ctrl->device,
1110                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1111                                          cur->nidl);
1112                                 goto free_data;
1113                         }
1114                         len = NVME_NIDT_UUID_LEN;
1115                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1116                         break;
1117                 default:
1118                         /* Skip unknown types */
1119                         len = cur->nidl;
1120                         break;
1121                 }
1122
1123                 len += sizeof(*cur);
1124         }
1125 free_data:
1126         kfree(data);
1127         return status;
1128 }
1129
1130 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1131 {
1132         struct nvme_command c = { };
1133
1134         c.identify.opcode = nvme_admin_identify;
1135         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1136         c.identify.nsid = cpu_to_le32(nsid);
1137         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1138                                     NVME_IDENTIFY_DATA_SIZE);
1139 }
1140
1141 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1142                 unsigned nsid, struct nvme_id_ns **id)
1143 {
1144         struct nvme_command c = { };
1145         int error;
1146
1147         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1148         c.identify.opcode = nvme_admin_identify;
1149         c.identify.nsid = cpu_to_le32(nsid);
1150         c.identify.cns = NVME_ID_CNS_NS;
1151
1152         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1153         if (!*id)
1154                 return -ENOMEM;
1155
1156         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1157         if (error) {
1158                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1159                 kfree(*id);
1160         }
1161
1162         return error;
1163 }
1164
1165 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1166                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1167 {
1168         struct nvme_command c;
1169         union nvme_result res;
1170         int ret;
1171
1172         memset(&c, 0, sizeof(c));
1173         c.features.opcode = op;
1174         c.features.fid = cpu_to_le32(fid);
1175         c.features.dword11 = cpu_to_le32(dword11);
1176
1177         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1178                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1179         if (ret >= 0 && result)
1180                 *result = le32_to_cpu(res.u32);
1181         return ret;
1182 }
1183
1184 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1185                       unsigned int dword11, void *buffer, size_t buflen,
1186                       u32 *result)
1187 {
1188         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1189                              buflen, result);
1190 }
1191 EXPORT_SYMBOL_GPL(nvme_set_features);
1192
1193 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1194                       unsigned int dword11, void *buffer, size_t buflen,
1195                       u32 *result)
1196 {
1197         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1198                              buflen, result);
1199 }
1200 EXPORT_SYMBOL_GPL(nvme_get_features);
1201
1202 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1203 {
1204         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1205         u32 result;
1206         int status, nr_io_queues;
1207
1208         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1209                         &result);
1210         if (status < 0)
1211                 return status;
1212
1213         /*
1214          * Degraded controllers might return an error when setting the queue
1215          * count.  We still want to be able to bring them online and offer
1216          * access to the admin queue, as that might be only way to fix them up.
1217          */
1218         if (status > 0) {
1219                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1220                 *count = 0;
1221         } else {
1222                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1223                 *count = min(*count, nr_io_queues);
1224         }
1225
1226         return 0;
1227 }
1228 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1229
1230 #define NVME_AEN_SUPPORTED \
1231         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1232          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1233
1234 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1235 {
1236         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1237         int status;
1238
1239         if (!supported_aens)
1240                 return;
1241
1242         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1243                         NULL, 0, &result);
1244         if (status)
1245                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1246                          supported_aens);
1247
1248         queue_work(nvme_wq, &ctrl->async_event_work);
1249 }
1250
1251 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1252 {
1253         struct nvme_user_io io;
1254         struct nvme_command c;
1255         unsigned length, meta_len;
1256         void __user *metadata;
1257
1258         if (copy_from_user(&io, uio, sizeof(io)))
1259                 return -EFAULT;
1260         if (io.flags)
1261                 return -EINVAL;
1262
1263         switch (io.opcode) {
1264         case nvme_cmd_write:
1265         case nvme_cmd_read:
1266         case nvme_cmd_compare:
1267                 break;
1268         default:
1269                 return -EINVAL;
1270         }
1271
1272         length = (io.nblocks + 1) << ns->lba_shift;
1273         meta_len = (io.nblocks + 1) * ns->ms;
1274         metadata = (void __user *)(uintptr_t)io.metadata;
1275
1276         if (ns->ext) {
1277                 length += meta_len;
1278                 meta_len = 0;
1279         } else if (meta_len) {
1280                 if ((io.metadata & 3) || !io.metadata)
1281                         return -EINVAL;
1282         }
1283
1284         memset(&c, 0, sizeof(c));
1285         c.rw.opcode = io.opcode;
1286         c.rw.flags = io.flags;
1287         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1288         c.rw.slba = cpu_to_le64(io.slba);
1289         c.rw.length = cpu_to_le16(io.nblocks);
1290         c.rw.control = cpu_to_le16(io.control);
1291         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1292         c.rw.reftag = cpu_to_le32(io.reftag);
1293         c.rw.apptag = cpu_to_le16(io.apptag);
1294         c.rw.appmask = cpu_to_le16(io.appmask);
1295
1296         return nvme_submit_user_cmd(ns->queue, &c,
1297                         (void __user *)(uintptr_t)io.addr, length,
1298                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1299 }
1300
1301 static u32 nvme_known_admin_effects(u8 opcode)
1302 {
1303         switch (opcode) {
1304         case nvme_admin_format_nvm:
1305                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1306                                         NVME_CMD_EFFECTS_CSE_MASK;
1307         case nvme_admin_sanitize_nvm:
1308                 return NVME_CMD_EFFECTS_CSE_MASK;
1309         default:
1310                 break;
1311         }
1312         return 0;
1313 }
1314
1315 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1316                                                                 u8 opcode)
1317 {
1318         u32 effects = 0;
1319
1320         if (ns) {
1321                 if (ctrl->effects)
1322                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1323                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1324                         dev_warn(ctrl->device,
1325                                  "IO command:%02x has unhandled effects:%08x\n",
1326                                  opcode, effects);
1327                 return 0;
1328         }
1329
1330         if (ctrl->effects)
1331                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1332         effects |= nvme_known_admin_effects(opcode);
1333
1334         /*
1335          * For simplicity, IO to all namespaces is quiesced even if the command
1336          * effects say only one namespace is affected.
1337          */
1338         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1339                 mutex_lock(&ctrl->scan_lock);
1340                 mutex_lock(&ctrl->subsys->lock);
1341                 nvme_mpath_start_freeze(ctrl->subsys);
1342                 nvme_mpath_wait_freeze(ctrl->subsys);
1343                 nvme_start_freeze(ctrl);
1344                 nvme_wait_freeze(ctrl);
1345         }
1346         return effects;
1347 }
1348
1349 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1350 {
1351         struct nvme_ns *ns;
1352
1353         down_read(&ctrl->namespaces_rwsem);
1354         list_for_each_entry(ns, &ctrl->namespaces, list)
1355                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1356                         nvme_set_queue_dying(ns);
1357         up_read(&ctrl->namespaces_rwsem);
1358 }
1359
1360 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1361 {
1362         /*
1363          * Revalidate LBA changes prior to unfreezing. This is necessary to
1364          * prevent memory corruption if a logical block size was changed by
1365          * this command.
1366          */
1367         if (effects & NVME_CMD_EFFECTS_LBCC)
1368                 nvme_update_formats(ctrl);
1369         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1370                 nvme_unfreeze(ctrl);
1371                 nvme_mpath_unfreeze(ctrl->subsys);
1372                 mutex_unlock(&ctrl->subsys->lock);
1373                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1374                 mutex_unlock(&ctrl->scan_lock);
1375         }
1376         if (effects & NVME_CMD_EFFECTS_CCC)
1377                 nvme_init_identify(ctrl);
1378         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1379                 nvme_queue_scan(ctrl);
1380 }
1381
1382 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1383                         struct nvme_passthru_cmd __user *ucmd)
1384 {
1385         struct nvme_passthru_cmd cmd;
1386         struct nvme_command c;
1387         unsigned timeout = 0;
1388         u32 effects;
1389         u64 result;
1390         int status;
1391
1392         if (!capable(CAP_SYS_ADMIN))
1393                 return -EACCES;
1394         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1395                 return -EFAULT;
1396         if (cmd.flags)
1397                 return -EINVAL;
1398
1399         memset(&c, 0, sizeof(c));
1400         c.common.opcode = cmd.opcode;
1401         c.common.flags = cmd.flags;
1402         c.common.nsid = cpu_to_le32(cmd.nsid);
1403         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1404         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1405         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1406         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1407         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1408         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1409         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1410         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1411
1412         if (cmd.timeout_ms)
1413                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1414
1415         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1416         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1417                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1418                         (void __user *)(uintptr_t)cmd.metadata,
1419                         cmd.metadata_len, 0, &result, timeout);
1420         nvme_passthru_end(ctrl, effects);
1421
1422         if (status >= 0) {
1423                 if (put_user(result, &ucmd->result))
1424                         return -EFAULT;
1425         }
1426
1427         return status;
1428 }
1429
1430 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1431                         struct nvme_passthru_cmd64 __user *ucmd)
1432 {
1433         struct nvme_passthru_cmd64 cmd;
1434         struct nvme_command c;
1435         unsigned timeout = 0;
1436         u32 effects;
1437         int status;
1438
1439         if (!capable(CAP_SYS_ADMIN))
1440                 return -EACCES;
1441         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1442                 return -EFAULT;
1443         if (cmd.flags)
1444                 return -EINVAL;
1445
1446         memset(&c, 0, sizeof(c));
1447         c.common.opcode = cmd.opcode;
1448         c.common.flags = cmd.flags;
1449         c.common.nsid = cpu_to_le32(cmd.nsid);
1450         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1451         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1452         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1453         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1454         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1455         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1456         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1457         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1458
1459         if (cmd.timeout_ms)
1460                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1461
1462         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1463         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1464                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1465                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1466                         0, &cmd.result, timeout);
1467         nvme_passthru_end(ctrl, effects);
1468
1469         if (status >= 0) {
1470                 if (put_user(cmd.result, &ucmd->result))
1471                         return -EFAULT;
1472         }
1473
1474         return status;
1475 }
1476
1477 /*
1478  * Issue ioctl requests on the first available path.  Note that unlike normal
1479  * block layer requests we will not retry failed request on another controller.
1480  */
1481 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1482                 struct nvme_ns_head **head, int *srcu_idx)
1483 {
1484 #ifdef CONFIG_NVME_MULTIPATH
1485         if (disk->fops == &nvme_ns_head_ops) {
1486                 struct nvme_ns *ns;
1487
1488                 *head = disk->private_data;
1489                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1490                 ns = nvme_find_path(*head);
1491                 if (!ns)
1492                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1493                 return ns;
1494         }
1495 #endif
1496         *head = NULL;
1497         *srcu_idx = -1;
1498         return disk->private_data;
1499 }
1500
1501 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1502 {
1503         if (head)
1504                 srcu_read_unlock(&head->srcu, idx);
1505 }
1506
1507 static bool is_ctrl_ioctl(unsigned int cmd)
1508 {
1509         if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1510                 return true;
1511         if (is_sed_ioctl(cmd))
1512                 return true;
1513         return false;
1514 }
1515
1516 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1517                                   void __user *argp,
1518                                   struct nvme_ns_head *head,
1519                                   int srcu_idx)
1520 {
1521         struct nvme_ctrl *ctrl = ns->ctrl;
1522         int ret;
1523
1524         nvme_get_ctrl(ns->ctrl);
1525         nvme_put_ns_from_disk(head, srcu_idx);
1526
1527         switch (cmd) {
1528         case NVME_IOCTL_ADMIN_CMD:
1529                 ret = nvme_user_cmd(ctrl, NULL, argp);
1530                 break;
1531         case NVME_IOCTL_ADMIN64_CMD:
1532                 ret = nvme_user_cmd64(ctrl, NULL, argp);
1533                 break;
1534         default:
1535                 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1536                 break;
1537         }
1538         nvme_put_ctrl(ctrl);
1539         return ret;
1540 }
1541
1542 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1543                 unsigned int cmd, unsigned long arg)
1544 {
1545         struct nvme_ns_head *head = NULL;
1546         void __user *argp = (void __user *)arg;
1547         struct nvme_ns *ns;
1548         int srcu_idx, ret;
1549
1550         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1551         if (unlikely(!ns))
1552                 return -EWOULDBLOCK;
1553
1554         /*
1555          * Handle ioctls that apply to the controller instead of the namespace
1556          * seperately and drop the ns SRCU reference early.  This avoids a
1557          * deadlock when deleting namespaces using the passthrough interface.
1558          */
1559         if (is_ctrl_ioctl(cmd))
1560                 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1561
1562         switch (cmd) {
1563         case NVME_IOCTL_ID:
1564                 force_successful_syscall_return();
1565                 ret = ns->head->ns_id;
1566                 break;
1567         case NVME_IOCTL_IO_CMD:
1568                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1569                 break;
1570         case NVME_IOCTL_SUBMIT_IO:
1571                 ret = nvme_submit_io(ns, argp);
1572                 break;
1573         case NVME_IOCTL_IO64_CMD:
1574                 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1575                 break;
1576         default:
1577                 if (ns->ndev)
1578                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1579                 else
1580                         ret = -ENOTTY;
1581         }
1582
1583         nvme_put_ns_from_disk(head, srcu_idx);
1584         return ret;
1585 }
1586
1587 static int nvme_open(struct block_device *bdev, fmode_t mode)
1588 {
1589         struct nvme_ns *ns = bdev->bd_disk->private_data;
1590
1591 #ifdef CONFIG_NVME_MULTIPATH
1592         /* should never be called due to GENHD_FL_HIDDEN */
1593         if (WARN_ON_ONCE(ns->head->disk))
1594                 goto fail;
1595 #endif
1596         if (!kref_get_unless_zero(&ns->kref))
1597                 goto fail;
1598         if (!try_module_get(ns->ctrl->ops->module))
1599                 goto fail_put_ns;
1600
1601         return 0;
1602
1603 fail_put_ns:
1604         nvme_put_ns(ns);
1605 fail:
1606         return -ENXIO;
1607 }
1608
1609 static void nvme_release(struct gendisk *disk, fmode_t mode)
1610 {
1611         struct nvme_ns *ns = disk->private_data;
1612
1613         module_put(ns->ctrl->ops->module);
1614         nvme_put_ns(ns);
1615 }
1616
1617 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1618 {
1619         /* some standard values */
1620         geo->heads = 1 << 6;
1621         geo->sectors = 1 << 5;
1622         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1623         return 0;
1624 }
1625
1626 #ifdef CONFIG_BLK_DEV_INTEGRITY
1627 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1628 {
1629         struct blk_integrity integrity;
1630
1631         memset(&integrity, 0, sizeof(integrity));
1632         switch (pi_type) {
1633         case NVME_NS_DPS_PI_TYPE3:
1634                 integrity.profile = &t10_pi_type3_crc;
1635                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1636                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1637                 break;
1638         case NVME_NS_DPS_PI_TYPE1:
1639         case NVME_NS_DPS_PI_TYPE2:
1640                 integrity.profile = &t10_pi_type1_crc;
1641                 integrity.tag_size = sizeof(u16);
1642                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1643                 break;
1644         default:
1645                 integrity.profile = NULL;
1646                 break;
1647         }
1648         integrity.tuple_size = ms;
1649         blk_integrity_register(disk, &integrity);
1650         blk_queue_max_integrity_segments(disk->queue, 1);
1651 }
1652 #else
1653 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1654 {
1655 }
1656 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1657
1658 static void nvme_set_chunk_size(struct nvme_ns *ns)
1659 {
1660         u32 chunk_size = nvme_lba_to_sect(ns, ns->noiob);
1661         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1662 }
1663
1664 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1665 {
1666         struct nvme_ctrl *ctrl = ns->ctrl;
1667         struct request_queue *queue = disk->queue;
1668         u32 size = queue_logical_block_size(queue);
1669
1670         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1671                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1672                 return;
1673         }
1674
1675         if (ctrl->nr_streams && ns->sws && ns->sgs)
1676                 size *= ns->sws * ns->sgs;
1677
1678         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1679                         NVME_DSM_MAX_RANGES);
1680
1681         queue->limits.discard_alignment = 0;
1682         queue->limits.discard_granularity = size;
1683
1684         /* If discard is already enabled, don't reset queue limits */
1685         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1686                 return;
1687
1688         blk_queue_max_discard_sectors(queue, UINT_MAX);
1689         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1690
1691         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1692                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1693 }
1694
1695 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1696 {
1697         u64 max_blocks;
1698
1699         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1700             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1701                 return;
1702         /*
1703          * Even though NVMe spec explicitly states that MDTS is not
1704          * applicable to the write-zeroes:- "The restriction does not apply to
1705          * commands that do not transfer data between the host and the
1706          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1707          * In order to be more cautious use controller's max_hw_sectors value
1708          * to configure the maximum sectors for the write-zeroes which is
1709          * configured based on the controller's MDTS field in the
1710          * nvme_init_identify() if available.
1711          */
1712         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1713                 max_blocks = (u64)USHRT_MAX + 1;
1714         else
1715                 max_blocks = ns->ctrl->max_hw_sectors + 1;
1716
1717         blk_queue_max_write_zeroes_sectors(disk->queue,
1718                                            nvme_lba_to_sect(ns, max_blocks));
1719 }
1720
1721 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1722                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1723 {
1724         int ret = 0;
1725
1726         memset(ids, 0, sizeof(*ids));
1727
1728         if (ctrl->vs >= NVME_VS(1, 1, 0))
1729                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1730         if (ctrl->vs >= NVME_VS(1, 2, 0))
1731                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1732         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1733                  /* Don't treat error as fatal we potentially
1734                   * already have a NGUID or EUI-64
1735                   */
1736                 ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1737                 if (ret)
1738                         dev_warn(ctrl->device,
1739                                  "Identify Descriptors failed (%d)\n", ret);
1740                 if (ret > 0)
1741                         ret = 0;
1742         }
1743         return ret;
1744 }
1745
1746 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1747 {
1748         return !uuid_is_null(&ids->uuid) ||
1749                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1750                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1751 }
1752
1753 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1754 {
1755         return uuid_equal(&a->uuid, &b->uuid) &&
1756                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1757                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1758 }
1759
1760 static void nvme_update_disk_info(struct gendisk *disk,
1761                 struct nvme_ns *ns, struct nvme_id_ns *id)
1762 {
1763         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1764         unsigned short bs = 1 << ns->lba_shift;
1765         u32 atomic_bs, phys_bs, io_opt;
1766
1767         if (ns->lba_shift > PAGE_SHIFT) {
1768                 /* unsupported block size, set capacity to 0 later */
1769                 bs = (1 << 9);
1770         }
1771         blk_mq_freeze_queue(disk->queue);
1772         blk_integrity_unregister(disk);
1773
1774         if (id->nabo == 0) {
1775                 /*
1776                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1777                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1778                  * 0 then AWUPF must be used instead.
1779                  */
1780                 if (id->nsfeat & (1 << 1) && id->nawupf)
1781                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1782                 else
1783                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1784         } else {
1785                 atomic_bs = bs;
1786         }
1787         phys_bs = bs;
1788         io_opt = bs;
1789         if (id->nsfeat & (1 << 4)) {
1790                 /* NPWG = Namespace Preferred Write Granularity */
1791                 phys_bs *= 1 + le16_to_cpu(id->npwg);
1792                 /* NOWS = Namespace Optimal Write Size */
1793                 io_opt *= 1 + le16_to_cpu(id->nows);
1794         }
1795
1796         blk_queue_logical_block_size(disk->queue, bs);
1797         /*
1798          * Linux filesystems assume writing a single physical block is
1799          * an atomic operation. Hence limit the physical block size to the
1800          * value of the Atomic Write Unit Power Fail parameter.
1801          */
1802         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1803         blk_queue_io_min(disk->queue, phys_bs);
1804         blk_queue_io_opt(disk->queue, io_opt);
1805
1806         if (ns->ms && !ns->ext &&
1807             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1808                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1809         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1810             ns->lba_shift > PAGE_SHIFT)
1811                 capacity = 0;
1812
1813         set_capacity(disk, capacity);
1814
1815         nvme_config_discard(disk, ns);
1816         nvme_config_write_zeroes(disk, ns);
1817
1818         if (id->nsattr & (1 << 0))
1819                 set_disk_ro(disk, true);
1820         else
1821                 set_disk_ro(disk, false);
1822
1823         blk_mq_unfreeze_queue(disk->queue);
1824 }
1825
1826 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1827 {
1828         struct nvme_ns *ns = disk->private_data;
1829
1830         /*
1831          * If identify namespace failed, use default 512 byte block size so
1832          * block layer can use before failing read/write for 0 capacity.
1833          */
1834         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1835         if (ns->lba_shift == 0)
1836                 ns->lba_shift = 9;
1837         ns->noiob = le16_to_cpu(id->noiob);
1838         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1839         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1840         /* the PI implementation requires metadata equal t10 pi tuple size */
1841         if (ns->ms == sizeof(struct t10_pi_tuple))
1842                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1843         else
1844                 ns->pi_type = 0;
1845
1846         if (ns->noiob)
1847                 nvme_set_chunk_size(ns);
1848         nvme_update_disk_info(disk, ns, id);
1849 #ifdef CONFIG_NVME_MULTIPATH
1850         if (ns->head->disk) {
1851                 nvme_update_disk_info(ns->head->disk, ns, id);
1852                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1853                 revalidate_disk(ns->head->disk);
1854         }
1855 #endif
1856 }
1857
1858 static int nvme_revalidate_disk(struct gendisk *disk)
1859 {
1860         struct nvme_ns *ns = disk->private_data;
1861         struct nvme_ctrl *ctrl = ns->ctrl;
1862         struct nvme_id_ns *id;
1863         struct nvme_ns_ids ids;
1864         int ret = 0;
1865
1866         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1867                 set_capacity(disk, 0);
1868                 return -ENODEV;
1869         }
1870
1871         ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1872         if (ret)
1873                 goto out;
1874
1875         if (id->ncap == 0) {
1876                 ret = -ENODEV;
1877                 goto free_id;
1878         }
1879
1880         __nvme_revalidate_disk(disk, id);
1881         ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1882         if (ret)
1883                 goto free_id;
1884
1885         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1886                 dev_err(ctrl->device,
1887                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1888                 ret = -ENODEV;
1889         }
1890
1891 free_id:
1892         kfree(id);
1893 out:
1894         /*
1895          * Only fail the function if we got a fatal error back from the
1896          * device, otherwise ignore the error and just move on.
1897          */
1898         if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1899                 ret = 0;
1900         else if (ret > 0)
1901                 ret = blk_status_to_errno(nvme_error_status(ret));
1902         return ret;
1903 }
1904
1905 static char nvme_pr_type(enum pr_type type)
1906 {
1907         switch (type) {
1908         case PR_WRITE_EXCLUSIVE:
1909                 return 1;
1910         case PR_EXCLUSIVE_ACCESS:
1911                 return 2;
1912         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1913                 return 3;
1914         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1915                 return 4;
1916         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1917                 return 5;
1918         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1919                 return 6;
1920         default:
1921                 return 0;
1922         }
1923 };
1924
1925 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1926                                 u64 key, u64 sa_key, u8 op)
1927 {
1928         struct nvme_ns_head *head = NULL;
1929         struct nvme_ns *ns;
1930         struct nvme_command c;
1931         int srcu_idx, ret;
1932         u8 data[16] = { 0, };
1933
1934         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1935         if (unlikely(!ns))
1936                 return -EWOULDBLOCK;
1937
1938         put_unaligned_le64(key, &data[0]);
1939         put_unaligned_le64(sa_key, &data[8]);
1940
1941         memset(&c, 0, sizeof(c));
1942         c.common.opcode = op;
1943         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1944         c.common.cdw10 = cpu_to_le32(cdw10);
1945
1946         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1947         nvme_put_ns_from_disk(head, srcu_idx);
1948         return ret;
1949 }
1950
1951 static int nvme_pr_register(struct block_device *bdev, u64 old,
1952                 u64 new, unsigned flags)
1953 {
1954         u32 cdw10;
1955
1956         if (flags & ~PR_FL_IGNORE_KEY)
1957                 return -EOPNOTSUPP;
1958
1959         cdw10 = old ? 2 : 0;
1960         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1961         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1962         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1963 }
1964
1965 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1966                 enum pr_type type, unsigned flags)
1967 {
1968         u32 cdw10;
1969
1970         if (flags & ~PR_FL_IGNORE_KEY)
1971                 return -EOPNOTSUPP;
1972
1973         cdw10 = nvme_pr_type(type) << 8;
1974         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1975         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1976 }
1977
1978 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1979                 enum pr_type type, bool abort)
1980 {
1981         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1982         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1983 }
1984
1985 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1986 {
1987         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1988         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1989 }
1990
1991 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1992 {
1993         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1994         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1995 }
1996
1997 static const struct pr_ops nvme_pr_ops = {
1998         .pr_register    = nvme_pr_register,
1999         .pr_reserve     = nvme_pr_reserve,
2000         .pr_release     = nvme_pr_release,
2001         .pr_preempt     = nvme_pr_preempt,
2002         .pr_clear       = nvme_pr_clear,
2003 };
2004
2005 #ifdef CONFIG_BLK_SED_OPAL
2006 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2007                 bool send)
2008 {
2009         struct nvme_ctrl *ctrl = data;
2010         struct nvme_command cmd;
2011
2012         memset(&cmd, 0, sizeof(cmd));
2013         if (send)
2014                 cmd.common.opcode = nvme_admin_security_send;
2015         else
2016                 cmd.common.opcode = nvme_admin_security_recv;
2017         cmd.common.nsid = 0;
2018         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2019         cmd.common.cdw11 = cpu_to_le32(len);
2020
2021         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2022                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2023 }
2024 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2025 #endif /* CONFIG_BLK_SED_OPAL */
2026
2027 static const struct block_device_operations nvme_fops = {
2028         .owner          = THIS_MODULE,
2029         .ioctl          = nvme_ioctl,
2030         .compat_ioctl   = nvme_ioctl,
2031         .open           = nvme_open,
2032         .release        = nvme_release,
2033         .getgeo         = nvme_getgeo,
2034         .revalidate_disk= nvme_revalidate_disk,
2035         .pr_ops         = &nvme_pr_ops,
2036 };
2037
2038 #ifdef CONFIG_NVME_MULTIPATH
2039 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2040 {
2041         struct nvme_ns_head *head = bdev->bd_disk->private_data;
2042
2043         if (!kref_get_unless_zero(&head->ref))
2044                 return -ENXIO;
2045         return 0;
2046 }
2047
2048 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2049 {
2050         nvme_put_ns_head(disk->private_data);
2051 }
2052
2053 const struct block_device_operations nvme_ns_head_ops = {
2054         .owner          = THIS_MODULE,
2055         .open           = nvme_ns_head_open,
2056         .release        = nvme_ns_head_release,
2057         .ioctl          = nvme_ioctl,
2058         .compat_ioctl   = nvme_ioctl,
2059         .getgeo         = nvme_getgeo,
2060         .pr_ops         = &nvme_pr_ops,
2061 };
2062 #endif /* CONFIG_NVME_MULTIPATH */
2063
2064 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2065 {
2066         unsigned long timeout =
2067                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2068         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2069         int ret;
2070
2071         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2072                 if (csts == ~0)
2073                         return -ENODEV;
2074                 if ((csts & NVME_CSTS_RDY) == bit)
2075                         break;
2076
2077                 msleep(100);
2078                 if (fatal_signal_pending(current))
2079                         return -EINTR;
2080                 if (time_after(jiffies, timeout)) {
2081                         dev_err(ctrl->device,
2082                                 "Device not ready; aborting %s\n", enabled ?
2083                                                 "initialisation" : "reset");
2084                         return -ENODEV;
2085                 }
2086         }
2087
2088         return ret;
2089 }
2090
2091 /*
2092  * If the device has been passed off to us in an enabled state, just clear
2093  * the enabled bit.  The spec says we should set the 'shutdown notification
2094  * bits', but doing so may cause the device to complete commands to the
2095  * admin queue ... and we don't know what memory that might be pointing at!
2096  */
2097 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2098 {
2099         int ret;
2100
2101         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2102         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2103
2104         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2105         if (ret)
2106                 return ret;
2107
2108         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2109                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2110
2111         return nvme_wait_ready(ctrl, ctrl->cap, false);
2112 }
2113 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2114
2115 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2116 {
2117         /*
2118          * Default to a 4K page size, with the intention to update this
2119          * path in the future to accomodate architectures with differing
2120          * kernel and IO page sizes.
2121          */
2122         unsigned dev_page_min, page_shift = 12;
2123         int ret;
2124
2125         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2126         if (ret) {
2127                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2128                 return ret;
2129         }
2130         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2131
2132         if (page_shift < dev_page_min) {
2133                 dev_err(ctrl->device,
2134                         "Minimum device page size %u too large for host (%u)\n",
2135                         1 << dev_page_min, 1 << page_shift);
2136                 return -ENODEV;
2137         }
2138
2139         ctrl->page_size = 1 << page_shift;
2140
2141         ctrl->ctrl_config = NVME_CC_CSS_NVM;
2142         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2143         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2144         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2145         ctrl->ctrl_config |= NVME_CC_ENABLE;
2146
2147         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2148         if (ret)
2149                 return ret;
2150         return nvme_wait_ready(ctrl, ctrl->cap, true);
2151 }
2152 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2153
2154 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2155 {
2156         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2157         u32 csts;
2158         int ret;
2159
2160         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2161         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2162
2163         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2164         if (ret)
2165                 return ret;
2166
2167         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2168                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2169                         break;
2170
2171                 msleep(100);
2172                 if (fatal_signal_pending(current))
2173                         return -EINTR;
2174                 if (time_after(jiffies, timeout)) {
2175                         dev_err(ctrl->device,
2176                                 "Device shutdown incomplete; abort shutdown\n");
2177                         return -ENODEV;
2178                 }
2179         }
2180
2181         return ret;
2182 }
2183 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2184
2185 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2186                 struct request_queue *q)
2187 {
2188         bool vwc = false;
2189
2190         if (ctrl->max_hw_sectors) {
2191                 u32 max_segments =
2192                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2193
2194                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2195                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2196                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2197         }
2198         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2199             is_power_of_2(ctrl->max_hw_sectors))
2200                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2201         blk_queue_virt_boundary(q, ctrl->page_size - 1);
2202         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2203                 vwc = true;
2204         blk_queue_write_cache(q, vwc, vwc);
2205 }
2206
2207 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2208 {
2209         __le64 ts;
2210         int ret;
2211
2212         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2213                 return 0;
2214
2215         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2216         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2217                         NULL);
2218         if (ret)
2219                 dev_warn_once(ctrl->device,
2220                         "could not set timestamp (%d)\n", ret);
2221         return ret;
2222 }
2223
2224 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2225 {
2226         struct nvme_feat_host_behavior *host;
2227         int ret;
2228
2229         /* Don't bother enabling the feature if retry delay is not reported */
2230         if (!ctrl->crdt[0])
2231                 return 0;
2232
2233         host = kzalloc(sizeof(*host), GFP_KERNEL);
2234         if (!host)
2235                 return 0;
2236
2237         host->acre = NVME_ENABLE_ACRE;
2238         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2239                                 host, sizeof(*host), NULL);
2240         kfree(host);
2241         return ret;
2242 }
2243
2244 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2245 {
2246         /*
2247          * APST (Autonomous Power State Transition) lets us program a
2248          * table of power state transitions that the controller will
2249          * perform automatically.  We configure it with a simple
2250          * heuristic: we are willing to spend at most 2% of the time
2251          * transitioning between power states.  Therefore, when running
2252          * in any given state, we will enter the next lower-power
2253          * non-operational state after waiting 50 * (enlat + exlat)
2254          * microseconds, as long as that state's exit latency is under
2255          * the requested maximum latency.
2256          *
2257          * We will not autonomously enter any non-operational state for
2258          * which the total latency exceeds ps_max_latency_us.  Users
2259          * can set ps_max_latency_us to zero to turn off APST.
2260          */
2261
2262         unsigned apste;
2263         struct nvme_feat_auto_pst *table;
2264         u64 max_lat_us = 0;
2265         int max_ps = -1;
2266         int ret;
2267
2268         /*
2269          * If APST isn't supported or if we haven't been initialized yet,
2270          * then don't do anything.
2271          */
2272         if (!ctrl->apsta)
2273                 return 0;
2274
2275         if (ctrl->npss > 31) {
2276                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2277                 return 0;
2278         }
2279
2280         table = kzalloc(sizeof(*table), GFP_KERNEL);
2281         if (!table)
2282                 return 0;
2283
2284         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2285                 /* Turn off APST. */
2286                 apste = 0;
2287                 dev_dbg(ctrl->device, "APST disabled\n");
2288         } else {
2289                 __le64 target = cpu_to_le64(0);
2290                 int state;
2291
2292                 /*
2293                  * Walk through all states from lowest- to highest-power.
2294                  * According to the spec, lower-numbered states use more
2295                  * power.  NPSS, despite the name, is the index of the
2296                  * lowest-power state, not the number of states.
2297                  */
2298                 for (state = (int)ctrl->npss; state >= 0; state--) {
2299                         u64 total_latency_us, exit_latency_us, transition_ms;
2300
2301                         if (target)
2302                                 table->entries[state] = target;
2303
2304                         /*
2305                          * Don't allow transitions to the deepest state
2306                          * if it's quirked off.
2307                          */
2308                         if (state == ctrl->npss &&
2309                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2310                                 continue;
2311
2312                         /*
2313                          * Is this state a useful non-operational state for
2314                          * higher-power states to autonomously transition to?
2315                          */
2316                         if (!(ctrl->psd[state].flags &
2317                               NVME_PS_FLAGS_NON_OP_STATE))
2318                                 continue;
2319
2320                         exit_latency_us =
2321                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2322                         if (exit_latency_us > ctrl->ps_max_latency_us)
2323                                 continue;
2324
2325                         total_latency_us =
2326                                 exit_latency_us +
2327                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2328
2329                         /*
2330                          * This state is good.  Use it as the APST idle
2331                          * target for higher power states.
2332                          */
2333                         transition_ms = total_latency_us + 19;
2334                         do_div(transition_ms, 20);
2335                         if (transition_ms > (1 << 24) - 1)
2336                                 transition_ms = (1 << 24) - 1;
2337
2338                         target = cpu_to_le64((state << 3) |
2339                                              (transition_ms << 8));
2340
2341                         if (max_ps == -1)
2342                                 max_ps = state;
2343
2344                         if (total_latency_us > max_lat_us)
2345                                 max_lat_us = total_latency_us;
2346                 }
2347
2348                 apste = 1;
2349
2350                 if (max_ps == -1) {
2351                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2352                 } else {
2353                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2354                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2355                 }
2356         }
2357
2358         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2359                                 table, sizeof(*table), NULL);
2360         if (ret)
2361                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2362
2363         kfree(table);
2364         return ret;
2365 }
2366
2367 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2368 {
2369         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2370         u64 latency;
2371
2372         switch (val) {
2373         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2374         case PM_QOS_LATENCY_ANY:
2375                 latency = U64_MAX;
2376                 break;
2377
2378         default:
2379                 latency = val;
2380         }
2381
2382         if (ctrl->ps_max_latency_us != latency) {
2383                 ctrl->ps_max_latency_us = latency;
2384                 nvme_configure_apst(ctrl);
2385         }
2386 }
2387
2388 struct nvme_core_quirk_entry {
2389         /*
2390          * NVMe model and firmware strings are padded with spaces.  For
2391          * simplicity, strings in the quirk table are padded with NULLs
2392          * instead.
2393          */
2394         u16 vid;
2395         const char *mn;
2396         const char *fr;
2397         unsigned long quirks;
2398 };
2399
2400 static const struct nvme_core_quirk_entry core_quirks[] = {
2401         {
2402                 /*
2403                  * This Toshiba device seems to die using any APST states.  See:
2404                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2405                  */
2406                 .vid = 0x1179,
2407                 .mn = "THNSF5256GPUK TOSHIBA",
2408                 .quirks = NVME_QUIRK_NO_APST,
2409         },
2410         {
2411                 /*
2412                  * This LiteON CL1-3D*-Q11 firmware version has a race
2413                  * condition associated with actions related to suspend to idle
2414                  * LiteON has resolved the problem in future firmware
2415                  */
2416                 .vid = 0x14a4,
2417                 .fr = "22301111",
2418                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2419         }
2420 };
2421
2422 /* match is null-terminated but idstr is space-padded. */
2423 static bool string_matches(const char *idstr, const char *match, size_t len)
2424 {
2425         size_t matchlen;
2426
2427         if (!match)
2428                 return true;
2429
2430         matchlen = strlen(match);
2431         WARN_ON_ONCE(matchlen > len);
2432
2433         if (memcmp(idstr, match, matchlen))
2434                 return false;
2435
2436         for (; matchlen < len; matchlen++)
2437                 if (idstr[matchlen] != ' ')
2438                         return false;
2439
2440         return true;
2441 }
2442
2443 static bool quirk_matches(const struct nvme_id_ctrl *id,
2444                           const struct nvme_core_quirk_entry *q)
2445 {
2446         return q->vid == le16_to_cpu(id->vid) &&
2447                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2448                 string_matches(id->fr, q->fr, sizeof(id->fr));
2449 }
2450
2451 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2452                 struct nvme_id_ctrl *id)
2453 {
2454         size_t nqnlen;
2455         int off;
2456
2457         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2458                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2459                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2460                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2461                         return;
2462                 }
2463
2464                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2465                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2466         }
2467
2468         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2469         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2470                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2471                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2472         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2473         off += sizeof(id->sn);
2474         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2475         off += sizeof(id->mn);
2476         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2477 }
2478
2479 static void nvme_release_subsystem(struct device *dev)
2480 {
2481         struct nvme_subsystem *subsys =
2482                 container_of(dev, struct nvme_subsystem, dev);
2483
2484         if (subsys->instance >= 0)
2485                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2486         kfree(subsys);
2487 }
2488
2489 static void nvme_destroy_subsystem(struct kref *ref)
2490 {
2491         struct nvme_subsystem *subsys =
2492                         container_of(ref, struct nvme_subsystem, ref);
2493
2494         mutex_lock(&nvme_subsystems_lock);
2495         list_del(&subsys->entry);
2496         mutex_unlock(&nvme_subsystems_lock);
2497
2498         ida_destroy(&subsys->ns_ida);
2499         device_del(&subsys->dev);
2500         put_device(&subsys->dev);
2501 }
2502
2503 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2504 {
2505         kref_put(&subsys->ref, nvme_destroy_subsystem);
2506 }
2507
2508 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2509 {
2510         struct nvme_subsystem *subsys;
2511
2512         lockdep_assert_held(&nvme_subsystems_lock);
2513
2514         /*
2515          * Fail matches for discovery subsystems. This results
2516          * in each discovery controller bound to a unique subsystem.
2517          * This avoids issues with validating controller values
2518          * that can only be true when there is a single unique subsystem.
2519          * There may be multiple and completely independent entities
2520          * that provide discovery controllers.
2521          */
2522         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2523                 return NULL;
2524
2525         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2526                 if (strcmp(subsys->subnqn, subsysnqn))
2527                         continue;
2528                 if (!kref_get_unless_zero(&subsys->ref))
2529                         continue;
2530                 return subsys;
2531         }
2532
2533         return NULL;
2534 }
2535
2536 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2537         struct device_attribute subsys_attr_##_name = \
2538                 __ATTR(_name, _mode, _show, NULL)
2539
2540 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2541                                     struct device_attribute *attr,
2542                                     char *buf)
2543 {
2544         struct nvme_subsystem *subsys =
2545                 container_of(dev, struct nvme_subsystem, dev);
2546
2547         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2548 }
2549 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2550
2551 #define nvme_subsys_show_str_function(field)                            \
2552 static ssize_t subsys_##field##_show(struct device *dev,                \
2553                             struct device_attribute *attr, char *buf)   \
2554 {                                                                       \
2555         struct nvme_subsystem *subsys =                                 \
2556                 container_of(dev, struct nvme_subsystem, dev);          \
2557         return sprintf(buf, "%.*s\n",                                   \
2558                        (int)sizeof(subsys->field), subsys->field);      \
2559 }                                                                       \
2560 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2561
2562 nvme_subsys_show_str_function(model);
2563 nvme_subsys_show_str_function(serial);
2564 nvme_subsys_show_str_function(firmware_rev);
2565
2566 static struct attribute *nvme_subsys_attrs[] = {
2567         &subsys_attr_model.attr,
2568         &subsys_attr_serial.attr,
2569         &subsys_attr_firmware_rev.attr,
2570         &subsys_attr_subsysnqn.attr,
2571 #ifdef CONFIG_NVME_MULTIPATH
2572         &subsys_attr_iopolicy.attr,
2573 #endif
2574         NULL,
2575 };
2576
2577 static struct attribute_group nvme_subsys_attrs_group = {
2578         .attrs = nvme_subsys_attrs,
2579 };
2580
2581 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2582         &nvme_subsys_attrs_group,
2583         NULL,
2584 };
2585
2586 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2587                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2588 {
2589         struct nvme_ctrl *tmp;
2590
2591         lockdep_assert_held(&nvme_subsystems_lock);
2592
2593         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2594                 if (tmp->state == NVME_CTRL_DELETING ||
2595                     tmp->state == NVME_CTRL_DEAD)
2596                         continue;
2597
2598                 if (tmp->cntlid == ctrl->cntlid) {
2599                         dev_err(ctrl->device,
2600                                 "Duplicate cntlid %u with %s, rejecting\n",
2601                                 ctrl->cntlid, dev_name(tmp->device));
2602                         return false;
2603                 }
2604
2605                 if ((id->cmic & (1 << 1)) ||
2606                     (ctrl->opts && ctrl->opts->discovery_nqn))
2607                         continue;
2608
2609                 dev_err(ctrl->device,
2610                         "Subsystem does not support multiple controllers\n");
2611                 return false;
2612         }
2613
2614         return true;
2615 }
2616
2617 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2618 {
2619         struct nvme_subsystem *subsys, *found;
2620         int ret;
2621
2622         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2623         if (!subsys)
2624                 return -ENOMEM;
2625
2626         subsys->instance = -1;
2627         mutex_init(&subsys->lock);
2628         kref_init(&subsys->ref);
2629         INIT_LIST_HEAD(&subsys->ctrls);
2630         INIT_LIST_HEAD(&subsys->nsheads);
2631         nvme_init_subnqn(subsys, ctrl, id);
2632         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2633         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2634         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2635         subsys->vendor_id = le16_to_cpu(id->vid);
2636         subsys->cmic = id->cmic;
2637         subsys->awupf = le16_to_cpu(id->awupf);
2638 #ifdef CONFIG_NVME_MULTIPATH
2639         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2640 #endif
2641
2642         subsys->dev.class = nvme_subsys_class;
2643         subsys->dev.release = nvme_release_subsystem;
2644         subsys->dev.groups = nvme_subsys_attrs_groups;
2645         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2646         device_initialize(&subsys->dev);
2647
2648         mutex_lock(&nvme_subsystems_lock);
2649         found = __nvme_find_get_subsystem(subsys->subnqn);
2650         if (found) {
2651                 put_device(&subsys->dev);
2652                 subsys = found;
2653
2654                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2655                         ret = -EINVAL;
2656                         goto out_put_subsystem;
2657                 }
2658         } else {
2659                 ret = device_add(&subsys->dev);
2660                 if (ret) {
2661                         dev_err(ctrl->device,
2662                                 "failed to register subsystem device.\n");
2663                         put_device(&subsys->dev);
2664                         goto out_unlock;
2665                 }
2666                 ida_init(&subsys->ns_ida);
2667                 list_add_tail(&subsys->entry, &nvme_subsystems);
2668         }
2669
2670         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2671                                 dev_name(ctrl->device));
2672         if (ret) {
2673                 dev_err(ctrl->device,
2674                         "failed to create sysfs link from subsystem.\n");
2675                 goto out_put_subsystem;
2676         }
2677
2678         if (!found)
2679                 subsys->instance = ctrl->instance;
2680         ctrl->subsys = subsys;
2681         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2682         mutex_unlock(&nvme_subsystems_lock);
2683         return 0;
2684
2685 out_put_subsystem:
2686         nvme_put_subsystem(subsys);
2687 out_unlock:
2688         mutex_unlock(&nvme_subsystems_lock);
2689         return ret;
2690 }
2691
2692 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2693                 void *log, size_t size, u64 offset)
2694 {
2695         struct nvme_command c = { };
2696         unsigned long dwlen = size / 4 - 1;
2697
2698         c.get_log_page.opcode = nvme_admin_get_log_page;
2699         c.get_log_page.nsid = cpu_to_le32(nsid);
2700         c.get_log_page.lid = log_page;
2701         c.get_log_page.lsp = lsp;
2702         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2703         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2704         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2705         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2706
2707         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2708 }
2709
2710 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2711 {
2712         int ret;
2713
2714         if (!ctrl->effects)
2715                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2716
2717         if (!ctrl->effects)
2718                 return 0;
2719
2720         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2721                         ctrl->effects, sizeof(*ctrl->effects), 0);
2722         if (ret) {
2723                 kfree(ctrl->effects);
2724                 ctrl->effects = NULL;
2725         }
2726         return ret;
2727 }
2728
2729 /*
2730  * Initialize the cached copies of the Identify data and various controller
2731  * register in our nvme_ctrl structure.  This should be called as soon as
2732  * the admin queue is fully up and running.
2733  */
2734 int nvme_init_identify(struct nvme_ctrl *ctrl)
2735 {
2736         struct nvme_id_ctrl *id;
2737         int ret, page_shift;
2738         u32 max_hw_sectors;
2739         bool prev_apst_enabled;
2740
2741         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2742         if (ret) {
2743                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2744                 return ret;
2745         }
2746         page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2747         ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2748
2749         if (ctrl->vs >= NVME_VS(1, 1, 0))
2750                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2751
2752         ret = nvme_identify_ctrl(ctrl, &id);
2753         if (ret) {
2754                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2755                 return -EIO;
2756         }
2757
2758         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2759                 ret = nvme_get_effects_log(ctrl);
2760                 if (ret < 0)
2761                         goto out_free;
2762         }
2763
2764         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2765                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2766
2767         if (!ctrl->identified) {
2768                 int i;
2769
2770                 ret = nvme_init_subsystem(ctrl, id);
2771                 if (ret)
2772                         goto out_free;
2773
2774                 /*
2775                  * Check for quirks.  Quirk can depend on firmware version,
2776                  * so, in principle, the set of quirks present can change
2777                  * across a reset.  As a possible future enhancement, we
2778                  * could re-scan for quirks every time we reinitialize
2779                  * the device, but we'd have to make sure that the driver
2780                  * behaves intelligently if the quirks change.
2781                  */
2782                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2783                         if (quirk_matches(id, &core_quirks[i]))
2784                                 ctrl->quirks |= core_quirks[i].quirks;
2785                 }
2786         }
2787
2788         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2789                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2790                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2791         }
2792
2793         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2794         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2795         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2796
2797         ctrl->oacs = le16_to_cpu(id->oacs);
2798         ctrl->oncs = le16_to_cpu(id->oncs);
2799         ctrl->mtfa = le16_to_cpu(id->mtfa);
2800         ctrl->oaes = le32_to_cpu(id->oaes);
2801         ctrl->wctemp = le16_to_cpu(id->wctemp);
2802         ctrl->cctemp = le16_to_cpu(id->cctemp);
2803
2804         atomic_set(&ctrl->abort_limit, id->acl + 1);
2805         ctrl->vwc = id->vwc;
2806         if (id->mdts)
2807                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2808         else
2809                 max_hw_sectors = UINT_MAX;
2810         ctrl->max_hw_sectors =
2811                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2812
2813         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2814         ctrl->sgls = le32_to_cpu(id->sgls);
2815         ctrl->kas = le16_to_cpu(id->kas);
2816         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2817         ctrl->ctratt = le32_to_cpu(id->ctratt);
2818
2819         if (id->rtd3e) {
2820                 /* us -> s */
2821                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2822
2823                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2824                                                  shutdown_timeout, 60);
2825
2826                 if (ctrl->shutdown_timeout != shutdown_timeout)
2827                         dev_info(ctrl->device,
2828                                  "Shutdown timeout set to %u seconds\n",
2829                                  ctrl->shutdown_timeout);
2830         } else
2831                 ctrl->shutdown_timeout = shutdown_timeout;
2832
2833         ctrl->npss = id->npss;
2834         ctrl->apsta = id->apsta;
2835         prev_apst_enabled = ctrl->apst_enabled;
2836         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2837                 if (force_apst && id->apsta) {
2838                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2839                         ctrl->apst_enabled = true;
2840                 } else {
2841                         ctrl->apst_enabled = false;
2842                 }
2843         } else {
2844                 ctrl->apst_enabled = id->apsta;
2845         }
2846         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2847
2848         if (ctrl->ops->flags & NVME_F_FABRICS) {
2849                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2850                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2851                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2852                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2853
2854                 /*
2855                  * In fabrics we need to verify the cntlid matches the
2856                  * admin connect
2857                  */
2858                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2859                         dev_err(ctrl->device,
2860                                 "Mismatching cntlid: Connect %u vs Identify "
2861                                 "%u, rejecting\n",
2862                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
2863                         ret = -EINVAL;
2864                         goto out_free;
2865                 }
2866
2867                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2868                         dev_err(ctrl->device,
2869                                 "keep-alive support is mandatory for fabrics\n");
2870                         ret = -EINVAL;
2871                         goto out_free;
2872                 }
2873         } else {
2874                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2875                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2876                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2877                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2878         }
2879
2880         ret = nvme_mpath_init(ctrl, id);
2881         kfree(id);
2882
2883         if (ret < 0)
2884                 return ret;
2885
2886         if (ctrl->apst_enabled && !prev_apst_enabled)
2887                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2888         else if (!ctrl->apst_enabled && prev_apst_enabled)
2889                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2890
2891         ret = nvme_configure_apst(ctrl);
2892         if (ret < 0)
2893                 return ret;
2894
2895         ret = nvme_configure_timestamp(ctrl);
2896         if (ret < 0)
2897                 return ret;
2898
2899         ret = nvme_configure_directives(ctrl);
2900         if (ret < 0)
2901                 return ret;
2902
2903         ret = nvme_configure_acre(ctrl);
2904         if (ret < 0)
2905                 return ret;
2906
2907         if (!ctrl->identified)
2908                 nvme_hwmon_init(ctrl);
2909
2910         ctrl->identified = true;
2911
2912         return 0;
2913
2914 out_free:
2915         kfree(id);
2916         return ret;
2917 }
2918 EXPORT_SYMBOL_GPL(nvme_init_identify);
2919
2920 static int nvme_dev_open(struct inode *inode, struct file *file)
2921 {
2922         struct nvme_ctrl *ctrl =
2923                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2924
2925         switch (ctrl->state) {
2926         case NVME_CTRL_LIVE:
2927                 break;
2928         default:
2929                 return -EWOULDBLOCK;
2930         }
2931
2932         file->private_data = ctrl;
2933         return 0;
2934 }
2935
2936 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2937 {
2938         struct nvme_ns *ns;
2939         int ret;
2940
2941         down_read(&ctrl->namespaces_rwsem);
2942         if (list_empty(&ctrl->namespaces)) {
2943                 ret = -ENOTTY;
2944                 goto out_unlock;
2945         }
2946
2947         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2948         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2949                 dev_warn(ctrl->device,
2950                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2951                 ret = -EINVAL;
2952                 goto out_unlock;
2953         }
2954
2955         dev_warn(ctrl->device,
2956                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2957         kref_get(&ns->kref);
2958         up_read(&ctrl->namespaces_rwsem);
2959
2960         ret = nvme_user_cmd(ctrl, ns, argp);
2961         nvme_put_ns(ns);
2962         return ret;
2963
2964 out_unlock:
2965         up_read(&ctrl->namespaces_rwsem);
2966         return ret;
2967 }
2968
2969 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2970                 unsigned long arg)
2971 {
2972         struct nvme_ctrl *ctrl = file->private_data;
2973         void __user *argp = (void __user *)arg;
2974
2975         switch (cmd) {
2976         case NVME_IOCTL_ADMIN_CMD:
2977                 return nvme_user_cmd(ctrl, NULL, argp);
2978         case NVME_IOCTL_ADMIN64_CMD:
2979                 return nvme_user_cmd64(ctrl, NULL, argp);
2980         case NVME_IOCTL_IO_CMD:
2981                 return nvme_dev_user_cmd(ctrl, argp);
2982         case NVME_IOCTL_RESET:
2983                 dev_warn(ctrl->device, "resetting controller\n");
2984                 return nvme_reset_ctrl_sync(ctrl);
2985         case NVME_IOCTL_SUBSYS_RESET:
2986                 return nvme_reset_subsystem(ctrl);
2987         case NVME_IOCTL_RESCAN:
2988                 nvme_queue_scan(ctrl);
2989                 return 0;
2990         default:
2991                 return -ENOTTY;
2992         }
2993 }
2994
2995 static const struct file_operations nvme_dev_fops = {
2996         .owner          = THIS_MODULE,
2997         .open           = nvme_dev_open,
2998         .unlocked_ioctl = nvme_dev_ioctl,
2999         .compat_ioctl   = compat_ptr_ioctl,
3000 };
3001
3002 static ssize_t nvme_sysfs_reset(struct device *dev,
3003                                 struct device_attribute *attr, const char *buf,
3004                                 size_t count)
3005 {
3006         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3007         int ret;
3008
3009         ret = nvme_reset_ctrl_sync(ctrl);
3010         if (ret < 0)
3011                 return ret;
3012         return count;
3013 }
3014 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3015
3016 static ssize_t nvme_sysfs_rescan(struct device *dev,
3017                                 struct device_attribute *attr, const char *buf,
3018                                 size_t count)
3019 {
3020         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3021
3022         nvme_queue_scan(ctrl);
3023         return count;
3024 }
3025 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3026
3027 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3028 {
3029         struct gendisk *disk = dev_to_disk(dev);
3030
3031         if (disk->fops == &nvme_fops)
3032                 return nvme_get_ns_from_dev(dev)->head;
3033         else
3034                 return disk->private_data;
3035 }
3036
3037 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3038                 char *buf)
3039 {
3040         struct nvme_ns_head *head = dev_to_ns_head(dev);
3041         struct nvme_ns_ids *ids = &head->ids;
3042         struct nvme_subsystem *subsys = head->subsys;
3043         int serial_len = sizeof(subsys->serial);
3044         int model_len = sizeof(subsys->model);
3045
3046         if (!uuid_is_null(&ids->uuid))
3047                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3048
3049         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3050                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3051
3052         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3053                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3054
3055         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3056                                   subsys->serial[serial_len - 1] == '\0'))
3057                 serial_len--;
3058         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3059                                  subsys->model[model_len - 1] == '\0'))
3060                 model_len--;
3061
3062         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3063                 serial_len, subsys->serial, model_len, subsys->model,
3064                 head->ns_id);
3065 }
3066 static DEVICE_ATTR_RO(wwid);
3067
3068 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3069                 char *buf)
3070 {
3071         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3072 }
3073 static DEVICE_ATTR_RO(nguid);
3074
3075 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3076                 char *buf)
3077 {
3078         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3079
3080         /* For backward compatibility expose the NGUID to userspace if
3081          * we have no UUID set
3082          */
3083         if (uuid_is_null(&ids->uuid)) {
3084                 printk_ratelimited(KERN_WARNING
3085                                    "No UUID available providing old NGUID\n");
3086                 return sprintf(buf, "%pU\n", ids->nguid);
3087         }
3088         return sprintf(buf, "%pU\n", &ids->uuid);
3089 }
3090 static DEVICE_ATTR_RO(uuid);
3091
3092 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3093                 char *buf)
3094 {
3095         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3096 }
3097 static DEVICE_ATTR_RO(eui);
3098
3099 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3100                 char *buf)
3101 {
3102         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3103 }
3104 static DEVICE_ATTR_RO(nsid);
3105
3106 static struct attribute *nvme_ns_id_attrs[] = {
3107         &dev_attr_wwid.attr,
3108         &dev_attr_uuid.attr,
3109         &dev_attr_nguid.attr,
3110         &dev_attr_eui.attr,
3111         &dev_attr_nsid.attr,
3112 #ifdef CONFIG_NVME_MULTIPATH
3113         &dev_attr_ana_grpid.attr,
3114         &dev_attr_ana_state.attr,
3115 #endif
3116         NULL,
3117 };
3118
3119 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3120                 struct attribute *a, int n)
3121 {
3122         struct device *dev = container_of(kobj, struct device, kobj);
3123         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3124
3125         if (a == &dev_attr_uuid.attr) {
3126                 if (uuid_is_null(&ids->uuid) &&
3127                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3128                         return 0;
3129         }
3130         if (a == &dev_attr_nguid.attr) {
3131                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3132                         return 0;
3133         }
3134         if (a == &dev_attr_eui.attr) {
3135                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3136                         return 0;
3137         }
3138 #ifdef CONFIG_NVME_MULTIPATH
3139         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3140                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3141                         return 0;
3142                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3143                         return 0;
3144         }
3145 #endif
3146         return a->mode;
3147 }
3148
3149 static const struct attribute_group nvme_ns_id_attr_group = {
3150         .attrs          = nvme_ns_id_attrs,
3151         .is_visible     = nvme_ns_id_attrs_are_visible,
3152 };
3153
3154 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3155         &nvme_ns_id_attr_group,
3156 #ifdef CONFIG_NVM
3157         &nvme_nvm_attr_group,
3158 #endif
3159         NULL,
3160 };
3161
3162 #define nvme_show_str_function(field)                                           \
3163 static ssize_t  field##_show(struct device *dev,                                \
3164                             struct device_attribute *attr, char *buf)           \
3165 {                                                                               \
3166         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3167         return sprintf(buf, "%.*s\n",                                           \
3168                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3169 }                                                                               \
3170 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3171
3172 nvme_show_str_function(model);
3173 nvme_show_str_function(serial);
3174 nvme_show_str_function(firmware_rev);
3175
3176 #define nvme_show_int_function(field)                                           \
3177 static ssize_t  field##_show(struct device *dev,                                \
3178                             struct device_attribute *attr, char *buf)           \
3179 {                                                                               \
3180         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3181         return sprintf(buf, "%d\n", ctrl->field);       \
3182 }                                                                               \
3183 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3184
3185 nvme_show_int_function(cntlid);
3186 nvme_show_int_function(numa_node);
3187 nvme_show_int_function(queue_count);
3188 nvme_show_int_function(sqsize);
3189
3190 static ssize_t nvme_sysfs_delete(struct device *dev,
3191                                 struct device_attribute *attr, const char *buf,
3192                                 size_t count)
3193 {
3194         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3195
3196         if (device_remove_file_self(dev, attr))
3197                 nvme_delete_ctrl_sync(ctrl);
3198         return count;
3199 }
3200 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3201
3202 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3203                                          struct device_attribute *attr,
3204                                          char *buf)
3205 {
3206         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3207
3208         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3209 }
3210 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3211
3212 static ssize_t nvme_sysfs_show_state(struct device *dev,
3213                                      struct device_attribute *attr,
3214                                      char *buf)
3215 {
3216         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3217         static const char *const state_name[] = {
3218                 [NVME_CTRL_NEW]         = "new",
3219                 [NVME_CTRL_LIVE]        = "live",
3220                 [NVME_CTRL_RESETTING]   = "resetting",
3221                 [NVME_CTRL_CONNECTING]  = "connecting",
3222                 [NVME_CTRL_DELETING]    = "deleting",
3223                 [NVME_CTRL_DEAD]        = "dead",
3224         };
3225
3226         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3227             state_name[ctrl->state])
3228                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3229
3230         return sprintf(buf, "unknown state\n");
3231 }
3232
3233 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3234
3235 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3236                                          struct device_attribute *attr,
3237                                          char *buf)
3238 {
3239         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3240
3241         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3242 }
3243 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3244
3245 static ssize_t nvme_sysfs_show_address(struct device *dev,
3246                                          struct device_attribute *attr,
3247                                          char *buf)
3248 {
3249         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3250
3251         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3252 }
3253 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3254
3255 static struct attribute *nvme_dev_attrs[] = {
3256         &dev_attr_reset_controller.attr,
3257         &dev_attr_rescan_controller.attr,
3258         &dev_attr_model.attr,
3259         &dev_attr_serial.attr,
3260         &dev_attr_firmware_rev.attr,
3261         &dev_attr_cntlid.attr,
3262         &dev_attr_delete_controller.attr,
3263         &dev_attr_transport.attr,
3264         &dev_attr_subsysnqn.attr,
3265         &dev_attr_address.attr,
3266         &dev_attr_state.attr,
3267         &dev_attr_numa_node.attr,
3268         &dev_attr_queue_count.attr,
3269         &dev_attr_sqsize.attr,
3270         NULL
3271 };
3272
3273 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3274                 struct attribute *a, int n)
3275 {
3276         struct device *dev = container_of(kobj, struct device, kobj);
3277         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3278
3279         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3280                 return 0;
3281         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3282                 return 0;
3283
3284         return a->mode;
3285 }
3286
3287 static struct attribute_group nvme_dev_attrs_group = {
3288         .attrs          = nvme_dev_attrs,
3289         .is_visible     = nvme_dev_attrs_are_visible,
3290 };
3291
3292 static const struct attribute_group *nvme_dev_attr_groups[] = {
3293         &nvme_dev_attrs_group,
3294         NULL,
3295 };
3296
3297 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3298                 unsigned nsid)
3299 {
3300         struct nvme_ns_head *h;
3301
3302         lockdep_assert_held(&subsys->lock);
3303
3304         list_for_each_entry(h, &subsys->nsheads, entry) {
3305                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3306                         return h;
3307         }
3308
3309         return NULL;
3310 }
3311
3312 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3313                 struct nvme_ns_head *new)
3314 {
3315         struct nvme_ns_head *h;
3316
3317         lockdep_assert_held(&subsys->lock);
3318
3319         list_for_each_entry(h, &subsys->nsheads, entry) {
3320                 if (nvme_ns_ids_valid(&new->ids) &&
3321                     !list_empty(&h->list) &&
3322                     nvme_ns_ids_equal(&new->ids, &h->ids))
3323                         return -EINVAL;
3324         }
3325
3326         return 0;
3327 }
3328
3329 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3330                 unsigned nsid, struct nvme_id_ns *id)
3331 {
3332         struct nvme_ns_head *head;
3333         size_t size = sizeof(*head);
3334         int ret = -ENOMEM;
3335
3336 #ifdef CONFIG_NVME_MULTIPATH
3337         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3338 #endif
3339
3340         head = kzalloc(size, GFP_KERNEL);
3341         if (!head)
3342                 goto out;
3343         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3344         if (ret < 0)
3345                 goto out_free_head;
3346         head->instance = ret;
3347         INIT_LIST_HEAD(&head->list);
3348         ret = init_srcu_struct(&head->srcu);
3349         if (ret)
3350                 goto out_ida_remove;
3351         head->subsys = ctrl->subsys;
3352         head->ns_id = nsid;
3353         kref_init(&head->ref);
3354
3355         ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3356         if (ret)
3357                 goto out_cleanup_srcu;
3358
3359         ret = __nvme_check_ids(ctrl->subsys, head);
3360         if (ret) {
3361                 dev_err(ctrl->device,
3362                         "duplicate IDs for nsid %d\n", nsid);
3363                 goto out_cleanup_srcu;
3364         }
3365
3366         ret = nvme_mpath_alloc_disk(ctrl, head);
3367         if (ret)
3368                 goto out_cleanup_srcu;
3369
3370         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3371
3372         kref_get(&ctrl->subsys->ref);
3373
3374         return head;
3375 out_cleanup_srcu:
3376         cleanup_srcu_struct(&head->srcu);
3377 out_ida_remove:
3378         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3379 out_free_head:
3380         kfree(head);
3381 out:
3382         if (ret > 0)
3383                 ret = blk_status_to_errno(nvme_error_status(ret));
3384         return ERR_PTR(ret);
3385 }
3386
3387 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3388                 struct nvme_id_ns *id)
3389 {
3390         struct nvme_ctrl *ctrl = ns->ctrl;
3391         bool is_shared = id->nmic & (1 << 0);
3392         struct nvme_ns_head *head = NULL;
3393         int ret = 0;
3394
3395         mutex_lock(&ctrl->subsys->lock);
3396         if (is_shared)
3397                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3398         if (!head) {
3399                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3400                 if (IS_ERR(head)) {
3401                         ret = PTR_ERR(head);
3402                         goto out_unlock;
3403                 }
3404         } else {
3405                 struct nvme_ns_ids ids;
3406
3407                 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3408                 if (ret)
3409                         goto out_unlock;
3410
3411                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3412                         dev_err(ctrl->device,
3413                                 "IDs don't match for shared namespace %d\n",
3414                                         nsid);
3415                         ret = -EINVAL;
3416                         goto out_unlock;
3417                 }
3418         }
3419
3420         list_add_tail(&ns->siblings, &head->list);
3421         ns->head = head;
3422
3423 out_unlock:
3424         mutex_unlock(&ctrl->subsys->lock);
3425         if (ret > 0)
3426                 ret = blk_status_to_errno(nvme_error_status(ret));
3427         return ret;
3428 }
3429
3430 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3431 {
3432         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3433         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3434
3435         return nsa->head->ns_id - nsb->head->ns_id;
3436 }
3437
3438 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3439 {
3440         struct nvme_ns *ns, *ret = NULL;
3441
3442         down_read(&ctrl->namespaces_rwsem);
3443         list_for_each_entry(ns, &ctrl->namespaces, list) {
3444                 if (ns->head->ns_id == nsid) {
3445                         if (!kref_get_unless_zero(&ns->kref))
3446                                 continue;
3447                         ret = ns;
3448                         break;
3449                 }
3450                 if (ns->head->ns_id > nsid)
3451                         break;
3452         }
3453         up_read(&ctrl->namespaces_rwsem);
3454         return ret;
3455 }
3456
3457 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3458 {
3459         struct streams_directive_params s;
3460         int ret;
3461
3462         if (!ctrl->nr_streams)
3463                 return 0;
3464
3465         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3466         if (ret)
3467                 return ret;
3468
3469         ns->sws = le32_to_cpu(s.sws);
3470         ns->sgs = le16_to_cpu(s.sgs);
3471
3472         if (ns->sws) {
3473                 unsigned int bs = 1 << ns->lba_shift;
3474
3475                 blk_queue_io_min(ns->queue, bs * ns->sws);
3476                 if (ns->sgs)
3477                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3478         }
3479
3480         return 0;
3481 }
3482
3483 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3484 {
3485         struct nvme_ns *ns;
3486         struct gendisk *disk;
3487         struct nvme_id_ns *id;
3488         char disk_name[DISK_NAME_LEN];
3489         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3490
3491         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3492         if (!ns)
3493                 return -ENOMEM;
3494
3495         ns->queue = blk_mq_init_queue(ctrl->tagset);
3496         if (IS_ERR(ns->queue)) {
3497                 ret = PTR_ERR(ns->queue);
3498                 goto out_free_ns;
3499         }
3500
3501         if (ctrl->opts && ctrl->opts->data_digest)
3502                 ns->queue->backing_dev_info->capabilities
3503                         |= BDI_CAP_STABLE_WRITES;
3504
3505         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3506         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3507                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3508
3509         ns->queue->queuedata = ns;
3510         ns->ctrl = ctrl;
3511
3512         kref_init(&ns->kref);
3513         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3514
3515         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3516         nvme_set_queue_limits(ctrl, ns->queue);
3517
3518         ret = nvme_identify_ns(ctrl, nsid, &id);
3519         if (ret)
3520                 goto out_free_queue;
3521
3522         if (id->ncap == 0) {
3523                 ret = -EINVAL;
3524                 goto out_free_id;
3525         }
3526
3527         ret = nvme_init_ns_head(ns, nsid, id);
3528         if (ret)
3529                 goto out_free_id;
3530         nvme_setup_streams_ns(ctrl, ns);
3531         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3532
3533         disk = alloc_disk_node(0, node);
3534         if (!disk) {
3535                 ret = -ENOMEM;
3536                 goto out_unlink_ns;
3537         }
3538
3539         disk->fops = &nvme_fops;
3540         disk->private_data = ns;
3541         disk->queue = ns->queue;
3542         disk->flags = flags;
3543         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3544         ns->disk = disk;
3545
3546         __nvme_revalidate_disk(disk, id);
3547
3548         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3549                 ret = nvme_nvm_register(ns, disk_name, node);
3550                 if (ret) {
3551                         dev_warn(ctrl->device, "LightNVM init failure\n");
3552                         goto out_put_disk;
3553                 }
3554         }
3555
3556         down_write(&ctrl->namespaces_rwsem);
3557         list_add_tail(&ns->list, &ctrl->namespaces);
3558         up_write(&ctrl->namespaces_rwsem);
3559
3560         nvme_get_ctrl(ctrl);
3561
3562         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3563
3564         nvme_mpath_add_disk(ns, id);
3565         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3566         kfree(id);
3567
3568         return 0;
3569  out_put_disk:
3570         put_disk(ns->disk);
3571  out_unlink_ns:
3572         mutex_lock(&ctrl->subsys->lock);
3573         list_del_rcu(&ns->siblings);
3574         mutex_unlock(&ctrl->subsys->lock);
3575         nvme_put_ns_head(ns->head);
3576  out_free_id:
3577         kfree(id);
3578  out_free_queue:
3579         blk_cleanup_queue(ns->queue);
3580  out_free_ns:
3581         kfree(ns);
3582         if (ret > 0)
3583                 ret = blk_status_to_errno(nvme_error_status(ret));
3584         return ret;
3585 }
3586
3587 static void nvme_ns_remove(struct nvme_ns *ns)
3588 {
3589         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3590                 return;
3591
3592         nvme_fault_inject_fini(&ns->fault_inject);
3593
3594         mutex_lock(&ns->ctrl->subsys->lock);
3595         list_del_rcu(&ns->siblings);
3596         mutex_unlock(&ns->ctrl->subsys->lock);
3597         synchronize_rcu(); /* guarantee not available in head->list */
3598         nvme_mpath_clear_current_path(ns);
3599         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3600
3601         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3602                 del_gendisk(ns->disk);
3603                 blk_cleanup_queue(ns->queue);
3604                 if (blk_get_integrity(ns->disk))
3605                         blk_integrity_unregister(ns->disk);
3606         }
3607
3608         down_write(&ns->ctrl->namespaces_rwsem);
3609         list_del_init(&ns->list);
3610         up_write(&ns->ctrl->namespaces_rwsem);
3611
3612         nvme_mpath_check_last_path(ns);
3613         nvme_put_ns(ns);
3614 }
3615
3616 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3617 {
3618         struct nvme_ns *ns;
3619
3620         ns = nvme_find_get_ns(ctrl, nsid);
3621         if (ns) {
3622                 if (ns->disk && revalidate_disk(ns->disk))
3623                         nvme_ns_remove(ns);
3624                 nvme_put_ns(ns);
3625         } else
3626                 nvme_alloc_ns(ctrl, nsid);
3627 }
3628
3629 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3630                                         unsigned nsid)
3631 {
3632         struct nvme_ns *ns, *next;
3633         LIST_HEAD(rm_list);
3634
3635         down_write(&ctrl->namespaces_rwsem);
3636         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3637                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3638                         list_move_tail(&ns->list, &rm_list);
3639         }
3640         up_write(&ctrl->namespaces_rwsem);
3641
3642         list_for_each_entry_safe(ns, next, &rm_list, list)
3643                 nvme_ns_remove(ns);
3644
3645 }
3646
3647 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3648 {
3649         struct nvme_ns *ns;
3650         __le32 *ns_list;
3651         unsigned i, j, nsid, prev = 0;
3652         unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3653         int ret = 0;
3654
3655         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3656         if (!ns_list)
3657                 return -ENOMEM;
3658
3659         for (i = 0; i < num_lists; i++) {
3660                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3661                 if (ret)
3662                         goto free;
3663
3664                 for (j = 0; j < min(nn, 1024U); j++) {
3665                         nsid = le32_to_cpu(ns_list[j]);
3666                         if (!nsid)
3667                                 goto out;
3668
3669                         nvme_validate_ns(ctrl, nsid);
3670
3671                         while (++prev < nsid) {
3672                                 ns = nvme_find_get_ns(ctrl, prev);
3673                                 if (ns) {
3674                                         nvme_ns_remove(ns);
3675                                         nvme_put_ns(ns);
3676                                 }
3677                         }
3678                 }
3679                 nn -= j;
3680         }
3681  out:
3682         nvme_remove_invalid_namespaces(ctrl, prev);
3683  free:
3684         kfree(ns_list);
3685         return ret;
3686 }
3687
3688 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3689 {
3690         unsigned i;
3691
3692         for (i = 1; i <= nn; i++)
3693                 nvme_validate_ns(ctrl, i);
3694
3695         nvme_remove_invalid_namespaces(ctrl, nn);
3696 }
3697
3698 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3699 {
3700         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3701         __le32 *log;
3702         int error;
3703
3704         log = kzalloc(log_size, GFP_KERNEL);
3705         if (!log)
3706                 return;
3707
3708         /*
3709          * We need to read the log to clear the AEN, but we don't want to rely
3710          * on it for the changed namespace information as userspace could have
3711          * raced with us in reading the log page, which could cause us to miss
3712          * updates.
3713          */
3714         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3715                         log_size, 0);
3716         if (error)
3717                 dev_warn(ctrl->device,
3718                         "reading changed ns log failed: %d\n", error);
3719
3720         kfree(log);
3721 }
3722
3723 static void nvme_scan_work(struct work_struct *work)
3724 {
3725         struct nvme_ctrl *ctrl =
3726                 container_of(work, struct nvme_ctrl, scan_work);
3727         struct nvme_id_ctrl *id;
3728         unsigned nn;
3729
3730         /* No tagset on a live ctrl means IO queues could not created */
3731         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3732                 return;
3733
3734         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3735                 dev_info(ctrl->device, "rescanning namespaces.\n");
3736                 nvme_clear_changed_ns_log(ctrl);
3737         }
3738
3739         if (nvme_identify_ctrl(ctrl, &id))
3740                 return;
3741
3742         mutex_lock(&ctrl->scan_lock);
3743         nn = le32_to_cpu(id->nn);
3744         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3745             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3746                 if (!nvme_scan_ns_list(ctrl, nn))
3747                         goto out_free_id;
3748         }
3749         nvme_scan_ns_sequential(ctrl, nn);
3750 out_free_id:
3751         mutex_unlock(&ctrl->scan_lock);
3752         kfree(id);
3753         down_write(&ctrl->namespaces_rwsem);
3754         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3755         up_write(&ctrl->namespaces_rwsem);
3756 }
3757
3758 /*
3759  * This function iterates the namespace list unlocked to allow recovery from
3760  * controller failure. It is up to the caller to ensure the namespace list is
3761  * not modified by scan work while this function is executing.
3762  */
3763 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3764 {
3765         struct nvme_ns *ns, *next;
3766         LIST_HEAD(ns_list);
3767
3768         /*
3769          * make sure to requeue I/O to all namespaces as these
3770          * might result from the scan itself and must complete
3771          * for the scan_work to make progress
3772          */
3773         nvme_mpath_clear_ctrl_paths(ctrl);
3774
3775         /* prevent racing with ns scanning */
3776         flush_work(&ctrl->scan_work);
3777
3778         /*
3779          * The dead states indicates the controller was not gracefully
3780          * disconnected. In that case, we won't be able to flush any data while
3781          * removing the namespaces' disks; fail all the queues now to avoid
3782          * potentially having to clean up the failed sync later.
3783          */
3784         if (ctrl->state == NVME_CTRL_DEAD)
3785                 nvme_kill_queues(ctrl);
3786
3787         down_write(&ctrl->namespaces_rwsem);
3788         list_splice_init(&ctrl->namespaces, &ns_list);
3789         up_write(&ctrl->namespaces_rwsem);
3790
3791         list_for_each_entry_safe(ns, next, &ns_list, list)
3792                 nvme_ns_remove(ns);
3793 }
3794 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3795
3796 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3797 {
3798         struct nvme_ctrl *ctrl =
3799                 container_of(dev, struct nvme_ctrl, ctrl_device);
3800         struct nvmf_ctrl_options *opts = ctrl->opts;
3801         int ret;
3802
3803         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3804         if (ret)
3805                 return ret;
3806
3807         if (opts) {
3808                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3809                 if (ret)
3810                         return ret;
3811
3812                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3813                                 opts->trsvcid ?: "none");
3814                 if (ret)
3815                         return ret;
3816
3817                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3818                                 opts->host_traddr ?: "none");
3819         }
3820         return ret;
3821 }
3822
3823 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3824 {
3825         char *envp[2] = { NULL, NULL };
3826         u32 aen_result = ctrl->aen_result;
3827
3828         ctrl->aen_result = 0;
3829         if (!aen_result)
3830                 return;
3831
3832         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3833         if (!envp[0])
3834                 return;
3835         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3836         kfree(envp[0]);
3837 }
3838
3839 static void nvme_async_event_work(struct work_struct *work)
3840 {
3841         struct nvme_ctrl *ctrl =
3842                 container_of(work, struct nvme_ctrl, async_event_work);
3843
3844         nvme_aen_uevent(ctrl);
3845         ctrl->ops->submit_async_event(ctrl);
3846 }
3847
3848 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3849 {
3850
3851         u32 csts;
3852
3853         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3854                 return false;
3855
3856         if (csts == ~0)
3857                 return false;
3858
3859         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3860 }
3861
3862 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3863 {
3864         struct nvme_fw_slot_info_log *log;
3865
3866         log = kmalloc(sizeof(*log), GFP_KERNEL);
3867         if (!log)
3868                 return;
3869
3870         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3871                         sizeof(*log), 0))
3872                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3873         kfree(log);
3874 }
3875
3876 static void nvme_fw_act_work(struct work_struct *work)
3877 {
3878         struct nvme_ctrl *ctrl = container_of(work,
3879                                 struct nvme_ctrl, fw_act_work);
3880         unsigned long fw_act_timeout;
3881
3882         if (ctrl->mtfa)
3883                 fw_act_timeout = jiffies +
3884                                 msecs_to_jiffies(ctrl->mtfa * 100);
3885         else
3886                 fw_act_timeout = jiffies +
3887                                 msecs_to_jiffies(admin_timeout * 1000);
3888
3889         nvme_stop_queues(ctrl);
3890         while (nvme_ctrl_pp_status(ctrl)) {
3891                 if (time_after(jiffies, fw_act_timeout)) {
3892                         dev_warn(ctrl->device,
3893                                 "Fw activation timeout, reset controller\n");
3894                         nvme_try_sched_reset(ctrl);
3895                         return;
3896                 }
3897                 msleep(100);
3898         }
3899
3900         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3901                 return;
3902
3903         nvme_start_queues(ctrl);
3904         /* read FW slot information to clear the AER */
3905         nvme_get_fw_slot_info(ctrl);
3906 }
3907
3908 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3909 {
3910         u32 aer_notice_type = (result & 0xff00) >> 8;
3911
3912         trace_nvme_async_event(ctrl, aer_notice_type);
3913
3914         switch (aer_notice_type) {
3915         case NVME_AER_NOTICE_NS_CHANGED:
3916                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3917                 nvme_queue_scan(ctrl);
3918                 break;
3919         case NVME_AER_NOTICE_FW_ACT_STARTING:
3920                 /*
3921                  * We are (ab)using the RESETTING state to prevent subsequent
3922                  * recovery actions from interfering with the controller's
3923                  * firmware activation.
3924                  */
3925                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
3926                         queue_work(nvme_wq, &ctrl->fw_act_work);
3927                 break;
3928 #ifdef CONFIG_NVME_MULTIPATH
3929         case NVME_AER_NOTICE_ANA:
3930                 if (!ctrl->ana_log_buf)
3931                         break;
3932                 queue_work(nvme_wq, &ctrl->ana_work);
3933                 break;
3934 #endif
3935         case NVME_AER_NOTICE_DISC_CHANGED:
3936                 ctrl->aen_result = result;
3937                 break;
3938         default:
3939                 dev_warn(ctrl->device, "async event result %08x\n", result);
3940         }
3941 }
3942
3943 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3944                 volatile union nvme_result *res)
3945 {
3946         u32 result = le32_to_cpu(res->u32);
3947         u32 aer_type = result & 0x07;
3948
3949         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3950                 return;
3951
3952         switch (aer_type) {
3953         case NVME_AER_NOTICE:
3954                 nvme_handle_aen_notice(ctrl, result);
3955                 break;
3956         case NVME_AER_ERROR:
3957         case NVME_AER_SMART:
3958         case NVME_AER_CSS:
3959         case NVME_AER_VS:
3960                 trace_nvme_async_event(ctrl, aer_type);
3961                 ctrl->aen_result = result;
3962                 break;
3963         default:
3964                 break;
3965         }
3966         queue_work(nvme_wq, &ctrl->async_event_work);
3967 }
3968 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3969
3970 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3971 {
3972         nvme_mpath_stop(ctrl);
3973         nvme_stop_keep_alive(ctrl);
3974         flush_work(&ctrl->async_event_work);
3975         cancel_work_sync(&ctrl->fw_act_work);
3976 }
3977 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3978
3979 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3980 {
3981         if (ctrl->kato)
3982                 nvme_start_keep_alive(ctrl);
3983
3984         nvme_enable_aen(ctrl);
3985
3986         if (ctrl->queue_count > 1) {
3987                 nvme_queue_scan(ctrl);
3988                 nvme_start_queues(ctrl);
3989         }
3990 }
3991 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3992
3993 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3994 {
3995         nvme_fault_inject_fini(&ctrl->fault_inject);
3996         dev_pm_qos_hide_latency_tolerance(ctrl->device);
3997         cdev_device_del(&ctrl->cdev, ctrl->device);
3998 }
3999 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4000
4001 static void nvme_free_ctrl(struct device *dev)
4002 {
4003         struct nvme_ctrl *ctrl =
4004                 container_of(dev, struct nvme_ctrl, ctrl_device);
4005         struct nvme_subsystem *subsys = ctrl->subsys;
4006
4007         if (subsys && ctrl->instance != subsys->instance)
4008                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4009
4010         kfree(ctrl->effects);
4011         nvme_mpath_uninit(ctrl);
4012         __free_page(ctrl->discard_page);
4013
4014         if (subsys) {
4015                 mutex_lock(&nvme_subsystems_lock);
4016                 list_del(&ctrl->subsys_entry);
4017                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4018                 mutex_unlock(&nvme_subsystems_lock);
4019         }
4020
4021         ctrl->ops->free_ctrl(ctrl);
4022
4023         if (subsys)
4024                 nvme_put_subsystem(subsys);
4025 }
4026
4027 /*
4028  * Initialize a NVMe controller structures.  This needs to be called during
4029  * earliest initialization so that we have the initialized structured around
4030  * during probing.
4031  */
4032 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4033                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4034 {
4035         int ret;
4036
4037         ctrl->state = NVME_CTRL_NEW;
4038         spin_lock_init(&ctrl->lock);
4039         mutex_init(&ctrl->scan_lock);
4040         INIT_LIST_HEAD(&ctrl->namespaces);
4041         init_rwsem(&ctrl->namespaces_rwsem);
4042         ctrl->dev = dev;
4043         ctrl->ops = ops;
4044         ctrl->quirks = quirks;
4045         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4046         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4047         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4048         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4049         init_waitqueue_head(&ctrl->state_wq);
4050
4051         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4052         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4053         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4054
4055         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4056                         PAGE_SIZE);
4057         ctrl->discard_page = alloc_page(GFP_KERNEL);
4058         if (!ctrl->discard_page) {
4059                 ret = -ENOMEM;
4060                 goto out;
4061         }
4062
4063         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4064         if (ret < 0)
4065                 goto out;
4066         ctrl->instance = ret;
4067
4068         device_initialize(&ctrl->ctrl_device);
4069         ctrl->device = &ctrl->ctrl_device;
4070         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4071         ctrl->device->class = nvme_class;
4072         ctrl->device->parent = ctrl->dev;
4073         ctrl->device->groups = nvme_dev_attr_groups;
4074         ctrl->device->release = nvme_free_ctrl;
4075         dev_set_drvdata(ctrl->device, ctrl);
4076         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4077         if (ret)
4078                 goto out_release_instance;
4079
4080         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4081         ctrl->cdev.owner = ops->module;
4082         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4083         if (ret)
4084                 goto out_free_name;
4085
4086         /*
4087          * Initialize latency tolerance controls.  The sysfs files won't
4088          * be visible to userspace unless the device actually supports APST.
4089          */
4090         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4091         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4092                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4093
4094         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4095
4096         return 0;
4097 out_free_name:
4098         kfree_const(ctrl->device->kobj.name);
4099 out_release_instance:
4100         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4101 out:
4102         if (ctrl->discard_page)
4103                 __free_page(ctrl->discard_page);
4104         return ret;
4105 }
4106 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4107
4108 /**
4109  * nvme_kill_queues(): Ends all namespace queues
4110  * @ctrl: the dead controller that needs to end
4111  *
4112  * Call this function when the driver determines it is unable to get the
4113  * controller in a state capable of servicing IO.
4114  */
4115 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4116 {
4117         struct nvme_ns *ns;
4118
4119         down_read(&ctrl->namespaces_rwsem);
4120
4121         /* Forcibly unquiesce queues to avoid blocking dispatch */
4122         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4123                 blk_mq_unquiesce_queue(ctrl->admin_q);
4124
4125         list_for_each_entry(ns, &ctrl->namespaces, list)
4126                 nvme_set_queue_dying(ns);
4127
4128         up_read(&ctrl->namespaces_rwsem);
4129 }
4130 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4131
4132 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4133 {
4134         struct nvme_ns *ns;
4135
4136         down_read(&ctrl->namespaces_rwsem);
4137         list_for_each_entry(ns, &ctrl->namespaces, list)
4138                 blk_mq_unfreeze_queue(ns->queue);
4139         up_read(&ctrl->namespaces_rwsem);
4140 }
4141 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4142
4143 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4144 {
4145         struct nvme_ns *ns;
4146
4147         down_read(&ctrl->namespaces_rwsem);
4148         list_for_each_entry(ns, &ctrl->namespaces, list) {
4149                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4150                 if (timeout <= 0)
4151                         break;
4152         }
4153         up_read(&ctrl->namespaces_rwsem);
4154 }
4155 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4156
4157 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4158 {
4159         struct nvme_ns *ns;
4160
4161         down_read(&ctrl->namespaces_rwsem);
4162         list_for_each_entry(ns, &ctrl->namespaces, list)
4163                 blk_mq_freeze_queue_wait(ns->queue);
4164         up_read(&ctrl->namespaces_rwsem);
4165 }
4166 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4167
4168 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4169 {
4170         struct nvme_ns *ns;
4171
4172         down_read(&ctrl->namespaces_rwsem);
4173         list_for_each_entry(ns, &ctrl->namespaces, list)
4174                 blk_freeze_queue_start(ns->queue);
4175         up_read(&ctrl->namespaces_rwsem);
4176 }
4177 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4178
4179 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4180 {
4181         struct nvme_ns *ns;
4182
4183         down_read(&ctrl->namespaces_rwsem);
4184         list_for_each_entry(ns, &ctrl->namespaces, list)
4185                 blk_mq_quiesce_queue(ns->queue);
4186         up_read(&ctrl->namespaces_rwsem);
4187 }
4188 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4189
4190 void nvme_start_queues(struct nvme_ctrl *ctrl)
4191 {
4192         struct nvme_ns *ns;
4193
4194         down_read(&ctrl->namespaces_rwsem);
4195         list_for_each_entry(ns, &ctrl->namespaces, list)
4196                 blk_mq_unquiesce_queue(ns->queue);
4197         up_read(&ctrl->namespaces_rwsem);
4198 }
4199 EXPORT_SYMBOL_GPL(nvme_start_queues);
4200
4201
4202 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4203 {
4204         struct nvme_ns *ns;
4205
4206         down_read(&ctrl->namespaces_rwsem);
4207         list_for_each_entry(ns, &ctrl->namespaces, list)
4208                 blk_sync_queue(ns->queue);
4209         up_read(&ctrl->namespaces_rwsem);
4210
4211         if (ctrl->admin_q)
4212                 blk_sync_queue(ctrl->admin_q);
4213 }
4214 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4215
4216 /*
4217  * Check we didn't inadvertently grow the command structure sizes:
4218  */
4219 static inline void _nvme_check_size(void)
4220 {
4221         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4222         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4223         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4224         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4225         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4226         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4227         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4228         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4229         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4230         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4231         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4232         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4233         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4234         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4235         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4236         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4237         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4238 }
4239
4240
4241 static int __init nvme_core_init(void)
4242 {
4243         int result = -ENOMEM;
4244
4245         _nvme_check_size();
4246
4247         nvme_wq = alloc_workqueue("nvme-wq",
4248                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4249         if (!nvme_wq)
4250                 goto out;
4251
4252         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4253                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4254         if (!nvme_reset_wq)
4255                 goto destroy_wq;
4256
4257         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4258                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4259         if (!nvme_delete_wq)
4260                 goto destroy_reset_wq;
4261
4262         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4263         if (result < 0)
4264                 goto destroy_delete_wq;
4265
4266         nvme_class = class_create(THIS_MODULE, "nvme");
4267         if (IS_ERR(nvme_class)) {
4268                 result = PTR_ERR(nvme_class);
4269                 goto unregister_chrdev;
4270         }
4271         nvme_class->dev_uevent = nvme_class_uevent;
4272
4273         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4274         if (IS_ERR(nvme_subsys_class)) {
4275                 result = PTR_ERR(nvme_subsys_class);
4276                 goto destroy_class;
4277         }
4278         return 0;
4279
4280 destroy_class:
4281         class_destroy(nvme_class);
4282 unregister_chrdev:
4283         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4284 destroy_delete_wq:
4285         destroy_workqueue(nvme_delete_wq);
4286 destroy_reset_wq:
4287         destroy_workqueue(nvme_reset_wq);
4288 destroy_wq:
4289         destroy_workqueue(nvme_wq);
4290 out:
4291         return result;
4292 }
4293
4294 static void __exit nvme_core_exit(void)
4295 {
4296         class_destroy(nvme_subsys_class);
4297         class_destroy(nvme_class);
4298         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4299         destroy_workqueue(nvme_delete_wq);
4300         destroy_workqueue(nvme_reset_wq);
4301         destroy_workqueue(nvme_wq);
4302 }
4303
4304 MODULE_LICENSE("GPL");
4305 MODULE_VERSION("1.0");
4306 module_init(nvme_core_init);
4307 module_exit(nvme_core_exit);