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/compat.h>
  10 #include <linux/delay.h>
  11 #include <linux/errno.h>
  12 #include <linux/hdreg.h>
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/list_sort.h>
  17 #include <linux/slab.h>
  18 #include <linux/types.h>
  19 #include <linux/pr.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/nvme_ioctl.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 as scan, aen handling, fw activation,
  70  * keep-alive, 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 void nvme_put_subsystem(struct nvme_subsystem *subsys);
  93 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  94                                            unsigned nsid);
  95
  96 /*
  97  * Prepare a queue for teardown.
  98  *
  99  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
 100  * the capacity to 0 after that to avoid blocking dispatchers that may be
 101  * holding bd_butex.  This will end buffered writers dirtying pages that can't
 102  * be synced.
 103  */
 104 static void nvme_set_queue_dying(struct nvme_ns *ns)
 105 {
 106         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 107                 return;
 108
 109         blk_set_queue_dying(ns->queue);
 110         blk_mq_unquiesce_queue(ns->queue);
 111
 112         set_capacity_and_notify(ns->disk, 0);
 113 }
 114
 115 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 116 {
 117         /*
 118          * Only new queue scan work when admin and IO queues are both alive
 119          */
 120         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 121                 queue_work(nvme_wq, &ctrl->scan_work);
 122 }
 123
 124 /*
 125  * Use this function to proceed with scheduling reset_work for a controller
 126  * that had previously been set to the resetting state. This is intended for
 127  * code paths that can't be interrupted by other reset attempts. A hot removal
 128  * may prevent this from succeeding.
 129  */
 130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 131 {
 132         if (ctrl->state != NVME_CTRL_RESETTING)
 133                 return -EBUSY;
 134         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 135                 return -EBUSY;
 136         return 0;
 137 }
 138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 139
 140 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 141 {
 142         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 143                 return -EBUSY;
 144         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 145                 return -EBUSY;
 146         return 0;
 147 }
 148 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 149
 150 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 151 {
 152         int ret;
 153
 154         ret = nvme_reset_ctrl(ctrl);
 155         if (!ret) {
 156                 flush_work(&ctrl->reset_work);
 157                 if (ctrl->state != NVME_CTRL_LIVE)
 158                         ret = -ENETRESET;
 159         }
 160
 161         return ret;
 162 }
 163 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 164
 165 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 166 {
 167         dev_info(ctrl->device,
 168                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 169
 170         flush_work(&ctrl->reset_work);
 171         nvme_stop_ctrl(ctrl);
 172         nvme_remove_namespaces(ctrl);
 173         ctrl->ops->delete_ctrl(ctrl);
 174         nvme_uninit_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 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 196 {
 197         /*
 198          * Keep a reference until nvme_do_delete_ctrl() complete,
 199          * since ->delete_ctrl can free the controller.
 200          */
 201         nvme_get_ctrl(ctrl);
 202         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 203                 nvme_do_delete_ctrl(ctrl);
 204         nvme_put_ctrl(ctrl);
 205 }
 206
 207 static blk_status_t nvme_error_status(u16 status)
 208 {
 209         switch (status & 0x7ff) {
 210         case NVME_SC_SUCCESS:
 211                 return BLK_STS_OK;
 212         case NVME_SC_CAP_EXCEEDED:
 213                 return BLK_STS_NOSPC;
 214         case NVME_SC_LBA_RANGE:
 215         case NVME_SC_CMD_INTERRUPTED:
 216         case NVME_SC_NS_NOT_READY:
 217                 return BLK_STS_TARGET;
 218         case NVME_SC_BAD_ATTRIBUTES:
 219         case NVME_SC_ONCS_NOT_SUPPORTED:
 220         case NVME_SC_INVALID_OPCODE:
 221         case NVME_SC_INVALID_FIELD:
 222         case NVME_SC_INVALID_NS:
 223                 return BLK_STS_NOTSUPP;
 224         case NVME_SC_WRITE_FAULT:
 225         case NVME_SC_READ_ERROR:
 226         case NVME_SC_UNWRITTEN_BLOCK:
 227         case NVME_SC_ACCESS_DENIED:
 228         case NVME_SC_READ_ONLY:
 229         case NVME_SC_COMPARE_FAILED:
 230                 return BLK_STS_MEDIUM;
 231         case NVME_SC_GUARD_CHECK:
 232         case NVME_SC_APPTAG_CHECK:
 233         case NVME_SC_REFTAG_CHECK:
 234         case NVME_SC_INVALID_PI:
 235                 return BLK_STS_PROTECTION;
 236         case NVME_SC_RESERVATION_CONFLICT:
 237                 return BLK_STS_NEXUS;
 238         case NVME_SC_HOST_PATH_ERROR:
 239                 return BLK_STS_TRANSPORT;
 240         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 241                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 242         case NVME_SC_ZONE_TOO_MANY_OPEN:
 243                 return BLK_STS_ZONE_OPEN_RESOURCE;
 244         default:
 245                 return BLK_STS_IOERR;
 246         }
 247 }
 248
 249 static void nvme_retry_req(struct request *req)
 250 {
 251         struct nvme_ns *ns = req->q->queuedata;
 252         unsigned long delay = 0;
 253         u16 crd;
 254
 255         /* The mask and shift result must be <= 3 */
 256         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 257         if (ns && crd)
 258                 delay = ns->ctrl->crdt[crd - 1] * 100;
 259
 260         nvme_req(req)->retries++;
 261         blk_mq_requeue_request(req, false);
 262         blk_mq_delay_kick_requeue_list(req->q, delay);
 263 }
 264
 265 enum nvme_disposition {
 266         COMPLETE,
 267         RETRY,
 268         FAILOVER,
 269 };
 270
 271 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 272 {
 273         if (likely(nvme_req(req)->status == 0))
 274                 return COMPLETE;
 275
 276         if (blk_noretry_request(req) ||
 277             (nvme_req(req)->status & NVME_SC_DNR) ||
 278             nvme_req(req)->retries >= nvme_max_retries)
 279                 return COMPLETE;
 280
 281         if (req->cmd_flags & REQ_NVME_MPATH) {
 282                 if (nvme_is_path_error(nvme_req(req)->status) ||
 283                     blk_queue_dying(req->q))
 284                         return FAILOVER;
 285         } else {
 286                 if (blk_queue_dying(req->q))
 287                         return COMPLETE;
 288         }
 289
 290         return RETRY;
 291 }
 292
 293 static inline void nvme_end_req(struct request *req)
 294 {
 295         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 296
 297         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 298             req_op(req) == REQ_OP_ZONE_APPEND)
 299                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 300                         le64_to_cpu(nvme_req(req)->result.u64));
 301
 302         nvme_trace_bio_complete(req, status);
 303         blk_mq_end_request(req, status);
 304 }
 305
 306 void nvme_complete_rq(struct request *req)
 307 {
 308         trace_nvme_complete_rq(req);
 309         nvme_cleanup_cmd(req);
 310
 311         if (nvme_req(req)->ctrl->kas)
 312                 nvme_req(req)->ctrl->comp_seen = true;
 313
 314         switch (nvme_decide_disposition(req)) {
 315         case COMPLETE:
 316                 nvme_end_req(req);
 317                 return;
 318         case RETRY:
 319                 nvme_retry_req(req);
 320                 return;
 321         case FAILOVER:
 322                 nvme_failover_req(req);
 323                 return;
 324         }
 325 }
 326 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 327
 328 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 329 {
 330         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 331                                 "Cancelling I/O %d", req->tag);
 332
 333         /* don't abort one completed request */
 334         if (blk_mq_request_completed(req))
 335                 return true;
 336
 337         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 338         blk_mq_complete_request(req);
 339         return true;
 340 }
 341 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 342
 343 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 344                 enum nvme_ctrl_state new_state)
 345 {
 346         enum nvme_ctrl_state old_state;
 347         unsigned long flags;
 348         bool changed = false;
 349
 350         spin_lock_irqsave(&ctrl->lock, flags);
 351
 352         old_state = ctrl->state;
 353         switch (new_state) {
 354         case NVME_CTRL_LIVE:
 355                 switch (old_state) {
 356                 case NVME_CTRL_NEW:
 357                 case NVME_CTRL_RESETTING:
 358                 case NVME_CTRL_CONNECTING:
 359                         changed = true;
 360                         fallthrough;
 361                 default:
 362                         break;
 363                 }
 364                 break;
 365         case NVME_CTRL_RESETTING:
 366                 switch (old_state) {
 367                 case NVME_CTRL_NEW:
 368                 case NVME_CTRL_LIVE:
 369                         changed = true;
 370                         fallthrough;
 371                 default:
 372                         break;
 373                 }
 374                 break;
 375         case NVME_CTRL_CONNECTING:
 376                 switch (old_state) {
 377                 case NVME_CTRL_NEW:
 378                 case NVME_CTRL_RESETTING:
 379                         changed = true;
 380                         fallthrough;
 381                 default:
 382                         break;
 383                 }
 384                 break;
 385         case NVME_CTRL_DELETING:
 386                 switch (old_state) {
 387                 case NVME_CTRL_LIVE:
 388                 case NVME_CTRL_RESETTING:
 389                 case NVME_CTRL_CONNECTING:
 390                         changed = true;
 391                         fallthrough;
 392                 default:
 393                         break;
 394                 }
 395                 break;
 396         case NVME_CTRL_DELETING_NOIO:
 397                 switch (old_state) {
 398                 case NVME_CTRL_DELETING:
 399                 case NVME_CTRL_DEAD:
 400                         changed = true;
 401                         fallthrough;
 402                 default:
 403                         break;
 404                 }
 405                 break;
 406         case NVME_CTRL_DEAD:
 407                 switch (old_state) {
 408                 case NVME_CTRL_DELETING:
 409                         changed = true;
 410                         fallthrough;
 411                 default:
 412                         break;
 413                 }
 414                 break;
 415         default:
 416                 break;
 417         }
 418
 419         if (changed) {
 420                 ctrl->state = new_state;
 421                 wake_up_all(&ctrl->state_wq);
 422         }
 423
 424         spin_unlock_irqrestore(&ctrl->lock, flags);
 425         if (changed && ctrl->state == NVME_CTRL_LIVE)
 426                 nvme_kick_requeue_lists(ctrl);
 427         return changed;
 428 }
 429 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 430
 431 /*
 432  * Returns true for sink states that can't ever transition back to live.
 433  */
 434 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 435 {
 436         switch (ctrl->state) {
 437         case NVME_CTRL_NEW:
 438         case NVME_CTRL_LIVE:
 439         case NVME_CTRL_RESETTING:
 440         case NVME_CTRL_CONNECTING:
 441                 return false;
 442         case NVME_CTRL_DELETING:
 443         case NVME_CTRL_DELETING_NOIO:
 444         case NVME_CTRL_DEAD:
 445                 return true;
 446         default:
 447                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 448                 return true;
 449         }
 450 }
 451
 452 /*
 453  * Waits for the controller state to be resetting, or returns false if it is
 454  * not possible to ever transition to that state.
 455  */
 456 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 457 {
 458         wait_event(ctrl->state_wq,
 459                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 460                    nvme_state_terminal(ctrl));
 461         return ctrl->state == NVME_CTRL_RESETTING;
 462 }
 463 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 464
 465 static void nvme_free_ns_head(struct kref *ref)
 466 {
 467         struct nvme_ns_head *head =
 468                 container_of(ref, struct nvme_ns_head, ref);
 469
 470         nvme_mpath_remove_disk(head);
 471         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 472         cleanup_srcu_struct(&head->srcu);
 473         nvme_put_subsystem(head->subsys);
 474         kfree(head);
 475 }
 476
 477 static void nvme_put_ns_head(struct nvme_ns_head *head)
 478 {
 479         kref_put(&head->ref, nvme_free_ns_head);
 480 }
 481
 482 static void nvme_free_ns(struct kref *kref)
 483 {
 484         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 485
 486         if (ns->ndev)
 487                 nvme_nvm_unregister(ns);
 488
 489         put_disk(ns->disk);
 490         nvme_put_ns_head(ns->head);
 491         nvme_put_ctrl(ns->ctrl);
 492         kfree(ns);
 493 }
 494
 495 void nvme_put_ns(struct nvme_ns *ns)
 496 {
 497         kref_put(&ns->kref, nvme_free_ns);
 498 }
 499 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 500
 501 static inline void nvme_clear_nvme_request(struct request *req)
 502 {
 503         if (!(req->rq_flags & RQF_DONTPREP)) {
 504                 nvme_req(req)->retries = 0;
 505                 nvme_req(req)->flags = 0;
 506                 req->rq_flags |= RQF_DONTPREP;
 507         }
 508 }
 509
 510 struct request *nvme_alloc_request(struct request_queue *q,
 511                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 512 {
 513         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 514         struct request *req;
 515
 516         if (qid == NVME_QID_ANY) {
 517                 req = blk_mq_alloc_request(q, op, flags);
 518         } else {
 519                 req = blk_mq_alloc_request_hctx(q, op, flags,
 520                                 qid ? qid - 1 : 0);
 521         }
 522         if (IS_ERR(req))
 523                 return req;
 524
 525         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 526         nvme_clear_nvme_request(req);
 527         nvme_req(req)->cmd = cmd;
 528
 529         return req;
 530 }
 531 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 532
 533 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 534 {
 535         struct nvme_command c;
 536
 537         memset(&c, 0, sizeof(c));
 538
 539         c.directive.opcode = nvme_admin_directive_send;
 540         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 541         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 542         c.directive.dtype = NVME_DIR_IDENTIFY;
 543         c.directive.tdtype = NVME_DIR_STREAMS;
 544         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 545
 546         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 547 }
 548
 549 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 550 {
 551         return nvme_toggle_streams(ctrl, false);
 552 }
 553
 554 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 555 {
 556         return nvme_toggle_streams(ctrl, true);
 557 }
 558
 559 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 560                                   struct streams_directive_params *s, u32 nsid)
 561 {
 562         struct nvme_command c;
 563
 564         memset(&c, 0, sizeof(c));
 565         memset(s, 0, sizeof(*s));
 566
 567         c.directive.opcode = nvme_admin_directive_recv;
 568         c.directive.nsid = cpu_to_le32(nsid);
 569         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 570         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 571         c.directive.dtype = NVME_DIR_STREAMS;
 572
 573         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 574 }
 575
 576 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 577 {
 578         struct streams_directive_params s;
 579         int ret;
 580
 581         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 582                 return 0;
 583         if (!streams)
 584                 return 0;
 585
 586         ret = nvme_enable_streams(ctrl);
 587         if (ret)
 588                 return ret;
 589
 590         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 591         if (ret)
 592                 goto out_disable_stream;
 593
 594         ctrl->nssa = le16_to_cpu(s.nssa);
 595         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 596                 dev_info(ctrl->device, "too few streams (%u) available\n",
 597                                         ctrl->nssa);
 598                 goto out_disable_stream;
 599         }
 600
 601         ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 602         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 603         return 0;
 604
 605 out_disable_stream:
 606         nvme_disable_streams(ctrl);
 607         return ret;
 608 }
 609
 610 /*
 611  * Check if 'req' has a write hint associated with it. If it does, assign
 612  * a valid namespace stream to the write.
 613  */
 614 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 615                                      struct request *req, u16 *control,
 616                                      u32 *dsmgmt)
 617 {
 618         enum rw_hint streamid = req->write_hint;
 619
 620         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 621                 streamid = 0;
 622         else {
 623                 streamid--;
 624                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 625                         return;
 626
 627                 *control |= NVME_RW_DTYPE_STREAMS;
 628                 *dsmgmt |= streamid << 16;
 629         }
 630
 631         if (streamid < ARRAY_SIZE(req->q->write_hints))
 632                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 633 }
 634
 635 static void nvme_setup_passthrough(struct request *req,
 636                 struct nvme_command *cmd)
 637 {
 638         memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 639         /* passthru commands should let the driver set the SGL flags */
 640         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 641 }
 642
 643 static inline void nvme_setup_flush(struct nvme_ns *ns,
 644                 struct nvme_command *cmnd)
 645 {
 646         cmnd->common.opcode = nvme_cmd_flush;
 647         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 648 }
 649
 650 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 651                 struct nvme_command *cmnd)
 652 {
 653         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 654         struct nvme_dsm_range *range;
 655         struct bio *bio;
 656
 657         /*
 658          * Some devices do not consider the DSM 'Number of Ranges' field when
 659          * determining how much data to DMA. Always allocate memory for maximum
 660          * number of segments to prevent device reading beyond end of buffer.
 661          */
 662         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 663
 664         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 665         if (!range) {
 666                 /*
 667                  * If we fail allocation our range, fallback to the controller
 668                  * discard page. If that's also busy, it's safe to return
 669                  * busy, as we know we can make progress once that's freed.
 670                  */
 671                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 672                         return BLK_STS_RESOURCE;
 673
 674                 range = page_address(ns->ctrl->discard_page);
 675         }
 676
 677         __rq_for_each_bio(bio, req) {
 678                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 679                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 680
 681                 if (n < segments) {
 682                         range[n].cattr = cpu_to_le32(0);
 683                         range[n].nlb = cpu_to_le32(nlb);
 684                         range[n].slba = cpu_to_le64(slba);
 685                 }
 686                 n++;
 687         }
 688
 689         if (WARN_ON_ONCE(n != segments)) {
 690                 if (virt_to_page(range) == ns->ctrl->discard_page)
 691                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 692                 else
 693                         kfree(range);
 694                 return BLK_STS_IOERR;
 695         }
 696
 697         cmnd->dsm.opcode = nvme_cmd_dsm;
 698         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 699         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 700         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 701
 702         req->special_vec.bv_page = virt_to_page(range);
 703         req->special_vec.bv_offset = offset_in_page(range);
 704         req->special_vec.bv_len = alloc_size;
 705         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 706
 707         return BLK_STS_OK;
 708 }
 709
 710 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 711                 struct request *req, struct nvme_command *cmnd)
 712 {
 713         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 714                 return nvme_setup_discard(ns, req, cmnd);
 715
 716         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 717         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 718         cmnd->write_zeroes.slba =
 719                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 720         cmnd->write_zeroes.length =
 721                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 722         cmnd->write_zeroes.control = 0;
 723         return BLK_STS_OK;
 724 }
 725
 726 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 727                 struct request *req, struct nvme_command *cmnd,
 728                 enum nvme_opcode op)
 729 {
 730         struct nvme_ctrl *ctrl = ns->ctrl;
 731         u16 control = 0;
 732         u32 dsmgmt = 0;
 733
 734         if (req->cmd_flags & REQ_FUA)
 735                 control |= NVME_RW_FUA;
 736         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 737                 control |= NVME_RW_LR;
 738
 739         if (req->cmd_flags & REQ_RAHEAD)
 740                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 741
 742         cmnd->rw.opcode = op;
 743         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 744         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 745         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 746
 747         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 748                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 749
 750         if (ns->ms) {
 751                 /*
 752                  * If formated with metadata, the block layer always provides a
 753                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 754                  * we enable the PRACT bit for protection information or set the
 755                  * namespace capacity to zero to prevent any I/O.
 756                  */
 757                 if (!blk_integrity_rq(req)) {
 758                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 759                                 return BLK_STS_NOTSUPP;
 760                         control |= NVME_RW_PRINFO_PRACT;
 761                 }
 762
 763                 switch (ns->pi_type) {
 764                 case NVME_NS_DPS_PI_TYPE3:
 765                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 766                         break;
 767                 case NVME_NS_DPS_PI_TYPE1:
 768                 case NVME_NS_DPS_PI_TYPE2:
 769                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 770                                         NVME_RW_PRINFO_PRCHK_REF;
 771                         if (op == nvme_cmd_zone_append)
 772                                 control |= NVME_RW_APPEND_PIREMAP;
 773                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 774                         break;
 775                 }
 776         }
 777
 778         cmnd->rw.control = cpu_to_le16(control);
 779         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 780         return 0;
 781 }
 782
 783 void nvme_cleanup_cmd(struct request *req)
 784 {
 785         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 786                 struct nvme_ns *ns = req->rq_disk->private_data;
 787                 struct page *page = req->special_vec.bv_page;
 788
 789                 if (page == ns->ctrl->discard_page)
 790                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 791                 else
 792                         kfree(page_address(page) + req->special_vec.bv_offset);
 793         }
 794 }
 795 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 796
 797 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 798                 struct nvme_command *cmd)
 799 {
 800         blk_status_t ret = BLK_STS_OK;
 801
 802         nvme_clear_nvme_request(req);
 803
 804         memset(cmd, 0, sizeof(*cmd));
 805         switch (req_op(req)) {
 806         case REQ_OP_DRV_IN:
 807         case REQ_OP_DRV_OUT:
 808                 nvme_setup_passthrough(req, cmd);
 809                 break;
 810         case REQ_OP_FLUSH:
 811                 nvme_setup_flush(ns, cmd);
 812                 break;
 813         case REQ_OP_ZONE_RESET_ALL:
 814         case REQ_OP_ZONE_RESET:
 815                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 816                 break;
 817         case REQ_OP_ZONE_OPEN:
 818                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
 819                 break;
 820         case REQ_OP_ZONE_CLOSE:
 821                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
 822                 break;
 823         case REQ_OP_ZONE_FINISH:
 824                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
 825                 break;
 826         case REQ_OP_WRITE_ZEROES:
 827                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 828                 break;
 829         case REQ_OP_DISCARD:
 830                 ret = nvme_setup_discard(ns, req, cmd);
 831                 break;
 832         case REQ_OP_READ:
 833                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
 834                 break;
 835         case REQ_OP_WRITE:
 836                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
 837                 break;
 838         case REQ_OP_ZONE_APPEND:
 839                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
 840                 break;
 841         default:
 842                 WARN_ON_ONCE(1);
 843                 return BLK_STS_IOERR;
 844         }
 845
 846         cmd->common.command_id = req->tag;
 847         trace_nvme_setup_cmd(req, cmd);
 848         return ret;
 849 }
 850 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 851
 852 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 853 {
 854         struct completion *waiting = rq->end_io_data;
 855
 856         rq->end_io_data = NULL;
 857         complete(waiting);
 858 }
 859
 860 static void nvme_execute_rq_polled(struct request_queue *q,
 861                 struct gendisk *bd_disk, struct request *rq, int at_head)
 862 {
 863         DECLARE_COMPLETION_ONSTACK(wait);
 864
 865         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 866
 867         rq->cmd_flags |= REQ_HIPRI;
 868         rq->end_io_data = &wait;
 869         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 870
 871         while (!completion_done(&wait)) {
 872                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 873                 cond_resched();
 874         }
 875 }
 876
 877 /*
 878  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 879  * if the result is positive, it's an NVM Express status code
 880  */
 881 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 882                 union nvme_result *result, void *buffer, unsigned bufflen,
 883                 unsigned timeout, int qid, int at_head,
 884                 blk_mq_req_flags_t flags, bool poll)
 885 {
 886         struct request *req;
 887         int ret;
 888
 889         req = nvme_alloc_request(q, cmd, flags, qid);
 890         if (IS_ERR(req))
 891                 return PTR_ERR(req);
 892
 893         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 894
 895         if (buffer && bufflen) {
 896                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 897                 if (ret)
 898                         goto out;
 899         }
 900
 901         if (poll)
 902                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 903         else
 904                 blk_execute_rq(req->q, NULL, req, at_head);
 905         if (result)
 906                 *result = nvme_req(req)->result;
 907         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 908                 ret = -EINTR;
 909         else
 910                 ret = nvme_req(req)->status;
 911  out:
 912         blk_mq_free_request(req);
 913         return ret;
 914 }
 915 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 916
 917 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 918                 void *buffer, unsigned bufflen)
 919 {
 920         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 921                         NVME_QID_ANY, 0, 0, false);
 922 }
 923 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 924
 925 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 926                 unsigned len, u32 seed, bool write)
 927 {
 928         struct bio_integrity_payload *bip;
 929         int ret = -ENOMEM;
 930         void *buf;
 931
 932         buf = kmalloc(len, GFP_KERNEL);
 933         if (!buf)
 934                 goto out;
 935
 936         ret = -EFAULT;
 937         if (write && copy_from_user(buf, ubuf, len))
 938                 goto out_free_meta;
 939
 940         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 941         if (IS_ERR(bip)) {
 942                 ret = PTR_ERR(bip);
 943                 goto out_free_meta;
 944         }
 945
 946         bip->bip_iter.bi_size = len;
 947         bip->bip_iter.bi_sector = seed;
 948         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 949                         offset_in_page(buf));
 950         if (ret == len)
 951                 return buf;
 952         ret = -ENOMEM;
 953 out_free_meta:
 954         kfree(buf);
 955 out:
 956         return ERR_PTR(ret);
 957 }
 958
 959 static u32 nvme_known_admin_effects(u8 opcode)
 960 {
 961         switch (opcode) {
 962         case nvme_admin_format_nvm:
 963                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
 964                         NVME_CMD_EFFECTS_CSE_MASK;
 965         case nvme_admin_sanitize_nvm:
 966                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
 967         default:
 968                 break;
 969         }
 970         return 0;
 971 }
 972
 973 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
 974 {
 975         u32 effects = 0;
 976
 977         if (ns) {
 978                 if (ns->head->effects)
 979                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
 980                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
 981                         dev_warn(ctrl->device,
 982                                  "IO command:%02x has unhandled effects:%08x\n",
 983                                  opcode, effects);
 984                 return 0;
 985         }
 986
 987         if (ctrl->effects)
 988                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
 989         effects |= nvme_known_admin_effects(opcode);
 990
 991         return effects;
 992 }
 993 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
 994
 995 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 996                                u8 opcode)
 997 {
 998         u32 effects = nvme_command_effects(ctrl, ns, opcode);
 999
1000         /*
1001          * For simplicity, IO to all namespaces is quiesced even if the command
1002          * effects say only one namespace is affected.
1003          */
1004         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1005                 mutex_lock(&ctrl->scan_lock);
1006                 mutex_lock(&ctrl->subsys->lock);
1007                 nvme_mpath_start_freeze(ctrl->subsys);
1008                 nvme_mpath_wait_freeze(ctrl->subsys);
1009                 nvme_start_freeze(ctrl);
1010                 nvme_wait_freeze(ctrl);
1011         }
1012         return effects;
1013 }
1014
1015 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1016 {
1017         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1018                 nvme_unfreeze(ctrl);
1019                 nvme_mpath_unfreeze(ctrl->subsys);
1020                 mutex_unlock(&ctrl->subsys->lock);
1021                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1022                 mutex_unlock(&ctrl->scan_lock);
1023         }
1024         if (effects & NVME_CMD_EFFECTS_CCC)
1025                 nvme_init_identify(ctrl);
1026         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1027                 nvme_queue_scan(ctrl);
1028                 flush_work(&ctrl->scan_work);
1029         }
1030 }
1031
1032 void nvme_execute_passthru_rq(struct request *rq)
1033 {
1034         struct nvme_command *cmd = nvme_req(rq)->cmd;
1035         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1036         struct nvme_ns *ns = rq->q->queuedata;
1037         struct gendisk *disk = ns ? ns->disk : NULL;
1038         u32 effects;
1039
1040         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1041         blk_execute_rq(rq->q, disk, rq, 0);
1042         nvme_passthru_end(ctrl, effects);
1043 }
1044 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1045
1046 static int nvme_submit_user_cmd(struct request_queue *q,
1047                 struct nvme_command *cmd, void __user *ubuffer,
1048                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1049                 u32 meta_seed, u64 *result, unsigned timeout)
1050 {
1051         bool write = nvme_is_write(cmd);
1052         struct nvme_ns *ns = q->queuedata;
1053         struct gendisk *disk = ns ? ns->disk : NULL;
1054         struct request *req;
1055         struct bio *bio = NULL;
1056         void *meta = NULL;
1057         int ret;
1058
1059         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
1060         if (IS_ERR(req))
1061                 return PTR_ERR(req);
1062
1063         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
1064         nvme_req(req)->flags |= NVME_REQ_USERCMD;
1065
1066         if (ubuffer && bufflen) {
1067                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1068                                 GFP_KERNEL);
1069                 if (ret)
1070                         goto out;
1071                 bio = req->bio;
1072                 bio->bi_disk = disk;
1073                 if (disk && meta_buffer && meta_len) {
1074                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1075                                         meta_seed, write);
1076                         if (IS_ERR(meta)) {
1077                                 ret = PTR_ERR(meta);
1078                                 goto out_unmap;
1079                         }
1080                         req->cmd_flags |= REQ_INTEGRITY;
1081                 }
1082         }
1083
1084         nvme_execute_passthru_rq(req);
1085         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1086                 ret = -EINTR;
1087         else
1088                 ret = nvme_req(req)->status;
1089         if (result)
1090                 *result = le64_to_cpu(nvme_req(req)->result.u64);
1091         if (meta && !ret && !write) {
1092                 if (copy_to_user(meta_buffer, meta, meta_len))
1093                         ret = -EFAULT;
1094         }
1095         kfree(meta);
1096  out_unmap:
1097         if (bio)
1098                 blk_rq_unmap_user(bio);
1099  out:
1100         blk_mq_free_request(req);
1101         return ret;
1102 }
1103
1104 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1105 {
1106         struct nvme_ctrl *ctrl = rq->end_io_data;
1107         unsigned long flags;
1108         bool startka = false;
1109
1110         blk_mq_free_request(rq);
1111
1112         if (status) {
1113                 dev_err(ctrl->device,
1114                         "failed nvme_keep_alive_end_io error=%d\n",
1115                                 status);
1116                 return;
1117         }
1118
1119         ctrl->comp_seen = false;
1120         spin_lock_irqsave(&ctrl->lock, flags);
1121         if (ctrl->state == NVME_CTRL_LIVE ||
1122             ctrl->state == NVME_CTRL_CONNECTING)
1123                 startka = true;
1124         spin_unlock_irqrestore(&ctrl->lock, flags);
1125         if (startka)
1126                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1127 }
1128
1129 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1130 {
1131         struct request *rq;
1132
1133         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
1134                         NVME_QID_ANY);
1135         if (IS_ERR(rq))
1136                 return PTR_ERR(rq);
1137
1138         rq->timeout = ctrl->kato * HZ;
1139         rq->end_io_data = ctrl;
1140
1141         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
1142
1143         return 0;
1144 }
1145
1146 static void nvme_keep_alive_work(struct work_struct *work)
1147 {
1148         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1149                         struct nvme_ctrl, ka_work);
1150         bool comp_seen = ctrl->comp_seen;
1151
1152         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1153                 dev_dbg(ctrl->device,
1154                         "reschedule traffic based keep-alive timer\n");
1155                 ctrl->comp_seen = false;
1156                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1157                 return;
1158         }
1159
1160         if (nvme_keep_alive(ctrl)) {
1161                 /* allocation failure, reset the controller */
1162                 dev_err(ctrl->device, "keep-alive failed\n");
1163                 nvme_reset_ctrl(ctrl);
1164                 return;
1165         }
1166 }
1167
1168 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1169 {
1170         if (unlikely(ctrl->kato == 0))
1171                 return;
1172
1173         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1174 }
1175
1176 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1177 {
1178         if (unlikely(ctrl->kato == 0))
1179                 return;
1180
1181         cancel_delayed_work_sync(&ctrl->ka_work);
1182 }
1183 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1184
1185 /*
1186  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1187  * flag, thus sending any new CNS opcodes has a big chance of not working.
1188  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1189  * (but not for any later version).
1190  */
1191 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1192 {
1193         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1194                 return ctrl->vs < NVME_VS(1, 2, 0);
1195         return ctrl->vs < NVME_VS(1, 1, 0);
1196 }
1197
1198 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1199 {
1200         struct nvme_command c = { };
1201         int error;
1202
1203         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1204         c.identify.opcode = nvme_admin_identify;
1205         c.identify.cns = NVME_ID_CNS_CTRL;
1206
1207         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1208         if (!*id)
1209                 return -ENOMEM;
1210
1211         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1212                         sizeof(struct nvme_id_ctrl));
1213         if (error)
1214                 kfree(*id);
1215         return error;
1216 }
1217
1218 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1219 {
1220         return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1221 }
1222
1223 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1224                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1225 {
1226         const char *warn_str = "ctrl returned bogus length:";
1227         void *data = cur;
1228
1229         switch (cur->nidt) {
1230         case NVME_NIDT_EUI64:
1231                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1232                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1233                                  warn_str, cur->nidl);
1234                         return -1;
1235                 }
1236                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1237                 return NVME_NIDT_EUI64_LEN;
1238         case NVME_NIDT_NGUID:
1239                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1240                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1241                                  warn_str, cur->nidl);
1242                         return -1;
1243                 }
1244                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1245                 return NVME_NIDT_NGUID_LEN;
1246         case NVME_NIDT_UUID:
1247                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1248                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1249                                  warn_str, cur->nidl);
1250                         return -1;
1251                 }
1252                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1253                 return NVME_NIDT_UUID_LEN;
1254         case NVME_NIDT_CSI:
1255                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1256                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1257                                  warn_str, cur->nidl);
1258                         return -1;
1259                 }
1260                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1261                 *csi_seen = true;
1262                 return NVME_NIDT_CSI_LEN;
1263         default:
1264                 /* Skip unknown types */
1265                 return cur->nidl;
1266         }
1267 }
1268
1269 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1270                 struct nvme_ns_ids *ids)
1271 {
1272         struct nvme_command c = { };
1273         bool csi_seen = false;
1274         int status, pos, len;
1275         void *data;
1276
1277         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1278                 return 0;
1279         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1280                 return 0;
1281
1282         c.identify.opcode = nvme_admin_identify;
1283         c.identify.nsid = cpu_to_le32(nsid);
1284         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1285
1286         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1287         if (!data)
1288                 return -ENOMEM;
1289
1290         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1291                                       NVME_IDENTIFY_DATA_SIZE);
1292         if (status) {
1293                 dev_warn(ctrl->device,
1294                         "Identify Descriptors failed (%d)\n", status);
1295                 goto free_data;
1296         }
1297
1298         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1299                 struct nvme_ns_id_desc *cur = data + pos;
1300
1301                 if (cur->nidl == 0)
1302                         break;
1303
1304                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1305                 if (len < 0)
1306                         break;
1307
1308                 len += sizeof(*cur);
1309         }
1310
1311         if (nvme_multi_css(ctrl) && !csi_seen) {
1312                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1313                          nsid);
1314                 status = -EINVAL;
1315         }
1316
1317 free_data:
1318         kfree(data);
1319         return status;
1320 }
1321
1322 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1323                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1324 {
1325         struct nvme_command c = { };
1326         int error;
1327
1328         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1329         c.identify.opcode = nvme_admin_identify;
1330         c.identify.nsid = cpu_to_le32(nsid);
1331         c.identify.cns = NVME_ID_CNS_NS;
1332
1333         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1334         if (!*id)
1335                 return -ENOMEM;
1336
1337         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1338         if (error) {
1339                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1340                 goto out_free_id;
1341         }
1342
1343         error = -ENODEV;
1344         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1345                 goto out_free_id;
1346
1347         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1348             !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1349                 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1350         if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1351             !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1352                 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1353
1354         return 0;
1355
1356 out_free_id:
1357         kfree(*id);
1358         return error;
1359 }
1360
1361 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1362                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1363 {
1364         union nvme_result res = { 0 };
1365         struct nvme_command c;
1366         int ret;
1367
1368         memset(&c, 0, sizeof(c));
1369         c.features.opcode = op;
1370         c.features.fid = cpu_to_le32(fid);
1371         c.features.dword11 = cpu_to_le32(dword11);
1372
1373         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1374                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1375         if (ret >= 0 && result)
1376                 *result = le32_to_cpu(res.u32);
1377         return ret;
1378 }
1379
1380 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1381                       unsigned int dword11, void *buffer, size_t buflen,
1382                       u32 *result)
1383 {
1384         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1385                              buflen, result);
1386 }
1387 EXPORT_SYMBOL_GPL(nvme_set_features);
1388
1389 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1390                       unsigned int dword11, void *buffer, size_t buflen,
1391                       u32 *result)
1392 {
1393         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1394                              buflen, result);
1395 }
1396 EXPORT_SYMBOL_GPL(nvme_get_features);
1397
1398 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1399 {
1400         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1401         u32 result;
1402         int status, nr_io_queues;
1403
1404         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1405                         &result);
1406         if (status < 0)
1407                 return status;
1408
1409         /*
1410          * Degraded controllers might return an error when setting the queue
1411          * count.  We still want to be able to bring them online and offer
1412          * access to the admin queue, as that might be only way to fix them up.
1413          */
1414         if (status > 0) {
1415                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1416                 *count = 0;
1417         } else {
1418                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1419                 *count = min(*count, nr_io_queues);
1420         }
1421
1422         return 0;
1423 }
1424 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1425
1426 #define NVME_AEN_SUPPORTED \
1427         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1428          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1429
1430 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1431 {
1432         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1433         int status;
1434
1435         if (!supported_aens)
1436                 return;
1437
1438         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1439                         NULL, 0, &result);
1440         if (status)
1441                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1442                          supported_aens);
1443
1444         queue_work(nvme_wq, &ctrl->async_event_work);
1445 }
1446
1447 /*
1448  * Convert integer values from ioctl structures to user pointers, silently
1449  * ignoring the upper bits in the compat case to match behaviour of 32-bit
1450  * kernels.
1451  */
1452 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1453 {
1454         if (in_compat_syscall())
1455                 ptrval = (compat_uptr_t)ptrval;
1456         return (void __user *)ptrval;
1457 }
1458
1459 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1460 {
1461         struct nvme_user_io io;
1462         struct nvme_command c;
1463         unsigned length, meta_len;
1464         void __user *metadata;
1465
1466         if (copy_from_user(&io, uio, sizeof(io)))
1467                 return -EFAULT;
1468         if (io.flags)
1469                 return -EINVAL;
1470
1471         switch (io.opcode) {
1472         case nvme_cmd_write:
1473         case nvme_cmd_read:
1474         case nvme_cmd_compare:
1475                 break;
1476         default:
1477                 return -EINVAL;
1478         }
1479
1480         length = (io.nblocks + 1) << ns->lba_shift;
1481         meta_len = (io.nblocks + 1) * ns->ms;
1482         metadata = nvme_to_user_ptr(io.metadata);
1483
1484         if (ns->features & NVME_NS_EXT_LBAS) {
1485                 length += meta_len;
1486                 meta_len = 0;
1487         } else if (meta_len) {
1488                 if ((io.metadata & 3) || !io.metadata)
1489                         return -EINVAL;
1490         }
1491
1492         memset(&c, 0, sizeof(c));
1493         c.rw.opcode = io.opcode;
1494         c.rw.flags = io.flags;
1495         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1496         c.rw.slba = cpu_to_le64(io.slba);
1497         c.rw.length = cpu_to_le16(io.nblocks);
1498         c.rw.control = cpu_to_le16(io.control);
1499         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1500         c.rw.reftag = cpu_to_le32(io.reftag);
1501         c.rw.apptag = cpu_to_le16(io.apptag);
1502         c.rw.appmask = cpu_to_le16(io.appmask);
1503
1504         return nvme_submit_user_cmd(ns->queue, &c,
1505                         nvme_to_user_ptr(io.addr), length,
1506                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1507 }
1508
1509 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1510                         struct nvme_passthru_cmd __user *ucmd)
1511 {
1512         struct nvme_passthru_cmd cmd;
1513         struct nvme_command c;
1514         unsigned timeout = 0;
1515         u64 result;
1516         int status;
1517
1518         if (!capable(CAP_SYS_ADMIN))
1519                 return -EACCES;
1520         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1521                 return -EFAULT;
1522         if (cmd.flags)
1523                 return -EINVAL;
1524
1525         memset(&c, 0, sizeof(c));
1526         c.common.opcode = cmd.opcode;
1527         c.common.flags = cmd.flags;
1528         c.common.nsid = cpu_to_le32(cmd.nsid);
1529         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1530         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1531         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1532         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1533         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1534         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1535         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1536         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1537
1538         if (cmd.timeout_ms)
1539                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1540
1541         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1542                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1543                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1544                         0, &result, timeout);
1545
1546         if (status >= 0) {
1547                 if (put_user(result, &ucmd->result))
1548                         return -EFAULT;
1549         }
1550
1551         return status;
1552 }
1553
1554 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1555                         struct nvme_passthru_cmd64 __user *ucmd)
1556 {
1557         struct nvme_passthru_cmd64 cmd;
1558         struct nvme_command c;
1559         unsigned timeout = 0;
1560         int status;
1561
1562         if (!capable(CAP_SYS_ADMIN))
1563                 return -EACCES;
1564         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1565                 return -EFAULT;
1566         if (cmd.flags)
1567                 return -EINVAL;
1568
1569         memset(&c, 0, sizeof(c));
1570         c.common.opcode = cmd.opcode;
1571         c.common.flags = cmd.flags;
1572         c.common.nsid = cpu_to_le32(cmd.nsid);
1573         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1574         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1575         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1576         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1577         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1578         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1579         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1580         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1581
1582         if (cmd.timeout_ms)
1583                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1584
1585         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1586                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1587                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1588                         0, &cmd.result, timeout);
1589
1590         if (status >= 0) {
1591                 if (put_user(cmd.result, &ucmd->result))
1592                         return -EFAULT;
1593         }
1594
1595         return status;
1596 }
1597
1598 /*
1599  * Issue ioctl requests on the first available path.  Note that unlike normal
1600  * block layer requests we will not retry failed request on another controller.
1601  */
1602 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1603                 struct nvme_ns_head **head, int *srcu_idx)
1604 {
1605 #ifdef CONFIG_NVME_MULTIPATH
1606         if (disk->fops == &nvme_ns_head_ops) {
1607                 struct nvme_ns *ns;
1608
1609                 *head = disk->private_data;
1610                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1611                 ns = nvme_find_path(*head);
1612                 if (!ns)
1613                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1614                 return ns;
1615         }
1616 #endif
1617         *head = NULL;
1618         *srcu_idx = -1;
1619         return disk->private_data;
1620 }
1621
1622 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1623 {
1624         if (head)
1625                 srcu_read_unlock(&head->srcu, idx);
1626 }
1627
1628 static bool is_ctrl_ioctl(unsigned int cmd)
1629 {
1630         if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1631                 return true;
1632         if (is_sed_ioctl(cmd))
1633                 return true;
1634         return false;
1635 }
1636
1637 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1638                                   void __user *argp,
1639                                   struct nvme_ns_head *head,
1640                                   int srcu_idx)
1641 {
1642         struct nvme_ctrl *ctrl = ns->ctrl;
1643         int ret;
1644
1645         nvme_get_ctrl(ns->ctrl);
1646         nvme_put_ns_from_disk(head, srcu_idx);
1647
1648         switch (cmd) {
1649         case NVME_IOCTL_ADMIN_CMD:
1650                 ret = nvme_user_cmd(ctrl, NULL, argp);
1651                 break;
1652         case NVME_IOCTL_ADMIN64_CMD:
1653                 ret = nvme_user_cmd64(ctrl, NULL, argp);
1654                 break;
1655         default:
1656                 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1657                 break;
1658         }
1659         nvme_put_ctrl(ctrl);
1660         return ret;
1661 }
1662
1663 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1664                 unsigned int cmd, unsigned long arg)
1665 {
1666         struct nvme_ns_head *head = NULL;
1667         void __user *argp = (void __user *)arg;
1668         struct nvme_ns *ns;
1669         int srcu_idx, ret;
1670
1671         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1672         if (unlikely(!ns))
1673                 return -EWOULDBLOCK;
1674
1675         /*
1676          * Handle ioctls that apply to the controller instead of the namespace
1677          * seperately and drop the ns SRCU reference early.  This avoids a
1678          * deadlock when deleting namespaces using the passthrough interface.
1679          */
1680         if (is_ctrl_ioctl(cmd))
1681                 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1682
1683         switch (cmd) {
1684         case NVME_IOCTL_ID:
1685                 force_successful_syscall_return();
1686                 ret = ns->head->ns_id;
1687                 break;
1688         case NVME_IOCTL_IO_CMD:
1689                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1690                 break;
1691         case NVME_IOCTL_SUBMIT_IO:
1692                 ret = nvme_submit_io(ns, argp);
1693                 break;
1694         case NVME_IOCTL_IO64_CMD:
1695                 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1696                 break;
1697         default:
1698                 if (ns->ndev)
1699                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1700                 else
1701                         ret = -ENOTTY;
1702         }
1703
1704         nvme_put_ns_from_disk(head, srcu_idx);
1705         return ret;
1706 }
1707
1708 #ifdef CONFIG_COMPAT
1709 struct nvme_user_io32 {
1710         __u8    opcode;
1711         __u8    flags;
1712         __u16   control;
1713         __u16   nblocks;
1714         __u16   rsvd;
1715         __u64   metadata;
1716         __u64   addr;
1717         __u64   slba;
1718         __u32   dsmgmt;
1719         __u32   reftag;
1720         __u16   apptag;
1721         __u16   appmask;
1722 } __attribute__((__packed__));
1723
1724 #define NVME_IOCTL_SUBMIT_IO32  _IOW('N', 0x42, struct nvme_user_io32)
1725
1726 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1727                 unsigned int cmd, unsigned long arg)
1728 {
1729         /*
1730          * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1731          * between 32 bit programs and 64 bit kernel.
1732          * The cause is that the results of sizeof(struct nvme_user_io),
1733          * which is used to define NVME_IOCTL_SUBMIT_IO,
1734          * are not same between 32 bit compiler and 64 bit compiler.
1735          * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1736          * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1737          * Other IOCTL numbers are same between 32 bit and 64 bit.
1738          * So there is nothing to do regarding to other IOCTL numbers.
1739          */
1740         if (cmd == NVME_IOCTL_SUBMIT_IO32)
1741                 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1742
1743         return nvme_ioctl(bdev, mode, cmd, arg);
1744 }
1745 #else
1746 #define nvme_compat_ioctl       NULL
1747 #endif /* CONFIG_COMPAT */
1748
1749 static int nvme_open(struct block_device *bdev, fmode_t mode)
1750 {
1751         struct nvme_ns *ns = bdev->bd_disk->private_data;
1752
1753 #ifdef CONFIG_NVME_MULTIPATH
1754         /* should never be called due to GENHD_FL_HIDDEN */
1755         if (WARN_ON_ONCE(ns->head->disk))
1756                 goto fail;
1757 #endif
1758         if (!kref_get_unless_zero(&ns->kref))
1759                 goto fail;
1760         if (!try_module_get(ns->ctrl->ops->module))
1761                 goto fail_put_ns;
1762
1763         return 0;
1764
1765 fail_put_ns:
1766         nvme_put_ns(ns);
1767 fail:
1768         return -ENXIO;
1769 }
1770
1771 static void nvme_release(struct gendisk *disk, fmode_t mode)
1772 {
1773         struct nvme_ns *ns = disk->private_data;
1774
1775         module_put(ns->ctrl->ops->module);
1776         nvme_put_ns(ns);
1777 }
1778
1779 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1780 {
1781         /* some standard values */
1782         geo->heads = 1 << 6;
1783         geo->sectors = 1 << 5;
1784         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1785         return 0;
1786 }
1787
1788 #ifdef CONFIG_BLK_DEV_INTEGRITY
1789 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1790                                 u32 max_integrity_segments)
1791 {
1792         struct blk_integrity integrity;
1793
1794         memset(&integrity, 0, sizeof(integrity));
1795         switch (pi_type) {
1796         case NVME_NS_DPS_PI_TYPE3:
1797                 integrity.profile = &t10_pi_type3_crc;
1798                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1799                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1800                 break;
1801         case NVME_NS_DPS_PI_TYPE1:
1802         case NVME_NS_DPS_PI_TYPE2:
1803                 integrity.profile = &t10_pi_type1_crc;
1804                 integrity.tag_size = sizeof(u16);
1805                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1806                 break;
1807         default:
1808                 integrity.profile = NULL;
1809                 break;
1810         }
1811         integrity.tuple_size = ms;
1812         blk_integrity_register(disk, &integrity);
1813         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1814 }
1815 #else
1816 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1817                                 u32 max_integrity_segments)
1818 {
1819 }
1820 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1821
1822 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1823 {
1824         struct nvme_ctrl *ctrl = ns->ctrl;
1825         struct request_queue *queue = disk->queue;
1826         u32 size = queue_logical_block_size(queue);
1827
1828         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1829                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1830                 return;
1831         }
1832
1833         if (ctrl->nr_streams && ns->sws && ns->sgs)
1834                 size *= ns->sws * ns->sgs;
1835
1836         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1837                         NVME_DSM_MAX_RANGES);
1838
1839         queue->limits.discard_alignment = 0;
1840         queue->limits.discard_granularity = size;
1841
1842         /* If discard is already enabled, don't reset queue limits */
1843         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1844                 return;
1845
1846         blk_queue_max_discard_sectors(queue, UINT_MAX);
1847         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1848
1849         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1850                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1851 }
1852
1853 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1854 {
1855         u64 max_blocks;
1856
1857         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1858             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1859                 return;
1860         /*
1861          * Even though NVMe spec explicitly states that MDTS is not
1862          * applicable to the write-zeroes:- "The restriction does not apply to
1863          * commands that do not transfer data between the host and the
1864          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1865          * In order to be more cautious use controller's max_hw_sectors value
1866          * to configure the maximum sectors for the write-zeroes which is
1867          * configured based on the controller's MDTS field in the
1868          * nvme_init_identify() if available.
1869          */
1870         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1871                 max_blocks = (u64)USHRT_MAX + 1;
1872         else
1873                 max_blocks = ns->ctrl->max_hw_sectors + 1;
1874
1875         blk_queue_max_write_zeroes_sectors(disk->queue,
1876                                            nvme_lba_to_sect(ns, max_blocks));
1877 }
1878
1879 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1880 {
1881         return !uuid_is_null(&ids->uuid) ||
1882                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1883                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1884 }
1885
1886 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1887 {
1888         return uuid_equal(&a->uuid, &b->uuid) &&
1889                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1890                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1891                 a->csi == b->csi;
1892 }
1893
1894 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1895                                  u32 *phys_bs, u32 *io_opt)
1896 {
1897         struct streams_directive_params s;
1898         int ret;
1899
1900         if (!ctrl->nr_streams)
1901                 return 0;
1902
1903         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1904         if (ret)
1905                 return ret;
1906
1907         ns->sws = le32_to_cpu(s.sws);
1908         ns->sgs = le16_to_cpu(s.sgs);
1909
1910         if (ns->sws) {
1911                 *phys_bs = ns->sws * (1 << ns->lba_shift);
1912                 if (ns->sgs)
1913                         *io_opt = *phys_bs * ns->sgs;
1914         }
1915
1916         return 0;
1917 }
1918
1919 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1920 {
1921         struct nvme_ctrl *ctrl = ns->ctrl;
1922
1923         /*
1924          * The PI implementation requires the metadata size to be equal to the
1925          * t10 pi tuple size.
1926          */
1927         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1928         if (ns->ms == sizeof(struct t10_pi_tuple))
1929                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1930         else
1931                 ns->pi_type = 0;
1932
1933         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1934         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1935                 return 0;
1936         if (ctrl->ops->flags & NVME_F_FABRICS) {
1937                 /*
1938                  * The NVMe over Fabrics specification only supports metadata as
1939                  * part of the extended data LBA.  We rely on HCA/HBA support to
1940                  * remap the separate metadata buffer from the block layer.
1941                  */
1942                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1943                         return -EINVAL;
1944                 if (ctrl->max_integrity_segments)
1945                         ns->features |=
1946                                 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1947         } else {
1948                 /*
1949                  * For PCIe controllers, we can't easily remap the separate
1950                  * metadata buffer from the block layer and thus require a
1951                  * separate metadata buffer for block layer metadata/PI support.
1952                  * We allow extended LBAs for the passthrough interface, though.
1953                  */
1954                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1955                         ns->features |= NVME_NS_EXT_LBAS;
1956                 else
1957                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1958         }
1959
1960         return 0;
1961 }
1962
1963 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1964                 struct request_queue *q)
1965 {
1966         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1967
1968         if (ctrl->max_hw_sectors) {
1969                 u32 max_segments =
1970                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1971
1972                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1973                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1974                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1975         }
1976         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1977         blk_queue_dma_alignment(q, 7);
1978         blk_queue_write_cache(q, vwc, vwc);
1979 }
1980
1981 static void nvme_update_disk_info(struct gendisk *disk,
1982                 struct nvme_ns *ns, struct nvme_id_ns *id)
1983 {
1984         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1985         unsigned short bs = 1 << ns->lba_shift;
1986         u32 atomic_bs, phys_bs, io_opt = 0;
1987
1988         /*
1989          * The block layer can't support LBA sizes larger than the page size
1990          * yet, so catch this early and don't allow block I/O.
1991          */
1992         if (ns->lba_shift > PAGE_SHIFT) {
1993                 capacity = 0;
1994                 bs = (1 << 9);
1995         }
1996
1997         blk_integrity_unregister(disk);
1998
1999         atomic_bs = phys_bs = bs;
2000         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2001         if (id->nabo == 0) {
2002                 /*
2003                  * Bit 1 indicates whether NAWUPF is defined for this namespace
2004                  * and whether it should be used instead of AWUPF. If NAWUPF ==
2005                  * 0 then AWUPF must be used instead.
2006                  */
2007                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2008                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2009                 else
2010                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2011         }
2012
2013         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2014                 /* NPWG = Namespace Preferred Write Granularity */
2015                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2016                 /* NOWS = Namespace Optimal Write Size */
2017                 io_opt = bs * (1 + le16_to_cpu(id->nows));
2018         }
2019
2020         blk_queue_logical_block_size(disk->queue, bs);
2021         /*
2022          * Linux filesystems assume writing a single physical block is
2023          * an atomic operation. Hence limit the physical block size to the
2024          * value of the Atomic Write Unit Power Fail parameter.
2025          */
2026         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2027         blk_queue_io_min(disk->queue, phys_bs);
2028         blk_queue_io_opt(disk->queue, io_opt);
2029
2030         /*
2031          * Register a metadata profile for PI, or the plain non-integrity NVMe
2032          * metadata masquerading as Type 0 if supported, otherwise reject block
2033          * I/O to namespaces with metadata except when the namespace supports
2034          * PI, as it can strip/insert in that case.
2035          */
2036         if (ns->ms) {
2037                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2038                     (ns->features & NVME_NS_METADATA_SUPPORTED))
2039                         nvme_init_integrity(disk, ns->ms, ns->pi_type,
2040                                             ns->ctrl->max_integrity_segments);
2041                 else if (!nvme_ns_has_pi(ns))
2042                         capacity = 0;
2043         }
2044
2045         set_capacity_and_notify(disk, capacity);
2046
2047         nvme_config_discard(disk, ns);
2048         nvme_config_write_zeroes(disk, ns);
2049
2050         if (id->nsattr & NVME_NS_ATTR_RO)
2051                 set_disk_ro(disk, true);
2052 }
2053
2054 static inline bool nvme_first_scan(struct gendisk *disk)
2055 {
2056         /* nvme_alloc_ns() scans the disk prior to adding it */
2057         return !(disk->flags & GENHD_FL_UP);
2058 }
2059
2060 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2061 {
2062         struct nvme_ctrl *ctrl = ns->ctrl;
2063         u32 iob;
2064
2065         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2066             is_power_of_2(ctrl->max_hw_sectors))
2067                 iob = ctrl->max_hw_sectors;
2068         else
2069                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2070
2071         if (!iob)
2072                 return;
2073
2074         if (!is_power_of_2(iob)) {
2075                 if (nvme_first_scan(ns->disk))
2076                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2077                                 ns->disk->disk_name, iob);
2078                 return;
2079         }
2080
2081         if (blk_queue_is_zoned(ns->disk->queue)) {
2082                 if (nvme_first_scan(ns->disk))
2083                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
2084                                 ns->disk->disk_name);
2085                 return;
2086         }
2087
2088         blk_queue_chunk_sectors(ns->queue, iob);
2089 }
2090
2091 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2092 {
2093         unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2094         int ret;
2095
2096         blk_mq_freeze_queue(ns->disk->queue);
2097         ns->lba_shift = id->lbaf[lbaf].ds;
2098         nvme_set_queue_limits(ns->ctrl, ns->queue);
2099
2100         if (ns->head->ids.csi == NVME_CSI_ZNS) {
2101                 ret = nvme_update_zone_info(ns, lbaf);
2102                 if (ret)
2103                         goto out_unfreeze;
2104         }
2105
2106         ret = nvme_configure_metadata(ns, id);
2107         if (ret)
2108                 goto out_unfreeze;
2109         nvme_set_chunk_sectors(ns, id);
2110         nvme_update_disk_info(ns->disk, ns, id);
2111         blk_mq_unfreeze_queue(ns->disk->queue);
2112
2113         if (blk_queue_is_zoned(ns->queue)) {
2114                 ret = nvme_revalidate_zones(ns);
2115                 if (ret && !nvme_first_scan(ns->disk))
2116                         return ret;
2117         }
2118
2119 #ifdef CONFIG_NVME_MULTIPATH
2120         if (ns->head->disk) {
2121                 blk_mq_freeze_queue(ns->head->disk->queue);
2122                 nvme_update_disk_info(ns->head->disk, ns, id);
2123                 blk_stack_limits(&ns->head->disk->queue->limits,
2124                                  &ns->queue->limits, 0);
2125                 blk_queue_update_readahead(ns->head->disk->queue);
2126                 blk_mq_unfreeze_queue(ns->head->disk->queue);
2127         }
2128 #endif
2129         return 0;
2130
2131 out_unfreeze:
2132         blk_mq_unfreeze_queue(ns->disk->queue);
2133         return ret;
2134 }
2135
2136 static char nvme_pr_type(enum pr_type type)
2137 {
2138         switch (type) {
2139         case PR_WRITE_EXCLUSIVE:
2140                 return 1;
2141         case PR_EXCLUSIVE_ACCESS:
2142                 return 2;
2143         case PR_WRITE_EXCLUSIVE_REG_ONLY:
2144                 return 3;
2145         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2146                 return 4;
2147         case PR_WRITE_EXCLUSIVE_ALL_REGS:
2148                 return 5;
2149         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2150                 return 6;
2151         default:
2152                 return 0;
2153         }
2154 };
2155
2156 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2157                                 u64 key, u64 sa_key, u8 op)
2158 {
2159         struct nvme_ns_head *head = NULL;
2160         struct nvme_ns *ns;
2161         struct nvme_command c;
2162         int srcu_idx, ret;
2163         u8 data[16] = { 0, };
2164
2165         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2166         if (unlikely(!ns))
2167                 return -EWOULDBLOCK;
2168
2169         put_unaligned_le64(key, &data[0]);
2170         put_unaligned_le64(sa_key, &data[8]);
2171
2172         memset(&c, 0, sizeof(c));
2173         c.common.opcode = op;
2174         c.common.nsid = cpu_to_le32(ns->head->ns_id);
2175         c.common.cdw10 = cpu_to_le32(cdw10);
2176
2177         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2178         nvme_put_ns_from_disk(head, srcu_idx);
2179         return ret;
2180 }
2181
2182 static int nvme_pr_register(struct block_device *bdev, u64 old,
2183                 u64 new, unsigned flags)
2184 {
2185         u32 cdw10;
2186
2187         if (flags & ~PR_FL_IGNORE_KEY)
2188                 return -EOPNOTSUPP;
2189
2190         cdw10 = old ? 2 : 0;
2191         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2192         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2193         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2194 }
2195
2196 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2197                 enum pr_type type, unsigned flags)
2198 {
2199         u32 cdw10;
2200
2201         if (flags & ~PR_FL_IGNORE_KEY)
2202                 return -EOPNOTSUPP;
2203
2204         cdw10 = nvme_pr_type(type) << 8;
2205         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2206         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2207 }
2208
2209 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2210                 enum pr_type type, bool abort)
2211 {
2212         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2213         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2214 }
2215
2216 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2217 {
2218         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2219         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2220 }
2221
2222 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2223 {
2224         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2225         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2226 }
2227
2228 static const struct pr_ops nvme_pr_ops = {
2229         .pr_register    = nvme_pr_register,
2230         .pr_reserve     = nvme_pr_reserve,
2231         .pr_release     = nvme_pr_release,
2232         .pr_preempt     = nvme_pr_preempt,
2233         .pr_clear       = nvme_pr_clear,
2234 };
2235
2236 #ifdef CONFIG_BLK_SED_OPAL
2237 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2238                 bool send)
2239 {
2240         struct nvme_ctrl *ctrl = data;
2241         struct nvme_command cmd;
2242
2243         memset(&cmd, 0, sizeof(cmd));
2244         if (send)
2245                 cmd.common.opcode = nvme_admin_security_send;
2246         else
2247                 cmd.common.opcode = nvme_admin_security_recv;
2248         cmd.common.nsid = 0;
2249         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2250         cmd.common.cdw11 = cpu_to_le32(len);
2251
2252         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2253                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2254 }
2255 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2256 #endif /* CONFIG_BLK_SED_OPAL */
2257
2258 static const struct block_device_operations nvme_fops = {
2259         .owner          = THIS_MODULE,
2260         .ioctl          = nvme_ioctl,
2261         .compat_ioctl   = nvme_compat_ioctl,
2262         .open           = nvme_open,
2263         .release        = nvme_release,
2264         .getgeo         = nvme_getgeo,
2265         .report_zones   = nvme_report_zones,
2266         .pr_ops         = &nvme_pr_ops,
2267 };
2268
2269 #ifdef CONFIG_NVME_MULTIPATH
2270 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2271 {
2272         struct nvme_ns_head *head = bdev->bd_disk->private_data;
2273
2274         if (!kref_get_unless_zero(&head->ref))
2275                 return -ENXIO;
2276         return 0;
2277 }
2278
2279 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2280 {
2281         nvme_put_ns_head(disk->private_data);
2282 }
2283
2284 const struct block_device_operations nvme_ns_head_ops = {
2285         .owner          = THIS_MODULE,
2286         .submit_bio     = nvme_ns_head_submit_bio,
2287         .open           = nvme_ns_head_open,
2288         .release        = nvme_ns_head_release,
2289         .ioctl          = nvme_ioctl,
2290         .compat_ioctl   = nvme_compat_ioctl,
2291         .getgeo         = nvme_getgeo,
2292         .report_zones   = nvme_report_zones,
2293         .pr_ops         = &nvme_pr_ops,
2294 };
2295 #endif /* CONFIG_NVME_MULTIPATH */
2296
2297 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2298 {
2299         unsigned long timeout =
2300                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2301         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2302         int ret;
2303
2304         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2305                 if (csts == ~0)
2306                         return -ENODEV;
2307                 if ((csts & NVME_CSTS_RDY) == bit)
2308                         break;
2309
2310                 usleep_range(1000, 2000);
2311                 if (fatal_signal_pending(current))
2312                         return -EINTR;
2313                 if (time_after(jiffies, timeout)) {
2314                         dev_err(ctrl->device,
2315                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2316                                 enabled ? "initialisation" : "reset", csts);
2317                         return -ENODEV;
2318                 }
2319         }
2320
2321         return ret;
2322 }
2323
2324 /*
2325  * If the device has been passed off to us in an enabled state, just clear
2326  * the enabled bit.  The spec says we should set the 'shutdown notification
2327  * bits', but doing so may cause the device to complete commands to the
2328  * admin queue ... and we don't know what memory that might be pointing at!
2329  */
2330 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2331 {
2332         int ret;
2333
2334         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2335         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2336
2337         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2338         if (ret)
2339                 return ret;
2340
2341         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2342                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2343
2344         return nvme_wait_ready(ctrl, ctrl->cap, false);
2345 }
2346 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2347
2348 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2349 {
2350         unsigned dev_page_min;
2351         int ret;
2352
2353         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2354         if (ret) {
2355                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2356                 return ret;
2357         }
2358         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2359
2360         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2361                 dev_err(ctrl->device,
2362                         "Minimum device page size %u too large for host (%u)\n",
2363                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2364                 return -ENODEV;
2365         }
2366
2367         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2368                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2369         else
2370                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2371         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2372         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2373         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2374         ctrl->ctrl_config |= NVME_CC_ENABLE;
2375
2376         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2377         if (ret)
2378                 return ret;
2379         return nvme_wait_ready(ctrl, ctrl->cap, true);
2380 }
2381 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2382
2383 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2384 {
2385         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2386         u32 csts;
2387         int ret;
2388
2389         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2390         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2391
2392         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2393         if (ret)
2394                 return ret;
2395
2396         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2397                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2398                         break;
2399
2400                 msleep(100);
2401                 if (fatal_signal_pending(current))
2402                         return -EINTR;
2403                 if (time_after(jiffies, timeout)) {
2404                         dev_err(ctrl->device,
2405                                 "Device shutdown incomplete; abort shutdown\n");
2406                         return -ENODEV;
2407                 }
2408         }
2409
2410         return ret;
2411 }
2412 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2413
2414 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2415 {
2416         __le64 ts;
2417         int ret;
2418
2419         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2420                 return 0;
2421
2422         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2423         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2424                         NULL);
2425         if (ret)
2426                 dev_warn_once(ctrl->device,
2427                         "could not set timestamp (%d)\n", ret);
2428         return ret;
2429 }
2430
2431 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2432 {
2433         struct nvme_feat_host_behavior *host;
2434         int ret;
2435
2436         /* Don't bother enabling the feature if retry delay is not reported */
2437         if (!ctrl->crdt[0])
2438                 return 0;
2439
2440         host = kzalloc(sizeof(*host), GFP_KERNEL);
2441         if (!host)
2442                 return 0;
2443
2444         host->acre = NVME_ENABLE_ACRE;
2445         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2446                                 host, sizeof(*host), NULL);
2447         kfree(host);
2448         return ret;
2449 }
2450
2451 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2452 {
2453         /*
2454          * APST (Autonomous Power State Transition) lets us program a
2455          * table of power state transitions that the controller will
2456          * perform automatically.  We configure it with a simple
2457          * heuristic: we are willing to spend at most 2% of the time
2458          * transitioning between power states.  Therefore, when running
2459          * in any given state, we will enter the next lower-power
2460          * non-operational state after waiting 50 * (enlat + exlat)
2461          * microseconds, as long as that state's exit latency is under
2462          * the requested maximum latency.
2463          *
2464          * We will not autonomously enter any non-operational state for
2465          * which the total latency exceeds ps_max_latency_us.  Users
2466          * can set ps_max_latency_us to zero to turn off APST.
2467          */
2468
2469         unsigned apste;
2470         struct nvme_feat_auto_pst *table;
2471         u64 max_lat_us = 0;
2472         int max_ps = -1;
2473         int ret;
2474
2475         /*
2476          * If APST isn't supported or if we haven't been initialized yet,
2477          * then don't do anything.
2478          */
2479         if (!ctrl->apsta)
2480                 return 0;
2481
2482         if (ctrl->npss > 31) {
2483                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2484                 return 0;
2485         }
2486
2487         table = kzalloc(sizeof(*table), GFP_KERNEL);
2488         if (!table)
2489                 return 0;
2490
2491         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2492                 /* Turn off APST. */
2493                 apste = 0;
2494                 dev_dbg(ctrl->device, "APST disabled\n");
2495         } else {
2496                 __le64 target = cpu_to_le64(0);
2497                 int state;
2498
2499                 /*
2500                  * Walk through all states from lowest- to highest-power.
2501                  * According to the spec, lower-numbered states use more
2502                  * power.  NPSS, despite the name, is the index of the
2503                  * lowest-power state, not the number of states.
2504                  */
2505                 for (state = (int)ctrl->npss; state >= 0; state--) {
2506                         u64 total_latency_us, exit_latency_us, transition_ms;
2507
2508                         if (target)
2509                                 table->entries[state] = target;
2510
2511                         /*
2512                          * Don't allow transitions to the deepest state
2513                          * if it's quirked off.
2514                          */
2515                         if (state == ctrl->npss &&
2516                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2517                                 continue;
2518
2519                         /*
2520                          * Is this state a useful non-operational state for
2521                          * higher-power states to autonomously transition to?
2522                          */
2523                         if (!(ctrl->psd[state].flags &
2524                               NVME_PS_FLAGS_NON_OP_STATE))
2525                                 continue;
2526
2527                         exit_latency_us =
2528                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2529                         if (exit_latency_us > ctrl->ps_max_latency_us)
2530                                 continue;
2531
2532                         total_latency_us =
2533                                 exit_latency_us +
2534                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2535
2536                         /*
2537                          * This state is good.  Use it as the APST idle
2538                          * target for higher power states.
2539                          */
2540                         transition_ms = total_latency_us + 19;
2541                         do_div(transition_ms, 20);
2542                         if (transition_ms > (1 << 24) - 1)
2543                                 transition_ms = (1 << 24) - 1;
2544
2545                         target = cpu_to_le64((state << 3) |
2546                                              (transition_ms << 8));
2547
2548                         if (max_ps == -1)
2549                                 max_ps = state;
2550
2551                         if (total_latency_us > max_lat_us)
2552                                 max_lat_us = total_latency_us;
2553                 }
2554
2555                 apste = 1;
2556
2557                 if (max_ps == -1) {
2558                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2559                 } else {
2560                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2561                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2562                 }
2563         }
2564
2565         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2566                                 table, sizeof(*table), NULL);
2567         if (ret)
2568                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2569
2570         kfree(table);
2571         return ret;
2572 }
2573
2574 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2575 {
2576         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2577         u64 latency;
2578
2579         switch (val) {
2580         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2581         case PM_QOS_LATENCY_ANY:
2582                 latency = U64_MAX;
2583                 break;
2584
2585         default:
2586                 latency = val;
2587         }
2588
2589         if (ctrl->ps_max_latency_us != latency) {
2590                 ctrl->ps_max_latency_us = latency;
2591                 nvme_configure_apst(ctrl);
2592         }
2593 }
2594
2595 struct nvme_core_quirk_entry {
2596         /*
2597          * NVMe model and firmware strings are padded with spaces.  For
2598          * simplicity, strings in the quirk table are padded with NULLs
2599          * instead.
2600          */
2601         u16 vid;
2602         const char *mn;
2603         const char *fr;
2604         unsigned long quirks;
2605 };
2606
2607 static const struct nvme_core_quirk_entry core_quirks[] = {
2608         {
2609                 /*
2610                  * This Toshiba device seems to die using any APST states.  See:
2611                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2612                  */
2613                 .vid = 0x1179,
2614                 .mn = "THNSF5256GPUK TOSHIBA",
2615                 .quirks = NVME_QUIRK_NO_APST,
2616         },
2617         {
2618                 /*
2619                  * This LiteON CL1-3D*-Q11 firmware version has a race
2620                  * condition associated with actions related to suspend to idle
2621                  * LiteON has resolved the problem in future firmware
2622                  */
2623                 .vid = 0x14a4,
2624                 .fr = "22301111",
2625                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2626         }
2627 };
2628
2629 /* match is null-terminated but idstr is space-padded. */
2630 static bool string_matches(const char *idstr, const char *match, size_t len)
2631 {
2632         size_t matchlen;
2633
2634         if (!match)
2635                 return true;
2636
2637         matchlen = strlen(match);
2638         WARN_ON_ONCE(matchlen > len);
2639
2640         if (memcmp(idstr, match, matchlen))
2641                 return false;
2642
2643         for (; matchlen < len; matchlen++)
2644                 if (idstr[matchlen] != ' ')
2645                         return false;
2646
2647         return true;
2648 }
2649
2650 static bool quirk_matches(const struct nvme_id_ctrl *id,
2651                           const struct nvme_core_quirk_entry *q)
2652 {
2653         return q->vid == le16_to_cpu(id->vid) &&
2654                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2655                 string_matches(id->fr, q->fr, sizeof(id->fr));
2656 }
2657
2658 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2659                 struct nvme_id_ctrl *id)
2660 {
2661         size_t nqnlen;
2662         int off;
2663
2664         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2665                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2666                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2667                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2668                         return;
2669                 }
2670
2671                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2672                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2673         }
2674
2675         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2676         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2677                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2678                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2679         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2680         off += sizeof(id->sn);
2681         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2682         off += sizeof(id->mn);
2683         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2684 }
2685
2686 static void nvme_release_subsystem(struct device *dev)
2687 {
2688         struct nvme_subsystem *subsys =
2689                 container_of(dev, struct nvme_subsystem, dev);
2690
2691         if (subsys->instance >= 0)
2692                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2693         kfree(subsys);
2694 }
2695
2696 static void nvme_destroy_subsystem(struct kref *ref)
2697 {
2698         struct nvme_subsystem *subsys =
2699                         container_of(ref, struct nvme_subsystem, ref);
2700
2701         mutex_lock(&nvme_subsystems_lock);
2702         list_del(&subsys->entry);
2703         mutex_unlock(&nvme_subsystems_lock);
2704
2705         ida_destroy(&subsys->ns_ida);
2706         device_del(&subsys->dev);
2707         put_device(&subsys->dev);
2708 }
2709
2710 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2711 {
2712         kref_put(&subsys->ref, nvme_destroy_subsystem);
2713 }
2714
2715 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2716 {
2717         struct nvme_subsystem *subsys;
2718
2719         lockdep_assert_held(&nvme_subsystems_lock);
2720
2721         /*
2722          * Fail matches for discovery subsystems. This results
2723          * in each discovery controller bound to a unique subsystem.
2724          * This avoids issues with validating controller values
2725          * that can only be true when there is a single unique subsystem.
2726          * There may be multiple and completely independent entities
2727          * that provide discovery controllers.
2728          */
2729         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2730                 return NULL;
2731
2732         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2733                 if (strcmp(subsys->subnqn, subsysnqn))
2734                         continue;
2735                 if (!kref_get_unless_zero(&subsys->ref))
2736                         continue;
2737                 return subsys;
2738         }
2739
2740         return NULL;
2741 }
2742
2743 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2744         struct device_attribute subsys_attr_##_name = \
2745                 __ATTR(_name, _mode, _show, NULL)
2746
2747 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2748                                     struct device_attribute *attr,
2749                                     char *buf)
2750 {
2751         struct nvme_subsystem *subsys =
2752                 container_of(dev, struct nvme_subsystem, dev);
2753
2754         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2755 }
2756 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2757
2758 #define nvme_subsys_show_str_function(field)                            \
2759 static ssize_t subsys_##field##_show(struct device *dev,                \
2760                             struct device_attribute *attr, char *buf)   \
2761 {                                                                       \
2762         struct nvme_subsystem *subsys =                                 \
2763                 container_of(dev, struct nvme_subsystem, dev);          \
2764         return sprintf(buf, "%.*s\n",                                   \
2765                        (int)sizeof(subsys->field), subsys->field);      \
2766 }                                                                       \
2767 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2768
2769 nvme_subsys_show_str_function(model);
2770 nvme_subsys_show_str_function(serial);
2771 nvme_subsys_show_str_function(firmware_rev);
2772
2773 static struct attribute *nvme_subsys_attrs[] = {
2774         &subsys_attr_model.attr,
2775         &subsys_attr_serial.attr,
2776         &subsys_attr_firmware_rev.attr,
2777         &subsys_attr_subsysnqn.attr,
2778 #ifdef CONFIG_NVME_MULTIPATH
2779         &subsys_attr_iopolicy.attr,
2780 #endif
2781         NULL,
2782 };
2783
2784 static struct attribute_group nvme_subsys_attrs_group = {
2785         .attrs = nvme_subsys_attrs,
2786 };
2787
2788 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2789         &nvme_subsys_attrs_group,
2790         NULL,
2791 };
2792
2793 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2794                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2795 {
2796         struct nvme_ctrl *tmp;
2797
2798         lockdep_assert_held(&nvme_subsystems_lock);
2799
2800         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2801                 if (nvme_state_terminal(tmp))
2802                         continue;
2803
2804                 if (tmp->cntlid == ctrl->cntlid) {
2805                         dev_err(ctrl->device,
2806                                 "Duplicate cntlid %u with %s, rejecting\n",
2807                                 ctrl->cntlid, dev_name(tmp->device));
2808                         return false;
2809                 }
2810
2811                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2812                     (ctrl->opts && ctrl->opts->discovery_nqn))
2813                         continue;
2814
2815                 dev_err(ctrl->device,
2816                         "Subsystem does not support multiple controllers\n");
2817                 return false;
2818         }
2819
2820         return true;
2821 }
2822
2823 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2824 {
2825         struct nvme_subsystem *subsys, *found;
2826         int ret;
2827
2828         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2829         if (!subsys)
2830                 return -ENOMEM;
2831
2832         subsys->instance = -1;
2833         mutex_init(&subsys->lock);
2834         kref_init(&subsys->ref);
2835         INIT_LIST_HEAD(&subsys->ctrls);
2836         INIT_LIST_HEAD(&subsys->nsheads);
2837         nvme_init_subnqn(subsys, ctrl, id);
2838         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2839         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2840         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2841         subsys->vendor_id = le16_to_cpu(id->vid);
2842         subsys->cmic = id->cmic;
2843         subsys->awupf = le16_to_cpu(id->awupf);
2844 #ifdef CONFIG_NVME_MULTIPATH
2845         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2846 #endif
2847
2848         subsys->dev.class = nvme_subsys_class;
2849         subsys->dev.release = nvme_release_subsystem;
2850         subsys->dev.groups = nvme_subsys_attrs_groups;
2851         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2852         device_initialize(&subsys->dev);
2853
2854         mutex_lock(&nvme_subsystems_lock);
2855         found = __nvme_find_get_subsystem(subsys->subnqn);
2856         if (found) {
2857                 put_device(&subsys->dev);
2858                 subsys = found;
2859
2860                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2861                         ret = -EINVAL;
2862                         goto out_put_subsystem;
2863                 }
2864         } else {
2865                 ret = device_add(&subsys->dev);
2866                 if (ret) {
2867                         dev_err(ctrl->device,
2868                                 "failed to register subsystem device.\n");
2869                         put_device(&subsys->dev);
2870                         goto out_unlock;
2871                 }
2872                 ida_init(&subsys->ns_ida);
2873                 list_add_tail(&subsys->entry, &nvme_subsystems);
2874         }
2875
2876         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2877                                 dev_name(ctrl->device));
2878         if (ret) {
2879                 dev_err(ctrl->device,
2880                         "failed to create sysfs link from subsystem.\n");
2881                 goto out_put_subsystem;
2882         }
2883
2884         if (!found)
2885                 subsys->instance = ctrl->instance;
2886         ctrl->subsys = subsys;
2887         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2888         mutex_unlock(&nvme_subsystems_lock);
2889         return 0;
2890
2891 out_put_subsystem:
2892         nvme_put_subsystem(subsys);
2893 out_unlock:
2894         mutex_unlock(&nvme_subsystems_lock);
2895         return ret;
2896 }
2897
2898 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2899                 void *log, size_t size, u64 offset)
2900 {
2901         struct nvme_command c = { };
2902         u32 dwlen = nvme_bytes_to_numd(size);
2903
2904         c.get_log_page.opcode = nvme_admin_get_log_page;
2905         c.get_log_page.nsid = cpu_to_le32(nsid);
2906         c.get_log_page.lid = log_page;
2907         c.get_log_page.lsp = lsp;
2908         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2909         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2910         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2911         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2912         c.get_log_page.csi = csi;
2913
2914         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2915 }
2916
2917 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2918                                 struct nvme_effects_log **log)
2919 {
2920         struct nvme_cel *cel = xa_load(&ctrl->cels, csi);
2921         int ret;
2922
2923         if (cel)
2924                 goto out;
2925
2926         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2927         if (!cel)
2928                 return -ENOMEM;
2929
2930         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2931                         &cel->log, sizeof(cel->log), 0);
2932         if (ret) {
2933                 kfree(cel);
2934                 return ret;
2935         }
2936
2937         cel->csi = csi;
2938         xa_store(&ctrl->cels, cel->csi, cel, GFP_KERNEL);
2939 out:
2940         *log = &cel->log;
2941         return 0;
2942 }
2943
2944 /*
2945  * Initialize the cached copies of the Identify data and various controller
2946  * register in our nvme_ctrl structure.  This should be called as soon as
2947  * the admin queue is fully up and running.
2948  */
2949 int nvme_init_identify(struct nvme_ctrl *ctrl)
2950 {
2951         struct nvme_id_ctrl *id;
2952         int ret, page_shift;
2953         u32 max_hw_sectors;
2954         bool prev_apst_enabled;
2955
2956         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2957         if (ret) {
2958                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2959                 return ret;
2960         }
2961         page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2962         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2963
2964         if (ctrl->vs >= NVME_VS(1, 1, 0))
2965                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2966
2967         ret = nvme_identify_ctrl(ctrl, &id);
2968         if (ret) {
2969                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2970                 return -EIO;
2971         }
2972
2973         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2974                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2975                 if (ret < 0)
2976                         goto out_free;
2977         }
2978
2979         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2980                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2981
2982         if (!ctrl->identified) {
2983                 int i;
2984
2985                 ret = nvme_init_subsystem(ctrl, id);
2986                 if (ret)
2987                         goto out_free;
2988
2989                 /*
2990                  * Check for quirks.  Quirk can depend on firmware version,
2991                  * so, in principle, the set of quirks present can change
2992                  * across a reset.  As a possible future enhancement, we
2993                  * could re-scan for quirks every time we reinitialize
2994                  * the device, but we'd have to make sure that the driver
2995                  * behaves intelligently if the quirks change.
2996                  */
2997                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2998                         if (quirk_matches(id, &core_quirks[i]))
2999                                 ctrl->quirks |= core_quirks[i].quirks;
3000                 }
3001         }
3002
3003         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3004                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3005                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3006         }
3007
3008         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3009         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3010         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3011
3012         ctrl->oacs = le16_to_cpu(id->oacs);
3013         ctrl->oncs = le16_to_cpu(id->oncs);
3014         ctrl->mtfa = le16_to_cpu(id->mtfa);
3015         ctrl->oaes = le32_to_cpu(id->oaes);
3016         ctrl->wctemp = le16_to_cpu(id->wctemp);
3017         ctrl->cctemp = le16_to_cpu(id->cctemp);
3018
3019         atomic_set(&ctrl->abort_limit, id->acl + 1);
3020         ctrl->vwc = id->vwc;
3021         if (id->mdts)
3022                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3023         else
3024                 max_hw_sectors = UINT_MAX;
3025         ctrl->max_hw_sectors =
3026                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3027
3028         nvme_set_queue_limits(ctrl, ctrl->admin_q);
3029         ctrl->sgls = le32_to_cpu(id->sgls);
3030         ctrl->kas = le16_to_cpu(id->kas);
3031         ctrl->max_namespaces = le32_to_cpu(id->mnan);
3032         ctrl->ctratt = le32_to_cpu(id->ctratt);
3033
3034         if (id->rtd3e) {
3035                 /* us -> s */
3036                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3037
3038                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3039                                                  shutdown_timeout, 60);
3040
3041                 if (ctrl->shutdown_timeout != shutdown_timeout)
3042                         dev_info(ctrl->device,
3043                                  "Shutdown timeout set to %u seconds\n",
3044                                  ctrl->shutdown_timeout);
3045         } else
3046                 ctrl->shutdown_timeout = shutdown_timeout;
3047
3048         ctrl->npss = id->npss;
3049         ctrl->apsta = id->apsta;
3050         prev_apst_enabled = ctrl->apst_enabled;
3051         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3052                 if (force_apst && id->apsta) {
3053                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3054                         ctrl->apst_enabled = true;
3055                 } else {
3056                         ctrl->apst_enabled = false;
3057                 }
3058         } else {
3059                 ctrl->apst_enabled = id->apsta;
3060         }
3061         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3062
3063         if (ctrl->ops->flags & NVME_F_FABRICS) {
3064                 ctrl->icdoff = le16_to_cpu(id->icdoff);
3065                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3066                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3067                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3068
3069                 /*
3070                  * In fabrics we need to verify the cntlid matches the
3071                  * admin connect
3072                  */
3073                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3074                         dev_err(ctrl->device,
3075                                 "Mismatching cntlid: Connect %u vs Identify "
3076                                 "%u, rejecting\n",
3077                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
3078                         ret = -EINVAL;
3079                         goto out_free;
3080                 }
3081
3082                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
3083                         dev_err(ctrl->device,
3084                                 "keep-alive support is mandatory for fabrics\n");
3085                         ret = -EINVAL;
3086                         goto out_free;
3087                 }
3088         } else {
3089                 ctrl->hmpre = le32_to_cpu(id->hmpre);
3090                 ctrl->hmmin = le32_to_cpu(id->hmmin);
3091                 ctrl->hmminds = le32_to_cpu(id->hmminds);
3092                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3093         }
3094
3095         ret = nvme_mpath_init(ctrl, id);
3096         kfree(id);
3097
3098         if (ret < 0)
3099                 return ret;
3100
3101         if (ctrl->apst_enabled && !prev_apst_enabled)
3102                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3103         else if (!ctrl->apst_enabled && prev_apst_enabled)
3104                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3105
3106         ret = nvme_configure_apst(ctrl);
3107         if (ret < 0)
3108                 return ret;
3109
3110         ret = nvme_configure_timestamp(ctrl);
3111         if (ret < 0)
3112                 return ret;
3113
3114         ret = nvme_configure_directives(ctrl);
3115         if (ret < 0)
3116                 return ret;
3117
3118         ret = nvme_configure_acre(ctrl);
3119         if (ret < 0)
3120                 return ret;
3121
3122         if (!ctrl->identified) {
3123                 ret = nvme_hwmon_init(ctrl);
3124                 if (ret < 0)
3125                         return ret;
3126         }
3127
3128         ctrl->identified = true;
3129
3130         return 0;
3131
3132 out_free:
3133         kfree(id);
3134         return ret;
3135 }
3136 EXPORT_SYMBOL_GPL(nvme_init_identify);
3137
3138 static int nvme_dev_open(struct inode *inode, struct file *file)
3139 {
3140         struct nvme_ctrl *ctrl =
3141                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3142
3143         switch (ctrl->state) {
3144         case NVME_CTRL_LIVE:
3145                 break;
3146         default:
3147                 return -EWOULDBLOCK;
3148         }
3149
3150         nvme_get_ctrl(ctrl);
3151         if (!try_module_get(ctrl->ops->module)) {
3152                 nvme_put_ctrl(ctrl);
3153                 return -EINVAL;
3154         }
3155
3156         file->private_data = ctrl;
3157         return 0;
3158 }
3159
3160 static int nvme_dev_release(struct inode *inode, struct file *file)
3161 {
3162         struct nvme_ctrl *ctrl =
3163                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3164
3165         module_put(ctrl->ops->module);
3166         nvme_put_ctrl(ctrl);
3167         return 0;
3168 }
3169
3170 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3171 {
3172         struct nvme_ns *ns;
3173         int ret;
3174
3175         down_read(&ctrl->namespaces_rwsem);
3176         if (list_empty(&ctrl->namespaces)) {
3177                 ret = -ENOTTY;
3178                 goto out_unlock;
3179         }
3180
3181         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3182         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3183                 dev_warn(ctrl->device,
3184                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3185                 ret = -EINVAL;
3186                 goto out_unlock;
3187         }
3188
3189         dev_warn(ctrl->device,
3190                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3191         kref_get(&ns->kref);
3192         up_read(&ctrl->namespaces_rwsem);
3193
3194         ret = nvme_user_cmd(ctrl, ns, argp);
3195         nvme_put_ns(ns);
3196         return ret;
3197
3198 out_unlock:
3199         up_read(&ctrl->namespaces_rwsem);
3200         return ret;
3201 }
3202
3203 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3204                 unsigned long arg)
3205 {
3206         struct nvme_ctrl *ctrl = file->private_data;
3207         void __user *argp = (void __user *)arg;
3208
3209         switch (cmd) {
3210         case NVME_IOCTL_ADMIN_CMD:
3211                 return nvme_user_cmd(ctrl, NULL, argp);
3212         case NVME_IOCTL_ADMIN64_CMD:
3213                 return nvme_user_cmd64(ctrl, NULL, argp);
3214         case NVME_IOCTL_IO_CMD:
3215                 return nvme_dev_user_cmd(ctrl, argp);
3216         case NVME_IOCTL_RESET:
3217                 dev_warn(ctrl->device, "resetting controller\n");
3218                 return nvme_reset_ctrl_sync(ctrl);
3219         case NVME_IOCTL_SUBSYS_RESET:
3220                 return nvme_reset_subsystem(ctrl);
3221         case NVME_IOCTL_RESCAN:
3222                 nvme_queue_scan(ctrl);
3223                 return 0;
3224         default:
3225                 return -ENOTTY;
3226         }
3227 }
3228
3229 static const struct file_operations nvme_dev_fops = {
3230         .owner          = THIS_MODULE,
3231         .open           = nvme_dev_open,
3232         .release        = nvme_dev_release,
3233         .unlocked_ioctl = nvme_dev_ioctl,
3234         .compat_ioctl   = compat_ptr_ioctl,
3235 };
3236
3237 static ssize_t nvme_sysfs_reset(struct device *dev,
3238                                 struct device_attribute *attr, const char *buf,
3239                                 size_t count)
3240 {
3241         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3242         int ret;
3243
3244         ret = nvme_reset_ctrl_sync(ctrl);
3245         if (ret < 0)
3246                 return ret;
3247         return count;
3248 }
3249 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3250
3251 static ssize_t nvme_sysfs_rescan(struct device *dev,
3252                                 struct device_attribute *attr, const char *buf,
3253                                 size_t count)
3254 {
3255         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3256
3257         nvme_queue_scan(ctrl);
3258         return count;
3259 }
3260 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3261
3262 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3263 {
3264         struct gendisk *disk = dev_to_disk(dev);
3265
3266         if (disk->fops == &nvme_fops)
3267                 return nvme_get_ns_from_dev(dev)->head;
3268         else
3269                 return disk->private_data;
3270 }
3271
3272 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3273                 char *buf)
3274 {
3275         struct nvme_ns_head *head = dev_to_ns_head(dev);
3276         struct nvme_ns_ids *ids = &head->ids;
3277         struct nvme_subsystem *subsys = head->subsys;
3278         int serial_len = sizeof(subsys->serial);
3279         int model_len = sizeof(subsys->model);
3280
3281         if (!uuid_is_null(&ids->uuid))
3282                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3283
3284         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3285                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3286
3287         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3288                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3289
3290         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3291                                   subsys->serial[serial_len - 1] == '\0'))
3292                 serial_len--;
3293         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3294                                  subsys->model[model_len - 1] == '\0'))
3295                 model_len--;
3296
3297         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3298                 serial_len, subsys->serial, model_len, subsys->model,
3299                 head->ns_id);
3300 }
3301 static DEVICE_ATTR_RO(wwid);
3302
3303 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3304                 char *buf)
3305 {
3306         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3307 }
3308 static DEVICE_ATTR_RO(nguid);
3309
3310 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3311                 char *buf)
3312 {
3313         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3314
3315         /* For backward compatibility expose the NGUID to userspace if
3316          * we have no UUID set
3317          */
3318         if (uuid_is_null(&ids->uuid)) {
3319                 printk_ratelimited(KERN_WARNING
3320                                    "No UUID available providing old NGUID\n");
3321                 return sprintf(buf, "%pU\n", ids->nguid);
3322         }
3323         return sprintf(buf, "%pU\n", &ids->uuid);
3324 }
3325 static DEVICE_ATTR_RO(uuid);
3326
3327 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3328                 char *buf)
3329 {
3330         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3331 }
3332 static DEVICE_ATTR_RO(eui);
3333
3334 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3335                 char *buf)
3336 {
3337         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3338 }
3339 static DEVICE_ATTR_RO(nsid);
3340
3341 static struct attribute *nvme_ns_id_attrs[] = {
3342         &dev_attr_wwid.attr,
3343         &dev_attr_uuid.attr,
3344         &dev_attr_nguid.attr,
3345         &dev_attr_eui.attr,
3346         &dev_attr_nsid.attr,
3347 #ifdef CONFIG_NVME_MULTIPATH
3348         &dev_attr_ana_grpid.attr,
3349         &dev_attr_ana_state.attr,
3350 #endif
3351         NULL,
3352 };
3353
3354 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3355                 struct attribute *a, int n)
3356 {
3357         struct device *dev = container_of(kobj, struct device, kobj);
3358         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3359
3360         if (a == &dev_attr_uuid.attr) {
3361                 if (uuid_is_null(&ids->uuid) &&
3362                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3363                         return 0;
3364         }
3365         if (a == &dev_attr_nguid.attr) {
3366                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3367                         return 0;
3368         }
3369         if (a == &dev_attr_eui.attr) {
3370                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3371                         return 0;
3372         }
3373 #ifdef CONFIG_NVME_MULTIPATH
3374         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3375                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3376                         return 0;
3377                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3378                         return 0;
3379         }
3380 #endif
3381         return a->mode;
3382 }
3383
3384 static const struct attribute_group nvme_ns_id_attr_group = {
3385         .attrs          = nvme_ns_id_attrs,
3386         .is_visible     = nvme_ns_id_attrs_are_visible,
3387 };
3388
3389 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3390         &nvme_ns_id_attr_group,
3391 #ifdef CONFIG_NVM
3392         &nvme_nvm_attr_group,
3393 #endif
3394         NULL,
3395 };
3396
3397 #define nvme_show_str_function(field)                                           \
3398 static ssize_t  field##_show(struct device *dev,                                \
3399                             struct device_attribute *attr, char *buf)           \
3400 {                                                                               \
3401         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3402         return sprintf(buf, "%.*s\n",                                           \
3403                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3404 }                                                                               \
3405 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3406
3407 nvme_show_str_function(model);
3408 nvme_show_str_function(serial);
3409 nvme_show_str_function(firmware_rev);
3410
3411 #define nvme_show_int_function(field)                                           \
3412 static ssize_t  field##_show(struct device *dev,                                \
3413                             struct device_attribute *attr, char *buf)           \
3414 {                                                                               \
3415         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3416         return sprintf(buf, "%d\n", ctrl->field);       \
3417 }                                                                               \
3418 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3419
3420 nvme_show_int_function(cntlid);
3421 nvme_show_int_function(numa_node);
3422 nvme_show_int_function(queue_count);
3423 nvme_show_int_function(sqsize);
3424
3425 static ssize_t nvme_sysfs_delete(struct device *dev,
3426                                 struct device_attribute *attr, const char *buf,
3427                                 size_t count)
3428 {
3429         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3430
3431         if (device_remove_file_self(dev, attr))
3432                 nvme_delete_ctrl_sync(ctrl);
3433         return count;
3434 }
3435 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3436
3437 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3438                                          struct device_attribute *attr,
3439                                          char *buf)
3440 {
3441         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3442
3443         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3444 }
3445 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3446
3447 static ssize_t nvme_sysfs_show_state(struct device *dev,
3448                                      struct device_attribute *attr,
3449                                      char *buf)
3450 {
3451         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3452         static const char *const state_name[] = {
3453                 [NVME_CTRL_NEW]         = "new",
3454                 [NVME_CTRL_LIVE]        = "live",
3455                 [NVME_CTRL_RESETTING]   = "resetting",
3456                 [NVME_CTRL_CONNECTING]  = "connecting",
3457                 [NVME_CTRL_DELETING]    = "deleting",
3458                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3459                 [NVME_CTRL_DEAD]        = "dead",
3460         };
3461
3462         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3463             state_name[ctrl->state])
3464                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3465
3466         return sprintf(buf, "unknown state\n");
3467 }
3468
3469 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3470
3471 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3472                                          struct device_attribute *attr,
3473                                          char *buf)
3474 {
3475         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3476
3477         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3478 }
3479 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3480
3481 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3482                                         struct device_attribute *attr,
3483                                         char *buf)
3484 {
3485         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3486
3487         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3488 }
3489 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3490
3491 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3492                                         struct device_attribute *attr,
3493                                         char *buf)
3494 {
3495         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3496
3497         return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3498 }
3499 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3500
3501 static ssize_t nvme_sysfs_show_address(struct device *dev,
3502                                          struct device_attribute *attr,
3503                                          char *buf)
3504 {
3505         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3506
3507         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3508 }
3509 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3510
3511 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3512                 struct device_attribute *attr, char *buf)
3513 {
3514         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3515         struct nvmf_ctrl_options *opts = ctrl->opts;
3516
3517         if (ctrl->opts->max_reconnects == -1)
3518                 return sprintf(buf, "off\n");
3519         return sprintf(buf, "%d\n",
3520                         opts->max_reconnects * opts->reconnect_delay);
3521 }
3522
3523 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3524                 struct device_attribute *attr, const char *buf, size_t count)
3525 {
3526         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3527         struct nvmf_ctrl_options *opts = ctrl->opts;
3528         int ctrl_loss_tmo, err;
3529
3530         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3531         if (err)
3532                 return -EINVAL;
3533
3534         else if (ctrl_loss_tmo < 0)
3535                 opts->max_reconnects = -1;
3536         else
3537                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3538                                                 opts->reconnect_delay);
3539         return count;
3540 }
3541 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3542         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3543
3544 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3545                 struct device_attribute *attr, char *buf)
3546 {
3547         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3548
3549         if (ctrl->opts->reconnect_delay == -1)
3550                 return sprintf(buf, "off\n");
3551         return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3552 }
3553
3554 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3555                 struct device_attribute *attr, const char *buf, size_t count)
3556 {
3557         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3558         unsigned int v;
3559         int err;
3560
3561         err = kstrtou32(buf, 10, &v);
3562         if (err)
3563                 return err;
3564
3565         ctrl->opts->reconnect_delay = v;
3566         return count;
3567 }
3568 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3569         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3570
3571 static struct attribute *nvme_dev_attrs[] = {
3572         &dev_attr_reset_controller.attr,
3573         &dev_attr_rescan_controller.attr,
3574         &dev_attr_model.attr,
3575         &dev_attr_serial.attr,
3576         &dev_attr_firmware_rev.attr,
3577         &dev_attr_cntlid.attr,
3578         &dev_attr_delete_controller.attr,
3579         &dev_attr_transport.attr,
3580         &dev_attr_subsysnqn.attr,
3581         &dev_attr_address.attr,
3582         &dev_attr_state.attr,
3583         &dev_attr_numa_node.attr,
3584         &dev_attr_queue_count.attr,
3585         &dev_attr_sqsize.attr,
3586         &dev_attr_hostnqn.attr,
3587         &dev_attr_hostid.attr,
3588         &dev_attr_ctrl_loss_tmo.attr,
3589         &dev_attr_reconnect_delay.attr,
3590         NULL
3591 };
3592
3593 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3594                 struct attribute *a, int n)
3595 {
3596         struct device *dev = container_of(kobj, struct device, kobj);
3597         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3598
3599         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3600                 return 0;
3601         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3602                 return 0;
3603         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3604                 return 0;
3605         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3606                 return 0;
3607         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3608                 return 0;
3609         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3610                 return 0;
3611
3612         return a->mode;
3613 }
3614
3615 static struct attribute_group nvme_dev_attrs_group = {
3616         .attrs          = nvme_dev_attrs,
3617         .is_visible     = nvme_dev_attrs_are_visible,
3618 };
3619
3620 static const struct attribute_group *nvme_dev_attr_groups[] = {
3621         &nvme_dev_attrs_group,
3622         NULL,
3623 };
3624
3625 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3626                 unsigned nsid)
3627 {
3628         struct nvme_ns_head *h;
3629
3630         lockdep_assert_held(&subsys->lock);
3631
3632         list_for_each_entry(h, &subsys->nsheads, entry) {
3633                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3634                         return h;
3635         }
3636
3637         return NULL;
3638 }
3639
3640 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3641                 struct nvme_ns_head *new)
3642 {
3643         struct nvme_ns_head *h;
3644
3645         lockdep_assert_held(&subsys->lock);
3646
3647         list_for_each_entry(h, &subsys->nsheads, entry) {
3648                 if (nvme_ns_ids_valid(&new->ids) &&
3649                     nvme_ns_ids_equal(&new->ids, &h->ids))
3650                         return -EINVAL;
3651         }
3652
3653         return 0;
3654 }
3655
3656 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3657                 unsigned nsid, struct nvme_ns_ids *ids)
3658 {
3659         struct nvme_ns_head *head;
3660         size_t size = sizeof(*head);
3661         int ret = -ENOMEM;
3662
3663 #ifdef CONFIG_NVME_MULTIPATH
3664         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3665 #endif
3666
3667         head = kzalloc(size, GFP_KERNEL);
3668         if (!head)
3669                 goto out;
3670         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3671         if (ret < 0)
3672                 goto out_free_head;
3673         head->instance = ret;
3674         INIT_LIST_HEAD(&head->list);
3675         ret = init_srcu_struct(&head->srcu);
3676         if (ret)
3677                 goto out_ida_remove;
3678         head->subsys = ctrl->subsys;
3679         head->ns_id = nsid;
3680         head->ids = *ids;
3681         kref_init(&head->ref);
3682
3683         ret = __nvme_check_ids(ctrl->subsys, head);
3684         if (ret) {
3685                 dev_err(ctrl->device,
3686                         "duplicate IDs for nsid %d\n", nsid);
3687                 goto out_cleanup_srcu;
3688         }
3689
3690         if (head->ids.csi) {
3691                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3692                 if (ret)
3693                         goto out_cleanup_srcu;
3694         } else
3695                 head->effects = ctrl->effects;
3696
3697         ret = nvme_mpath_alloc_disk(ctrl, head);
3698         if (ret)
3699                 goto out_cleanup_srcu;
3700
3701         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3702
3703         kref_get(&ctrl->subsys->ref);
3704
3705         return head;
3706 out_cleanup_srcu:
3707         cleanup_srcu_struct(&head->srcu);
3708 out_ida_remove:
3709         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3710 out_free_head:
3711         kfree(head);
3712 out:
3713         if (ret > 0)
3714                 ret = blk_status_to_errno(nvme_error_status(ret));
3715         return ERR_PTR(ret);
3716 }
3717
3718 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3719                 struct nvme_ns_ids *ids, bool is_shared)
3720 {
3721         struct nvme_ctrl *ctrl = ns->ctrl;
3722         struct nvme_ns_head *head = NULL;
3723         int ret = 0;
3724
3725         mutex_lock(&ctrl->subsys->lock);
3726         head = nvme_find_ns_head(ctrl->subsys, nsid);
3727         if (!head) {
3728                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3729                 if (IS_ERR(head)) {
3730                         ret = PTR_ERR(head);
3731                         goto out_unlock;
3732                 }
3733                 head->shared = is_shared;
3734         } else {
3735                 ret = -EINVAL;
3736                 if (!is_shared || !head->shared) {
3737                         dev_err(ctrl->device,
3738                                 "Duplicate unshared namespace %d\n", nsid);
3739                         goto out_put_ns_head;
3740                 }
3741                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3742                         dev_err(ctrl->device,
3743                                 "IDs don't match for shared namespace %d\n",
3744                                         nsid);
3745                         goto out_put_ns_head;
3746                 }
3747         }
3748
3749         list_add_tail(&ns->siblings, &head->list);
3750         ns->head = head;
3751         mutex_unlock(&ctrl->subsys->lock);
3752         return 0;
3753
3754 out_put_ns_head:
3755         nvme_put_ns_head(head);
3756 out_unlock:
3757         mutex_unlock(&ctrl->subsys->lock);
3758         return ret;
3759 }
3760
3761 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3762 {
3763         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3764         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3765
3766         return nsa->head->ns_id - nsb->head->ns_id;
3767 }
3768
3769 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3770 {
3771         struct nvme_ns *ns, *ret = NULL;
3772
3773         down_read(&ctrl->namespaces_rwsem);
3774         list_for_each_entry(ns, &ctrl->namespaces, list) {
3775                 if (ns->head->ns_id == nsid) {
3776                         if (!kref_get_unless_zero(&ns->kref))
3777                                 continue;
3778                         ret = ns;
3779                         break;
3780                 }
3781                 if (ns->head->ns_id > nsid)
3782                         break;
3783         }
3784         up_read(&ctrl->namespaces_rwsem);
3785         return ret;
3786 }
3787 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3788
3789 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3790                 struct nvme_ns_ids *ids)
3791 {
3792         struct nvme_ns *ns;
3793         struct gendisk *disk;
3794         struct nvme_id_ns *id;
3795         char disk_name[DISK_NAME_LEN];
3796         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3797
3798         if (nvme_identify_ns(ctrl, nsid, ids, &id))
3799                 return;
3800
3801         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3802         if (!ns)
3803                 goto out_free_id;
3804
3805         ns->queue = blk_mq_init_queue(ctrl->tagset);
3806         if (IS_ERR(ns->queue))
3807                 goto out_free_ns;
3808
3809         if (ctrl->opts && ctrl->opts->data_digest)
3810                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3811
3812         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3813         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3814                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3815
3816         ns->queue->queuedata = ns;
3817         ns->ctrl = ctrl;
3818         kref_init(&ns->kref);
3819
3820         ret = nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED);
3821         if (ret)
3822                 goto out_free_queue;
3823         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3824
3825         disk = alloc_disk_node(0, node);
3826         if (!disk)
3827                 goto out_unlink_ns;
3828
3829         disk->fops = &nvme_fops;
3830         disk->private_data = ns;
3831         disk->queue = ns->queue;
3832         disk->flags = flags;
3833         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3834         ns->disk = disk;
3835
3836         if (nvme_update_ns_info(ns, id))
3837                 goto out_put_disk;
3838
3839         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3840                 ret = nvme_nvm_register(ns, disk_name, node);
3841                 if (ret) {
3842                         dev_warn(ctrl->device, "LightNVM init failure\n");
3843                         goto out_put_disk;
3844                 }
3845         }
3846
3847         down_write(&ctrl->namespaces_rwsem);
3848         list_add_tail(&ns->list, &ctrl->namespaces);
3849         up_write(&ctrl->namespaces_rwsem);
3850
3851         nvme_get_ctrl(ctrl);
3852
3853         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3854
3855         nvme_mpath_add_disk(ns, id);
3856         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3857         kfree(id);
3858
3859         return;
3860  out_put_disk:
3861         /* prevent double queue cleanup */
3862         ns->disk->queue = NULL;
3863         put_disk(ns->disk);
3864  out_unlink_ns:
3865         mutex_lock(&ctrl->subsys->lock);
3866         list_del_rcu(&ns->siblings);
3867         if (list_empty(&ns->head->list))
3868                 list_del_init(&ns->head->entry);
3869         mutex_unlock(&ctrl->subsys->lock);
3870         nvme_put_ns_head(ns->head);
3871  out_free_queue:
3872         blk_cleanup_queue(ns->queue);
3873  out_free_ns:
3874         kfree(ns);
3875  out_free_id:
3876         kfree(id);
3877 }
3878
3879 static void nvme_ns_remove(struct nvme_ns *ns)
3880 {
3881         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3882                 return;
3883
3884         set_capacity(ns->disk, 0);
3885         nvme_fault_inject_fini(&ns->fault_inject);
3886
3887         mutex_lock(&ns->ctrl->subsys->lock);
3888         list_del_rcu(&ns->siblings);
3889         if (list_empty(&ns->head->list))
3890                 list_del_init(&ns->head->entry);
3891         mutex_unlock(&ns->ctrl->subsys->lock);
3892
3893         synchronize_rcu(); /* guarantee not available in head->list */
3894         nvme_mpath_clear_current_path(ns);
3895         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3896
3897         if (ns->disk->flags & GENHD_FL_UP) {
3898                 del_gendisk(ns->disk);
3899                 blk_cleanup_queue(ns->queue);
3900                 if (blk_get_integrity(ns->disk))
3901                         blk_integrity_unregister(ns->disk);
3902         }
3903
3904         down_write(&ns->ctrl->namespaces_rwsem);
3905         list_del_init(&ns->list);
3906         up_write(&ns->ctrl->namespaces_rwsem);
3907
3908         nvme_mpath_check_last_path(ns);
3909         nvme_put_ns(ns);
3910 }
3911
3912 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3913 {
3914         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3915
3916         if (ns) {
3917                 nvme_ns_remove(ns);
3918                 nvme_put_ns(ns);
3919         }
3920 }
3921
3922 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3923 {
3924         struct nvme_id_ns *id;
3925         int ret = -ENODEV;
3926
3927         if (test_bit(NVME_NS_DEAD, &ns->flags))
3928                 goto out;
3929
3930         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3931         if (ret)
3932                 goto out;
3933
3934         ret = -ENODEV;
3935         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3936                 dev_err(ns->ctrl->device,
3937                         "identifiers changed for nsid %d\n", ns->head->ns_id);
3938                 goto out_free_id;
3939         }
3940
3941         ret = nvme_update_ns_info(ns, id);
3942
3943 out_free_id:
3944         kfree(id);
3945 out:
3946         /*
3947          * Only remove the namespace if we got a fatal error back from the
3948          * device, otherwise ignore the error and just move on.
3949          *
3950          * TODO: we should probably schedule a delayed retry here.
3951          */
3952         if (ret && ret != -ENOMEM && !(ret > 0 && !(ret & NVME_SC_DNR)))
3953                 nvme_ns_remove(ns);
3954 }
3955
3956 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3957 {
3958         struct nvme_ns_ids ids = { };
3959         struct nvme_ns *ns;
3960
3961         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
3962                 return;
3963
3964         ns = nvme_find_get_ns(ctrl, nsid);
3965         if (ns) {
3966                 nvme_validate_ns(ns, &ids);
3967                 nvme_put_ns(ns);
3968                 return;
3969         }
3970
3971         switch (ids.csi) {
3972         case NVME_CSI_NVM:
3973                 nvme_alloc_ns(ctrl, nsid, &ids);
3974                 break;
3975         case NVME_CSI_ZNS:
3976                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
3977                         dev_warn(ctrl->device,
3978                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
3979                                 nsid);
3980                         break;
3981                 }
3982                 nvme_alloc_ns(ctrl, nsid, &ids);
3983                 break;
3984         default:
3985                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
3986                         ids.csi, nsid);
3987                 break;
3988         }
3989 }
3990
3991 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3992                                         unsigned nsid)
3993 {
3994         struct nvme_ns *ns, *next;
3995         LIST_HEAD(rm_list);
3996
3997         down_write(&ctrl->namespaces_rwsem);
3998         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3999                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4000                         list_move_tail(&ns->list, &rm_list);
4001         }
4002         up_write(&ctrl->namespaces_rwsem);
4003
4004         list_for_each_entry_safe(ns, next, &rm_list, list)
4005                 nvme_ns_remove(ns);
4006
4007 }
4008
4009 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4010 {
4011         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4012         __le32 *ns_list;
4013         u32 prev = 0;
4014         int ret = 0, i;
4015
4016         if (nvme_ctrl_limited_cns(ctrl))
4017                 return -EOPNOTSUPP;
4018
4019         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4020         if (!ns_list)
4021                 return -ENOMEM;
4022
4023         for (;;) {
4024                 struct nvme_command cmd = {
4025                         .identify.opcode        = nvme_admin_identify,
4026                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
4027                         .identify.nsid          = cpu_to_le32(prev),
4028                 };
4029
4030                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4031                                             NVME_IDENTIFY_DATA_SIZE);
4032                 if (ret)
4033                         goto free;
4034
4035                 for (i = 0; i < nr_entries; i++) {
4036                         u32 nsid = le32_to_cpu(ns_list[i]);
4037
4038                         if (!nsid)      /* end of the list? */
4039                                 goto out;
4040                         nvme_validate_or_alloc_ns(ctrl, nsid);
4041                         while (++prev < nsid)
4042                                 nvme_ns_remove_by_nsid(ctrl, prev);
4043                 }
4044         }
4045  out:
4046         nvme_remove_invalid_namespaces(ctrl, prev);
4047  free:
4048         kfree(ns_list);
4049         return ret;
4050 }
4051
4052 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4053 {
4054         struct nvme_id_ctrl *id;
4055         u32 nn, i;
4056
4057         if (nvme_identify_ctrl(ctrl, &id))
4058                 return;
4059         nn = le32_to_cpu(id->nn);
4060         kfree(id);
4061
4062         for (i = 1; i <= nn; i++)
4063                 nvme_validate_or_alloc_ns(ctrl, i);
4064
4065         nvme_remove_invalid_namespaces(ctrl, nn);
4066 }
4067
4068 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4069 {
4070         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4071         __le32 *log;
4072         int error;
4073
4074         log = kzalloc(log_size, GFP_KERNEL);
4075         if (!log)
4076                 return;
4077
4078         /*
4079          * We need to read the log to clear the AEN, but we don't want to rely
4080          * on it for the changed namespace information as userspace could have
4081          * raced with us in reading the log page, which could cause us to miss
4082          * updates.
4083          */
4084         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4085                         NVME_CSI_NVM, log, log_size, 0);
4086         if (error)
4087                 dev_warn(ctrl->device,
4088                         "reading changed ns log failed: %d\n", error);
4089
4090         kfree(log);
4091 }
4092
4093 static void nvme_scan_work(struct work_struct *work)
4094 {
4095         struct nvme_ctrl *ctrl =
4096                 container_of(work, struct nvme_ctrl, scan_work);
4097
4098         /* No tagset on a live ctrl means IO queues could not created */
4099         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4100                 return;
4101
4102         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4103                 dev_info(ctrl->device, "rescanning namespaces.\n");
4104                 nvme_clear_changed_ns_log(ctrl);
4105         }
4106
4107         mutex_lock(&ctrl->scan_lock);
4108         if (nvme_scan_ns_list(ctrl) != 0)
4109                 nvme_scan_ns_sequential(ctrl);
4110         mutex_unlock(&ctrl->scan_lock);
4111
4112         down_write(&ctrl->namespaces_rwsem);
4113         list_sort(NULL, &ctrl->namespaces, ns_cmp);
4114         up_write(&ctrl->namespaces_rwsem);
4115 }
4116
4117 /*
4118  * This function iterates the namespace list unlocked to allow recovery from
4119  * controller failure. It is up to the caller to ensure the namespace list is
4120  * not modified by scan work while this function is executing.
4121  */
4122 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4123 {
4124         struct nvme_ns *ns, *next;
4125         LIST_HEAD(ns_list);
4126
4127         /*
4128          * make sure to requeue I/O to all namespaces as these
4129          * might result from the scan itself and must complete
4130          * for the scan_work to make progress
4131          */
4132         nvme_mpath_clear_ctrl_paths(ctrl);
4133
4134         /* prevent racing with ns scanning */
4135         flush_work(&ctrl->scan_work);
4136
4137         /*
4138          * The dead states indicates the controller was not gracefully
4139          * disconnected. In that case, we won't be able to flush any data while
4140          * removing the namespaces' disks; fail all the queues now to avoid
4141          * potentially having to clean up the failed sync later.
4142          */
4143         if (ctrl->state == NVME_CTRL_DEAD)
4144                 nvme_kill_queues(ctrl);
4145
4146         /* this is a no-op when called from the controller reset handler */
4147         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4148
4149         down_write(&ctrl->namespaces_rwsem);
4150         list_splice_init(&ctrl->namespaces, &ns_list);
4151         up_write(&ctrl->namespaces_rwsem);
4152
4153         list_for_each_entry_safe(ns, next, &ns_list, list)
4154                 nvme_ns_remove(ns);
4155 }
4156 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4157
4158 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4159 {
4160         struct nvme_ctrl *ctrl =
4161                 container_of(dev, struct nvme_ctrl, ctrl_device);
4162         struct nvmf_ctrl_options *opts = ctrl->opts;
4163         int ret;
4164
4165         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4166         if (ret)
4167                 return ret;
4168
4169         if (opts) {
4170                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4171                 if (ret)
4172                         return ret;
4173
4174                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4175                                 opts->trsvcid ?: "none");
4176                 if (ret)
4177                         return ret;
4178
4179                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4180                                 opts->host_traddr ?: "none");
4181         }
4182         return ret;
4183 }
4184
4185 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4186 {
4187         char *envp[2] = { NULL, NULL };
4188         u32 aen_result = ctrl->aen_result;
4189
4190         ctrl->aen_result = 0;
4191         if (!aen_result)
4192                 return;
4193
4194         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4195         if (!envp[0])
4196                 return;
4197         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4198         kfree(envp[0]);
4199 }
4200
4201 static void nvme_async_event_work(struct work_struct *work)
4202 {
4203         struct nvme_ctrl *ctrl =
4204                 container_of(work, struct nvme_ctrl, async_event_work);
4205
4206         nvme_aen_uevent(ctrl);
4207         ctrl->ops->submit_async_event(ctrl);
4208 }
4209
4210 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4211 {
4212
4213         u32 csts;
4214
4215         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4216                 return false;
4217
4218         if (csts == ~0)
4219                 return false;
4220
4221         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4222 }
4223
4224 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4225 {
4226         struct nvme_fw_slot_info_log *log;
4227
4228         log = kmalloc(sizeof(*log), GFP_KERNEL);
4229         if (!log)
4230                 return;
4231
4232         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4233                         log, sizeof(*log), 0))
4234                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4235         kfree(log);
4236 }
4237
4238 static void nvme_fw_act_work(struct work_struct *work)
4239 {
4240         struct nvme_ctrl *ctrl = container_of(work,
4241                                 struct nvme_ctrl, fw_act_work);
4242         unsigned long fw_act_timeout;
4243
4244         if (ctrl->mtfa)
4245                 fw_act_timeout = jiffies +
4246                                 msecs_to_jiffies(ctrl->mtfa * 100);
4247         else
4248                 fw_act_timeout = jiffies +
4249                                 msecs_to_jiffies(admin_timeout * 1000);
4250
4251         nvme_stop_queues(ctrl);
4252         while (nvme_ctrl_pp_status(ctrl)) {
4253                 if (time_after(jiffies, fw_act_timeout)) {
4254                         dev_warn(ctrl->device,
4255                                 "Fw activation timeout, reset controller\n");
4256                         nvme_try_sched_reset(ctrl);
4257                         return;
4258                 }
4259                 msleep(100);
4260         }
4261
4262         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4263                 return;
4264
4265         nvme_start_queues(ctrl);
4266         /* read FW slot information to clear the AER */
4267         nvme_get_fw_slot_info(ctrl);
4268 }
4269
4270 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4271 {
4272         u32 aer_notice_type = (result & 0xff00) >> 8;
4273
4274         trace_nvme_async_event(ctrl, aer_notice_type);
4275
4276         switch (aer_notice_type) {
4277         case NVME_AER_NOTICE_NS_CHANGED:
4278                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4279                 nvme_queue_scan(ctrl);
4280                 break;
4281         case NVME_AER_NOTICE_FW_ACT_STARTING:
4282                 /*
4283                  * We are (ab)using the RESETTING state to prevent subsequent
4284                  * recovery actions from interfering with the controller's
4285                  * firmware activation.
4286                  */
4287                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4288                         queue_work(nvme_wq, &ctrl->fw_act_work);
4289                 break;
4290 #ifdef CONFIG_NVME_MULTIPATH
4291         case NVME_AER_NOTICE_ANA:
4292                 if (!ctrl->ana_log_buf)
4293                         break;
4294                 queue_work(nvme_wq, &ctrl->ana_work);
4295                 break;
4296 #endif
4297         case NVME_AER_NOTICE_DISC_CHANGED:
4298                 ctrl->aen_result = result;
4299                 break;
4300         default:
4301                 dev_warn(ctrl->device, "async event result %08x\n", result);
4302         }
4303 }
4304
4305 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4306                 volatile union nvme_result *res)
4307 {
4308         u32 result = le32_to_cpu(res->u32);
4309         u32 aer_type = result & 0x07;
4310
4311         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4312                 return;
4313
4314         switch (aer_type) {
4315         case NVME_AER_NOTICE:
4316                 nvme_handle_aen_notice(ctrl, result);
4317                 break;
4318         case NVME_AER_ERROR:
4319         case NVME_AER_SMART:
4320         case NVME_AER_CSS:
4321         case NVME_AER_VS:
4322                 trace_nvme_async_event(ctrl, aer_type);
4323                 ctrl->aen_result = result;
4324                 break;
4325         default:
4326                 break;
4327         }
4328         queue_work(nvme_wq, &ctrl->async_event_work);
4329 }
4330 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4331
4332 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4333 {
4334         nvme_mpath_stop(ctrl);
4335         nvme_stop_keep_alive(ctrl);
4336         flush_work(&ctrl->async_event_work);
4337         cancel_work_sync(&ctrl->fw_act_work);
4338 }
4339 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4340
4341 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4342 {
4343         nvme_start_keep_alive(ctrl);
4344
4345         nvme_enable_aen(ctrl);
4346
4347         if (ctrl->queue_count > 1) {
4348                 nvme_queue_scan(ctrl);
4349                 nvme_start_queues(ctrl);
4350         }
4351 }
4352 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4353
4354 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4355 {
4356         nvme_fault_inject_fini(&ctrl->fault_inject);
4357         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4358         cdev_device_del(&ctrl->cdev, ctrl->device);
4359         nvme_put_ctrl(ctrl);
4360 }
4361 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4362
4363 static void nvme_free_ctrl(struct device *dev)
4364 {
4365         struct nvme_ctrl *ctrl =
4366                 container_of(dev, struct nvme_ctrl, ctrl_device);
4367         struct nvme_subsystem *subsys = ctrl->subsys;
4368
4369         if (!subsys || ctrl->instance != subsys->instance)
4370                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4371
4372         xa_destroy(&ctrl->cels);
4373
4374         nvme_mpath_uninit(ctrl);
4375         __free_page(ctrl->discard_page);
4376
4377         if (subsys) {
4378                 mutex_lock(&nvme_subsystems_lock);
4379                 list_del(&ctrl->subsys_entry);
4380                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4381                 mutex_unlock(&nvme_subsystems_lock);
4382         }
4383
4384         ctrl->ops->free_ctrl(ctrl);
4385
4386         if (subsys)
4387                 nvme_put_subsystem(subsys);
4388 }
4389
4390 /*
4391  * Initialize a NVMe controller structures.  This needs to be called during
4392  * earliest initialization so that we have the initialized structured around
4393  * during probing.
4394  */
4395 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4396                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4397 {
4398         int ret;
4399
4400         ctrl->state = NVME_CTRL_NEW;
4401         spin_lock_init(&ctrl->lock);
4402         mutex_init(&ctrl->scan_lock);
4403         INIT_LIST_HEAD(&ctrl->namespaces);
4404         xa_init(&ctrl->cels);
4405         init_rwsem(&ctrl->namespaces_rwsem);
4406         ctrl->dev = dev;
4407         ctrl->ops = ops;
4408         ctrl->quirks = quirks;
4409         ctrl->numa_node = NUMA_NO_NODE;
4410         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4411         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4412         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4413         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4414         init_waitqueue_head(&ctrl->state_wq);
4415
4416         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4417         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4418         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4419
4420         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4421                         PAGE_SIZE);
4422         ctrl->discard_page = alloc_page(GFP_KERNEL);
4423         if (!ctrl->discard_page) {
4424                 ret = -ENOMEM;
4425                 goto out;
4426         }
4427
4428         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4429         if (ret < 0)
4430                 goto out;
4431         ctrl->instance = ret;
4432
4433         device_initialize(&ctrl->ctrl_device);
4434         ctrl->device = &ctrl->ctrl_device;
4435         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4436         ctrl->device->class = nvme_class;
4437         ctrl->device->parent = ctrl->dev;
4438         ctrl->device->groups = nvme_dev_attr_groups;
4439         ctrl->device->release = nvme_free_ctrl;
4440         dev_set_drvdata(ctrl->device, ctrl);
4441         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4442         if (ret)
4443                 goto out_release_instance;
4444
4445         nvme_get_ctrl(ctrl);
4446         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4447         ctrl->cdev.owner = ops->module;
4448         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4449         if (ret)
4450                 goto out_free_name;
4451
4452         /*
4453          * Initialize latency tolerance controls.  The sysfs files won't
4454          * be visible to userspace unless the device actually supports APST.
4455          */
4456         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4457         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4458                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4459
4460         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4461
4462         return 0;
4463 out_free_name:
4464         nvme_put_ctrl(ctrl);
4465         kfree_const(ctrl->device->kobj.name);
4466 out_release_instance:
4467         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4468 out:
4469         if (ctrl->discard_page)
4470                 __free_page(ctrl->discard_page);
4471         return ret;
4472 }
4473 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4474
4475 /**
4476  * nvme_kill_queues(): Ends all namespace queues
4477  * @ctrl: the dead controller that needs to end
4478  *
4479  * Call this function when the driver determines it is unable to get the
4480  * controller in a state capable of servicing IO.
4481  */
4482 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4483 {
4484         struct nvme_ns *ns;
4485
4486         down_read(&ctrl->namespaces_rwsem);
4487
4488         /* Forcibly unquiesce queues to avoid blocking dispatch */
4489         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4490                 blk_mq_unquiesce_queue(ctrl->admin_q);
4491
4492         list_for_each_entry(ns, &ctrl->namespaces, list)
4493                 nvme_set_queue_dying(ns);
4494
4495         up_read(&ctrl->namespaces_rwsem);
4496 }
4497 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4498
4499 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4500 {
4501         struct nvme_ns *ns;
4502
4503         down_read(&ctrl->namespaces_rwsem);
4504         list_for_each_entry(ns, &ctrl->namespaces, list)
4505                 blk_mq_unfreeze_queue(ns->queue);
4506         up_read(&ctrl->namespaces_rwsem);
4507 }
4508 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4509
4510 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4511 {
4512         struct nvme_ns *ns;
4513
4514         down_read(&ctrl->namespaces_rwsem);
4515         list_for_each_entry(ns, &ctrl->namespaces, list) {
4516                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4517                 if (timeout <= 0)
4518                         break;
4519         }
4520         up_read(&ctrl->namespaces_rwsem);
4521         return timeout;
4522 }
4523 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4524
4525 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4526 {
4527         struct nvme_ns *ns;
4528
4529         down_read(&ctrl->namespaces_rwsem);
4530         list_for_each_entry(ns, &ctrl->namespaces, list)
4531                 blk_mq_freeze_queue_wait(ns->queue);
4532         up_read(&ctrl->namespaces_rwsem);
4533 }
4534 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4535
4536 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4537 {
4538         struct nvme_ns *ns;
4539
4540         down_read(&ctrl->namespaces_rwsem);
4541         list_for_each_entry(ns, &ctrl->namespaces, list)
4542                 blk_freeze_queue_start(ns->queue);
4543         up_read(&ctrl->namespaces_rwsem);
4544 }
4545 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4546
4547 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4548 {
4549         struct nvme_ns *ns;
4550
4551         down_read(&ctrl->namespaces_rwsem);
4552         list_for_each_entry(ns, &ctrl->namespaces, list)
4553                 blk_mq_quiesce_queue(ns->queue);
4554         up_read(&ctrl->namespaces_rwsem);
4555 }
4556 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4557
4558 void nvme_start_queues(struct nvme_ctrl *ctrl)
4559 {
4560         struct nvme_ns *ns;
4561
4562         down_read(&ctrl->namespaces_rwsem);
4563         list_for_each_entry(ns, &ctrl->namespaces, list)
4564                 blk_mq_unquiesce_queue(ns->queue);
4565         up_read(&ctrl->namespaces_rwsem);
4566 }
4567 EXPORT_SYMBOL_GPL(nvme_start_queues);
4568
4569 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4570 {
4571         struct nvme_ns *ns;
4572
4573         down_read(&ctrl->namespaces_rwsem);
4574         list_for_each_entry(ns, &ctrl->namespaces, list)
4575                 blk_sync_queue(ns->queue);
4576         up_read(&ctrl->namespaces_rwsem);
4577 }
4578 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4579
4580 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4581 {
4582         nvme_sync_io_queues(ctrl);
4583         if (ctrl->admin_q)
4584                 blk_sync_queue(ctrl->admin_q);
4585 }
4586 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4587
4588 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4589 {
4590         if (file->f_op != &nvme_dev_fops)
4591                 return NULL;
4592         return file->private_data;
4593 }
4594 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4595
4596 /*
4597  * Check we didn't inadvertently grow the command structure sizes:
4598  */
4599 static inline void _nvme_check_size(void)
4600 {
4601         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4602         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4603         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4604         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4605         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4606         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4607         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4608         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4609         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4610         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4611         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4612         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4613         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4614         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4615         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4616         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4617         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4618         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4619         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4620 }
4621
4622
4623 static int __init nvme_core_init(void)
4624 {
4625         int result = -ENOMEM;
4626
4627         _nvme_check_size();
4628
4629         nvme_wq = alloc_workqueue("nvme-wq",
4630                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4631         if (!nvme_wq)
4632                 goto out;
4633
4634         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4635                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4636         if (!nvme_reset_wq)
4637                 goto destroy_wq;
4638
4639         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4640                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4641         if (!nvme_delete_wq)
4642                 goto destroy_reset_wq;
4643
4644         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4645         if (result < 0)
4646                 goto destroy_delete_wq;
4647
4648         nvme_class = class_create(THIS_MODULE, "nvme");
4649         if (IS_ERR(nvme_class)) {
4650                 result = PTR_ERR(nvme_class);
4651                 goto unregister_chrdev;
4652         }
4653         nvme_class->dev_uevent = nvme_class_uevent;
4654
4655         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4656         if (IS_ERR(nvme_subsys_class)) {
4657                 result = PTR_ERR(nvme_subsys_class);
4658                 goto destroy_class;
4659         }
4660         return 0;
4661
4662 destroy_class:
4663         class_destroy(nvme_class);
4664 unregister_chrdev:
4665         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4666 destroy_delete_wq:
4667         destroy_workqueue(nvme_delete_wq);
4668 destroy_reset_wq:
4669         destroy_workqueue(nvme_reset_wq);
4670 destroy_wq:
4671         destroy_workqueue(nvme_wq);
4672 out:
4673         return result;
4674 }
4675
4676 static void __exit nvme_core_exit(void)
4677 {
4678         class_destroy(nvme_subsys_class);
4679         class_destroy(nvme_class);
4680         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4681         destroy_workqueue(nvme_delete_wq);
4682         destroy_workqueue(nvme_reset_wq);
4683         destroy_workqueue(nvme_wq);
4684         ida_destroy(&nvme_instance_ida);
4685 }
4686
4687 MODULE_LICENSE("GPL");
4688 MODULE_VERSION("1.0");
4689 module_init(nvme_core_init);
4690 module_exit(nvme_core_exit);