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_ctrl_base_chr_devt;
  89 static struct class *nvme_class;
  90 static struct class *nvme_subsys_class;
  91
  92 static DEFINE_IDA(nvme_ns_chr_minor_ida);
  93 static dev_t nvme_ns_chr_devt;
  94 static struct class *nvme_ns_chr_class;
  95
  96 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  97 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  98                                            unsigned nsid);
  99
 100 /*
 101  * Prepare a queue for teardown.
 102  *
 103  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
 104  * the capacity to 0 after that to avoid blocking dispatchers that may be
 105  * holding bd_butex.  This will end buffered writers dirtying pages that can't
 106  * be synced.
 107  */
 108 static void nvme_set_queue_dying(struct nvme_ns *ns)
 109 {
 110         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 111                 return;
 112
 113         blk_set_queue_dying(ns->queue);
 114         blk_mq_unquiesce_queue(ns->queue);
 115
 116         set_capacity_and_notify(ns->disk, 0);
 117 }
 118
 119 void nvme_queue_scan(struct nvme_ctrl *ctrl)
 120 {
 121         /*
 122          * Only new queue scan work when admin and IO queues are both alive
 123          */
 124         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 125                 queue_work(nvme_wq, &ctrl->scan_work);
 126 }
 127
 128 /*
 129  * Use this function to proceed with scheduling reset_work for a controller
 130  * that had previously been set to the resetting state. This is intended for
 131  * code paths that can't be interrupted by other reset attempts. A hot removal
 132  * may prevent this from succeeding.
 133  */
 134 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 135 {
 136         if (ctrl->state != NVME_CTRL_RESETTING)
 137                 return -EBUSY;
 138         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 139                 return -EBUSY;
 140         return 0;
 141 }
 142 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 143
 144 static void nvme_failfast_work(struct work_struct *work)
 145 {
 146         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 147                         struct nvme_ctrl, failfast_work);
 148
 149         if (ctrl->state != NVME_CTRL_CONNECTING)
 150                 return;
 151
 152         set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 153         dev_info(ctrl->device, "failfast expired\n");
 154         nvme_kick_requeue_lists(ctrl);
 155 }
 156
 157 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 158 {
 159         if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 160                 return;
 161
 162         schedule_delayed_work(&ctrl->failfast_work,
 163                               ctrl->opts->fast_io_fail_tmo * HZ);
 164 }
 165
 166 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 167 {
 168         if (!ctrl->opts)
 169                 return;
 170
 171         cancel_delayed_work_sync(&ctrl->failfast_work);
 172         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 173 }
 174
 175
 176 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 177 {
 178         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 179                 return -EBUSY;
 180         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 181                 return -EBUSY;
 182         return 0;
 183 }
 184 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 185
 186 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 187 {
 188         int ret;
 189
 190         ret = nvme_reset_ctrl(ctrl);
 191         if (!ret) {
 192                 flush_work(&ctrl->reset_work);
 193                 if (ctrl->state != NVME_CTRL_LIVE)
 194                         ret = -ENETRESET;
 195         }
 196
 197         return ret;
 198 }
 199
 200 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 201 {
 202         dev_info(ctrl->device,
 203                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 204
 205         flush_work(&ctrl->reset_work);
 206         nvme_stop_ctrl(ctrl);
 207         nvme_remove_namespaces(ctrl);
 208         ctrl->ops->delete_ctrl(ctrl);
 209         nvme_uninit_ctrl(ctrl);
 210 }
 211
 212 static void nvme_delete_ctrl_work(struct work_struct *work)
 213 {
 214         struct nvme_ctrl *ctrl =
 215                 container_of(work, struct nvme_ctrl, delete_work);
 216
 217         nvme_do_delete_ctrl(ctrl);
 218 }
 219
 220 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 221 {
 222         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 223                 return -EBUSY;
 224         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 225                 return -EBUSY;
 226         return 0;
 227 }
 228 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 229
 230 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 231 {
 232         /*
 233          * Keep a reference until nvme_do_delete_ctrl() complete,
 234          * since ->delete_ctrl can free the controller.
 235          */
 236         nvme_get_ctrl(ctrl);
 237         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 238                 nvme_do_delete_ctrl(ctrl);
 239         nvme_put_ctrl(ctrl);
 240 }
 241
 242 static blk_status_t nvme_error_status(u16 status)
 243 {
 244         switch (status & 0x7ff) {
 245         case NVME_SC_SUCCESS:
 246                 return BLK_STS_OK;
 247         case NVME_SC_CAP_EXCEEDED:
 248                 return BLK_STS_NOSPC;
 249         case NVME_SC_LBA_RANGE:
 250         case NVME_SC_CMD_INTERRUPTED:
 251         case NVME_SC_NS_NOT_READY:
 252                 return BLK_STS_TARGET;
 253         case NVME_SC_BAD_ATTRIBUTES:
 254         case NVME_SC_ONCS_NOT_SUPPORTED:
 255         case NVME_SC_INVALID_OPCODE:
 256         case NVME_SC_INVALID_FIELD:
 257         case NVME_SC_INVALID_NS:
 258                 return BLK_STS_NOTSUPP;
 259         case NVME_SC_WRITE_FAULT:
 260         case NVME_SC_READ_ERROR:
 261         case NVME_SC_UNWRITTEN_BLOCK:
 262         case NVME_SC_ACCESS_DENIED:
 263         case NVME_SC_READ_ONLY:
 264         case NVME_SC_COMPARE_FAILED:
 265                 return BLK_STS_MEDIUM;
 266         case NVME_SC_GUARD_CHECK:
 267         case NVME_SC_APPTAG_CHECK:
 268         case NVME_SC_REFTAG_CHECK:
 269         case NVME_SC_INVALID_PI:
 270                 return BLK_STS_PROTECTION;
 271         case NVME_SC_RESERVATION_CONFLICT:
 272                 return BLK_STS_NEXUS;
 273         case NVME_SC_HOST_PATH_ERROR:
 274                 return BLK_STS_TRANSPORT;
 275         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 276                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 277         case NVME_SC_ZONE_TOO_MANY_OPEN:
 278                 return BLK_STS_ZONE_OPEN_RESOURCE;
 279         default:
 280                 return BLK_STS_IOERR;
 281         }
 282 }
 283
 284 static void nvme_retry_req(struct request *req)
 285 {
 286         unsigned long delay = 0;
 287         u16 crd;
 288
 289         /* The mask and shift result must be <= 3 */
 290         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 291         if (crd)
 292                 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 293
 294         nvme_req(req)->retries++;
 295         blk_mq_requeue_request(req, false);
 296         blk_mq_delay_kick_requeue_list(req->q, delay);
 297 }
 298
 299 enum nvme_disposition {
 300         COMPLETE,
 301         RETRY,
 302         FAILOVER,
 303 };
 304
 305 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 306 {
 307         if (likely(nvme_req(req)->status == 0))
 308                 return COMPLETE;
 309
 310         if (blk_noretry_request(req) ||
 311             (nvme_req(req)->status & NVME_SC_DNR) ||
 312             nvme_req(req)->retries >= nvme_max_retries)
 313                 return COMPLETE;
 314
 315         if (req->cmd_flags & REQ_NVME_MPATH) {
 316                 if (nvme_is_path_error(nvme_req(req)->status) ||
 317                     blk_queue_dying(req->q))
 318                         return FAILOVER;
 319         } else {
 320                 if (blk_queue_dying(req->q))
 321                         return COMPLETE;
 322         }
 323
 324         return RETRY;
 325 }
 326
 327 static inline void nvme_end_req(struct request *req)
 328 {
 329         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 330
 331         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 332             req_op(req) == REQ_OP_ZONE_APPEND)
 333                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 334                         le64_to_cpu(nvme_req(req)->result.u64));
 335
 336         nvme_trace_bio_complete(req);
 337         blk_mq_end_request(req, status);
 338 }
 339
 340 void nvme_complete_rq(struct request *req)
 341 {
 342         trace_nvme_complete_rq(req);
 343         nvme_cleanup_cmd(req);
 344
 345         if (nvme_req(req)->ctrl->kas)
 346                 nvme_req(req)->ctrl->comp_seen = true;
 347
 348         switch (nvme_decide_disposition(req)) {
 349         case COMPLETE:
 350                 nvme_end_req(req);
 351                 return;
 352         case RETRY:
 353                 nvme_retry_req(req);
 354                 return;
 355         case FAILOVER:
 356                 nvme_failover_req(req);
 357                 return;
 358         }
 359 }
 360 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 361
 362 /*
 363  * Called to unwind from ->queue_rq on a failed command submission so that the
 364  * multipathing code gets called to potentially failover to another path.
 365  * The caller needs to unwind all transport specific resource allocations and
 366  * must return propagate the return value.
 367  */
 368 blk_status_t nvme_host_path_error(struct request *req)
 369 {
 370         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 371         blk_mq_set_request_complete(req);
 372         nvme_complete_rq(req);
 373         return BLK_STS_OK;
 374 }
 375 EXPORT_SYMBOL_GPL(nvme_host_path_error);
 376
 377 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 378 {
 379         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 380                                 "Cancelling I/O %d", req->tag);
 381
 382         /* don't abort one completed request */
 383         if (blk_mq_request_completed(req))
 384                 return true;
 385
 386         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 387         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 388         blk_mq_complete_request(req);
 389         return true;
 390 }
 391 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 392
 393 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 394 {
 395         if (ctrl->tagset) {
 396                 blk_mq_tagset_busy_iter(ctrl->tagset,
 397                                 nvme_cancel_request, ctrl);
 398                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 399         }
 400 }
 401 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 402
 403 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 404 {
 405         if (ctrl->admin_tagset) {
 406                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 407                                 nvme_cancel_request, ctrl);
 408                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 409         }
 410 }
 411 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 412
 413 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 414                 enum nvme_ctrl_state new_state)
 415 {
 416         enum nvme_ctrl_state old_state;
 417         unsigned long flags;
 418         bool changed = false;
 419
 420         spin_lock_irqsave(&ctrl->lock, flags);
 421
 422         old_state = ctrl->state;
 423         switch (new_state) {
 424         case NVME_CTRL_LIVE:
 425                 switch (old_state) {
 426                 case NVME_CTRL_NEW:
 427                 case NVME_CTRL_RESETTING:
 428                 case NVME_CTRL_CONNECTING:
 429                         changed = true;
 430                         fallthrough;
 431                 default:
 432                         break;
 433                 }
 434                 break;
 435         case NVME_CTRL_RESETTING:
 436                 switch (old_state) {
 437                 case NVME_CTRL_NEW:
 438                 case NVME_CTRL_LIVE:
 439                         changed = true;
 440                         fallthrough;
 441                 default:
 442                         break;
 443                 }
 444                 break;
 445         case NVME_CTRL_CONNECTING:
 446                 switch (old_state) {
 447                 case NVME_CTRL_NEW:
 448                 case NVME_CTRL_RESETTING:
 449                         changed = true;
 450                         fallthrough;
 451                 default:
 452                         break;
 453                 }
 454                 break;
 455         case NVME_CTRL_DELETING:
 456                 switch (old_state) {
 457                 case NVME_CTRL_LIVE:
 458                 case NVME_CTRL_RESETTING:
 459                 case NVME_CTRL_CONNECTING:
 460                         changed = true;
 461                         fallthrough;
 462                 default:
 463                         break;
 464                 }
 465                 break;
 466         case NVME_CTRL_DELETING_NOIO:
 467                 switch (old_state) {
 468                 case NVME_CTRL_DELETING:
 469                 case NVME_CTRL_DEAD:
 470                         changed = true;
 471                         fallthrough;
 472                 default:
 473                         break;
 474                 }
 475                 break;
 476         case NVME_CTRL_DEAD:
 477                 switch (old_state) {
 478                 case NVME_CTRL_DELETING:
 479                         changed = true;
 480                         fallthrough;
 481                 default:
 482                         break;
 483                 }
 484                 break;
 485         default:
 486                 break;
 487         }
 488
 489         if (changed) {
 490                 ctrl->state = new_state;
 491                 wake_up_all(&ctrl->state_wq);
 492         }
 493
 494         spin_unlock_irqrestore(&ctrl->lock, flags);
 495         if (!changed)
 496                 return false;
 497
 498         if (ctrl->state == NVME_CTRL_LIVE) {
 499                 if (old_state == NVME_CTRL_CONNECTING)
 500                         nvme_stop_failfast_work(ctrl);
 501                 nvme_kick_requeue_lists(ctrl);
 502         } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 503                 old_state == NVME_CTRL_RESETTING) {
 504                 nvme_start_failfast_work(ctrl);
 505         }
 506         return changed;
 507 }
 508 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 509
 510 /*
 511  * Returns true for sink states that can't ever transition back to live.
 512  */
 513 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 514 {
 515         switch (ctrl->state) {
 516         case NVME_CTRL_NEW:
 517         case NVME_CTRL_LIVE:
 518         case NVME_CTRL_RESETTING:
 519         case NVME_CTRL_CONNECTING:
 520                 return false;
 521         case NVME_CTRL_DELETING:
 522         case NVME_CTRL_DELETING_NOIO:
 523         case NVME_CTRL_DEAD:
 524                 return true;
 525         default:
 526                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 527                 return true;
 528         }
 529 }
 530
 531 /*
 532  * Waits for the controller state to be resetting, or returns false if it is
 533  * not possible to ever transition to that state.
 534  */
 535 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 536 {
 537         wait_event(ctrl->state_wq,
 538                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 539                    nvme_state_terminal(ctrl));
 540         return ctrl->state == NVME_CTRL_RESETTING;
 541 }
 542 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 543
 544 static void nvme_free_ns_head(struct kref *ref)
 545 {
 546         struct nvme_ns_head *head =
 547                 container_of(ref, struct nvme_ns_head, ref);
 548
 549         nvme_mpath_remove_disk(head);
 550         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 551         cleanup_srcu_struct(&head->srcu);
 552         nvme_put_subsystem(head->subsys);
 553         kfree(head);
 554 }
 555
 556 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
 557 {
 558         return kref_get_unless_zero(&head->ref);
 559 }
 560
 561 void nvme_put_ns_head(struct nvme_ns_head *head)
 562 {
 563         kref_put(&head->ref, nvme_free_ns_head);
 564 }
 565
 566 static void nvme_free_ns(struct kref *kref)
 567 {
 568         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 569
 570         if (ns->ndev)
 571                 nvme_nvm_unregister(ns);
 572
 573         put_disk(ns->disk);
 574         nvme_put_ns_head(ns->head);
 575         nvme_put_ctrl(ns->ctrl);
 576         kfree(ns);
 577 }
 578
 579 void nvme_put_ns(struct nvme_ns *ns)
 580 {
 581         kref_put(&ns->kref, nvme_free_ns);
 582 }
 583 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 584
 585 static inline void nvme_clear_nvme_request(struct request *req)
 586 {
 587         struct nvme_command *cmd = nvme_req(req)->cmd;
 588
 589         memset(cmd, 0, sizeof(*cmd));
 590         nvme_req(req)->retries = 0;
 591         nvme_req(req)->flags = 0;
 592         req->rq_flags |= RQF_DONTPREP;
 593 }
 594
 595 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
 596 {
 597         return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 598 }
 599
 600 static inline void nvme_init_request(struct request *req,
 601                 struct nvme_command *cmd)
 602 {
 603         if (req->q->queuedata)
 604                 req->timeout = NVME_IO_TIMEOUT;
 605         else /* no queuedata implies admin queue */
 606                 req->timeout = NVME_ADMIN_TIMEOUT;
 607
 608         /* passthru commands should let the driver set the SGL flags */
 609         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 610
 611         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 612         nvme_clear_nvme_request(req);
 613         memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
 614 }
 615
 616 struct request *nvme_alloc_request(struct request_queue *q,
 617                 struct nvme_command *cmd, blk_mq_req_flags_t flags)
 618 {
 619         struct request *req;
 620
 621         req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
 622         if (!IS_ERR(req))
 623                 nvme_init_request(req, cmd);
 624         return req;
 625 }
 626 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 627
 628 static struct request *nvme_alloc_request_qid(struct request_queue *q,
 629                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 630 {
 631         struct request *req;
 632
 633         req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
 634                         qid ? qid - 1 : 0);
 635         if (!IS_ERR(req))
 636                 nvme_init_request(req, cmd);
 637         return req;
 638 }
 639
 640 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 641 {
 642         struct nvme_command c;
 643
 644         memset(&c, 0, sizeof(c));
 645
 646         c.directive.opcode = nvme_admin_directive_send;
 647         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 648         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 649         c.directive.dtype = NVME_DIR_IDENTIFY;
 650         c.directive.tdtype = NVME_DIR_STREAMS;
 651         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 652
 653         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 654 }
 655
 656 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 657 {
 658         return nvme_toggle_streams(ctrl, false);
 659 }
 660
 661 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 662 {
 663         return nvme_toggle_streams(ctrl, true);
 664 }
 665
 666 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 667                                   struct streams_directive_params *s, u32 nsid)
 668 {
 669         struct nvme_command c;
 670
 671         memset(&c, 0, sizeof(c));
 672         memset(s, 0, sizeof(*s));
 673
 674         c.directive.opcode = nvme_admin_directive_recv;
 675         c.directive.nsid = cpu_to_le32(nsid);
 676         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 677         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 678         c.directive.dtype = NVME_DIR_STREAMS;
 679
 680         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 681 }
 682
 683 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 684 {
 685         struct streams_directive_params s;
 686         int ret;
 687
 688         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 689                 return 0;
 690         if (!streams)
 691                 return 0;
 692
 693         ret = nvme_enable_streams(ctrl);
 694         if (ret)
 695                 return ret;
 696
 697         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 698         if (ret)
 699                 goto out_disable_stream;
 700
 701         ctrl->nssa = le16_to_cpu(s.nssa);
 702         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 703                 dev_info(ctrl->device, "too few streams (%u) available\n",
 704                                         ctrl->nssa);
 705                 goto out_disable_stream;
 706         }
 707
 708         ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 709         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 710         return 0;
 711
 712 out_disable_stream:
 713         nvme_disable_streams(ctrl);
 714         return ret;
 715 }
 716
 717 /*
 718  * Check if 'req' has a write hint associated with it. If it does, assign
 719  * a valid namespace stream to the write.
 720  */
 721 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 722                                      struct request *req, u16 *control,
 723                                      u32 *dsmgmt)
 724 {
 725         enum rw_hint streamid = req->write_hint;
 726
 727         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 728                 streamid = 0;
 729         else {
 730                 streamid--;
 731                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 732                         return;
 733
 734                 *control |= NVME_RW_DTYPE_STREAMS;
 735                 *dsmgmt |= streamid << 16;
 736         }
 737
 738         if (streamid < ARRAY_SIZE(req->q->write_hints))
 739                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 740 }
 741
 742 static inline void nvme_setup_flush(struct nvme_ns *ns,
 743                 struct nvme_command *cmnd)
 744 {
 745         cmnd->common.opcode = nvme_cmd_flush;
 746         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 747 }
 748
 749 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 750                 struct nvme_command *cmnd)
 751 {
 752         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 753         struct nvme_dsm_range *range;
 754         struct bio *bio;
 755
 756         /*
 757          * Some devices do not consider the DSM 'Number of Ranges' field when
 758          * determining how much data to DMA. Always allocate memory for maximum
 759          * number of segments to prevent device reading beyond end of buffer.
 760          */
 761         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 762
 763         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 764         if (!range) {
 765                 /*
 766                  * If we fail allocation our range, fallback to the controller
 767                  * discard page. If that's also busy, it's safe to return
 768                  * busy, as we know we can make progress once that's freed.
 769                  */
 770                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 771                         return BLK_STS_RESOURCE;
 772
 773                 range = page_address(ns->ctrl->discard_page);
 774         }
 775
 776         __rq_for_each_bio(bio, req) {
 777                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 778                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 779
 780                 if (n < segments) {
 781                         range[n].cattr = cpu_to_le32(0);
 782                         range[n].nlb = cpu_to_le32(nlb);
 783                         range[n].slba = cpu_to_le64(slba);
 784                 }
 785                 n++;
 786         }
 787
 788         if (WARN_ON_ONCE(n != segments)) {
 789                 if (virt_to_page(range) == ns->ctrl->discard_page)
 790                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 791                 else
 792                         kfree(range);
 793                 return BLK_STS_IOERR;
 794         }
 795
 796         cmnd->dsm.opcode = nvme_cmd_dsm;
 797         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 798         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 799         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 800
 801         req->special_vec.bv_page = virt_to_page(range);
 802         req->special_vec.bv_offset = offset_in_page(range);
 803         req->special_vec.bv_len = alloc_size;
 804         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 805
 806         return BLK_STS_OK;
 807 }
 808
 809 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 810                 struct request *req, struct nvme_command *cmnd)
 811 {
 812         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 813                 return nvme_setup_discard(ns, req, cmnd);
 814
 815         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 816         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 817         cmnd->write_zeroes.slba =
 818                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 819         cmnd->write_zeroes.length =
 820                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 821         cmnd->write_zeroes.control = 0;
 822         return BLK_STS_OK;
 823 }
 824
 825 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 826                 struct request *req, struct nvme_command *cmnd,
 827                 enum nvme_opcode op)
 828 {
 829         struct nvme_ctrl *ctrl = ns->ctrl;
 830         u16 control = 0;
 831         u32 dsmgmt = 0;
 832
 833         if (req->cmd_flags & REQ_FUA)
 834                 control |= NVME_RW_FUA;
 835         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 836                 control |= NVME_RW_LR;
 837
 838         if (req->cmd_flags & REQ_RAHEAD)
 839                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 840
 841         cmnd->rw.opcode = op;
 842         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 843         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 844         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 845
 846         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 847                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 848
 849         if (ns->ms) {
 850                 /*
 851                  * If formated with metadata, the block layer always provides a
 852                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 853                  * we enable the PRACT bit for protection information or set the
 854                  * namespace capacity to zero to prevent any I/O.
 855                  */
 856                 if (!blk_integrity_rq(req)) {
 857                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 858                                 return BLK_STS_NOTSUPP;
 859                         control |= NVME_RW_PRINFO_PRACT;
 860                 }
 861
 862                 switch (ns->pi_type) {
 863                 case NVME_NS_DPS_PI_TYPE3:
 864                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 865                         break;
 866                 case NVME_NS_DPS_PI_TYPE1:
 867                 case NVME_NS_DPS_PI_TYPE2:
 868                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 869                                         NVME_RW_PRINFO_PRCHK_REF;
 870                         if (op == nvme_cmd_zone_append)
 871                                 control |= NVME_RW_APPEND_PIREMAP;
 872                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 873                         break;
 874                 }
 875         }
 876
 877         cmnd->rw.control = cpu_to_le16(control);
 878         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 879         return 0;
 880 }
 881
 882 void nvme_cleanup_cmd(struct request *req)
 883 {
 884         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 885                 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 886                 struct page *page = req->special_vec.bv_page;
 887
 888                 if (page == ctrl->discard_page)
 889                         clear_bit_unlock(0, &ctrl->discard_page_busy);
 890                 else
 891                         kfree(page_address(page) + req->special_vec.bv_offset);
 892         }
 893 }
 894 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 895
 896 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
 897 {
 898         struct nvme_command *cmd = nvme_req(req)->cmd;
 899         blk_status_t ret = BLK_STS_OK;
 900
 901         if (!(req->rq_flags & RQF_DONTPREP))
 902                 nvme_clear_nvme_request(req);
 903
 904         switch (req_op(req)) {
 905         case REQ_OP_DRV_IN:
 906         case REQ_OP_DRV_OUT:
 907                 /* these are setup prior to execution in nvme_init_request() */
 908                 break;
 909         case REQ_OP_FLUSH:
 910                 nvme_setup_flush(ns, cmd);
 911                 break;
 912         case REQ_OP_ZONE_RESET_ALL:
 913         case REQ_OP_ZONE_RESET:
 914                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 915                 break;
 916         case REQ_OP_ZONE_OPEN:
 917                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
 918                 break;
 919         case REQ_OP_ZONE_CLOSE:
 920                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
 921                 break;
 922         case REQ_OP_ZONE_FINISH:
 923                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
 924                 break;
 925         case REQ_OP_WRITE_ZEROES:
 926                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 927                 break;
 928         case REQ_OP_DISCARD:
 929                 ret = nvme_setup_discard(ns, req, cmd);
 930                 break;
 931         case REQ_OP_READ:
 932                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
 933                 break;
 934         case REQ_OP_WRITE:
 935                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
 936                 break;
 937         case REQ_OP_ZONE_APPEND:
 938                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
 939                 break;
 940         default:
 941                 WARN_ON_ONCE(1);
 942                 return BLK_STS_IOERR;
 943         }
 944
 945         cmd->common.command_id = req->tag;
 946         trace_nvme_setup_cmd(req, cmd);
 947         return ret;
 948 }
 949 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 950
 951 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 952 {
 953         struct completion *waiting = rq->end_io_data;
 954
 955         rq->end_io_data = NULL;
 956         complete(waiting);
 957 }
 958
 959 static void nvme_execute_rq_polled(struct request_queue *q,
 960                 struct gendisk *bd_disk, struct request *rq, int at_head)
 961 {
 962         DECLARE_COMPLETION_ONSTACK(wait);
 963
 964         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 965
 966         rq->cmd_flags |= REQ_HIPRI;
 967         rq->end_io_data = &wait;
 968         blk_execute_rq_nowait(bd_disk, rq, at_head, nvme_end_sync_rq);
 969
 970         while (!completion_done(&wait)) {
 971                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 972                 cond_resched();
 973         }
 974 }
 975
 976 /*
 977  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 978  * if the result is positive, it's an NVM Express status code
 979  */
 980 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 981                 union nvme_result *result, void *buffer, unsigned bufflen,
 982                 unsigned timeout, int qid, int at_head,
 983                 blk_mq_req_flags_t flags, bool poll)
 984 {
 985         struct request *req;
 986         int ret;
 987
 988         if (qid == NVME_QID_ANY)
 989                 req = nvme_alloc_request(q, cmd, flags);
 990         else
 991                 req = nvme_alloc_request_qid(q, cmd, flags, qid);
 992         if (IS_ERR(req))
 993                 return PTR_ERR(req);
 994
 995         if (timeout)
 996                 req->timeout = timeout;
 997
 998         if (buffer && bufflen) {
 999                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1000                 if (ret)
1001                         goto out;
1002         }
1003
1004         if (poll)
1005                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
1006         else
1007                 blk_execute_rq(NULL, req, at_head);
1008         if (result)
1009                 *result = nvme_req(req)->result;
1010         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1011                 ret = -EINTR;
1012         else
1013                 ret = nvme_req(req)->status;
1014  out:
1015         blk_mq_free_request(req);
1016         return ret;
1017 }
1018 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1019
1020 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1021                 void *buffer, unsigned bufflen)
1022 {
1023         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1024                         NVME_QID_ANY, 0, 0, false);
1025 }
1026 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1027
1028 static u32 nvme_known_admin_effects(u8 opcode)
1029 {
1030         switch (opcode) {
1031         case nvme_admin_format_nvm:
1032                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1033                         NVME_CMD_EFFECTS_CSE_MASK;
1034         case nvme_admin_sanitize_nvm:
1035                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1036         default:
1037                 break;
1038         }
1039         return 0;
1040 }
1041
1042 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1043 {
1044         u32 effects = 0;
1045
1046         if (ns) {
1047                 if (ns->head->effects)
1048                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1049                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1050                         dev_warn_once(ctrl->device,
1051                                 "IO command:%02x has unhandled effects:%08x\n",
1052                                 opcode, effects);
1053                 return 0;
1054         }
1055
1056         if (ctrl->effects)
1057                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1058         effects |= nvme_known_admin_effects(opcode);
1059
1060         return effects;
1061 }
1062 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1063
1064 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1065                                u8 opcode)
1066 {
1067         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1068
1069         /*
1070          * For simplicity, IO to all namespaces is quiesced even if the command
1071          * effects say only one namespace is affected.
1072          */
1073         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1074                 mutex_lock(&ctrl->scan_lock);
1075                 mutex_lock(&ctrl->subsys->lock);
1076                 nvme_mpath_start_freeze(ctrl->subsys);
1077                 nvme_mpath_wait_freeze(ctrl->subsys);
1078                 nvme_start_freeze(ctrl);
1079                 nvme_wait_freeze(ctrl);
1080         }
1081         return effects;
1082 }
1083
1084 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1085 {
1086         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1087                 nvme_unfreeze(ctrl);
1088                 nvme_mpath_unfreeze(ctrl->subsys);
1089                 mutex_unlock(&ctrl->subsys->lock);
1090                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1091                 mutex_unlock(&ctrl->scan_lock);
1092         }
1093         if (effects & NVME_CMD_EFFECTS_CCC)
1094                 nvme_init_ctrl_finish(ctrl);
1095         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1096                 nvme_queue_scan(ctrl);
1097                 flush_work(&ctrl->scan_work);
1098         }
1099 }
1100
1101 void nvme_execute_passthru_rq(struct request *rq)
1102 {
1103         struct nvme_command *cmd = nvme_req(rq)->cmd;
1104         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1105         struct nvme_ns *ns = rq->q->queuedata;
1106         struct gendisk *disk = ns ? ns->disk : NULL;
1107         u32 effects;
1108
1109         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1110         blk_execute_rq(disk, rq, 0);
1111         if (effects) /* nothing to be done for zero cmd effects */
1112                 nvme_passthru_end(ctrl, effects);
1113 }
1114 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1115
1116 /*
1117  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1118  *
1119  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1120  *   accounting for transport roundtrip times [..].
1121  */
1122 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1123 {
1124         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1125 }
1126
1127 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1128 {
1129         struct nvme_ctrl *ctrl = rq->end_io_data;
1130         unsigned long flags;
1131         bool startka = false;
1132
1133         blk_mq_free_request(rq);
1134
1135         if (status) {
1136                 dev_err(ctrl->device,
1137                         "failed nvme_keep_alive_end_io error=%d\n",
1138                                 status);
1139                 return;
1140         }
1141
1142         ctrl->comp_seen = false;
1143         spin_lock_irqsave(&ctrl->lock, flags);
1144         if (ctrl->state == NVME_CTRL_LIVE ||
1145             ctrl->state == NVME_CTRL_CONNECTING)
1146                 startka = true;
1147         spin_unlock_irqrestore(&ctrl->lock, flags);
1148         if (startka)
1149                 nvme_queue_keep_alive_work(ctrl);
1150 }
1151
1152 static void nvme_keep_alive_work(struct work_struct *work)
1153 {
1154         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1155                         struct nvme_ctrl, ka_work);
1156         bool comp_seen = ctrl->comp_seen;
1157         struct request *rq;
1158
1159         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1160                 dev_dbg(ctrl->device,
1161                         "reschedule traffic based keep-alive timer\n");
1162                 ctrl->comp_seen = false;
1163                 nvme_queue_keep_alive_work(ctrl);
1164                 return;
1165         }
1166
1167         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1168                                 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1169         if (IS_ERR(rq)) {
1170                 /* allocation failure, reset the controller */
1171                 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1172                 nvme_reset_ctrl(ctrl);
1173                 return;
1174         }
1175
1176         rq->timeout = ctrl->kato * HZ;
1177         rq->end_io_data = ctrl;
1178         blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1179 }
1180
1181 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1182 {
1183         if (unlikely(ctrl->kato == 0))
1184                 return;
1185
1186         nvme_queue_keep_alive_work(ctrl);
1187 }
1188
1189 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1190 {
1191         if (unlikely(ctrl->kato == 0))
1192                 return;
1193
1194         cancel_delayed_work_sync(&ctrl->ka_work);
1195 }
1196 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1197
1198 /*
1199  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1200  * flag, thus sending any new CNS opcodes has a big chance of not working.
1201  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1202  * (but not for any later version).
1203  */
1204 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1205 {
1206         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1207                 return ctrl->vs < NVME_VS(1, 2, 0);
1208         return ctrl->vs < NVME_VS(1, 1, 0);
1209 }
1210
1211 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1212 {
1213         struct nvme_command c = { };
1214         int error;
1215
1216         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1217         c.identify.opcode = nvme_admin_identify;
1218         c.identify.cns = NVME_ID_CNS_CTRL;
1219
1220         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1221         if (!*id)
1222                 return -ENOMEM;
1223
1224         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1225                         sizeof(struct nvme_id_ctrl));
1226         if (error)
1227                 kfree(*id);
1228         return error;
1229 }
1230
1231 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1232 {
1233         return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1234 }
1235
1236 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1237                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1238 {
1239         const char *warn_str = "ctrl returned bogus length:";
1240         void *data = cur;
1241
1242         switch (cur->nidt) {
1243         case NVME_NIDT_EUI64:
1244                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1245                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1246                                  warn_str, cur->nidl);
1247                         return -1;
1248                 }
1249                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1250                 return NVME_NIDT_EUI64_LEN;
1251         case NVME_NIDT_NGUID:
1252                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1253                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1254                                  warn_str, cur->nidl);
1255                         return -1;
1256                 }
1257                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1258                 return NVME_NIDT_NGUID_LEN;
1259         case NVME_NIDT_UUID:
1260                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1261                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1262                                  warn_str, cur->nidl);
1263                         return -1;
1264                 }
1265                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1266                 return NVME_NIDT_UUID_LEN;
1267         case NVME_NIDT_CSI:
1268                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1269                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1270                                  warn_str, cur->nidl);
1271                         return -1;
1272                 }
1273                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1274                 *csi_seen = true;
1275                 return NVME_NIDT_CSI_LEN;
1276         default:
1277                 /* Skip unknown types */
1278                 return cur->nidl;
1279         }
1280 }
1281
1282 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1283                 struct nvme_ns_ids *ids)
1284 {
1285         struct nvme_command c = { };
1286         bool csi_seen = false;
1287         int status, pos, len;
1288         void *data;
1289
1290         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1291                 return 0;
1292         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1293                 return 0;
1294
1295         c.identify.opcode = nvme_admin_identify;
1296         c.identify.nsid = cpu_to_le32(nsid);
1297         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1298
1299         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1300         if (!data)
1301                 return -ENOMEM;
1302
1303         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1304                                       NVME_IDENTIFY_DATA_SIZE);
1305         if (status) {
1306                 dev_warn(ctrl->device,
1307                         "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1308                         nsid, status);
1309                 goto free_data;
1310         }
1311
1312         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1313                 struct nvme_ns_id_desc *cur = data + pos;
1314
1315                 if (cur->nidl == 0)
1316                         break;
1317
1318                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1319                 if (len < 0)
1320                         break;
1321
1322                 len += sizeof(*cur);
1323         }
1324
1325         if (nvme_multi_css(ctrl) && !csi_seen) {
1326                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1327                          nsid);
1328                 status = -EINVAL;
1329         }
1330
1331 free_data:
1332         kfree(data);
1333         return status;
1334 }
1335
1336 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1337                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1338 {
1339         struct nvme_command c = { };
1340         int error;
1341
1342         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1343         c.identify.opcode = nvme_admin_identify;
1344         c.identify.nsid = cpu_to_le32(nsid);
1345         c.identify.cns = NVME_ID_CNS_NS;
1346
1347         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1348         if (!*id)
1349                 return -ENOMEM;
1350
1351         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1352         if (error) {
1353                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1354                 goto out_free_id;
1355         }
1356
1357         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1358         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1359                 goto out_free_id;
1360
1361         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1362             !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1363                 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1364         if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1365             !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1366                 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1367
1368         return 0;
1369
1370 out_free_id:
1371         kfree(*id);
1372         return error;
1373 }
1374
1375 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1376                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1377 {
1378         union nvme_result res = { 0 };
1379         struct nvme_command c;
1380         int ret;
1381
1382         memset(&c, 0, sizeof(c));
1383         c.features.opcode = op;
1384         c.features.fid = cpu_to_le32(fid);
1385         c.features.dword11 = cpu_to_le32(dword11);
1386
1387         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1388                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1389         if (ret >= 0 && result)
1390                 *result = le32_to_cpu(res.u32);
1391         return ret;
1392 }
1393
1394 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1395                       unsigned int dword11, void *buffer, size_t buflen,
1396                       u32 *result)
1397 {
1398         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1399                              buflen, result);
1400 }
1401 EXPORT_SYMBOL_GPL(nvme_set_features);
1402
1403 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1404                       unsigned int dword11, void *buffer, size_t buflen,
1405                       u32 *result)
1406 {
1407         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1408                              buflen, result);
1409 }
1410 EXPORT_SYMBOL_GPL(nvme_get_features);
1411
1412 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1413 {
1414         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1415         u32 result;
1416         int status, nr_io_queues;
1417
1418         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1419                         &result);
1420         if (status < 0)
1421                 return status;
1422
1423         /*
1424          * Degraded controllers might return an error when setting the queue
1425          * count.  We still want to be able to bring them online and offer
1426          * access to the admin queue, as that might be only way to fix them up.
1427          */
1428         if (status > 0) {
1429                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1430                 *count = 0;
1431         } else {
1432                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1433                 *count = min(*count, nr_io_queues);
1434         }
1435
1436         return 0;
1437 }
1438 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1439
1440 #define NVME_AEN_SUPPORTED \
1441         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1442          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1443
1444 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1445 {
1446         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1447         int status;
1448
1449         if (!supported_aens)
1450                 return;
1451
1452         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1453                         NULL, 0, &result);
1454         if (status)
1455                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1456                          supported_aens);
1457
1458         queue_work(nvme_wq, &ctrl->async_event_work);
1459 }
1460
1461 /*
1462  * Issue ioctl requests on the first available path.  Note that unlike normal
1463  * block layer requests we will not retry failed request on another controller.
1464  */
1465 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1466                 struct nvme_ns_head **head, int *srcu_idx)
1467 {
1468 #ifdef CONFIG_NVME_MULTIPATH
1469         if (disk->fops == &nvme_ns_head_ops) {
1470                 struct nvme_ns *ns;
1471
1472                 *head = disk->private_data;
1473                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1474                 ns = nvme_find_path(*head);
1475                 if (!ns)
1476                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1477                 return ns;
1478         }
1479 #endif
1480         *head = NULL;
1481         *srcu_idx = -1;
1482         return disk->private_data;
1483 }
1484
1485 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1486 {
1487         if (head)
1488                 srcu_read_unlock(&head->srcu, idx);
1489 }
1490
1491 static int nvme_ns_open(struct nvme_ns *ns)
1492 {
1493
1494         /* should never be called due to GENHD_FL_HIDDEN */
1495         if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1496                 goto fail;
1497         if (!kref_get_unless_zero(&ns->kref))
1498                 goto fail;
1499         if (!try_module_get(ns->ctrl->ops->module))
1500                 goto fail_put_ns;
1501
1502         return 0;
1503
1504 fail_put_ns:
1505         nvme_put_ns(ns);
1506 fail:
1507         return -ENXIO;
1508 }
1509
1510 static void nvme_ns_release(struct nvme_ns *ns)
1511 {
1512
1513         module_put(ns->ctrl->ops->module);
1514         nvme_put_ns(ns);
1515 }
1516
1517 static int nvme_open(struct block_device *bdev, fmode_t mode)
1518 {
1519         return nvme_ns_open(bdev->bd_disk->private_data);
1520 }
1521
1522 static void nvme_release(struct gendisk *disk, fmode_t mode)
1523 {
1524         nvme_ns_release(disk->private_data);
1525 }
1526
1527 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1528 {
1529         /* some standard values */
1530         geo->heads = 1 << 6;
1531         geo->sectors = 1 << 5;
1532         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1533         return 0;
1534 }
1535
1536 #ifdef CONFIG_BLK_DEV_INTEGRITY
1537 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1538                                 u32 max_integrity_segments)
1539 {
1540         struct blk_integrity integrity;
1541
1542         memset(&integrity, 0, sizeof(integrity));
1543         switch (pi_type) {
1544         case NVME_NS_DPS_PI_TYPE3:
1545                 integrity.profile = &t10_pi_type3_crc;
1546                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1547                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1548                 break;
1549         case NVME_NS_DPS_PI_TYPE1:
1550         case NVME_NS_DPS_PI_TYPE2:
1551                 integrity.profile = &t10_pi_type1_crc;
1552                 integrity.tag_size = sizeof(u16);
1553                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1554                 break;
1555         default:
1556                 integrity.profile = NULL;
1557                 break;
1558         }
1559         integrity.tuple_size = ms;
1560         blk_integrity_register(disk, &integrity);
1561         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1562 }
1563 #else
1564 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1565                                 u32 max_integrity_segments)
1566 {
1567 }
1568 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1569
1570 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1571 {
1572         struct nvme_ctrl *ctrl = ns->ctrl;
1573         struct request_queue *queue = disk->queue;
1574         u32 size = queue_logical_block_size(queue);
1575
1576         if (ctrl->max_discard_sectors == 0) {
1577                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1578                 return;
1579         }
1580
1581         if (ctrl->nr_streams && ns->sws && ns->sgs)
1582                 size *= ns->sws * ns->sgs;
1583
1584         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1585                         NVME_DSM_MAX_RANGES);
1586
1587         queue->limits.discard_alignment = 0;
1588         queue->limits.discard_granularity = size;
1589
1590         /* If discard is already enabled, don't reset queue limits */
1591         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1592                 return;
1593
1594         blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1595         blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1596
1597         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1598                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1599 }
1600
1601 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1602 {
1603         return !uuid_is_null(&ids->uuid) ||
1604                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1605                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1606 }
1607
1608 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1609 {
1610         return uuid_equal(&a->uuid, &b->uuid) &&
1611                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1612                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1613                 a->csi == b->csi;
1614 }
1615
1616 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1617                                  u32 *phys_bs, u32 *io_opt)
1618 {
1619         struct streams_directive_params s;
1620         int ret;
1621
1622         if (!ctrl->nr_streams)
1623                 return 0;
1624
1625         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1626         if (ret)
1627                 return ret;
1628
1629         ns->sws = le32_to_cpu(s.sws);
1630         ns->sgs = le16_to_cpu(s.sgs);
1631
1632         if (ns->sws) {
1633                 *phys_bs = ns->sws * (1 << ns->lba_shift);
1634                 if (ns->sgs)
1635                         *io_opt = *phys_bs * ns->sgs;
1636         }
1637
1638         return 0;
1639 }
1640
1641 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1642 {
1643         struct nvme_ctrl *ctrl = ns->ctrl;
1644
1645         /*
1646          * The PI implementation requires the metadata size to be equal to the
1647          * t10 pi tuple size.
1648          */
1649         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1650         if (ns->ms == sizeof(struct t10_pi_tuple))
1651                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1652         else
1653                 ns->pi_type = 0;
1654
1655         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1656         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1657                 return 0;
1658         if (ctrl->ops->flags & NVME_F_FABRICS) {
1659                 /*
1660                  * The NVMe over Fabrics specification only supports metadata as
1661                  * part of the extended data LBA.  We rely on HCA/HBA support to
1662                  * remap the separate metadata buffer from the block layer.
1663                  */
1664                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1665                         return -EINVAL;
1666                 if (ctrl->max_integrity_segments)
1667                         ns->features |=
1668                                 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1669         } else {
1670                 /*
1671                  * For PCIe controllers, we can't easily remap the separate
1672                  * metadata buffer from the block layer and thus require a
1673                  * separate metadata buffer for block layer metadata/PI support.
1674                  * We allow extended LBAs for the passthrough interface, though.
1675                  */
1676                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1677                         ns->features |= NVME_NS_EXT_LBAS;
1678                 else
1679                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1680         }
1681
1682         return 0;
1683 }
1684
1685 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1686                 struct request_queue *q)
1687 {
1688         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1689
1690         if (ctrl->max_hw_sectors) {
1691                 u32 max_segments =
1692                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1693
1694                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1695                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1696                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1697         }
1698         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1699         blk_queue_dma_alignment(q, 7);
1700         blk_queue_write_cache(q, vwc, vwc);
1701 }
1702
1703 static void nvme_update_disk_info(struct gendisk *disk,
1704                 struct nvme_ns *ns, struct nvme_id_ns *id)
1705 {
1706         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1707         unsigned short bs = 1 << ns->lba_shift;
1708         u32 atomic_bs, phys_bs, io_opt = 0;
1709
1710         /*
1711          * The block layer can't support LBA sizes larger than the page size
1712          * yet, so catch this early and don't allow block I/O.
1713          */
1714         if (ns->lba_shift > PAGE_SHIFT) {
1715                 capacity = 0;
1716                 bs = (1 << 9);
1717         }
1718
1719         blk_integrity_unregister(disk);
1720
1721         atomic_bs = phys_bs = bs;
1722         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1723         if (id->nabo == 0) {
1724                 /*
1725                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1726                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1727                  * 0 then AWUPF must be used instead.
1728                  */
1729                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1730                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1731                 else
1732                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1733         }
1734
1735         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1736                 /* NPWG = Namespace Preferred Write Granularity */
1737                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1738                 /* NOWS = Namespace Optimal Write Size */
1739                 io_opt = bs * (1 + le16_to_cpu(id->nows));
1740         }
1741
1742         blk_queue_logical_block_size(disk->queue, bs);
1743         /*
1744          * Linux filesystems assume writing a single physical block is
1745          * an atomic operation. Hence limit the physical block size to the
1746          * value of the Atomic Write Unit Power Fail parameter.
1747          */
1748         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1749         blk_queue_io_min(disk->queue, phys_bs);
1750         blk_queue_io_opt(disk->queue, io_opt);
1751
1752         /*
1753          * Register a metadata profile for PI, or the plain non-integrity NVMe
1754          * metadata masquerading as Type 0 if supported, otherwise reject block
1755          * I/O to namespaces with metadata except when the namespace supports
1756          * PI, as it can strip/insert in that case.
1757          */
1758         if (ns->ms) {
1759                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1760                     (ns->features & NVME_NS_METADATA_SUPPORTED))
1761                         nvme_init_integrity(disk, ns->ms, ns->pi_type,
1762                                             ns->ctrl->max_integrity_segments);
1763                 else if (!nvme_ns_has_pi(ns))
1764                         capacity = 0;
1765         }
1766
1767         set_capacity_and_notify(disk, capacity);
1768
1769         nvme_config_discard(disk, ns);
1770         blk_queue_max_write_zeroes_sectors(disk->queue,
1771                                            ns->ctrl->max_zeroes_sectors);
1772
1773         set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1774                 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1775 }
1776
1777 static inline bool nvme_first_scan(struct gendisk *disk)
1778 {
1779         /* nvme_alloc_ns() scans the disk prior to adding it */
1780         return !(disk->flags & GENHD_FL_UP);
1781 }
1782
1783 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1784 {
1785         struct nvme_ctrl *ctrl = ns->ctrl;
1786         u32 iob;
1787
1788         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1789             is_power_of_2(ctrl->max_hw_sectors))
1790                 iob = ctrl->max_hw_sectors;
1791         else
1792                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1793
1794         if (!iob)
1795                 return;
1796
1797         if (!is_power_of_2(iob)) {
1798                 if (nvme_first_scan(ns->disk))
1799                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1800                                 ns->disk->disk_name, iob);
1801                 return;
1802         }
1803
1804         if (blk_queue_is_zoned(ns->disk->queue)) {
1805                 if (nvme_first_scan(ns->disk))
1806                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
1807                                 ns->disk->disk_name);
1808                 return;
1809         }
1810
1811         blk_queue_chunk_sectors(ns->queue, iob);
1812 }
1813
1814 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1815 {
1816         unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1817         int ret;
1818
1819         blk_mq_freeze_queue(ns->disk->queue);
1820         ns->lba_shift = id->lbaf[lbaf].ds;
1821         nvme_set_queue_limits(ns->ctrl, ns->queue);
1822
1823         ret = nvme_configure_metadata(ns, id);
1824         if (ret)
1825                 goto out_unfreeze;
1826         nvme_set_chunk_sectors(ns, id);
1827         nvme_update_disk_info(ns->disk, ns, id);
1828
1829         if (ns->head->ids.csi == NVME_CSI_ZNS) {
1830                 ret = nvme_update_zone_info(ns, lbaf);
1831                 if (ret)
1832                         goto out_unfreeze;
1833         }
1834
1835         blk_mq_unfreeze_queue(ns->disk->queue);
1836
1837         if (blk_queue_is_zoned(ns->queue)) {
1838                 ret = nvme_revalidate_zones(ns);
1839                 if (ret && !nvme_first_scan(ns->disk))
1840                         goto out;
1841         }
1842
1843         if (nvme_ns_head_multipath(ns->head)) {
1844                 blk_mq_freeze_queue(ns->head->disk->queue);
1845                 nvme_update_disk_info(ns->head->disk, ns, id);
1846                 blk_stack_limits(&ns->head->disk->queue->limits,
1847                                  &ns->queue->limits, 0);
1848                 blk_queue_update_readahead(ns->head->disk->queue);
1849                 blk_mq_unfreeze_queue(ns->head->disk->queue);
1850         }
1851         return 0;
1852
1853 out_unfreeze:
1854         blk_mq_unfreeze_queue(ns->disk->queue);
1855 out:
1856         /*
1857          * If probing fails due an unsupported feature, hide the block device,
1858          * but still allow other access.
1859          */
1860         if (ret == -ENODEV) {
1861                 ns->disk->flags |= GENHD_FL_HIDDEN;
1862                 ret = 0;
1863         }
1864         return ret;
1865 }
1866
1867 static char nvme_pr_type(enum pr_type type)
1868 {
1869         switch (type) {
1870         case PR_WRITE_EXCLUSIVE:
1871                 return 1;
1872         case PR_EXCLUSIVE_ACCESS:
1873                 return 2;
1874         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1875                 return 3;
1876         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1877                 return 4;
1878         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1879                 return 5;
1880         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1881                 return 6;
1882         default:
1883                 return 0;
1884         }
1885 };
1886
1887 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1888                                 u64 key, u64 sa_key, u8 op)
1889 {
1890         struct nvme_ns_head *head = NULL;
1891         struct nvme_ns *ns;
1892         struct nvme_command c;
1893         int srcu_idx, ret;
1894         u8 data[16] = { 0, };
1895
1896         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1897         if (unlikely(!ns))
1898                 return -EWOULDBLOCK;
1899
1900         put_unaligned_le64(key, &data[0]);
1901         put_unaligned_le64(sa_key, &data[8]);
1902
1903         memset(&c, 0, sizeof(c));
1904         c.common.opcode = op;
1905         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1906         c.common.cdw10 = cpu_to_le32(cdw10);
1907
1908         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1909         nvme_put_ns_from_disk(head, srcu_idx);
1910         return ret;
1911 }
1912
1913 static int nvme_pr_register(struct block_device *bdev, u64 old,
1914                 u64 new, unsigned flags)
1915 {
1916         u32 cdw10;
1917
1918         if (flags & ~PR_FL_IGNORE_KEY)
1919                 return -EOPNOTSUPP;
1920
1921         cdw10 = old ? 2 : 0;
1922         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1923         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1924         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1925 }
1926
1927 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1928                 enum pr_type type, unsigned flags)
1929 {
1930         u32 cdw10;
1931
1932         if (flags & ~PR_FL_IGNORE_KEY)
1933                 return -EOPNOTSUPP;
1934
1935         cdw10 = nvme_pr_type(type) << 8;
1936         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1937         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1938 }
1939
1940 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1941                 enum pr_type type, bool abort)
1942 {
1943         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1944
1945         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1946 }
1947
1948 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1949 {
1950         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1951
1952         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1953 }
1954
1955 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1956 {
1957         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1958
1959         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1960 }
1961
1962 const struct pr_ops nvme_pr_ops = {
1963         .pr_register    = nvme_pr_register,
1964         .pr_reserve     = nvme_pr_reserve,
1965         .pr_release     = nvme_pr_release,
1966         .pr_preempt     = nvme_pr_preempt,
1967         .pr_clear       = nvme_pr_clear,
1968 };
1969
1970 #ifdef CONFIG_BLK_SED_OPAL
1971 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1972                 bool send)
1973 {
1974         struct nvme_ctrl *ctrl = data;
1975         struct nvme_command cmd;
1976
1977         memset(&cmd, 0, sizeof(cmd));
1978         if (send)
1979                 cmd.common.opcode = nvme_admin_security_send;
1980         else
1981                 cmd.common.opcode = nvme_admin_security_recv;
1982         cmd.common.nsid = 0;
1983         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1984         cmd.common.cdw11 = cpu_to_le32(len);
1985
1986         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
1987                         NVME_QID_ANY, 1, 0, false);
1988 }
1989 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1990 #endif /* CONFIG_BLK_SED_OPAL */
1991
1992 static const struct block_device_operations nvme_bdev_ops = {
1993         .owner          = THIS_MODULE,
1994         .ioctl          = nvme_ioctl,
1995         .open           = nvme_open,
1996         .release        = nvme_release,
1997         .getgeo         = nvme_getgeo,
1998         .report_zones   = nvme_report_zones,
1999         .pr_ops         = &nvme_pr_ops,
2000 };
2001
2002 #ifdef CONFIG_NVME_MULTIPATH
2003 struct nvme_ctrl *nvme_find_get_live_ctrl(struct nvme_subsystem *subsys)
2004 {
2005         struct nvme_ctrl *ctrl;
2006         int ret;
2007
2008         ret = mutex_lock_killable(&nvme_subsystems_lock);
2009         if (ret)
2010                 return ERR_PTR(ret);
2011         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2012                 if (ctrl->state == NVME_CTRL_LIVE)
2013                         goto found;
2014         }
2015         mutex_unlock(&nvme_subsystems_lock);
2016         return ERR_PTR(-EWOULDBLOCK);
2017 found:
2018         nvme_get_ctrl(ctrl);
2019         mutex_unlock(&nvme_subsystems_lock);
2020         return ctrl;
2021 }
2022 #endif /* CONFIG_NVME_MULTIPATH */
2023
2024 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2025 {
2026         unsigned long timeout =
2027                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2028         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2029         int ret;
2030
2031         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2032                 if (csts == ~0)
2033                         return -ENODEV;
2034                 if ((csts & NVME_CSTS_RDY) == bit)
2035                         break;
2036
2037                 usleep_range(1000, 2000);
2038                 if (fatal_signal_pending(current))
2039                         return -EINTR;
2040                 if (time_after(jiffies, timeout)) {
2041                         dev_err(ctrl->device,
2042                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2043                                 enabled ? "initialisation" : "reset", csts);
2044                         return -ENODEV;
2045                 }
2046         }
2047
2048         return ret;
2049 }
2050
2051 /*
2052  * If the device has been passed off to us in an enabled state, just clear
2053  * the enabled bit.  The spec says we should set the 'shutdown notification
2054  * bits', but doing so may cause the device to complete commands to the
2055  * admin queue ... and we don't know what memory that might be pointing at!
2056  */
2057 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2058 {
2059         int ret;
2060
2061         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2062         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2063
2064         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2065         if (ret)
2066                 return ret;
2067
2068         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2069                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2070
2071         return nvme_wait_ready(ctrl, ctrl->cap, false);
2072 }
2073 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2074
2075 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2076 {
2077         unsigned dev_page_min;
2078         int ret;
2079
2080         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2081         if (ret) {
2082                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2083                 return ret;
2084         }
2085         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2086
2087         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2088                 dev_err(ctrl->device,
2089                         "Minimum device page size %u too large for host (%u)\n",
2090                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2091                 return -ENODEV;
2092         }
2093
2094         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2095                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2096         else
2097                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2098         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2099         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2100         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2101         ctrl->ctrl_config |= NVME_CC_ENABLE;
2102
2103         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2104         if (ret)
2105                 return ret;
2106         return nvme_wait_ready(ctrl, ctrl->cap, true);
2107 }
2108 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2109
2110 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2111 {
2112         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2113         u32 csts;
2114         int ret;
2115
2116         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2117         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2118
2119         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2120         if (ret)
2121                 return ret;
2122
2123         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2124                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2125                         break;
2126
2127                 msleep(100);
2128                 if (fatal_signal_pending(current))
2129                         return -EINTR;
2130                 if (time_after(jiffies, timeout)) {
2131                         dev_err(ctrl->device,
2132                                 "Device shutdown incomplete; abort shutdown\n");
2133                         return -ENODEV;
2134                 }
2135         }
2136
2137         return ret;
2138 }
2139 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2140
2141 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2142 {
2143         __le64 ts;
2144         int ret;
2145
2146         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2147                 return 0;
2148
2149         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2150         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2151                         NULL);
2152         if (ret)
2153                 dev_warn_once(ctrl->device,
2154                         "could not set timestamp (%d)\n", ret);
2155         return ret;
2156 }
2157
2158 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2159 {
2160         struct nvme_feat_host_behavior *host;
2161         int ret;
2162
2163         /* Don't bother enabling the feature if retry delay is not reported */
2164         if (!ctrl->crdt[0])
2165                 return 0;
2166
2167         host = kzalloc(sizeof(*host), GFP_KERNEL);
2168         if (!host)
2169                 return 0;
2170
2171         host->acre = NVME_ENABLE_ACRE;
2172         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2173                                 host, sizeof(*host), NULL);
2174         kfree(host);
2175         return ret;
2176 }
2177
2178 /*
2179  * APST (Autonomous Power State Transition) lets us program a table of power
2180  * state transitions that the controller will perform automatically.
2181  * We configure it with a simple heuristic: we are willing to spend at most 2%
2182  * of the time transitioning between power states.  Therefore, when running in
2183  * any given state, we will enter the next lower-power non-operational state
2184  * after waiting 50 * (enlat + exlat) microseconds, as long as that state's exit
2185  * latency is under the requested maximum latency.
2186  *
2187  * We will not autonomously enter any non-operational state for which the total
2188  * latency exceeds ps_max_latency_us.
2189  *
2190  * Users can set ps_max_latency_us to zero to turn off APST.
2191  */
2192 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2193 {
2194         struct nvme_feat_auto_pst *table;
2195         unsigned apste = 0;
2196         u64 max_lat_us = 0;
2197         __le64 target = 0;
2198         int max_ps = -1;
2199         int state;
2200         int ret;
2201
2202         /*
2203          * If APST isn't supported or if we haven't been initialized yet,
2204          * then don't do anything.
2205          */
2206         if (!ctrl->apsta)
2207                 return 0;
2208
2209         if (ctrl->npss > 31) {
2210                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2211                 return 0;
2212         }
2213
2214         table = kzalloc(sizeof(*table), GFP_KERNEL);
2215         if (!table)
2216                 return 0;
2217
2218         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2219                 /* Turn off APST. */
2220                 dev_dbg(ctrl->device, "APST disabled\n");
2221                 goto done;
2222         }
2223
2224         /*
2225          * Walk through all states from lowest- to highest-power.
2226          * According to the spec, lower-numbered states use more power.  NPSS,
2227          * despite the name, is the index of the lowest-power state, not the
2228          * number of states.
2229          */
2230         for (state = (int)ctrl->npss; state >= 0; state--) {
2231                 u64 total_latency_us, exit_latency_us, transition_ms;
2232
2233                 if (target)
2234                         table->entries[state] = target;
2235
2236                 /*
2237                  * Don't allow transitions to the deepest state if it's quirked
2238                  * off.
2239                  */
2240                 if (state == ctrl->npss &&
2241                     (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2242                         continue;
2243
2244                 /*
2245                  * Is this state a useful non-operational state for higher-power
2246                  * states to autonomously transition to?
2247                  */
2248                 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2249                         continue;
2250
2251                 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2252                 if (exit_latency_us > ctrl->ps_max_latency_us)
2253                         continue;
2254
2255                 total_latency_us = exit_latency_us +
2256                         le32_to_cpu(ctrl->psd[state].entry_lat);
2257
2258                 /*
2259                  * This state is good.  Use it as the APST idle target for
2260                  * higher power states.
2261                  */
2262                 transition_ms = total_latency_us + 19;
2263                 do_div(transition_ms, 20);
2264                 if (transition_ms > (1 << 24) - 1)
2265                         transition_ms = (1 << 24) - 1;
2266
2267                 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2268                 if (max_ps == -1)
2269                         max_ps = state;
2270                 if (total_latency_us > max_lat_us)
2271                         max_lat_us = total_latency_us;
2272         }
2273
2274         if (max_ps == -1)
2275                 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2276         else
2277                 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2278                         max_ps, max_lat_us, (int)sizeof(*table), table);
2279         apste = 1;
2280
2281 done:
2282         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2283                                 table, sizeof(*table), NULL);
2284         if (ret)
2285                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2286         kfree(table);
2287         return ret;
2288 }
2289
2290 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2291 {
2292         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2293         u64 latency;
2294
2295         switch (val) {
2296         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2297         case PM_QOS_LATENCY_ANY:
2298                 latency = U64_MAX;
2299                 break;
2300
2301         default:
2302                 latency = val;
2303         }
2304
2305         if (ctrl->ps_max_latency_us != latency) {
2306                 ctrl->ps_max_latency_us = latency;
2307                 if (ctrl->state == NVME_CTRL_LIVE)
2308                         nvme_configure_apst(ctrl);
2309         }
2310 }
2311
2312 struct nvme_core_quirk_entry {
2313         /*
2314          * NVMe model and firmware strings are padded with spaces.  For
2315          * simplicity, strings in the quirk table are padded with NULLs
2316          * instead.
2317          */
2318         u16 vid;
2319         const char *mn;
2320         const char *fr;
2321         unsigned long quirks;
2322 };
2323
2324 static const struct nvme_core_quirk_entry core_quirks[] = {
2325         {
2326                 /*
2327                  * This Toshiba device seems to die using any APST states.  See:
2328                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2329                  */
2330                 .vid = 0x1179,
2331                 .mn = "THNSF5256GPUK TOSHIBA",
2332                 .quirks = NVME_QUIRK_NO_APST,
2333         },
2334         {
2335                 /*
2336                  * This LiteON CL1-3D*-Q11 firmware version has a race
2337                  * condition associated with actions related to suspend to idle
2338                  * LiteON has resolved the problem in future firmware
2339                  */
2340                 .vid = 0x14a4,
2341                 .fr = "22301111",
2342                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2343         }
2344 };
2345
2346 /* match is null-terminated but idstr is space-padded. */
2347 static bool string_matches(const char *idstr, const char *match, size_t len)
2348 {
2349         size_t matchlen;
2350
2351         if (!match)
2352                 return true;
2353
2354         matchlen = strlen(match);
2355         WARN_ON_ONCE(matchlen > len);
2356
2357         if (memcmp(idstr, match, matchlen))
2358                 return false;
2359
2360         for (; matchlen < len; matchlen++)
2361                 if (idstr[matchlen] != ' ')
2362                         return false;
2363
2364         return true;
2365 }
2366
2367 static bool quirk_matches(const struct nvme_id_ctrl *id,
2368                           const struct nvme_core_quirk_entry *q)
2369 {
2370         return q->vid == le16_to_cpu(id->vid) &&
2371                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2372                 string_matches(id->fr, q->fr, sizeof(id->fr));
2373 }
2374
2375 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2376                 struct nvme_id_ctrl *id)
2377 {
2378         size_t nqnlen;
2379         int off;
2380
2381         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2382                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2383                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2384                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2385                         return;
2386                 }
2387
2388                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2389                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2390         }
2391
2392         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2393         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2394                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2395                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2396         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2397         off += sizeof(id->sn);
2398         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2399         off += sizeof(id->mn);
2400         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2401 }
2402
2403 static void nvme_release_subsystem(struct device *dev)
2404 {
2405         struct nvme_subsystem *subsys =
2406                 container_of(dev, struct nvme_subsystem, dev);
2407
2408         if (subsys->instance >= 0)
2409                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2410         kfree(subsys);
2411 }
2412
2413 static void nvme_destroy_subsystem(struct kref *ref)
2414 {
2415         struct nvme_subsystem *subsys =
2416                         container_of(ref, struct nvme_subsystem, ref);
2417
2418         mutex_lock(&nvme_subsystems_lock);
2419         list_del(&subsys->entry);
2420         mutex_unlock(&nvme_subsystems_lock);
2421
2422         ida_destroy(&subsys->ns_ida);
2423         device_del(&subsys->dev);
2424         put_device(&subsys->dev);
2425 }
2426
2427 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2428 {
2429         kref_put(&subsys->ref, nvme_destroy_subsystem);
2430 }
2431
2432 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2433 {
2434         struct nvme_subsystem *subsys;
2435
2436         lockdep_assert_held(&nvme_subsystems_lock);
2437
2438         /*
2439          * Fail matches for discovery subsystems. This results
2440          * in each discovery controller bound to a unique subsystem.
2441          * This avoids issues with validating controller values
2442          * that can only be true when there is a single unique subsystem.
2443          * There may be multiple and completely independent entities
2444          * that provide discovery controllers.
2445          */
2446         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2447                 return NULL;
2448
2449         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2450                 if (strcmp(subsys->subnqn, subsysnqn))
2451                         continue;
2452                 if (!kref_get_unless_zero(&subsys->ref))
2453                         continue;
2454                 return subsys;
2455         }
2456
2457         return NULL;
2458 }
2459
2460 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2461         struct device_attribute subsys_attr_##_name = \
2462                 __ATTR(_name, _mode, _show, NULL)
2463
2464 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2465                                     struct device_attribute *attr,
2466                                     char *buf)
2467 {
2468         struct nvme_subsystem *subsys =
2469                 container_of(dev, struct nvme_subsystem, dev);
2470
2471         return sysfs_emit(buf, "%s\n", subsys->subnqn);
2472 }
2473 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2474
2475 #define nvme_subsys_show_str_function(field)                            \
2476 static ssize_t subsys_##field##_show(struct device *dev,                \
2477                             struct device_attribute *attr, char *buf)   \
2478 {                                                                       \
2479         struct nvme_subsystem *subsys =                                 \
2480                 container_of(dev, struct nvme_subsystem, dev);          \
2481         return sysfs_emit(buf, "%.*s\n",                                \
2482                            (int)sizeof(subsys->field), subsys->field);  \
2483 }                                                                       \
2484 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2485
2486 nvme_subsys_show_str_function(model);
2487 nvme_subsys_show_str_function(serial);
2488 nvme_subsys_show_str_function(firmware_rev);
2489
2490 static struct attribute *nvme_subsys_attrs[] = {
2491         &subsys_attr_model.attr,
2492         &subsys_attr_serial.attr,
2493         &subsys_attr_firmware_rev.attr,
2494         &subsys_attr_subsysnqn.attr,
2495 #ifdef CONFIG_NVME_MULTIPATH
2496         &subsys_attr_iopolicy.attr,
2497 #endif
2498         NULL,
2499 };
2500
2501 static const struct attribute_group nvme_subsys_attrs_group = {
2502         .attrs = nvme_subsys_attrs,
2503 };
2504
2505 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2506         &nvme_subsys_attrs_group,
2507         NULL,
2508 };
2509
2510 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2511 {
2512         return ctrl->opts && ctrl->opts->discovery_nqn;
2513 }
2514
2515 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2516                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2517 {
2518         struct nvme_ctrl *tmp;
2519
2520         lockdep_assert_held(&nvme_subsystems_lock);
2521
2522         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2523                 if (nvme_state_terminal(tmp))
2524                         continue;
2525
2526                 if (tmp->cntlid == ctrl->cntlid) {
2527                         dev_err(ctrl->device,
2528                                 "Duplicate cntlid %u with %s, rejecting\n",
2529                                 ctrl->cntlid, dev_name(tmp->device));
2530                         return false;
2531                 }
2532
2533                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2534                     nvme_discovery_ctrl(ctrl))
2535                         continue;
2536
2537                 dev_err(ctrl->device,
2538                         "Subsystem does not support multiple controllers\n");
2539                 return false;
2540         }
2541
2542         return true;
2543 }
2544
2545 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2546 {
2547         struct nvme_subsystem *subsys, *found;
2548         int ret;
2549
2550         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2551         if (!subsys)
2552                 return -ENOMEM;
2553
2554         subsys->instance = -1;
2555         mutex_init(&subsys->lock);
2556         kref_init(&subsys->ref);
2557         INIT_LIST_HEAD(&subsys->ctrls);
2558         INIT_LIST_HEAD(&subsys->nsheads);
2559         nvme_init_subnqn(subsys, ctrl, id);
2560         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2561         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2562         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2563         subsys->vendor_id = le16_to_cpu(id->vid);
2564         subsys->cmic = id->cmic;
2565         subsys->awupf = le16_to_cpu(id->awupf);
2566 #ifdef CONFIG_NVME_MULTIPATH
2567         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2568 #endif
2569
2570         subsys->dev.class = nvme_subsys_class;
2571         subsys->dev.release = nvme_release_subsystem;
2572         subsys->dev.groups = nvme_subsys_attrs_groups;
2573         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2574         device_initialize(&subsys->dev);
2575
2576         mutex_lock(&nvme_subsystems_lock);
2577         found = __nvme_find_get_subsystem(subsys->subnqn);
2578         if (found) {
2579                 put_device(&subsys->dev);
2580                 subsys = found;
2581
2582                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2583                         ret = -EINVAL;
2584                         goto out_put_subsystem;
2585                 }
2586         } else {
2587                 ret = device_add(&subsys->dev);
2588                 if (ret) {
2589                         dev_err(ctrl->device,
2590                                 "failed to register subsystem device.\n");
2591                         put_device(&subsys->dev);
2592                         goto out_unlock;
2593                 }
2594                 ida_init(&subsys->ns_ida);
2595                 list_add_tail(&subsys->entry, &nvme_subsystems);
2596         }
2597
2598         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2599                                 dev_name(ctrl->device));
2600         if (ret) {
2601                 dev_err(ctrl->device,
2602                         "failed to create sysfs link from subsystem.\n");
2603                 goto out_put_subsystem;
2604         }
2605
2606         if (!found)
2607                 subsys->instance = ctrl->instance;
2608         ctrl->subsys = subsys;
2609         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2610         mutex_unlock(&nvme_subsystems_lock);
2611         return 0;
2612
2613 out_put_subsystem:
2614         nvme_put_subsystem(subsys);
2615 out_unlock:
2616         mutex_unlock(&nvme_subsystems_lock);
2617         return ret;
2618 }
2619
2620 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2621                 void *log, size_t size, u64 offset)
2622 {
2623         struct nvme_command c = { };
2624         u32 dwlen = nvme_bytes_to_numd(size);
2625
2626         c.get_log_page.opcode = nvme_admin_get_log_page;
2627         c.get_log_page.nsid = cpu_to_le32(nsid);
2628         c.get_log_page.lid = log_page;
2629         c.get_log_page.lsp = lsp;
2630         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2631         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2632         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2633         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2634         c.get_log_page.csi = csi;
2635
2636         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2637 }
2638
2639 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2640                                 struct nvme_effects_log **log)
2641 {
2642         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2643         int ret;
2644
2645         if (cel)
2646                 goto out;
2647
2648         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2649         if (!cel)
2650                 return -ENOMEM;
2651
2652         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2653                         cel, sizeof(*cel), 0);
2654         if (ret) {
2655                 kfree(cel);
2656                 return ret;
2657         }
2658
2659         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2660 out:
2661         *log = cel;
2662         return 0;
2663 }
2664
2665 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2666 {
2667         u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2668
2669         if (check_shl_overflow(1U, units + page_shift - 9, &val))
2670                 return UINT_MAX;
2671         return val;
2672 }
2673
2674 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2675 {
2676         struct nvme_command c = { };
2677         struct nvme_id_ctrl_nvm *id;
2678         int ret;
2679
2680         if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2681                 ctrl->max_discard_sectors = UINT_MAX;
2682                 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2683         } else {
2684                 ctrl->max_discard_sectors = 0;
2685                 ctrl->max_discard_segments = 0;
2686         }
2687
2688         /*
2689          * Even though NVMe spec explicitly states that MDTS is not applicable
2690          * to the write-zeroes, we are cautious and limit the size to the
2691          * controllers max_hw_sectors value, which is based on the MDTS field
2692          * and possibly other limiting factors.
2693          */
2694         if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2695             !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2696                 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2697         else
2698                 ctrl->max_zeroes_sectors = 0;
2699
2700         if (nvme_ctrl_limited_cns(ctrl))
2701                 return 0;
2702
2703         id = kzalloc(sizeof(*id), GFP_KERNEL);
2704         if (!id)
2705                 return 0;
2706
2707         c.identify.opcode = nvme_admin_identify;
2708         c.identify.cns = NVME_ID_CNS_CS_CTRL;
2709         c.identify.csi = NVME_CSI_NVM;
2710
2711         ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2712         if (ret)
2713                 goto free_data;
2714
2715         if (id->dmrl)
2716                 ctrl->max_discard_segments = id->dmrl;
2717         if (id->dmrsl)
2718                 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2719         if (id->wzsl)
2720                 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2721
2722 free_data:
2723         kfree(id);
2724         return ret;
2725 }
2726
2727 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2728 {
2729         struct nvme_id_ctrl *id;
2730         u32 max_hw_sectors;
2731         bool prev_apst_enabled;
2732         int ret;
2733
2734         ret = nvme_identify_ctrl(ctrl, &id);
2735         if (ret) {
2736                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2737                 return -EIO;
2738         }
2739
2740         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2741                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2742                 if (ret < 0)
2743                         goto out_free;
2744         }
2745
2746         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2747                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2748
2749         if (!ctrl->identified) {
2750                 unsigned int i;
2751
2752                 ret = nvme_init_subsystem(ctrl, id);
2753                 if (ret)
2754                         goto out_free;
2755
2756                 /*
2757                  * Check for quirks.  Quirk can depend on firmware version,
2758                  * so, in principle, the set of quirks present can change
2759                  * across a reset.  As a possible future enhancement, we
2760                  * could re-scan for quirks every time we reinitialize
2761                  * the device, but we'd have to make sure that the driver
2762                  * behaves intelligently if the quirks change.
2763                  */
2764                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2765                         if (quirk_matches(id, &core_quirks[i]))
2766                                 ctrl->quirks |= core_quirks[i].quirks;
2767                 }
2768         }
2769
2770         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2771                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2772                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2773         }
2774
2775         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2776         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2777         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2778
2779         ctrl->oacs = le16_to_cpu(id->oacs);
2780         ctrl->oncs = le16_to_cpu(id->oncs);
2781         ctrl->mtfa = le16_to_cpu(id->mtfa);
2782         ctrl->oaes = le32_to_cpu(id->oaes);
2783         ctrl->wctemp = le16_to_cpu(id->wctemp);
2784         ctrl->cctemp = le16_to_cpu(id->cctemp);
2785
2786         atomic_set(&ctrl->abort_limit, id->acl + 1);
2787         ctrl->vwc = id->vwc;
2788         if (id->mdts)
2789                 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2790         else
2791                 max_hw_sectors = UINT_MAX;
2792         ctrl->max_hw_sectors =
2793                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2794
2795         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2796         ctrl->sgls = le32_to_cpu(id->sgls);
2797         ctrl->kas = le16_to_cpu(id->kas);
2798         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2799         ctrl->ctratt = le32_to_cpu(id->ctratt);
2800
2801         if (id->rtd3e) {
2802                 /* us -> s */
2803                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2804
2805                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2806                                                  shutdown_timeout, 60);
2807
2808                 if (ctrl->shutdown_timeout != shutdown_timeout)
2809                         dev_info(ctrl->device,
2810                                  "Shutdown timeout set to %u seconds\n",
2811                                  ctrl->shutdown_timeout);
2812         } else
2813                 ctrl->shutdown_timeout = shutdown_timeout;
2814
2815         ctrl->npss = id->npss;
2816         ctrl->apsta = id->apsta;
2817         prev_apst_enabled = ctrl->apst_enabled;
2818         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2819                 if (force_apst && id->apsta) {
2820                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2821                         ctrl->apst_enabled = true;
2822                 } else {
2823                         ctrl->apst_enabled = false;
2824                 }
2825         } else {
2826                 ctrl->apst_enabled = id->apsta;
2827         }
2828         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2829
2830         if (ctrl->ops->flags & NVME_F_FABRICS) {
2831                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2832                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2833                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2834                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2835
2836                 /*
2837                  * In fabrics we need to verify the cntlid matches the
2838                  * admin connect
2839                  */
2840                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2841                         dev_err(ctrl->device,
2842                                 "Mismatching cntlid: Connect %u vs Identify "
2843                                 "%u, rejecting\n",
2844                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
2845                         ret = -EINVAL;
2846                         goto out_free;
2847                 }
2848
2849                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
2850                         dev_err(ctrl->device,
2851                                 "keep-alive support is mandatory for fabrics\n");
2852                         ret = -EINVAL;
2853                         goto out_free;
2854                 }
2855         } else {
2856                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2857                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2858                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2859                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2860         }
2861
2862         ret = nvme_mpath_init(ctrl, id);
2863         if (ret < 0)
2864                 goto out_free;
2865
2866         if (ctrl->apst_enabled && !prev_apst_enabled)
2867                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2868         else if (!ctrl->apst_enabled && prev_apst_enabled)
2869                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2870
2871 out_free:
2872         kfree(id);
2873         return ret;
2874 }
2875
2876 /*
2877  * Initialize the cached copies of the Identify data and various controller
2878  * register in our nvme_ctrl structure.  This should be called as soon as
2879  * the admin queue is fully up and running.
2880  */
2881 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
2882 {
2883         int ret;
2884
2885         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2886         if (ret) {
2887                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2888                 return ret;
2889         }
2890
2891         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2892
2893         if (ctrl->vs >= NVME_VS(1, 1, 0))
2894                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2895
2896         ret = nvme_init_identify(ctrl);
2897         if (ret)
2898                 return ret;
2899
2900         ret = nvme_init_non_mdts_limits(ctrl);
2901         if (ret < 0)
2902                 return ret;
2903
2904         ret = nvme_configure_apst(ctrl);
2905         if (ret < 0)
2906                 return ret;
2907
2908         ret = nvme_configure_timestamp(ctrl);
2909         if (ret < 0)
2910                 return ret;
2911
2912         ret = nvme_configure_directives(ctrl);
2913         if (ret < 0)
2914                 return ret;
2915
2916         ret = nvme_configure_acre(ctrl);
2917         if (ret < 0)
2918                 return ret;
2919
2920         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
2921                 ret = nvme_hwmon_init(ctrl);
2922                 if (ret < 0)
2923                         return ret;
2924         }
2925
2926         ctrl->identified = true;
2927
2928         return 0;
2929 }
2930 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
2931
2932 static int nvme_dev_open(struct inode *inode, struct file *file)
2933 {
2934         struct nvme_ctrl *ctrl =
2935                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2936
2937         switch (ctrl->state) {
2938         case NVME_CTRL_LIVE:
2939                 break;
2940         default:
2941                 return -EWOULDBLOCK;
2942         }
2943
2944         nvme_get_ctrl(ctrl);
2945         if (!try_module_get(ctrl->ops->module)) {
2946                 nvme_put_ctrl(ctrl);
2947                 return -EINVAL;
2948         }
2949
2950         file->private_data = ctrl;
2951         return 0;
2952 }
2953
2954 static int nvme_dev_release(struct inode *inode, struct file *file)
2955 {
2956         struct nvme_ctrl *ctrl =
2957                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2958
2959         module_put(ctrl->ops->module);
2960         nvme_put_ctrl(ctrl);
2961         return 0;
2962 }
2963
2964 static const struct file_operations nvme_dev_fops = {
2965         .owner          = THIS_MODULE,
2966         .open           = nvme_dev_open,
2967         .release        = nvme_dev_release,
2968         .unlocked_ioctl = nvme_dev_ioctl,
2969         .compat_ioctl   = compat_ptr_ioctl,
2970 };
2971
2972 static ssize_t nvme_sysfs_reset(struct device *dev,
2973                                 struct device_attribute *attr, const char *buf,
2974                                 size_t count)
2975 {
2976         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2977         int ret;
2978
2979         ret = nvme_reset_ctrl_sync(ctrl);
2980         if (ret < 0)
2981                 return ret;
2982         return count;
2983 }
2984 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2985
2986 static ssize_t nvme_sysfs_rescan(struct device *dev,
2987                                 struct device_attribute *attr, const char *buf,
2988                                 size_t count)
2989 {
2990         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2991
2992         nvme_queue_scan(ctrl);
2993         return count;
2994 }
2995 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2996
2997 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2998 {
2999         struct gendisk *disk = dev_to_disk(dev);
3000
3001         if (disk->fops == &nvme_bdev_ops)
3002                 return nvme_get_ns_from_dev(dev)->head;
3003         else
3004                 return disk->private_data;
3005 }
3006
3007 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3008                 char *buf)
3009 {
3010         struct nvme_ns_head *head = dev_to_ns_head(dev);
3011         struct nvme_ns_ids *ids = &head->ids;
3012         struct nvme_subsystem *subsys = head->subsys;
3013         int serial_len = sizeof(subsys->serial);
3014         int model_len = sizeof(subsys->model);
3015
3016         if (!uuid_is_null(&ids->uuid))
3017                 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3018
3019         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3020                 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3021
3022         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3023                 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3024
3025         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3026                                   subsys->serial[serial_len - 1] == '\0'))
3027                 serial_len--;
3028         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3029                                  subsys->model[model_len - 1] == '\0'))
3030                 model_len--;
3031
3032         return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3033                 serial_len, subsys->serial, model_len, subsys->model,
3034                 head->ns_id);
3035 }
3036 static DEVICE_ATTR_RO(wwid);
3037
3038 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3039                 char *buf)
3040 {
3041         return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3042 }
3043 static DEVICE_ATTR_RO(nguid);
3044
3045 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3046                 char *buf)
3047 {
3048         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3049
3050         /* For backward compatibility expose the NGUID to userspace if
3051          * we have no UUID set
3052          */
3053         if (uuid_is_null(&ids->uuid)) {
3054                 printk_ratelimited(KERN_WARNING
3055                                    "No UUID available providing old NGUID\n");
3056                 return sysfs_emit(buf, "%pU\n", ids->nguid);
3057         }
3058         return sysfs_emit(buf, "%pU\n", &ids->uuid);
3059 }
3060 static DEVICE_ATTR_RO(uuid);
3061
3062 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3063                 char *buf)
3064 {
3065         return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3066 }
3067 static DEVICE_ATTR_RO(eui);
3068
3069 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3070                 char *buf)
3071 {
3072         return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3073 }
3074 static DEVICE_ATTR_RO(nsid);
3075
3076 static struct attribute *nvme_ns_id_attrs[] = {
3077         &dev_attr_wwid.attr,
3078         &dev_attr_uuid.attr,
3079         &dev_attr_nguid.attr,
3080         &dev_attr_eui.attr,
3081         &dev_attr_nsid.attr,
3082 #ifdef CONFIG_NVME_MULTIPATH
3083         &dev_attr_ana_grpid.attr,
3084         &dev_attr_ana_state.attr,
3085 #endif
3086         NULL,
3087 };
3088
3089 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3090                 struct attribute *a, int n)
3091 {
3092         struct device *dev = container_of(kobj, struct device, kobj);
3093         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3094
3095         if (a == &dev_attr_uuid.attr) {
3096                 if (uuid_is_null(&ids->uuid) &&
3097                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3098                         return 0;
3099         }
3100         if (a == &dev_attr_nguid.attr) {
3101                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3102                         return 0;
3103         }
3104         if (a == &dev_attr_eui.attr) {
3105                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3106                         return 0;
3107         }
3108 #ifdef CONFIG_NVME_MULTIPATH
3109         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3110                 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3111                         return 0;
3112                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3113                         return 0;
3114         }
3115 #endif
3116         return a->mode;
3117 }
3118
3119 static const struct attribute_group nvme_ns_id_attr_group = {
3120         .attrs          = nvme_ns_id_attrs,
3121         .is_visible     = nvme_ns_id_attrs_are_visible,
3122 };
3123
3124 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3125         &nvme_ns_id_attr_group,
3126 #ifdef CONFIG_NVM
3127         &nvme_nvm_attr_group,
3128 #endif
3129         NULL,
3130 };
3131
3132 #define nvme_show_str_function(field)                                           \
3133 static ssize_t  field##_show(struct device *dev,                                \
3134                             struct device_attribute *attr, char *buf)           \
3135 {                                                                               \
3136         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3137         return sysfs_emit(buf, "%.*s\n",                                        \
3138                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3139 }                                                                               \
3140 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3141
3142 nvme_show_str_function(model);
3143 nvme_show_str_function(serial);
3144 nvme_show_str_function(firmware_rev);
3145
3146 #define nvme_show_int_function(field)                                           \
3147 static ssize_t  field##_show(struct device *dev,                                \
3148                             struct device_attribute *attr, char *buf)           \
3149 {                                                                               \
3150         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3151         return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3152 }                                                                               \
3153 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3154
3155 nvme_show_int_function(cntlid);
3156 nvme_show_int_function(numa_node);
3157 nvme_show_int_function(queue_count);
3158 nvme_show_int_function(sqsize);
3159 nvme_show_int_function(kato);
3160
3161 static ssize_t nvme_sysfs_delete(struct device *dev,
3162                                 struct device_attribute *attr, const char *buf,
3163                                 size_t count)
3164 {
3165         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3166
3167         if (device_remove_file_self(dev, attr))
3168                 nvme_delete_ctrl_sync(ctrl);
3169         return count;
3170 }
3171 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3172
3173 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3174                                          struct device_attribute *attr,
3175                                          char *buf)
3176 {
3177         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3178
3179         return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3180 }
3181 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3182
3183 static ssize_t nvme_sysfs_show_state(struct device *dev,
3184                                      struct device_attribute *attr,
3185                                      char *buf)
3186 {
3187         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3188         static const char *const state_name[] = {
3189                 [NVME_CTRL_NEW]         = "new",
3190                 [NVME_CTRL_LIVE]        = "live",
3191                 [NVME_CTRL_RESETTING]   = "resetting",
3192                 [NVME_CTRL_CONNECTING]  = "connecting",
3193                 [NVME_CTRL_DELETING]    = "deleting",
3194                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3195                 [NVME_CTRL_DEAD]        = "dead",
3196         };
3197
3198         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3199             state_name[ctrl->state])
3200                 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3201
3202         return sysfs_emit(buf, "unknown state\n");
3203 }
3204
3205 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3206
3207 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3208                                          struct device_attribute *attr,
3209                                          char *buf)
3210 {
3211         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3212
3213         return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3214 }
3215 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3216
3217 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3218                                         struct device_attribute *attr,
3219                                         char *buf)
3220 {
3221         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3222
3223         return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3224 }
3225 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3226
3227 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3228                                         struct device_attribute *attr,
3229                                         char *buf)
3230 {
3231         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3232
3233         return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3234 }
3235 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3236
3237 static ssize_t nvme_sysfs_show_address(struct device *dev,
3238                                          struct device_attribute *attr,
3239                                          char *buf)
3240 {
3241         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3242
3243         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3244 }
3245 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3246
3247 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3248                 struct device_attribute *attr, char *buf)
3249 {
3250         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3251         struct nvmf_ctrl_options *opts = ctrl->opts;
3252
3253         if (ctrl->opts->max_reconnects == -1)
3254                 return sysfs_emit(buf, "off\n");
3255         return sysfs_emit(buf, "%d\n",
3256                           opts->max_reconnects * opts->reconnect_delay);
3257 }
3258
3259 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3260                 struct device_attribute *attr, const char *buf, size_t count)
3261 {
3262         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3263         struct nvmf_ctrl_options *opts = ctrl->opts;
3264         int ctrl_loss_tmo, err;
3265
3266         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3267         if (err)
3268                 return -EINVAL;
3269
3270         if (ctrl_loss_tmo < 0)
3271                 opts->max_reconnects = -1;
3272         else
3273                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3274                                                 opts->reconnect_delay);
3275         return count;
3276 }
3277 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3278         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3279
3280 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3281                 struct device_attribute *attr, char *buf)
3282 {
3283         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3284
3285         if (ctrl->opts->reconnect_delay == -1)
3286                 return sysfs_emit(buf, "off\n");
3287         return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3288 }
3289
3290 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3291                 struct device_attribute *attr, const char *buf, size_t count)
3292 {
3293         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3294         unsigned int v;
3295         int err;
3296
3297         err = kstrtou32(buf, 10, &v);
3298         if (err)
3299                 return err;
3300
3301         ctrl->opts->reconnect_delay = v;
3302         return count;
3303 }
3304 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3305         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3306
3307 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3308                 struct device_attribute *attr, char *buf)
3309 {
3310         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3311
3312         if (ctrl->opts->fast_io_fail_tmo == -1)
3313                 return sysfs_emit(buf, "off\n");
3314         return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3315 }
3316
3317 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3318                 struct device_attribute *attr, const char *buf, size_t count)
3319 {
3320         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3321         struct nvmf_ctrl_options *opts = ctrl->opts;
3322         int fast_io_fail_tmo, err;
3323
3324         err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3325         if (err)
3326                 return -EINVAL;
3327
3328         if (fast_io_fail_tmo < 0)
3329                 opts->fast_io_fail_tmo = -1;
3330         else
3331                 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3332         return count;
3333 }
3334 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3335         nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3336
3337 static struct attribute *nvme_dev_attrs[] = {
3338         &dev_attr_reset_controller.attr,
3339         &dev_attr_rescan_controller.attr,
3340         &dev_attr_model.attr,
3341         &dev_attr_serial.attr,
3342         &dev_attr_firmware_rev.attr,
3343         &dev_attr_cntlid.attr,
3344         &dev_attr_delete_controller.attr,
3345         &dev_attr_transport.attr,
3346         &dev_attr_subsysnqn.attr,
3347         &dev_attr_address.attr,
3348         &dev_attr_state.attr,
3349         &dev_attr_numa_node.attr,
3350         &dev_attr_queue_count.attr,
3351         &dev_attr_sqsize.attr,
3352         &dev_attr_hostnqn.attr,
3353         &dev_attr_hostid.attr,
3354         &dev_attr_ctrl_loss_tmo.attr,
3355         &dev_attr_reconnect_delay.attr,
3356         &dev_attr_fast_io_fail_tmo.attr,
3357         &dev_attr_kato.attr,
3358         NULL
3359 };
3360
3361 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3362                 struct attribute *a, int n)
3363 {
3364         struct device *dev = container_of(kobj, struct device, kobj);
3365         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3366
3367         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3368                 return 0;
3369         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3370                 return 0;
3371         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3372                 return 0;
3373         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3374                 return 0;
3375         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3376                 return 0;
3377         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3378                 return 0;
3379         if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3380                 return 0;
3381
3382         return a->mode;
3383 }
3384
3385 static const struct attribute_group nvme_dev_attrs_group = {
3386         .attrs          = nvme_dev_attrs,
3387         .is_visible     = nvme_dev_attrs_are_visible,
3388 };
3389
3390 static const struct attribute_group *nvme_dev_attr_groups[] = {
3391         &nvme_dev_attrs_group,
3392         NULL,
3393 };
3394
3395 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3396                 unsigned nsid)
3397 {
3398         struct nvme_ns_head *h;
3399
3400         lockdep_assert_held(&subsys->lock);
3401
3402         list_for_each_entry(h, &subsys->nsheads, entry) {
3403                 if (h->ns_id == nsid && nvme_tryget_ns_head(h))
3404                         return h;
3405         }
3406
3407         return NULL;
3408 }
3409
3410 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3411                 struct nvme_ns_head *new)
3412 {
3413         struct nvme_ns_head *h;
3414
3415         lockdep_assert_held(&subsys->lock);
3416
3417         list_for_each_entry(h, &subsys->nsheads, entry) {
3418                 if (nvme_ns_ids_valid(&new->ids) &&
3419                     nvme_ns_ids_equal(&new->ids, &h->ids))
3420                         return -EINVAL;
3421         }
3422
3423         return 0;
3424 }
3425
3426 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3427 {
3428         cdev_device_del(cdev, cdev_device);
3429         ida_simple_remove(&nvme_ns_chr_minor_ida, MINOR(cdev_device->devt));
3430 }
3431
3432 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3433                 const struct file_operations *fops, struct module *owner)
3434 {
3435         int minor, ret;
3436
3437         minor = ida_simple_get(&nvme_ns_chr_minor_ida, 0, 0, GFP_KERNEL);
3438         if (minor < 0)
3439                 return minor;
3440         cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3441         cdev_device->class = nvme_ns_chr_class;
3442         device_initialize(cdev_device);
3443         cdev_init(cdev, fops);
3444         cdev->owner = owner;
3445         ret = cdev_device_add(cdev, cdev_device);
3446         if (ret)
3447                 ida_simple_remove(&nvme_ns_chr_minor_ida, minor);
3448         return ret;
3449 }
3450
3451 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3452 {
3453         return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3454 }
3455
3456 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3457 {
3458         nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3459         return 0;
3460 }
3461
3462 static const struct file_operations nvme_ns_chr_fops = {
3463         .owner          = THIS_MODULE,
3464         .open           = nvme_ns_chr_open,
3465         .release        = nvme_ns_chr_release,
3466         .unlocked_ioctl = nvme_ns_chr_ioctl,
3467         .compat_ioctl   = compat_ptr_ioctl,
3468 };
3469
3470 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3471 {
3472         int ret;
3473
3474         ns->cdev_device.parent = ns->ctrl->device;
3475         ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3476                            ns->ctrl->instance, ns->head->instance);
3477         if (ret)
3478                 return ret;
3479         ret = nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3480                             ns->ctrl->ops->module);
3481         if (ret)
3482                 kfree_const(ns->cdev_device.kobj.name);
3483         return ret;
3484 }
3485
3486 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3487                 unsigned nsid, struct nvme_ns_ids *ids)
3488 {
3489         struct nvme_ns_head *head;
3490         size_t size = sizeof(*head);
3491         int ret = -ENOMEM;
3492
3493 #ifdef CONFIG_NVME_MULTIPATH
3494         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3495 #endif
3496
3497         head = kzalloc(size, GFP_KERNEL);
3498         if (!head)
3499                 goto out;
3500         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3501         if (ret < 0)
3502                 goto out_free_head;
3503         head->instance = ret;
3504         INIT_LIST_HEAD(&head->list);
3505         ret = init_srcu_struct(&head->srcu);
3506         if (ret)
3507                 goto out_ida_remove;
3508         head->subsys = ctrl->subsys;
3509         head->ns_id = nsid;
3510         head->ids = *ids;
3511         kref_init(&head->ref);
3512
3513         ret = __nvme_check_ids(ctrl->subsys, head);
3514         if (ret) {
3515                 dev_err(ctrl->device,
3516                         "duplicate IDs for nsid %d\n", nsid);
3517                 goto out_cleanup_srcu;
3518         }
3519
3520         if (head->ids.csi) {
3521                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3522                 if (ret)
3523                         goto out_cleanup_srcu;
3524         } else
3525                 head->effects = ctrl->effects;
3526
3527         ret = nvme_mpath_alloc_disk(ctrl, head);
3528         if (ret)
3529                 goto out_cleanup_srcu;
3530
3531         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3532
3533         kref_get(&ctrl->subsys->ref);
3534
3535         return head;
3536 out_cleanup_srcu:
3537         cleanup_srcu_struct(&head->srcu);
3538 out_ida_remove:
3539         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3540 out_free_head:
3541         kfree(head);
3542 out:
3543         if (ret > 0)
3544                 ret = blk_status_to_errno(nvme_error_status(ret));
3545         return ERR_PTR(ret);
3546 }
3547
3548 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3549                 struct nvme_ns_ids *ids, bool is_shared)
3550 {
3551         struct nvme_ctrl *ctrl = ns->ctrl;
3552         struct nvme_ns_head *head = NULL;
3553         int ret = 0;
3554
3555         mutex_lock(&ctrl->subsys->lock);
3556         head = nvme_find_ns_head(ctrl->subsys, nsid);
3557         if (!head) {
3558                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3559                 if (IS_ERR(head)) {
3560                         ret = PTR_ERR(head);
3561                         goto out_unlock;
3562                 }
3563                 head->shared = is_shared;
3564         } else {
3565                 ret = -EINVAL;
3566                 if (!is_shared || !head->shared) {
3567                         dev_err(ctrl->device,
3568                                 "Duplicate unshared namespace %d\n", nsid);
3569                         goto out_put_ns_head;
3570                 }
3571                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3572                         dev_err(ctrl->device,
3573                                 "IDs don't match for shared namespace %d\n",
3574                                         nsid);
3575                         goto out_put_ns_head;
3576                 }
3577         }
3578
3579         list_add_tail_rcu(&ns->siblings, &head->list);
3580         ns->head = head;
3581         mutex_unlock(&ctrl->subsys->lock);
3582         return 0;
3583
3584 out_put_ns_head:
3585         nvme_put_ns_head(head);
3586 out_unlock:
3587         mutex_unlock(&ctrl->subsys->lock);
3588         return ret;
3589 }
3590
3591 static int ns_cmp(void *priv, const struct list_head *a,
3592                 const struct list_head *b)
3593 {
3594         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3595         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3596
3597         return nsa->head->ns_id - nsb->head->ns_id;
3598 }
3599
3600 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3601 {
3602         struct nvme_ns *ns, *ret = NULL;
3603
3604         down_read(&ctrl->namespaces_rwsem);
3605         list_for_each_entry(ns, &ctrl->namespaces, list) {
3606                 if (ns->head->ns_id == nsid) {
3607                         if (!kref_get_unless_zero(&ns->kref))
3608                                 continue;
3609                         ret = ns;
3610                         break;
3611                 }
3612                 if (ns->head->ns_id > nsid)
3613                         break;
3614         }
3615         up_read(&ctrl->namespaces_rwsem);
3616         return ret;
3617 }
3618 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3619
3620 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3621                 struct nvme_ns_ids *ids)
3622 {
3623         struct nvme_ns *ns;
3624         struct gendisk *disk;
3625         struct nvme_id_ns *id;
3626         int node = ctrl->numa_node;
3627
3628         if (nvme_identify_ns(ctrl, nsid, ids, &id))
3629                 return;
3630
3631         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3632         if (!ns)
3633                 goto out_free_id;
3634
3635         ns->queue = blk_mq_init_queue(ctrl->tagset);
3636         if (IS_ERR(ns->queue))
3637                 goto out_free_ns;
3638
3639         if (ctrl->opts && ctrl->opts->data_digest)
3640                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3641
3642         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3643         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3644                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3645
3646         ns->queue->queuedata = ns;
3647         ns->ctrl = ctrl;
3648         kref_init(&ns->kref);
3649
3650         if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3651                 goto out_free_queue;
3652
3653         disk = alloc_disk_node(0, node);
3654         if (!disk)
3655                 goto out_unlink_ns;
3656
3657         disk->fops = &nvme_bdev_ops;
3658         disk->private_data = ns;
3659         disk->queue = ns->queue;
3660         disk->flags = GENHD_FL_EXT_DEVT;
3661         /*
3662          * Without the multipath code enabled, multiple controller per
3663          * subsystems are visible as devices and thus we cannot use the
3664          * subsystem instance.
3665          */
3666         if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
3667                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3668                         ns->head->instance);
3669         ns->disk = disk;
3670
3671         if (nvme_update_ns_info(ns, id))
3672                 goto out_put_disk;
3673
3674         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3675                 if (nvme_nvm_register(ns, disk->disk_name, node)) {
3676                         dev_warn(ctrl->device, "LightNVM init failure\n");
3677                         goto out_put_disk;
3678                 }
3679         }
3680
3681         down_write(&ctrl->namespaces_rwsem);
3682         list_add_tail(&ns->list, &ctrl->namespaces);
3683         up_write(&ctrl->namespaces_rwsem);
3684
3685         nvme_get_ctrl(ctrl);
3686
3687         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3688         if (!nvme_ns_head_multipath(ns->head))
3689                 nvme_add_ns_cdev(ns);
3690
3691         nvme_mpath_add_disk(ns, id);
3692         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3693         kfree(id);
3694
3695         return;
3696  out_put_disk:
3697         /* prevent double queue cleanup */
3698         ns->disk->queue = NULL;
3699         put_disk(ns->disk);
3700  out_unlink_ns:
3701         mutex_lock(&ctrl->subsys->lock);
3702         list_del_rcu(&ns->siblings);
3703         if (list_empty(&ns->head->list))
3704                 list_del_init(&ns->head->entry);
3705         mutex_unlock(&ctrl->subsys->lock);
3706         nvme_put_ns_head(ns->head);
3707  out_free_queue:
3708         blk_cleanup_queue(ns->queue);
3709  out_free_ns:
3710         kfree(ns);
3711  out_free_id:
3712         kfree(id);
3713 }
3714
3715 static void nvme_ns_remove(struct nvme_ns *ns)
3716 {
3717         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3718                 return;
3719
3720         set_capacity(ns->disk, 0);
3721         nvme_fault_inject_fini(&ns->fault_inject);
3722
3723         mutex_lock(&ns->ctrl->subsys->lock);
3724         list_del_rcu(&ns->siblings);
3725         if (list_empty(&ns->head->list))
3726                 list_del_init(&ns->head->entry);
3727         mutex_unlock(&ns->ctrl->subsys->lock);
3728
3729         synchronize_rcu(); /* guarantee not available in head->list */
3730         nvme_mpath_clear_current_path(ns);
3731         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3732
3733         if (ns->disk->flags & GENHD_FL_UP) {
3734                 if (!nvme_ns_head_multipath(ns->head))
3735                         nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3736                 del_gendisk(ns->disk);
3737                 blk_cleanup_queue(ns->queue);
3738                 if (blk_get_integrity(ns->disk))
3739                         blk_integrity_unregister(ns->disk);
3740         }
3741
3742         down_write(&ns->ctrl->namespaces_rwsem);
3743         list_del_init(&ns->list);
3744         up_write(&ns->ctrl->namespaces_rwsem);
3745
3746         nvme_mpath_check_last_path(ns);
3747         nvme_put_ns(ns);
3748 }
3749
3750 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3751 {
3752         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3753
3754         if (ns) {
3755                 nvme_ns_remove(ns);
3756                 nvme_put_ns(ns);
3757         }
3758 }
3759
3760 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3761 {
3762         struct nvme_id_ns *id;
3763         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3764
3765         if (test_bit(NVME_NS_DEAD, &ns->flags))
3766                 goto out;
3767
3768         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3769         if (ret)
3770                 goto out;
3771
3772         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3773         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3774                 dev_err(ns->ctrl->device,
3775                         "identifiers changed for nsid %d\n", ns->head->ns_id);
3776                 goto out_free_id;
3777         }
3778
3779         ret = nvme_update_ns_info(ns, id);
3780
3781 out_free_id:
3782         kfree(id);
3783 out:
3784         /*
3785          * Only remove the namespace if we got a fatal error back from the
3786          * device, otherwise ignore the error and just move on.
3787          *
3788          * TODO: we should probably schedule a delayed retry here.
3789          */
3790         if (ret > 0 && (ret & NVME_SC_DNR))
3791                 nvme_ns_remove(ns);
3792 }
3793
3794 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3795 {
3796         struct nvme_ns_ids ids = { };
3797         struct nvme_ns *ns;
3798
3799         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
3800                 return;
3801
3802         ns = nvme_find_get_ns(ctrl, nsid);
3803         if (ns) {
3804                 nvme_validate_ns(ns, &ids);
3805                 nvme_put_ns(ns);
3806                 return;
3807         }
3808
3809         switch (ids.csi) {
3810         case NVME_CSI_NVM:
3811                 nvme_alloc_ns(ctrl, nsid, &ids);
3812                 break;
3813         case NVME_CSI_ZNS:
3814                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
3815                         dev_warn(ctrl->device,
3816                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
3817                                 nsid);
3818                         break;
3819                 }
3820                 if (!nvme_multi_css(ctrl)) {
3821                         dev_warn(ctrl->device,
3822                                 "command set not reported for nsid: %d\n",
3823                                 nsid);
3824                         break;
3825                 }
3826                 nvme_alloc_ns(ctrl, nsid, &ids);
3827                 break;
3828         default:
3829                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
3830                         ids.csi, nsid);
3831                 break;
3832         }
3833 }
3834
3835 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3836                                         unsigned nsid)
3837 {
3838         struct nvme_ns *ns, *next;
3839         LIST_HEAD(rm_list);
3840
3841         down_write(&ctrl->namespaces_rwsem);
3842         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3843                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3844                         list_move_tail(&ns->list, &rm_list);
3845         }
3846         up_write(&ctrl->namespaces_rwsem);
3847
3848         list_for_each_entry_safe(ns, next, &rm_list, list)
3849                 nvme_ns_remove(ns);
3850
3851 }
3852
3853 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3854 {
3855         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3856         __le32 *ns_list;
3857         u32 prev = 0;
3858         int ret = 0, i;
3859
3860         if (nvme_ctrl_limited_cns(ctrl))
3861                 return -EOPNOTSUPP;
3862
3863         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3864         if (!ns_list)
3865                 return -ENOMEM;
3866
3867         for (;;) {
3868                 struct nvme_command cmd = {
3869                         .identify.opcode        = nvme_admin_identify,
3870                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
3871                         .identify.nsid          = cpu_to_le32(prev),
3872                 };
3873
3874                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3875                                             NVME_IDENTIFY_DATA_SIZE);
3876                 if (ret) {
3877                         dev_warn(ctrl->device,
3878                                 "Identify NS List failed (status=0x%x)\n", ret);
3879                         goto free;
3880                 }
3881
3882                 for (i = 0; i < nr_entries; i++) {
3883                         u32 nsid = le32_to_cpu(ns_list[i]);
3884
3885                         if (!nsid)      /* end of the list? */
3886                                 goto out;
3887                         nvme_validate_or_alloc_ns(ctrl, nsid);
3888                         while (++prev < nsid)
3889                                 nvme_ns_remove_by_nsid(ctrl, prev);
3890                 }
3891         }
3892  out:
3893         nvme_remove_invalid_namespaces(ctrl, prev);
3894  free:
3895         kfree(ns_list);
3896         return ret;
3897 }
3898
3899 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3900 {
3901         struct nvme_id_ctrl *id;
3902         u32 nn, i;
3903
3904         if (nvme_identify_ctrl(ctrl, &id))
3905                 return;
3906         nn = le32_to_cpu(id->nn);
3907         kfree(id);
3908
3909         for (i = 1; i <= nn; i++)
3910                 nvme_validate_or_alloc_ns(ctrl, i);
3911
3912         nvme_remove_invalid_namespaces(ctrl, nn);
3913 }
3914
3915 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3916 {
3917         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3918         __le32 *log;
3919         int error;
3920
3921         log = kzalloc(log_size, GFP_KERNEL);
3922         if (!log)
3923                 return;
3924
3925         /*
3926          * We need to read the log to clear the AEN, but we don't want to rely
3927          * on it for the changed namespace information as userspace could have
3928          * raced with us in reading the log page, which could cause us to miss
3929          * updates.
3930          */
3931         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
3932                         NVME_CSI_NVM, log, log_size, 0);
3933         if (error)
3934                 dev_warn(ctrl->device,
3935                         "reading changed ns log failed: %d\n", error);
3936
3937         kfree(log);
3938 }
3939
3940 static void nvme_scan_work(struct work_struct *work)
3941 {
3942         struct nvme_ctrl *ctrl =
3943                 container_of(work, struct nvme_ctrl, scan_work);
3944
3945         /* No tagset on a live ctrl means IO queues could not created */
3946         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3947                 return;
3948
3949         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3950                 dev_info(ctrl->device, "rescanning namespaces.\n");
3951                 nvme_clear_changed_ns_log(ctrl);
3952         }
3953
3954         mutex_lock(&ctrl->scan_lock);
3955         if (nvme_scan_ns_list(ctrl) != 0)
3956                 nvme_scan_ns_sequential(ctrl);
3957         mutex_unlock(&ctrl->scan_lock);
3958
3959         down_write(&ctrl->namespaces_rwsem);
3960         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3961         up_write(&ctrl->namespaces_rwsem);
3962 }
3963
3964 /*
3965  * This function iterates the namespace list unlocked to allow recovery from
3966  * controller failure. It is up to the caller to ensure the namespace list is
3967  * not modified by scan work while this function is executing.
3968  */
3969 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3970 {
3971         struct nvme_ns *ns, *next;
3972         LIST_HEAD(ns_list);
3973
3974         /*
3975          * make sure to requeue I/O to all namespaces as these
3976          * might result from the scan itself and must complete
3977          * for the scan_work to make progress
3978          */
3979         nvme_mpath_clear_ctrl_paths(ctrl);
3980
3981         /* prevent racing with ns scanning */
3982         flush_work(&ctrl->scan_work);
3983
3984         /*
3985          * The dead states indicates the controller was not gracefully
3986          * disconnected. In that case, we won't be able to flush any data while
3987          * removing the namespaces' disks; fail all the queues now to avoid
3988          * potentially having to clean up the failed sync later.
3989          */
3990         if (ctrl->state == NVME_CTRL_DEAD)
3991                 nvme_kill_queues(ctrl);
3992
3993         /* this is a no-op when called from the controller reset handler */
3994         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
3995
3996         down_write(&ctrl->namespaces_rwsem);
3997         list_splice_init(&ctrl->namespaces, &ns_list);
3998         up_write(&ctrl->namespaces_rwsem);
3999
4000         list_for_each_entry_safe(ns, next, &ns_list, list)
4001                 nvme_ns_remove(ns);
4002 }
4003 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4004
4005 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4006 {
4007         struct nvme_ctrl *ctrl =
4008                 container_of(dev, struct nvme_ctrl, ctrl_device);
4009         struct nvmf_ctrl_options *opts = ctrl->opts;
4010         int ret;
4011
4012         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4013         if (ret)
4014                 return ret;
4015
4016         if (opts) {
4017                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4018                 if (ret)
4019                         return ret;
4020
4021                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4022                                 opts->trsvcid ?: "none");
4023                 if (ret)
4024                         return ret;
4025
4026                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4027                                 opts->host_traddr ?: "none");
4028         }
4029         return ret;
4030 }
4031
4032 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4033 {
4034         char *envp[2] = { NULL, NULL };
4035         u32 aen_result = ctrl->aen_result;
4036
4037         ctrl->aen_result = 0;
4038         if (!aen_result)
4039                 return;
4040
4041         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4042         if (!envp[0])
4043                 return;
4044         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4045         kfree(envp[0]);
4046 }
4047
4048 static void nvme_async_event_work(struct work_struct *work)
4049 {
4050         struct nvme_ctrl *ctrl =
4051                 container_of(work, struct nvme_ctrl, async_event_work);
4052
4053         nvme_aen_uevent(ctrl);
4054         ctrl->ops->submit_async_event(ctrl);
4055 }
4056
4057 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4058 {
4059
4060         u32 csts;
4061
4062         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4063                 return false;
4064
4065         if (csts == ~0)
4066                 return false;
4067
4068         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4069 }
4070
4071 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4072 {
4073         struct nvme_fw_slot_info_log *log;
4074
4075         log = kmalloc(sizeof(*log), GFP_KERNEL);
4076         if (!log)
4077                 return;
4078
4079         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4080                         log, sizeof(*log), 0))
4081                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4082         kfree(log);
4083 }
4084
4085 static void nvme_fw_act_work(struct work_struct *work)
4086 {
4087         struct nvme_ctrl *ctrl = container_of(work,
4088                                 struct nvme_ctrl, fw_act_work);
4089         unsigned long fw_act_timeout;
4090
4091         if (ctrl->mtfa)
4092                 fw_act_timeout = jiffies +
4093                                 msecs_to_jiffies(ctrl->mtfa * 100);
4094         else
4095                 fw_act_timeout = jiffies +
4096                                 msecs_to_jiffies(admin_timeout * 1000);
4097
4098         nvme_stop_queues(ctrl);
4099         while (nvme_ctrl_pp_status(ctrl)) {
4100                 if (time_after(jiffies, fw_act_timeout)) {
4101                         dev_warn(ctrl->device,
4102                                 "Fw activation timeout, reset controller\n");
4103                         nvme_try_sched_reset(ctrl);
4104                         return;
4105                 }
4106                 msleep(100);
4107         }
4108
4109         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4110                 return;
4111
4112         nvme_start_queues(ctrl);
4113         /* read FW slot information to clear the AER */
4114         nvme_get_fw_slot_info(ctrl);
4115 }
4116
4117 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4118 {
4119         u32 aer_notice_type = (result & 0xff00) >> 8;
4120
4121         trace_nvme_async_event(ctrl, aer_notice_type);
4122
4123         switch (aer_notice_type) {
4124         case NVME_AER_NOTICE_NS_CHANGED:
4125                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4126                 nvme_queue_scan(ctrl);
4127                 break;
4128         case NVME_AER_NOTICE_FW_ACT_STARTING:
4129                 /*
4130                  * We are (ab)using the RESETTING state to prevent subsequent
4131                  * recovery actions from interfering with the controller's
4132                  * firmware activation.
4133                  */
4134                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4135                         queue_work(nvme_wq, &ctrl->fw_act_work);
4136                 break;
4137 #ifdef CONFIG_NVME_MULTIPATH
4138         case NVME_AER_NOTICE_ANA:
4139                 if (!ctrl->ana_log_buf)
4140                         break;
4141                 queue_work(nvme_wq, &ctrl->ana_work);
4142                 break;
4143 #endif
4144         case NVME_AER_NOTICE_DISC_CHANGED:
4145                 ctrl->aen_result = result;
4146                 break;
4147         default:
4148                 dev_warn(ctrl->device, "async event result %08x\n", result);
4149         }
4150 }
4151
4152 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4153                 volatile union nvme_result *res)
4154 {
4155         u32 result = le32_to_cpu(res->u32);
4156         u32 aer_type = result & 0x07;
4157
4158         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4159                 return;
4160
4161         switch (aer_type) {
4162         case NVME_AER_NOTICE:
4163                 nvme_handle_aen_notice(ctrl, result);
4164                 break;
4165         case NVME_AER_ERROR:
4166         case NVME_AER_SMART:
4167         case NVME_AER_CSS:
4168         case NVME_AER_VS:
4169                 trace_nvme_async_event(ctrl, aer_type);
4170                 ctrl->aen_result = result;
4171                 break;
4172         default:
4173                 break;
4174         }
4175         queue_work(nvme_wq, &ctrl->async_event_work);
4176 }
4177 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4178
4179 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4180 {
4181         nvme_mpath_stop(ctrl);
4182         nvme_stop_keep_alive(ctrl);
4183         nvme_stop_failfast_work(ctrl);
4184         flush_work(&ctrl->async_event_work);
4185         cancel_work_sync(&ctrl->fw_act_work);
4186 }
4187 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4188
4189 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4190 {
4191         nvme_start_keep_alive(ctrl);
4192
4193         nvme_enable_aen(ctrl);
4194
4195         if (ctrl->queue_count > 1) {
4196                 nvme_queue_scan(ctrl);
4197                 nvme_start_queues(ctrl);
4198         }
4199 }
4200 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4201
4202 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4203 {
4204         nvme_hwmon_exit(ctrl);
4205         nvme_fault_inject_fini(&ctrl->fault_inject);
4206         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4207         cdev_device_del(&ctrl->cdev, ctrl->device);
4208         nvme_put_ctrl(ctrl);
4209 }
4210 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4211
4212 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4213 {
4214         struct nvme_effects_log *cel;
4215         unsigned long i;
4216
4217         xa_for_each(&ctrl->cels, i, cel) {
4218                 xa_erase(&ctrl->cels, i);
4219                 kfree(cel);
4220         }
4221
4222         xa_destroy(&ctrl->cels);
4223 }
4224
4225 static void nvme_free_ctrl(struct device *dev)
4226 {
4227         struct nvme_ctrl *ctrl =
4228                 container_of(dev, struct nvme_ctrl, ctrl_device);
4229         struct nvme_subsystem *subsys = ctrl->subsys;
4230
4231         if (!subsys || ctrl->instance != subsys->instance)
4232                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4233
4234         nvme_free_cels(ctrl);
4235         nvme_mpath_uninit(ctrl);
4236         __free_page(ctrl->discard_page);
4237
4238         if (subsys) {
4239                 mutex_lock(&nvme_subsystems_lock);
4240                 list_del(&ctrl->subsys_entry);
4241                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4242                 mutex_unlock(&nvme_subsystems_lock);
4243         }
4244
4245         ctrl->ops->free_ctrl(ctrl);
4246
4247         if (subsys)
4248                 nvme_put_subsystem(subsys);
4249 }
4250
4251 /*
4252  * Initialize a NVMe controller structures.  This needs to be called during
4253  * earliest initialization so that we have the initialized structured around
4254  * during probing.
4255  */
4256 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4257                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4258 {
4259         int ret;
4260
4261         ctrl->state = NVME_CTRL_NEW;
4262         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4263         spin_lock_init(&ctrl->lock);
4264         mutex_init(&ctrl->scan_lock);
4265         INIT_LIST_HEAD(&ctrl->namespaces);
4266         xa_init(&ctrl->cels);
4267         init_rwsem(&ctrl->namespaces_rwsem);
4268         ctrl->dev = dev;
4269         ctrl->ops = ops;
4270         ctrl->quirks = quirks;
4271         ctrl->numa_node = NUMA_NO_NODE;
4272         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4273         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4274         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4275         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4276         init_waitqueue_head(&ctrl->state_wq);
4277
4278         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4279         INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4280         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4281         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4282
4283         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4284                         PAGE_SIZE);
4285         ctrl->discard_page = alloc_page(GFP_KERNEL);
4286         if (!ctrl->discard_page) {
4287                 ret = -ENOMEM;
4288                 goto out;
4289         }
4290
4291         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4292         if (ret < 0)
4293                 goto out;
4294         ctrl->instance = ret;
4295
4296         device_initialize(&ctrl->ctrl_device);
4297         ctrl->device = &ctrl->ctrl_device;
4298         ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4299                         ctrl->instance);
4300         ctrl->device->class = nvme_class;
4301         ctrl->device->parent = ctrl->dev;
4302         ctrl->device->groups = nvme_dev_attr_groups;
4303         ctrl->device->release = nvme_free_ctrl;
4304         dev_set_drvdata(ctrl->device, ctrl);
4305         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4306         if (ret)
4307                 goto out_release_instance;
4308
4309         nvme_get_ctrl(ctrl);
4310         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4311         ctrl->cdev.owner = ops->module;
4312         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4313         if (ret)
4314                 goto out_free_name;
4315
4316         /*
4317          * Initialize latency tolerance controls.  The sysfs files won't
4318          * be visible to userspace unless the device actually supports APST.
4319          */
4320         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4321         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4322                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4323
4324         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4325
4326         return 0;
4327 out_free_name:
4328         nvme_put_ctrl(ctrl);
4329         kfree_const(ctrl->device->kobj.name);
4330 out_release_instance:
4331         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4332 out:
4333         if (ctrl->discard_page)
4334                 __free_page(ctrl->discard_page);
4335         return ret;
4336 }
4337 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4338
4339 /**
4340  * nvme_kill_queues(): Ends all namespace queues
4341  * @ctrl: the dead controller that needs to end
4342  *
4343  * Call this function when the driver determines it is unable to get the
4344  * controller in a state capable of servicing IO.
4345  */
4346 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4347 {
4348         struct nvme_ns *ns;
4349
4350         down_read(&ctrl->namespaces_rwsem);
4351
4352         /* Forcibly unquiesce queues to avoid blocking dispatch */
4353         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4354                 blk_mq_unquiesce_queue(ctrl->admin_q);
4355
4356         list_for_each_entry(ns, &ctrl->namespaces, list)
4357                 nvme_set_queue_dying(ns);
4358
4359         up_read(&ctrl->namespaces_rwsem);
4360 }
4361 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4362
4363 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4364 {
4365         struct nvme_ns *ns;
4366
4367         down_read(&ctrl->namespaces_rwsem);
4368         list_for_each_entry(ns, &ctrl->namespaces, list)
4369                 blk_mq_unfreeze_queue(ns->queue);
4370         up_read(&ctrl->namespaces_rwsem);
4371 }
4372 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4373
4374 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4375 {
4376         struct nvme_ns *ns;
4377
4378         down_read(&ctrl->namespaces_rwsem);
4379         list_for_each_entry(ns, &ctrl->namespaces, list) {
4380                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4381                 if (timeout <= 0)
4382                         break;
4383         }
4384         up_read(&ctrl->namespaces_rwsem);
4385         return timeout;
4386 }
4387 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4388
4389 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4390 {
4391         struct nvme_ns *ns;
4392
4393         down_read(&ctrl->namespaces_rwsem);
4394         list_for_each_entry(ns, &ctrl->namespaces, list)
4395                 blk_mq_freeze_queue_wait(ns->queue);
4396         up_read(&ctrl->namespaces_rwsem);
4397 }
4398 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4399
4400 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4401 {
4402         struct nvme_ns *ns;
4403
4404         down_read(&ctrl->namespaces_rwsem);
4405         list_for_each_entry(ns, &ctrl->namespaces, list)
4406                 blk_freeze_queue_start(ns->queue);
4407         up_read(&ctrl->namespaces_rwsem);
4408 }
4409 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4410
4411 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4412 {
4413         struct nvme_ns *ns;
4414
4415         down_read(&ctrl->namespaces_rwsem);
4416         list_for_each_entry(ns, &ctrl->namespaces, list)
4417                 blk_mq_quiesce_queue(ns->queue);
4418         up_read(&ctrl->namespaces_rwsem);
4419 }
4420 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4421
4422 void nvme_start_queues(struct nvme_ctrl *ctrl)
4423 {
4424         struct nvme_ns *ns;
4425
4426         down_read(&ctrl->namespaces_rwsem);
4427         list_for_each_entry(ns, &ctrl->namespaces, list)
4428                 blk_mq_unquiesce_queue(ns->queue);
4429         up_read(&ctrl->namespaces_rwsem);
4430 }
4431 EXPORT_SYMBOL_GPL(nvme_start_queues);
4432
4433 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4434 {
4435         struct nvme_ns *ns;
4436
4437         down_read(&ctrl->namespaces_rwsem);
4438         list_for_each_entry(ns, &ctrl->namespaces, list)
4439                 blk_sync_queue(ns->queue);
4440         up_read(&ctrl->namespaces_rwsem);
4441 }
4442 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4443
4444 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4445 {
4446         nvme_sync_io_queues(ctrl);
4447         if (ctrl->admin_q)
4448                 blk_sync_queue(ctrl->admin_q);
4449 }
4450 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4451
4452 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4453 {
4454         if (file->f_op != &nvme_dev_fops)
4455                 return NULL;
4456         return file->private_data;
4457 }
4458 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4459
4460 /*
4461  * Check we didn't inadvertently grow the command structure sizes:
4462  */
4463 static inline void _nvme_check_size(void)
4464 {
4465         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4466         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4467         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4468         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4469         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4470         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4471         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4472         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4473         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4474         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4475         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4476         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4477         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4478         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4479         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4480         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4481         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4482         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4483         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4484         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4485 }
4486
4487
4488 static int __init nvme_core_init(void)
4489 {
4490         int result = -ENOMEM;
4491
4492         _nvme_check_size();
4493
4494         nvme_wq = alloc_workqueue("nvme-wq",
4495                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4496         if (!nvme_wq)
4497                 goto out;
4498
4499         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4500                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4501         if (!nvme_reset_wq)
4502                 goto destroy_wq;
4503
4504         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4505                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4506         if (!nvme_delete_wq)
4507                 goto destroy_reset_wq;
4508
4509         result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4510                         NVME_MINORS, "nvme");
4511         if (result < 0)
4512                 goto destroy_delete_wq;
4513
4514         nvme_class = class_create(THIS_MODULE, "nvme");
4515         if (IS_ERR(nvme_class)) {
4516                 result = PTR_ERR(nvme_class);
4517                 goto unregister_chrdev;
4518         }
4519         nvme_class->dev_uevent = nvme_class_uevent;
4520
4521         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4522         if (IS_ERR(nvme_subsys_class)) {
4523                 result = PTR_ERR(nvme_subsys_class);
4524                 goto destroy_class;
4525         }
4526
4527         result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4528                                      "nvme-generic");
4529         if (result < 0)
4530                 goto destroy_subsys_class;
4531
4532         nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4533         if (IS_ERR(nvme_ns_chr_class)) {
4534                 result = PTR_ERR(nvme_ns_chr_class);
4535                 goto unregister_generic_ns;
4536         }
4537
4538         return 0;
4539
4540 unregister_generic_ns:
4541         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4542 destroy_subsys_class:
4543         class_destroy(nvme_subsys_class);
4544 destroy_class:
4545         class_destroy(nvme_class);
4546 unregister_chrdev:
4547         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4548 destroy_delete_wq:
4549         destroy_workqueue(nvme_delete_wq);
4550 destroy_reset_wq:
4551         destroy_workqueue(nvme_reset_wq);
4552 destroy_wq:
4553         destroy_workqueue(nvme_wq);
4554 out:
4555         return result;
4556 }
4557
4558 static void __exit nvme_core_exit(void)
4559 {
4560         class_destroy(nvme_ns_chr_class);
4561         class_destroy(nvme_subsys_class);
4562         class_destroy(nvme_class);
4563         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4564         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4565         destroy_workqueue(nvme_delete_wq);
4566         destroy_workqueue(nvme_reset_wq);
4567         destroy_workqueue(nvme_wq);
4568         ida_destroy(&nvme_ns_chr_minor_ida);
4569         ida_destroy(&nvme_instance_ida);
4570 }
4571
4572 MODULE_LICENSE("GPL");
4573 MODULE_VERSION("1.0");
4574 module_init(nvme_core_init);
4575 module_exit(nvme_core_exit);