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 void nvme_put_subsystem(struct nvme_subsystem *subsys);
  93 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  94                                            unsigned nsid);
  95
  96 /*
  97  * Prepare a queue for teardown.
  98  *
  99  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
 100  * the capacity to 0 after that to avoid blocking dispatchers that may be
 101  * holding bd_butex.  This will end buffered writers dirtying pages that can't
 102  * be synced.
 103  */
 104 static void nvme_set_queue_dying(struct nvme_ns *ns)
 105 {
 106         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 107                 return;
 108
 109         blk_set_queue_dying(ns->queue);
 110         blk_mq_unquiesce_queue(ns->queue);
 111
 112         set_capacity_and_notify(ns->disk, 0);
 113 }
 114
 115 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 116 {
 117         /*
 118          * Only new queue scan work when admin and IO queues are both alive
 119          */
 120         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 121                 queue_work(nvme_wq, &ctrl->scan_work);
 122 }
 123
 124 /*
 125  * Use this function to proceed with scheduling reset_work for a controller
 126  * that had previously been set to the resetting state. This is intended for
 127  * code paths that can't be interrupted by other reset attempts. A hot removal
 128  * may prevent this from succeeding.
 129  */
 130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 131 {
 132         if (ctrl->state != NVME_CTRL_RESETTING)
 133                 return -EBUSY;
 134         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 135                 return -EBUSY;
 136         return 0;
 137 }
 138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 139
 140 static void nvme_failfast_work(struct work_struct *work)
 141 {
 142         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 143                         struct nvme_ctrl, failfast_work);
 144
 145         if (ctrl->state != NVME_CTRL_CONNECTING)
 146                 return;
 147
 148         set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 149         dev_info(ctrl->device, "failfast expired\n");
 150         nvme_kick_requeue_lists(ctrl);
 151 }
 152
 153 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 154 {
 155         if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 156                 return;
 157
 158         schedule_delayed_work(&ctrl->failfast_work,
 159                               ctrl->opts->fast_io_fail_tmo * HZ);
 160 }
 161
 162 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 163 {
 164         if (!ctrl->opts)
 165                 return;
 166
 167         cancel_delayed_work_sync(&ctrl->failfast_work);
 168         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 169 }
 170
 171
 172 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 173 {
 174         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 175                 return -EBUSY;
 176         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 177                 return -EBUSY;
 178         return 0;
 179 }
 180 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 181
 182 static int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 183 {
 184         int ret;
 185
 186         ret = nvme_reset_ctrl(ctrl);
 187         if (!ret) {
 188                 flush_work(&ctrl->reset_work);
 189                 if (ctrl->state != NVME_CTRL_LIVE)
 190                         ret = -ENETRESET;
 191         }
 192
 193         return ret;
 194 }
 195
 196 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 197 {
 198         dev_info(ctrl->device,
 199                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 200
 201         flush_work(&ctrl->reset_work);
 202         nvme_stop_ctrl(ctrl);
 203         nvme_remove_namespaces(ctrl);
 204         ctrl->ops->delete_ctrl(ctrl);
 205         nvme_uninit_ctrl(ctrl);
 206 }
 207
 208 static void nvme_delete_ctrl_work(struct work_struct *work)
 209 {
 210         struct nvme_ctrl *ctrl =
 211                 container_of(work, struct nvme_ctrl, delete_work);
 212
 213         nvme_do_delete_ctrl(ctrl);
 214 }
 215
 216 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 217 {
 218         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 219                 return -EBUSY;
 220         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 221                 return -EBUSY;
 222         return 0;
 223 }
 224 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 225
 226 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 227 {
 228         /*
 229          * Keep a reference until nvme_do_delete_ctrl() complete,
 230          * since ->delete_ctrl can free the controller.
 231          */
 232         nvme_get_ctrl(ctrl);
 233         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 234                 nvme_do_delete_ctrl(ctrl);
 235         nvme_put_ctrl(ctrl);
 236 }
 237
 238 static blk_status_t nvme_error_status(u16 status)
 239 {
 240         switch (status & 0x7ff) {
 241         case NVME_SC_SUCCESS:
 242                 return BLK_STS_OK;
 243         case NVME_SC_CAP_EXCEEDED:
 244                 return BLK_STS_NOSPC;
 245         case NVME_SC_LBA_RANGE:
 246         case NVME_SC_CMD_INTERRUPTED:
 247         case NVME_SC_NS_NOT_READY:
 248                 return BLK_STS_TARGET;
 249         case NVME_SC_BAD_ATTRIBUTES:
 250         case NVME_SC_ONCS_NOT_SUPPORTED:
 251         case NVME_SC_INVALID_OPCODE:
 252         case NVME_SC_INVALID_FIELD:
 253         case NVME_SC_INVALID_NS:
 254                 return BLK_STS_NOTSUPP;
 255         case NVME_SC_WRITE_FAULT:
 256         case NVME_SC_READ_ERROR:
 257         case NVME_SC_UNWRITTEN_BLOCK:
 258         case NVME_SC_ACCESS_DENIED:
 259         case NVME_SC_READ_ONLY:
 260         case NVME_SC_COMPARE_FAILED:
 261                 return BLK_STS_MEDIUM;
 262         case NVME_SC_GUARD_CHECK:
 263         case NVME_SC_APPTAG_CHECK:
 264         case NVME_SC_REFTAG_CHECK:
 265         case NVME_SC_INVALID_PI:
 266                 return BLK_STS_PROTECTION;
 267         case NVME_SC_RESERVATION_CONFLICT:
 268                 return BLK_STS_NEXUS;
 269         case NVME_SC_HOST_PATH_ERROR:
 270                 return BLK_STS_TRANSPORT;
 271         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 272                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 273         case NVME_SC_ZONE_TOO_MANY_OPEN:
 274                 return BLK_STS_ZONE_OPEN_RESOURCE;
 275         default:
 276                 return BLK_STS_IOERR;
 277         }
 278 }
 279
 280 static void nvme_retry_req(struct request *req)
 281 {
 282         unsigned long delay = 0;
 283         u16 crd;
 284
 285         /* The mask and shift result must be <= 3 */
 286         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 287         if (crd)
 288                 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 289
 290         nvme_req(req)->retries++;
 291         blk_mq_requeue_request(req, false);
 292         blk_mq_delay_kick_requeue_list(req->q, delay);
 293 }
 294
 295 enum nvme_disposition {
 296         COMPLETE,
 297         RETRY,
 298         FAILOVER,
 299 };
 300
 301 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 302 {
 303         if (likely(nvme_req(req)->status == 0))
 304                 return COMPLETE;
 305
 306         if (blk_noretry_request(req) ||
 307             (nvme_req(req)->status & NVME_SC_DNR) ||
 308             nvme_req(req)->retries >= nvme_max_retries)
 309                 return COMPLETE;
 310
 311         if (req->cmd_flags & REQ_NVME_MPATH) {
 312                 if (nvme_is_path_error(nvme_req(req)->status) ||
 313                     blk_queue_dying(req->q))
 314                         return FAILOVER;
 315         } else {
 316                 if (blk_queue_dying(req->q))
 317                         return COMPLETE;
 318         }
 319
 320         return RETRY;
 321 }
 322
 323 static inline void nvme_end_req(struct request *req)
 324 {
 325         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 326
 327         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 328             req_op(req) == REQ_OP_ZONE_APPEND)
 329                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 330                         le64_to_cpu(nvme_req(req)->result.u64));
 331
 332         nvme_trace_bio_complete(req);
 333         blk_mq_end_request(req, status);
 334 }
 335
 336 void nvme_complete_rq(struct request *req)
 337 {
 338         trace_nvme_complete_rq(req);
 339         nvme_cleanup_cmd(req);
 340
 341         if (nvme_req(req)->ctrl->kas)
 342                 nvme_req(req)->ctrl->comp_seen = true;
 343
 344         switch (nvme_decide_disposition(req)) {
 345         case COMPLETE:
 346                 nvme_end_req(req);
 347                 return;
 348         case RETRY:
 349                 nvme_retry_req(req);
 350                 return;
 351         case FAILOVER:
 352                 nvme_failover_req(req);
 353                 return;
 354         }
 355 }
 356 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 357
 358 /*
 359  * Called to unwind from ->queue_rq on a failed command submission so that the
 360  * multipathing code gets called to potentially failover to another path.
 361  * The caller needs to unwind all transport specific resource allocations and
 362  * must return propagate the return value.
 363  */
 364 blk_status_t nvme_host_path_error(struct request *req)
 365 {
 366         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 367         blk_mq_set_request_complete(req);
 368         nvme_complete_rq(req);
 369         return BLK_STS_OK;
 370 }
 371 EXPORT_SYMBOL_GPL(nvme_host_path_error);
 372
 373 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 374 {
 375         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 376                                 "Cancelling I/O %d", req->tag);
 377
 378         /* don't abort one completed request */
 379         if (blk_mq_request_completed(req))
 380                 return true;
 381
 382         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 383         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 384         blk_mq_complete_request(req);
 385         return true;
 386 }
 387 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 388
 389 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 390 {
 391         if (ctrl->tagset) {
 392                 blk_mq_tagset_busy_iter(ctrl->tagset,
 393                                 nvme_cancel_request, ctrl);
 394                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 395         }
 396 }
 397 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 398
 399 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 400 {
 401         if (ctrl->admin_tagset) {
 402                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 403                                 nvme_cancel_request, ctrl);
 404                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 405         }
 406 }
 407 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 408
 409 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 410                 enum nvme_ctrl_state new_state)
 411 {
 412         enum nvme_ctrl_state old_state;
 413         unsigned long flags;
 414         bool changed = false;
 415
 416         spin_lock_irqsave(&ctrl->lock, flags);
 417
 418         old_state = ctrl->state;
 419         switch (new_state) {
 420         case NVME_CTRL_LIVE:
 421                 switch (old_state) {
 422                 case NVME_CTRL_NEW:
 423                 case NVME_CTRL_RESETTING:
 424                 case NVME_CTRL_CONNECTING:
 425                         changed = true;
 426                         fallthrough;
 427                 default:
 428                         break;
 429                 }
 430                 break;
 431         case NVME_CTRL_RESETTING:
 432                 switch (old_state) {
 433                 case NVME_CTRL_NEW:
 434                 case NVME_CTRL_LIVE:
 435                         changed = true;
 436                         fallthrough;
 437                 default:
 438                         break;
 439                 }
 440                 break;
 441         case NVME_CTRL_CONNECTING:
 442                 switch (old_state) {
 443                 case NVME_CTRL_NEW:
 444                 case NVME_CTRL_RESETTING:
 445                         changed = true;
 446                         fallthrough;
 447                 default:
 448                         break;
 449                 }
 450                 break;
 451         case NVME_CTRL_DELETING:
 452                 switch (old_state) {
 453                 case NVME_CTRL_LIVE:
 454                 case NVME_CTRL_RESETTING:
 455                 case NVME_CTRL_CONNECTING:
 456                         changed = true;
 457                         fallthrough;
 458                 default:
 459                         break;
 460                 }
 461                 break;
 462         case NVME_CTRL_DELETING_NOIO:
 463                 switch (old_state) {
 464                 case NVME_CTRL_DELETING:
 465                 case NVME_CTRL_DEAD:
 466                         changed = true;
 467                         fallthrough;
 468                 default:
 469                         break;
 470                 }
 471                 break;
 472         case NVME_CTRL_DEAD:
 473                 switch (old_state) {
 474                 case NVME_CTRL_DELETING:
 475                         changed = true;
 476                         fallthrough;
 477                 default:
 478                         break;
 479                 }
 480                 break;
 481         default:
 482                 break;
 483         }
 484
 485         if (changed) {
 486                 ctrl->state = new_state;
 487                 wake_up_all(&ctrl->state_wq);
 488         }
 489
 490         spin_unlock_irqrestore(&ctrl->lock, flags);
 491         if (!changed)
 492                 return false;
 493
 494         if (ctrl->state == NVME_CTRL_LIVE) {
 495                 if (old_state == NVME_CTRL_CONNECTING)
 496                         nvme_stop_failfast_work(ctrl);
 497                 nvme_kick_requeue_lists(ctrl);
 498         } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 499                 old_state == NVME_CTRL_RESETTING) {
 500                 nvme_start_failfast_work(ctrl);
 501         }
 502         return changed;
 503 }
 504 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 505
 506 /*
 507  * Returns true for sink states that can't ever transition back to live.
 508  */
 509 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 510 {
 511         switch (ctrl->state) {
 512         case NVME_CTRL_NEW:
 513         case NVME_CTRL_LIVE:
 514         case NVME_CTRL_RESETTING:
 515         case NVME_CTRL_CONNECTING:
 516                 return false;
 517         case NVME_CTRL_DELETING:
 518         case NVME_CTRL_DELETING_NOIO:
 519         case NVME_CTRL_DEAD:
 520                 return true;
 521         default:
 522                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 523                 return true;
 524         }
 525 }
 526
 527 /*
 528  * Waits for the controller state to be resetting, or returns false if it is
 529  * not possible to ever transition to that state.
 530  */
 531 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 532 {
 533         wait_event(ctrl->state_wq,
 534                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 535                    nvme_state_terminal(ctrl));
 536         return ctrl->state == NVME_CTRL_RESETTING;
 537 }
 538 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 539
 540 static void nvme_free_ns_head(struct kref *ref)
 541 {
 542         struct nvme_ns_head *head =
 543                 container_of(ref, struct nvme_ns_head, ref);
 544
 545         nvme_mpath_remove_disk(head);
 546         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 547         cleanup_srcu_struct(&head->srcu);
 548         nvme_put_subsystem(head->subsys);
 549         kfree(head);
 550 }
 551
 552 static void nvme_put_ns_head(struct nvme_ns_head *head)
 553 {
 554         kref_put(&head->ref, nvme_free_ns_head);
 555 }
 556
 557 static void nvme_free_ns(struct kref *kref)
 558 {
 559         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 560
 561         if (ns->ndev)
 562                 nvme_nvm_unregister(ns);
 563
 564         put_disk(ns->disk);
 565         nvme_put_ns_head(ns->head);
 566         nvme_put_ctrl(ns->ctrl);
 567         kfree(ns);
 568 }
 569
 570 void nvme_put_ns(struct nvme_ns *ns)
 571 {
 572         kref_put(&ns->kref, nvme_free_ns);
 573 }
 574 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 575
 576 static inline void nvme_clear_nvme_request(struct request *req)
 577 {
 578         struct nvme_command *cmd = nvme_req(req)->cmd;
 579
 580         memset(cmd, 0, sizeof(*cmd));
 581         nvme_req(req)->retries = 0;
 582         nvme_req(req)->flags = 0;
 583         req->rq_flags |= RQF_DONTPREP;
 584 }
 585
 586 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
 587 {
 588         return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 589 }
 590
 591 static inline void nvme_init_request(struct request *req,
 592                 struct nvme_command *cmd)
 593 {
 594         if (req->q->queuedata)
 595                 req->timeout = NVME_IO_TIMEOUT;
 596         else /* no queuedata implies admin queue */
 597                 req->timeout = NVME_ADMIN_TIMEOUT;
 598
 599         /* passthru commands should let the driver set the SGL flags */
 600         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 601
 602         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 603         nvme_clear_nvme_request(req);
 604         memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
 605 }
 606
 607 struct request *nvme_alloc_request(struct request_queue *q,
 608                 struct nvme_command *cmd, blk_mq_req_flags_t flags)
 609 {
 610         struct request *req;
 611
 612         req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
 613         if (!IS_ERR(req))
 614                 nvme_init_request(req, cmd);
 615         return req;
 616 }
 617 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 618
 619 static struct request *nvme_alloc_request_qid(struct request_queue *q,
 620                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 621 {
 622         struct request *req;
 623
 624         req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
 625                         qid ? qid - 1 : 0);
 626         if (!IS_ERR(req))
 627                 nvme_init_request(req, cmd);
 628         return req;
 629 }
 630
 631 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 632 {
 633         struct nvme_command c;
 634
 635         memset(&c, 0, sizeof(c));
 636
 637         c.directive.opcode = nvme_admin_directive_send;
 638         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 639         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 640         c.directive.dtype = NVME_DIR_IDENTIFY;
 641         c.directive.tdtype = NVME_DIR_STREAMS;
 642         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 643
 644         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 645 }
 646
 647 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 648 {
 649         return nvme_toggle_streams(ctrl, false);
 650 }
 651
 652 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 653 {
 654         return nvme_toggle_streams(ctrl, true);
 655 }
 656
 657 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 658                                   struct streams_directive_params *s, u32 nsid)
 659 {
 660         struct nvme_command c;
 661
 662         memset(&c, 0, sizeof(c));
 663         memset(s, 0, sizeof(*s));
 664
 665         c.directive.opcode = nvme_admin_directive_recv;
 666         c.directive.nsid = cpu_to_le32(nsid);
 667         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 668         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 669         c.directive.dtype = NVME_DIR_STREAMS;
 670
 671         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 672 }
 673
 674 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 675 {
 676         struct streams_directive_params s;
 677         int ret;
 678
 679         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 680                 return 0;
 681         if (!streams)
 682                 return 0;
 683
 684         ret = nvme_enable_streams(ctrl);
 685         if (ret)
 686                 return ret;
 687
 688         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 689         if (ret)
 690                 goto out_disable_stream;
 691
 692         ctrl->nssa = le16_to_cpu(s.nssa);
 693         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 694                 dev_info(ctrl->device, "too few streams (%u) available\n",
 695                                         ctrl->nssa);
 696                 goto out_disable_stream;
 697         }
 698
 699         ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 700         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 701         return 0;
 702
 703 out_disable_stream:
 704         nvme_disable_streams(ctrl);
 705         return ret;
 706 }
 707
 708 /*
 709  * Check if 'req' has a write hint associated with it. If it does, assign
 710  * a valid namespace stream to the write.
 711  */
 712 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 713                                      struct request *req, u16 *control,
 714                                      u32 *dsmgmt)
 715 {
 716         enum rw_hint streamid = req->write_hint;
 717
 718         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 719                 streamid = 0;
 720         else {
 721                 streamid--;
 722                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 723                         return;
 724
 725                 *control |= NVME_RW_DTYPE_STREAMS;
 726                 *dsmgmt |= streamid << 16;
 727         }
 728
 729         if (streamid < ARRAY_SIZE(req->q->write_hints))
 730                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 731 }
 732
 733 static inline void nvme_setup_flush(struct nvme_ns *ns,
 734                 struct nvme_command *cmnd)
 735 {
 736         cmnd->common.opcode = nvme_cmd_flush;
 737         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 738 }
 739
 740 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 741                 struct nvme_command *cmnd)
 742 {
 743         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 744         struct nvme_dsm_range *range;
 745         struct bio *bio;
 746
 747         /*
 748          * Some devices do not consider the DSM 'Number of Ranges' field when
 749          * determining how much data to DMA. Always allocate memory for maximum
 750          * number of segments to prevent device reading beyond end of buffer.
 751          */
 752         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 753
 754         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 755         if (!range) {
 756                 /*
 757                  * If we fail allocation our range, fallback to the controller
 758                  * discard page. If that's also busy, it's safe to return
 759                  * busy, as we know we can make progress once that's freed.
 760                  */
 761                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 762                         return BLK_STS_RESOURCE;
 763
 764                 range = page_address(ns->ctrl->discard_page);
 765         }
 766
 767         __rq_for_each_bio(bio, req) {
 768                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 769                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 770
 771                 if (n < segments) {
 772                         range[n].cattr = cpu_to_le32(0);
 773                         range[n].nlb = cpu_to_le32(nlb);
 774                         range[n].slba = cpu_to_le64(slba);
 775                 }
 776                 n++;
 777         }
 778
 779         if (WARN_ON_ONCE(n != segments)) {
 780                 if (virt_to_page(range) == ns->ctrl->discard_page)
 781                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 782                 else
 783                         kfree(range);
 784                 return BLK_STS_IOERR;
 785         }
 786
 787         cmnd->dsm.opcode = nvme_cmd_dsm;
 788         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 789         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 790         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 791
 792         req->special_vec.bv_page = virt_to_page(range);
 793         req->special_vec.bv_offset = offset_in_page(range);
 794         req->special_vec.bv_len = alloc_size;
 795         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 796
 797         return BLK_STS_OK;
 798 }
 799
 800 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 801                 struct request *req, struct nvme_command *cmnd)
 802 {
 803         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 804                 return nvme_setup_discard(ns, req, cmnd);
 805
 806         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 807         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 808         cmnd->write_zeroes.slba =
 809                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 810         cmnd->write_zeroes.length =
 811                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 812         cmnd->write_zeroes.control = 0;
 813         return BLK_STS_OK;
 814 }
 815
 816 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 817                 struct request *req, struct nvme_command *cmnd,
 818                 enum nvme_opcode op)
 819 {
 820         struct nvme_ctrl *ctrl = ns->ctrl;
 821         u16 control = 0;
 822         u32 dsmgmt = 0;
 823
 824         if (req->cmd_flags & REQ_FUA)
 825                 control |= NVME_RW_FUA;
 826         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 827                 control |= NVME_RW_LR;
 828
 829         if (req->cmd_flags & REQ_RAHEAD)
 830                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 831
 832         cmnd->rw.opcode = op;
 833         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 834         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 835         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 836
 837         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 838                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 839
 840         if (ns->ms) {
 841                 /*
 842                  * If formated with metadata, the block layer always provides a
 843                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 844                  * we enable the PRACT bit for protection information or set the
 845                  * namespace capacity to zero to prevent any I/O.
 846                  */
 847                 if (!blk_integrity_rq(req)) {
 848                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 849                                 return BLK_STS_NOTSUPP;
 850                         control |= NVME_RW_PRINFO_PRACT;
 851                 }
 852
 853                 switch (ns->pi_type) {
 854                 case NVME_NS_DPS_PI_TYPE3:
 855                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 856                         break;
 857                 case NVME_NS_DPS_PI_TYPE1:
 858                 case NVME_NS_DPS_PI_TYPE2:
 859                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 860                                         NVME_RW_PRINFO_PRCHK_REF;
 861                         if (op == nvme_cmd_zone_append)
 862                                 control |= NVME_RW_APPEND_PIREMAP;
 863                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 864                         break;
 865                 }
 866         }
 867
 868         cmnd->rw.control = cpu_to_le16(control);
 869         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 870         return 0;
 871 }
 872
 873 void nvme_cleanup_cmd(struct request *req)
 874 {
 875         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 876                 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 877                 struct page *page = req->special_vec.bv_page;
 878
 879                 if (page == ctrl->discard_page)
 880                         clear_bit_unlock(0, &ctrl->discard_page_busy);
 881                 else
 882                         kfree(page_address(page) + req->special_vec.bv_offset);
 883         }
 884 }
 885 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 886
 887 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
 888 {
 889         struct nvme_command *cmd = nvme_req(req)->cmd;
 890         blk_status_t ret = BLK_STS_OK;
 891
 892         if (!(req->rq_flags & RQF_DONTPREP))
 893                 nvme_clear_nvme_request(req);
 894
 895         switch (req_op(req)) {
 896         case REQ_OP_DRV_IN:
 897         case REQ_OP_DRV_OUT:
 898                 /* these are setup prior to execution in nvme_init_request() */
 899                 break;
 900         case REQ_OP_FLUSH:
 901                 nvme_setup_flush(ns, cmd);
 902                 break;
 903         case REQ_OP_ZONE_RESET_ALL:
 904         case REQ_OP_ZONE_RESET:
 905                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 906                 break;
 907         case REQ_OP_ZONE_OPEN:
 908                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
 909                 break;
 910         case REQ_OP_ZONE_CLOSE:
 911                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
 912                 break;
 913         case REQ_OP_ZONE_FINISH:
 914                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
 915                 break;
 916         case REQ_OP_WRITE_ZEROES:
 917                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 918                 break;
 919         case REQ_OP_DISCARD:
 920                 ret = nvme_setup_discard(ns, req, cmd);
 921                 break;
 922         case REQ_OP_READ:
 923                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
 924                 break;
 925         case REQ_OP_WRITE:
 926                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
 927                 break;
 928         case REQ_OP_ZONE_APPEND:
 929                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
 930                 break;
 931         default:
 932                 WARN_ON_ONCE(1);
 933                 return BLK_STS_IOERR;
 934         }
 935
 936         cmd->common.command_id = req->tag;
 937         trace_nvme_setup_cmd(req, cmd);
 938         return ret;
 939 }
 940 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 941
 942 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 943 {
 944         struct completion *waiting = rq->end_io_data;
 945
 946         rq->end_io_data = NULL;
 947         complete(waiting);
 948 }
 949
 950 static void nvme_execute_rq_polled(struct request_queue *q,
 951                 struct gendisk *bd_disk, struct request *rq, int at_head)
 952 {
 953         DECLARE_COMPLETION_ONSTACK(wait);
 954
 955         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 956
 957         rq->cmd_flags |= REQ_HIPRI;
 958         rq->end_io_data = &wait;
 959         blk_execute_rq_nowait(bd_disk, rq, at_head, nvme_end_sync_rq);
 960
 961         while (!completion_done(&wait)) {
 962                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 963                 cond_resched();
 964         }
 965 }
 966
 967 /*
 968  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 969  * if the result is positive, it's an NVM Express status code
 970  */
 971 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 972                 union nvme_result *result, void *buffer, unsigned bufflen,
 973                 unsigned timeout, int qid, int at_head,
 974                 blk_mq_req_flags_t flags, bool poll)
 975 {
 976         struct request *req;
 977         int ret;
 978
 979         if (qid == NVME_QID_ANY)
 980                 req = nvme_alloc_request(q, cmd, flags);
 981         else
 982                 req = nvme_alloc_request_qid(q, cmd, flags, qid);
 983         if (IS_ERR(req))
 984                 return PTR_ERR(req);
 985
 986         if (timeout)
 987                 req->timeout = timeout;
 988
 989         if (buffer && bufflen) {
 990                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 991                 if (ret)
 992                         goto out;
 993         }
 994
 995         if (poll)
 996                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 997         else
 998                 blk_execute_rq(NULL, req, at_head);
 999         if (result)
1000                 *result = nvme_req(req)->result;
1001         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1002                 ret = -EINTR;
1003         else
1004                 ret = nvme_req(req)->status;
1005  out:
1006         blk_mq_free_request(req);
1007         return ret;
1008 }
1009 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1010
1011 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1012                 void *buffer, unsigned bufflen)
1013 {
1014         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1015                         NVME_QID_ANY, 0, 0, false);
1016 }
1017 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1018
1019 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
1020                 unsigned len, u32 seed, bool write)
1021 {
1022         struct bio_integrity_payload *bip;
1023         int ret = -ENOMEM;
1024         void *buf;
1025
1026         buf = kmalloc(len, GFP_KERNEL);
1027         if (!buf)
1028                 goto out;
1029
1030         ret = -EFAULT;
1031         if (write && copy_from_user(buf, ubuf, len))
1032                 goto out_free_meta;
1033
1034         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1035         if (IS_ERR(bip)) {
1036                 ret = PTR_ERR(bip);
1037                 goto out_free_meta;
1038         }
1039
1040         bip->bip_iter.bi_size = len;
1041         bip->bip_iter.bi_sector = seed;
1042         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1043                         offset_in_page(buf));
1044         if (ret == len)
1045                 return buf;
1046         ret = -ENOMEM;
1047 out_free_meta:
1048         kfree(buf);
1049 out:
1050         return ERR_PTR(ret);
1051 }
1052
1053 static u32 nvme_known_admin_effects(u8 opcode)
1054 {
1055         switch (opcode) {
1056         case nvme_admin_format_nvm:
1057                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1058                         NVME_CMD_EFFECTS_CSE_MASK;
1059         case nvme_admin_sanitize_nvm:
1060                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1061         default:
1062                 break;
1063         }
1064         return 0;
1065 }
1066
1067 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1068 {
1069         u32 effects = 0;
1070
1071         if (ns) {
1072                 if (ns->head->effects)
1073                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1074                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1075                         dev_warn_once(ctrl->device,
1076                                 "IO command:%02x has unhandled effects:%08x\n",
1077                                 opcode, effects);
1078                 return 0;
1079         }
1080
1081         if (ctrl->effects)
1082                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1083         effects |= nvme_known_admin_effects(opcode);
1084
1085         return effects;
1086 }
1087 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1088
1089 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1090                                u8 opcode)
1091 {
1092         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1093
1094         /*
1095          * For simplicity, IO to all namespaces is quiesced even if the command
1096          * effects say only one namespace is affected.
1097          */
1098         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1099                 mutex_lock(&ctrl->scan_lock);
1100                 mutex_lock(&ctrl->subsys->lock);
1101                 nvme_mpath_start_freeze(ctrl->subsys);
1102                 nvme_mpath_wait_freeze(ctrl->subsys);
1103                 nvme_start_freeze(ctrl);
1104                 nvme_wait_freeze(ctrl);
1105         }
1106         return effects;
1107 }
1108
1109 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1110 {
1111         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1112                 nvme_unfreeze(ctrl);
1113                 nvme_mpath_unfreeze(ctrl->subsys);
1114                 mutex_unlock(&ctrl->subsys->lock);
1115                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1116                 mutex_unlock(&ctrl->scan_lock);
1117         }
1118         if (effects & NVME_CMD_EFFECTS_CCC)
1119                 nvme_init_ctrl_finish(ctrl);
1120         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1121                 nvme_queue_scan(ctrl);
1122                 flush_work(&ctrl->scan_work);
1123         }
1124 }
1125
1126 void nvme_execute_passthru_rq(struct request *rq)
1127 {
1128         struct nvme_command *cmd = nvme_req(rq)->cmd;
1129         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1130         struct nvme_ns *ns = rq->q->queuedata;
1131         struct gendisk *disk = ns ? ns->disk : NULL;
1132         u32 effects;
1133
1134         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1135         blk_execute_rq(disk, rq, 0);
1136         if (effects) /* nothing to be done for zero cmd effects */
1137                 nvme_passthru_end(ctrl, effects);
1138 }
1139 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1140
1141 static int nvme_submit_user_cmd(struct request_queue *q,
1142                 struct nvme_command *cmd, void __user *ubuffer,
1143                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1144                 u32 meta_seed, u64 *result, unsigned timeout)
1145 {
1146         bool write = nvme_is_write(cmd);
1147         struct nvme_ns *ns = q->queuedata;
1148         struct block_device *bdev = ns ? ns->disk->part0 : NULL;
1149         struct request *req;
1150         struct bio *bio = NULL;
1151         void *meta = NULL;
1152         int ret;
1153
1154         req = nvme_alloc_request(q, cmd, 0);
1155         if (IS_ERR(req))
1156                 return PTR_ERR(req);
1157
1158         if (timeout)
1159                 req->timeout = timeout;
1160         nvme_req(req)->flags |= NVME_REQ_USERCMD;
1161
1162         if (ubuffer && bufflen) {
1163                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1164                                 GFP_KERNEL);
1165                 if (ret)
1166                         goto out;
1167                 bio = req->bio;
1168                 if (bdev)
1169                         bio_set_dev(bio, bdev);
1170                 if (bdev && meta_buffer && meta_len) {
1171                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1172                                         meta_seed, write);
1173                         if (IS_ERR(meta)) {
1174                                 ret = PTR_ERR(meta);
1175                                 goto out_unmap;
1176                         }
1177                         req->cmd_flags |= REQ_INTEGRITY;
1178                 }
1179         }
1180
1181         nvme_execute_passthru_rq(req);
1182         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1183                 ret = -EINTR;
1184         else
1185                 ret = nvme_req(req)->status;
1186         if (result)
1187                 *result = le64_to_cpu(nvme_req(req)->result.u64);
1188         if (meta && !ret && !write) {
1189                 if (copy_to_user(meta_buffer, meta, meta_len))
1190                         ret = -EFAULT;
1191         }
1192         kfree(meta);
1193  out_unmap:
1194         if (bio)
1195                 blk_rq_unmap_user(bio);
1196  out:
1197         blk_mq_free_request(req);
1198         return ret;
1199 }
1200
1201 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1202 {
1203         struct nvme_ctrl *ctrl = rq->end_io_data;
1204         unsigned long flags;
1205         bool startka = false;
1206
1207         blk_mq_free_request(rq);
1208
1209         if (status) {
1210                 dev_err(ctrl->device,
1211                         "failed nvme_keep_alive_end_io error=%d\n",
1212                                 status);
1213                 return;
1214         }
1215
1216         ctrl->comp_seen = false;
1217         spin_lock_irqsave(&ctrl->lock, flags);
1218         if (ctrl->state == NVME_CTRL_LIVE ||
1219             ctrl->state == NVME_CTRL_CONNECTING)
1220                 startka = true;
1221         spin_unlock_irqrestore(&ctrl->lock, flags);
1222         if (startka)
1223                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1224 }
1225
1226 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1227 {
1228         struct request *rq;
1229
1230         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1231                         BLK_MQ_REQ_RESERVED);
1232         if (IS_ERR(rq))
1233                 return PTR_ERR(rq);
1234
1235         rq->timeout = ctrl->kato * HZ;
1236         rq->end_io_data = ctrl;
1237
1238         blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1239
1240         return 0;
1241 }
1242
1243 static void nvme_keep_alive_work(struct work_struct *work)
1244 {
1245         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1246                         struct nvme_ctrl, ka_work);
1247         bool comp_seen = ctrl->comp_seen;
1248
1249         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1250                 dev_dbg(ctrl->device,
1251                         "reschedule traffic based keep-alive timer\n");
1252                 ctrl->comp_seen = false;
1253                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1254                 return;
1255         }
1256
1257         if (nvme_keep_alive(ctrl)) {
1258                 /* allocation failure, reset the controller */
1259                 dev_err(ctrl->device, "keep-alive failed\n");
1260                 nvme_reset_ctrl(ctrl);
1261                 return;
1262         }
1263 }
1264
1265 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1266 {
1267         if (unlikely(ctrl->kato == 0))
1268                 return;
1269
1270         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1271 }
1272
1273 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1274 {
1275         if (unlikely(ctrl->kato == 0))
1276                 return;
1277
1278         cancel_delayed_work_sync(&ctrl->ka_work);
1279 }
1280 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1281
1282 /*
1283  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1284  * flag, thus sending any new CNS opcodes has a big chance of not working.
1285  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1286  * (but not for any later version).
1287  */
1288 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1289 {
1290         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1291                 return ctrl->vs < NVME_VS(1, 2, 0);
1292         return ctrl->vs < NVME_VS(1, 1, 0);
1293 }
1294
1295 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1296 {
1297         struct nvme_command c = { };
1298         int error;
1299
1300         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1301         c.identify.opcode = nvme_admin_identify;
1302         c.identify.cns = NVME_ID_CNS_CTRL;
1303
1304         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1305         if (!*id)
1306                 return -ENOMEM;
1307
1308         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1309                         sizeof(struct nvme_id_ctrl));
1310         if (error)
1311                 kfree(*id);
1312         return error;
1313 }
1314
1315 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1316 {
1317         return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1318 }
1319
1320 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1321                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1322 {
1323         const char *warn_str = "ctrl returned bogus length:";
1324         void *data = cur;
1325
1326         switch (cur->nidt) {
1327         case NVME_NIDT_EUI64:
1328                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1329                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1330                                  warn_str, cur->nidl);
1331                         return -1;
1332                 }
1333                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1334                 return NVME_NIDT_EUI64_LEN;
1335         case NVME_NIDT_NGUID:
1336                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1337                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1338                                  warn_str, cur->nidl);
1339                         return -1;
1340                 }
1341                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1342                 return NVME_NIDT_NGUID_LEN;
1343         case NVME_NIDT_UUID:
1344                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1345                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1346                                  warn_str, cur->nidl);
1347                         return -1;
1348                 }
1349                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1350                 return NVME_NIDT_UUID_LEN;
1351         case NVME_NIDT_CSI:
1352                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1353                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1354                                  warn_str, cur->nidl);
1355                         return -1;
1356                 }
1357                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1358                 *csi_seen = true;
1359                 return NVME_NIDT_CSI_LEN;
1360         default:
1361                 /* Skip unknown types */
1362                 return cur->nidl;
1363         }
1364 }
1365
1366 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1367                 struct nvme_ns_ids *ids)
1368 {
1369         struct nvme_command c = { };
1370         bool csi_seen = false;
1371         int status, pos, len;
1372         void *data;
1373
1374         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1375                 return 0;
1376         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1377                 return 0;
1378
1379         c.identify.opcode = nvme_admin_identify;
1380         c.identify.nsid = cpu_to_le32(nsid);
1381         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1382
1383         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1384         if (!data)
1385                 return -ENOMEM;
1386
1387         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1388                                       NVME_IDENTIFY_DATA_SIZE);
1389         if (status) {
1390                 dev_warn(ctrl->device,
1391                         "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1392                         nsid, status);
1393                 goto free_data;
1394         }
1395
1396         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1397                 struct nvme_ns_id_desc *cur = data + pos;
1398
1399                 if (cur->nidl == 0)
1400                         break;
1401
1402                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1403                 if (len < 0)
1404                         break;
1405
1406                 len += sizeof(*cur);
1407         }
1408
1409         if (nvme_multi_css(ctrl) && !csi_seen) {
1410                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1411                          nsid);
1412                 status = -EINVAL;
1413         }
1414
1415 free_data:
1416         kfree(data);
1417         return status;
1418 }
1419
1420 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1421                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1422 {
1423         struct nvme_command c = { };
1424         int error;
1425
1426         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1427         c.identify.opcode = nvme_admin_identify;
1428         c.identify.nsid = cpu_to_le32(nsid);
1429         c.identify.cns = NVME_ID_CNS_NS;
1430
1431         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1432         if (!*id)
1433                 return -ENOMEM;
1434
1435         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1436         if (error) {
1437                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1438                 goto out_free_id;
1439         }
1440
1441         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1442         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1443                 goto out_free_id;
1444
1445         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1446             !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1447                 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1448         if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1449             !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1450                 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1451
1452         return 0;
1453
1454 out_free_id:
1455         kfree(*id);
1456         return error;
1457 }
1458
1459 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1460                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1461 {
1462         union nvme_result res = { 0 };
1463         struct nvme_command c;
1464         int ret;
1465
1466         memset(&c, 0, sizeof(c));
1467         c.features.opcode = op;
1468         c.features.fid = cpu_to_le32(fid);
1469         c.features.dword11 = cpu_to_le32(dword11);
1470
1471         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1472                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1473         if (ret >= 0 && result)
1474                 *result = le32_to_cpu(res.u32);
1475         return ret;
1476 }
1477
1478 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1479                       unsigned int dword11, void *buffer, size_t buflen,
1480                       u32 *result)
1481 {
1482         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1483                              buflen, result);
1484 }
1485 EXPORT_SYMBOL_GPL(nvme_set_features);
1486
1487 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1488                       unsigned int dword11, void *buffer, size_t buflen,
1489                       u32 *result)
1490 {
1491         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1492                              buflen, result);
1493 }
1494 EXPORT_SYMBOL_GPL(nvme_get_features);
1495
1496 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1497 {
1498         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1499         u32 result;
1500         int status, nr_io_queues;
1501
1502         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1503                         &result);
1504         if (status < 0)
1505                 return status;
1506
1507         /*
1508          * Degraded controllers might return an error when setting the queue
1509          * count.  We still want to be able to bring them online and offer
1510          * access to the admin queue, as that might be only way to fix them up.
1511          */
1512         if (status > 0) {
1513                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1514                 *count = 0;
1515         } else {
1516                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1517                 *count = min(*count, nr_io_queues);
1518         }
1519
1520         return 0;
1521 }
1522 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1523
1524 #define NVME_AEN_SUPPORTED \
1525         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1526          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1527
1528 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1529 {
1530         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1531         int status;
1532
1533         if (!supported_aens)
1534                 return;
1535
1536         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1537                         NULL, 0, &result);
1538         if (status)
1539                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1540                          supported_aens);
1541
1542         queue_work(nvme_wq, &ctrl->async_event_work);
1543 }
1544
1545 /*
1546  * Convert integer values from ioctl structures to user pointers, silently
1547  * ignoring the upper bits in the compat case to match behaviour of 32-bit
1548  * kernels.
1549  */
1550 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1551 {
1552         if (in_compat_syscall())
1553                 ptrval = (compat_uptr_t)ptrval;
1554         return (void __user *)ptrval;
1555 }
1556
1557 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1558 {
1559         struct nvme_user_io io;
1560         struct nvme_command c;
1561         unsigned length, meta_len;
1562         void __user *metadata;
1563
1564         if (copy_from_user(&io, uio, sizeof(io)))
1565                 return -EFAULT;
1566         if (io.flags)
1567                 return -EINVAL;
1568
1569         switch (io.opcode) {
1570         case nvme_cmd_write:
1571         case nvme_cmd_read:
1572         case nvme_cmd_compare:
1573                 break;
1574         default:
1575                 return -EINVAL;
1576         }
1577
1578         length = (io.nblocks + 1) << ns->lba_shift;
1579
1580         if ((io.control & NVME_RW_PRINFO_PRACT) &&
1581             ns->ms == sizeof(struct t10_pi_tuple)) {
1582                 /*
1583                  * Protection information is stripped/inserted by the
1584                  * controller.
1585                  */
1586                 if (nvme_to_user_ptr(io.metadata))
1587                         return -EINVAL;
1588                 meta_len = 0;
1589                 metadata = NULL;
1590         } else {
1591                 meta_len = (io.nblocks + 1) * ns->ms;
1592                 metadata = nvme_to_user_ptr(io.metadata);
1593         }
1594
1595         if (ns->features & NVME_NS_EXT_LBAS) {
1596                 length += meta_len;
1597                 meta_len = 0;
1598         } else if (meta_len) {
1599                 if ((io.metadata & 3) || !io.metadata)
1600                         return -EINVAL;
1601         }
1602
1603         memset(&c, 0, sizeof(c));
1604         c.rw.opcode = io.opcode;
1605         c.rw.flags = io.flags;
1606         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1607         c.rw.slba = cpu_to_le64(io.slba);
1608         c.rw.length = cpu_to_le16(io.nblocks);
1609         c.rw.control = cpu_to_le16(io.control);
1610         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1611         c.rw.reftag = cpu_to_le32(io.reftag);
1612         c.rw.apptag = cpu_to_le16(io.apptag);
1613         c.rw.appmask = cpu_to_le16(io.appmask);
1614
1615         return nvme_submit_user_cmd(ns->queue, &c,
1616                         nvme_to_user_ptr(io.addr), length,
1617                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1618 }
1619
1620 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1621                         struct nvme_passthru_cmd __user *ucmd)
1622 {
1623         struct nvme_passthru_cmd cmd;
1624         struct nvme_command c;
1625         unsigned timeout = 0;
1626         u64 result;
1627         int status;
1628
1629         if (!capable(CAP_SYS_ADMIN))
1630                 return -EACCES;
1631         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1632                 return -EFAULT;
1633         if (cmd.flags)
1634                 return -EINVAL;
1635         if (ns && cmd.nsid != ns->head->ns_id) {
1636                 dev_err(ctrl->device,
1637                         "%s: nsid (%u) in cmd does not match nsid (%u) of namespace\n",
1638                         current->comm, cmd.nsid, ns->head->ns_id);
1639                 return -EINVAL;
1640         }
1641
1642         memset(&c, 0, sizeof(c));
1643         c.common.opcode = cmd.opcode;
1644         c.common.flags = cmd.flags;
1645         c.common.nsid = cpu_to_le32(cmd.nsid);
1646         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1647         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1648         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1649         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1650         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1651         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1652         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1653         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1654
1655         if (cmd.timeout_ms)
1656                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1657
1658         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1659                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1660                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1661                         0, &result, timeout);
1662
1663         if (status >= 0) {
1664                 if (put_user(result, &ucmd->result))
1665                         return -EFAULT;
1666         }
1667
1668         return status;
1669 }
1670
1671 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1672                         struct nvme_passthru_cmd64 __user *ucmd)
1673 {
1674         struct nvme_passthru_cmd64 cmd;
1675         struct nvme_command c;
1676         unsigned timeout = 0;
1677         int status;
1678
1679         if (!capable(CAP_SYS_ADMIN))
1680                 return -EACCES;
1681         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1682                 return -EFAULT;
1683         if (cmd.flags)
1684                 return -EINVAL;
1685         if (ns && cmd.nsid != ns->head->ns_id) {
1686                 dev_err(ctrl->device,
1687                         "%s: nsid (%u) in cmd does not match nsid (%u) of namespace\n",
1688                         current->comm, cmd.nsid, ns->head->ns_id);
1689                 return -EINVAL;
1690         }
1691
1692         memset(&c, 0, sizeof(c));
1693         c.common.opcode = cmd.opcode;
1694         c.common.flags = cmd.flags;
1695         c.common.nsid = cpu_to_le32(cmd.nsid);
1696         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1697         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1698         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1699         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1700         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1701         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1702         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1703         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1704
1705         if (cmd.timeout_ms)
1706                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1707
1708         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1709                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1710                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1711                         0, &cmd.result, timeout);
1712
1713         if (status >= 0) {
1714                 if (put_user(cmd.result, &ucmd->result))
1715                         return -EFAULT;
1716         }
1717
1718         return status;
1719 }
1720
1721 /*
1722  * Issue ioctl requests on the first available path.  Note that unlike normal
1723  * block layer requests we will not retry failed request on another controller.
1724  */
1725 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1726                 struct nvme_ns_head **head, int *srcu_idx)
1727 {
1728 #ifdef CONFIG_NVME_MULTIPATH
1729         if (disk->fops == &nvme_ns_head_ops) {
1730                 struct nvme_ns *ns;
1731
1732                 *head = disk->private_data;
1733                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1734                 ns = nvme_find_path(*head);
1735                 if (!ns)
1736                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1737                 return ns;
1738         }
1739 #endif
1740         *head = NULL;
1741         *srcu_idx = -1;
1742         return disk->private_data;
1743 }
1744
1745 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1746 {
1747         if (head)
1748                 srcu_read_unlock(&head->srcu, idx);
1749 }
1750
1751 static bool is_ctrl_ioctl(unsigned int cmd)
1752 {
1753         if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1754                 return true;
1755         if (is_sed_ioctl(cmd))
1756                 return true;
1757         return false;
1758 }
1759
1760 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1761                                   void __user *argp,
1762                                   struct nvme_ns_head *head,
1763                                   int srcu_idx)
1764 {
1765         struct nvme_ctrl *ctrl = ns->ctrl;
1766         int ret;
1767
1768         nvme_get_ctrl(ns->ctrl);
1769         nvme_put_ns_from_disk(head, srcu_idx);
1770
1771         switch (cmd) {
1772         case NVME_IOCTL_ADMIN_CMD:
1773                 ret = nvme_user_cmd(ctrl, NULL, argp);
1774                 break;
1775         case NVME_IOCTL_ADMIN64_CMD:
1776                 ret = nvme_user_cmd64(ctrl, NULL, argp);
1777                 break;
1778         default:
1779                 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1780                 break;
1781         }
1782         nvme_put_ctrl(ctrl);
1783         return ret;
1784 }
1785
1786 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1787                 unsigned int cmd, unsigned long arg)
1788 {
1789         struct nvme_ns_head *head = NULL;
1790         void __user *argp = (void __user *)arg;
1791         struct nvme_ns *ns;
1792         int srcu_idx, ret;
1793
1794         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1795         if (unlikely(!ns))
1796                 return -EWOULDBLOCK;
1797
1798         /*
1799          * Handle ioctls that apply to the controller instead of the namespace
1800          * seperately and drop the ns SRCU reference early.  This avoids a
1801          * deadlock when deleting namespaces using the passthrough interface.
1802          */
1803         if (is_ctrl_ioctl(cmd))
1804                 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1805
1806         switch (cmd) {
1807         case NVME_IOCTL_ID:
1808                 force_successful_syscall_return();
1809                 ret = ns->head->ns_id;
1810                 break;
1811         case NVME_IOCTL_IO_CMD:
1812                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1813                 break;
1814         case NVME_IOCTL_SUBMIT_IO:
1815                 ret = nvme_submit_io(ns, argp);
1816                 break;
1817         case NVME_IOCTL_IO64_CMD:
1818                 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1819                 break;
1820         default:
1821                 if (ns->ndev)
1822                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1823                 else
1824                         ret = -ENOTTY;
1825         }
1826
1827         nvme_put_ns_from_disk(head, srcu_idx);
1828         return ret;
1829 }
1830
1831 #ifdef CONFIG_COMPAT
1832 struct nvme_user_io32 {
1833         __u8    opcode;
1834         __u8    flags;
1835         __u16   control;
1836         __u16   nblocks;
1837         __u16   rsvd;
1838         __u64   metadata;
1839         __u64   addr;
1840         __u64   slba;
1841         __u32   dsmgmt;
1842         __u32   reftag;
1843         __u16   apptag;
1844         __u16   appmask;
1845 } __attribute__((__packed__));
1846
1847 #define NVME_IOCTL_SUBMIT_IO32  _IOW('N', 0x42, struct nvme_user_io32)
1848
1849 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1850                 unsigned int cmd, unsigned long arg)
1851 {
1852         /*
1853          * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1854          * between 32 bit programs and 64 bit kernel.
1855          * The cause is that the results of sizeof(struct nvme_user_io),
1856          * which is used to define NVME_IOCTL_SUBMIT_IO,
1857          * are not same between 32 bit compiler and 64 bit compiler.
1858          * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1859          * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1860          * Other IOCTL numbers are same between 32 bit and 64 bit.
1861          * So there is nothing to do regarding to other IOCTL numbers.
1862          */
1863         if (cmd == NVME_IOCTL_SUBMIT_IO32)
1864                 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1865
1866         return nvme_ioctl(bdev, mode, cmd, arg);
1867 }
1868 #else
1869 #define nvme_compat_ioctl       NULL
1870 #endif /* CONFIG_COMPAT */
1871
1872 static int nvme_open(struct block_device *bdev, fmode_t mode)
1873 {
1874         struct nvme_ns *ns = bdev->bd_disk->private_data;
1875
1876         /* should never be called due to GENHD_FL_HIDDEN */
1877         if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1878                 goto fail;
1879         if (!kref_get_unless_zero(&ns->kref))
1880                 goto fail;
1881         if (!try_module_get(ns->ctrl->ops->module))
1882                 goto fail_put_ns;
1883
1884         return 0;
1885
1886 fail_put_ns:
1887         nvme_put_ns(ns);
1888 fail:
1889         return -ENXIO;
1890 }
1891
1892 static void nvme_release(struct gendisk *disk, fmode_t mode)
1893 {
1894         struct nvme_ns *ns = disk->private_data;
1895
1896         module_put(ns->ctrl->ops->module);
1897         nvme_put_ns(ns);
1898 }
1899
1900 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1901 {
1902         /* some standard values */
1903         geo->heads = 1 << 6;
1904         geo->sectors = 1 << 5;
1905         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1906         return 0;
1907 }
1908
1909 #ifdef CONFIG_BLK_DEV_INTEGRITY
1910 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1911                                 u32 max_integrity_segments)
1912 {
1913         struct blk_integrity integrity;
1914
1915         memset(&integrity, 0, sizeof(integrity));
1916         switch (pi_type) {
1917         case NVME_NS_DPS_PI_TYPE3:
1918                 integrity.profile = &t10_pi_type3_crc;
1919                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1920                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1921                 break;
1922         case NVME_NS_DPS_PI_TYPE1:
1923         case NVME_NS_DPS_PI_TYPE2:
1924                 integrity.profile = &t10_pi_type1_crc;
1925                 integrity.tag_size = sizeof(u16);
1926                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1927                 break;
1928         default:
1929                 integrity.profile = NULL;
1930                 break;
1931         }
1932         integrity.tuple_size = ms;
1933         blk_integrity_register(disk, &integrity);
1934         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1935 }
1936 #else
1937 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1938                                 u32 max_integrity_segments)
1939 {
1940 }
1941 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1942
1943 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1944 {
1945         struct nvme_ctrl *ctrl = ns->ctrl;
1946         struct request_queue *queue = disk->queue;
1947         u32 size = queue_logical_block_size(queue);
1948
1949         if (ctrl->max_discard_sectors == 0) {
1950                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1951                 return;
1952         }
1953
1954         if (ctrl->nr_streams && ns->sws && ns->sgs)
1955                 size *= ns->sws * ns->sgs;
1956
1957         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1958                         NVME_DSM_MAX_RANGES);
1959
1960         queue->limits.discard_alignment = 0;
1961         queue->limits.discard_granularity = size;
1962
1963         /* If discard is already enabled, don't reset queue limits */
1964         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1965                 return;
1966
1967         blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1968         blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1969
1970         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1971                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1972 }
1973
1974 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1975 {
1976         return !uuid_is_null(&ids->uuid) ||
1977                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1978                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1979 }
1980
1981 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1982 {
1983         return uuid_equal(&a->uuid, &b->uuid) &&
1984                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1985                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1986                 a->csi == b->csi;
1987 }
1988
1989 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1990                                  u32 *phys_bs, u32 *io_opt)
1991 {
1992         struct streams_directive_params s;
1993         int ret;
1994
1995         if (!ctrl->nr_streams)
1996                 return 0;
1997
1998         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1999         if (ret)
2000                 return ret;
2001
2002         ns->sws = le32_to_cpu(s.sws);
2003         ns->sgs = le16_to_cpu(s.sgs);
2004
2005         if (ns->sws) {
2006                 *phys_bs = ns->sws * (1 << ns->lba_shift);
2007                 if (ns->sgs)
2008                         *io_opt = *phys_bs * ns->sgs;
2009         }
2010
2011         return 0;
2012 }
2013
2014 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
2015 {
2016         struct nvme_ctrl *ctrl = ns->ctrl;
2017
2018         /*
2019          * The PI implementation requires the metadata size to be equal to the
2020          * t10 pi tuple size.
2021          */
2022         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2023         if (ns->ms == sizeof(struct t10_pi_tuple))
2024                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2025         else
2026                 ns->pi_type = 0;
2027
2028         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2029         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2030                 return 0;
2031         if (ctrl->ops->flags & NVME_F_FABRICS) {
2032                 /*
2033                  * The NVMe over Fabrics specification only supports metadata as
2034                  * part of the extended data LBA.  We rely on HCA/HBA support to
2035                  * remap the separate metadata buffer from the block layer.
2036                  */
2037                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2038                         return -EINVAL;
2039                 if (ctrl->max_integrity_segments)
2040                         ns->features |=
2041                                 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2042         } else {
2043                 /*
2044                  * For PCIe controllers, we can't easily remap the separate
2045                  * metadata buffer from the block layer and thus require a
2046                  * separate metadata buffer for block layer metadata/PI support.
2047                  * We allow extended LBAs for the passthrough interface, though.
2048                  */
2049                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2050                         ns->features |= NVME_NS_EXT_LBAS;
2051                 else
2052                         ns->features |= NVME_NS_METADATA_SUPPORTED;
2053         }
2054
2055         return 0;
2056 }
2057
2058 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2059                 struct request_queue *q)
2060 {
2061         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2062
2063         if (ctrl->max_hw_sectors) {
2064                 u32 max_segments =
2065                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2066
2067                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2068                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2069                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2070         }
2071         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2072         blk_queue_dma_alignment(q, 7);
2073         blk_queue_write_cache(q, vwc, vwc);
2074 }
2075
2076 static void nvme_update_disk_info(struct gendisk *disk,
2077                 struct nvme_ns *ns, struct nvme_id_ns *id)
2078 {
2079         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2080         unsigned short bs = 1 << ns->lba_shift;
2081         u32 atomic_bs, phys_bs, io_opt = 0;
2082
2083         /*
2084          * The block layer can't support LBA sizes larger than the page size
2085          * yet, so catch this early and don't allow block I/O.
2086          */
2087         if (ns->lba_shift > PAGE_SHIFT) {
2088                 capacity = 0;
2089                 bs = (1 << 9);
2090         }
2091
2092         blk_integrity_unregister(disk);
2093
2094         atomic_bs = phys_bs = bs;
2095         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2096         if (id->nabo == 0) {
2097                 /*
2098                  * Bit 1 indicates whether NAWUPF is defined for this namespace
2099                  * and whether it should be used instead of AWUPF. If NAWUPF ==
2100                  * 0 then AWUPF must be used instead.
2101                  */
2102                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2103                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2104                 else
2105                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2106         }
2107
2108         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2109                 /* NPWG = Namespace Preferred Write Granularity */
2110                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2111                 /* NOWS = Namespace Optimal Write Size */
2112                 io_opt = bs * (1 + le16_to_cpu(id->nows));
2113         }
2114
2115         blk_queue_logical_block_size(disk->queue, bs);
2116         /*
2117          * Linux filesystems assume writing a single physical block is
2118          * an atomic operation. Hence limit the physical block size to the
2119          * value of the Atomic Write Unit Power Fail parameter.
2120          */
2121         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2122         blk_queue_io_min(disk->queue, phys_bs);
2123         blk_queue_io_opt(disk->queue, io_opt);
2124
2125         /*
2126          * Register a metadata profile for PI, or the plain non-integrity NVMe
2127          * metadata masquerading as Type 0 if supported, otherwise reject block
2128          * I/O to namespaces with metadata except when the namespace supports
2129          * PI, as it can strip/insert in that case.
2130          */
2131         if (ns->ms) {
2132                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2133                     (ns->features & NVME_NS_METADATA_SUPPORTED))
2134                         nvme_init_integrity(disk, ns->ms, ns->pi_type,
2135                                             ns->ctrl->max_integrity_segments);
2136                 else if (!nvme_ns_has_pi(ns))
2137                         capacity = 0;
2138         }
2139
2140         set_capacity_and_notify(disk, capacity);
2141
2142         nvme_config_discard(disk, ns);
2143         blk_queue_max_write_zeroes_sectors(disk->queue,
2144                                            ns->ctrl->max_zeroes_sectors);
2145
2146         set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
2147                 test_bit(NVME_NS_FORCE_RO, &ns->flags));
2148 }
2149
2150 static inline bool nvme_first_scan(struct gendisk *disk)
2151 {
2152         /* nvme_alloc_ns() scans the disk prior to adding it */
2153         return !(disk->flags & GENHD_FL_UP);
2154 }
2155
2156 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2157 {
2158         struct nvme_ctrl *ctrl = ns->ctrl;
2159         u32 iob;
2160
2161         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2162             is_power_of_2(ctrl->max_hw_sectors))
2163                 iob = ctrl->max_hw_sectors;
2164         else
2165                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2166
2167         if (!iob)
2168                 return;
2169
2170         if (!is_power_of_2(iob)) {
2171                 if (nvme_first_scan(ns->disk))
2172                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2173                                 ns->disk->disk_name, iob);
2174                 return;
2175         }
2176
2177         if (blk_queue_is_zoned(ns->disk->queue)) {
2178                 if (nvme_first_scan(ns->disk))
2179                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
2180                                 ns->disk->disk_name);
2181                 return;
2182         }
2183
2184         blk_queue_chunk_sectors(ns->queue, iob);
2185 }
2186
2187 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2188 {
2189         unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2190         int ret;
2191
2192         blk_mq_freeze_queue(ns->disk->queue);
2193         ns->lba_shift = id->lbaf[lbaf].ds;
2194         nvme_set_queue_limits(ns->ctrl, ns->queue);
2195
2196         ret = nvme_configure_metadata(ns, id);
2197         if (ret)
2198                 goto out_unfreeze;
2199         nvme_set_chunk_sectors(ns, id);
2200         nvme_update_disk_info(ns->disk, ns, id);
2201
2202         if (ns->head->ids.csi == NVME_CSI_ZNS) {
2203                 ret = nvme_update_zone_info(ns, lbaf);
2204                 if (ret)
2205                         goto out_unfreeze;
2206         }
2207
2208         blk_mq_unfreeze_queue(ns->disk->queue);
2209
2210         if (blk_queue_is_zoned(ns->queue)) {
2211                 ret = nvme_revalidate_zones(ns);
2212                 if (ret && !nvme_first_scan(ns->disk))
2213                         return ret;
2214         }
2215
2216         if (nvme_ns_head_multipath(ns->head)) {
2217                 blk_mq_freeze_queue(ns->head->disk->queue);
2218                 nvme_update_disk_info(ns->head->disk, ns, id);
2219                 blk_stack_limits(&ns->head->disk->queue->limits,
2220                                  &ns->queue->limits, 0);
2221                 blk_queue_update_readahead(ns->head->disk->queue);
2222                 blk_mq_unfreeze_queue(ns->head->disk->queue);
2223         }
2224         return 0;
2225
2226 out_unfreeze:
2227         blk_mq_unfreeze_queue(ns->disk->queue);
2228         return ret;
2229 }
2230
2231 static char nvme_pr_type(enum pr_type type)
2232 {
2233         switch (type) {
2234         case PR_WRITE_EXCLUSIVE:
2235                 return 1;
2236         case PR_EXCLUSIVE_ACCESS:
2237                 return 2;
2238         case PR_WRITE_EXCLUSIVE_REG_ONLY:
2239                 return 3;
2240         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2241                 return 4;
2242         case PR_WRITE_EXCLUSIVE_ALL_REGS:
2243                 return 5;
2244         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2245                 return 6;
2246         default:
2247                 return 0;
2248         }
2249 };
2250
2251 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2252                                 u64 key, u64 sa_key, u8 op)
2253 {
2254         struct nvme_ns_head *head = NULL;
2255         struct nvme_ns *ns;
2256         struct nvme_command c;
2257         int srcu_idx, ret;
2258         u8 data[16] = { 0, };
2259
2260         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2261         if (unlikely(!ns))
2262                 return -EWOULDBLOCK;
2263
2264         put_unaligned_le64(key, &data[0]);
2265         put_unaligned_le64(sa_key, &data[8]);
2266
2267         memset(&c, 0, sizeof(c));
2268         c.common.opcode = op;
2269         c.common.nsid = cpu_to_le32(ns->head->ns_id);
2270         c.common.cdw10 = cpu_to_le32(cdw10);
2271
2272         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2273         nvme_put_ns_from_disk(head, srcu_idx);
2274         return ret;
2275 }
2276
2277 static int nvme_pr_register(struct block_device *bdev, u64 old,
2278                 u64 new, unsigned flags)
2279 {
2280         u32 cdw10;
2281
2282         if (flags & ~PR_FL_IGNORE_KEY)
2283                 return -EOPNOTSUPP;
2284
2285         cdw10 = old ? 2 : 0;
2286         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2287         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2288         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2289 }
2290
2291 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2292                 enum pr_type type, unsigned flags)
2293 {
2294         u32 cdw10;
2295
2296         if (flags & ~PR_FL_IGNORE_KEY)
2297                 return -EOPNOTSUPP;
2298
2299         cdw10 = nvme_pr_type(type) << 8;
2300         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2301         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2302 }
2303
2304 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2305                 enum pr_type type, bool abort)
2306 {
2307         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2308
2309         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2310 }
2311
2312 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2313 {
2314         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2315
2316         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2317 }
2318
2319 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2320 {
2321         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2322
2323         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2324 }
2325
2326 static const struct pr_ops nvme_pr_ops = {
2327         .pr_register    = nvme_pr_register,
2328         .pr_reserve     = nvme_pr_reserve,
2329         .pr_release     = nvme_pr_release,
2330         .pr_preempt     = nvme_pr_preempt,
2331         .pr_clear       = nvme_pr_clear,
2332 };
2333
2334 #ifdef CONFIG_BLK_SED_OPAL
2335 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2336                 bool send)
2337 {
2338         struct nvme_ctrl *ctrl = data;
2339         struct nvme_command cmd;
2340
2341         memset(&cmd, 0, sizeof(cmd));
2342         if (send)
2343                 cmd.common.opcode = nvme_admin_security_send;
2344         else
2345                 cmd.common.opcode = nvme_admin_security_recv;
2346         cmd.common.nsid = 0;
2347         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2348         cmd.common.cdw11 = cpu_to_le32(len);
2349
2350         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2351                         NVME_QID_ANY, 1, 0, false);
2352 }
2353 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2354 #endif /* CONFIG_BLK_SED_OPAL */
2355
2356 static const struct block_device_operations nvme_bdev_ops = {
2357         .owner          = THIS_MODULE,
2358         .ioctl          = nvme_ioctl,
2359         .compat_ioctl   = nvme_compat_ioctl,
2360         .open           = nvme_open,
2361         .release        = nvme_release,
2362         .getgeo         = nvme_getgeo,
2363         .report_zones   = nvme_report_zones,
2364         .pr_ops         = &nvme_pr_ops,
2365 };
2366
2367 #ifdef CONFIG_NVME_MULTIPATH
2368 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2369 {
2370         struct nvme_ns_head *head = bdev->bd_disk->private_data;
2371
2372         if (!kref_get_unless_zero(&head->ref))
2373                 return -ENXIO;
2374         return 0;
2375 }
2376
2377 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2378 {
2379         nvme_put_ns_head(disk->private_data);
2380 }
2381
2382 const struct block_device_operations nvme_ns_head_ops = {
2383         .owner          = THIS_MODULE,
2384         .submit_bio     = nvme_ns_head_submit_bio,
2385         .open           = nvme_ns_head_open,
2386         .release        = nvme_ns_head_release,
2387         .ioctl          = nvme_ioctl,
2388         .compat_ioctl   = nvme_compat_ioctl,
2389         .getgeo         = nvme_getgeo,
2390         .report_zones   = nvme_report_zones,
2391         .pr_ops         = &nvme_pr_ops,
2392 };
2393 #endif /* CONFIG_NVME_MULTIPATH */
2394
2395 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2396 {
2397         unsigned long timeout =
2398                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2399         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2400         int ret;
2401
2402         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2403                 if (csts == ~0)
2404                         return -ENODEV;
2405                 if ((csts & NVME_CSTS_RDY) == bit)
2406                         break;
2407
2408                 usleep_range(1000, 2000);
2409                 if (fatal_signal_pending(current))
2410                         return -EINTR;
2411                 if (time_after(jiffies, timeout)) {
2412                         dev_err(ctrl->device,
2413                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2414                                 enabled ? "initialisation" : "reset", csts);
2415                         return -ENODEV;
2416                 }
2417         }
2418
2419         return ret;
2420 }
2421
2422 /*
2423  * If the device has been passed off to us in an enabled state, just clear
2424  * the enabled bit.  The spec says we should set the 'shutdown notification
2425  * bits', but doing so may cause the device to complete commands to the
2426  * admin queue ... and we don't know what memory that might be pointing at!
2427  */
2428 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2429 {
2430         int ret;
2431
2432         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2433         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2434
2435         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2436         if (ret)
2437                 return ret;
2438
2439         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2440                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2441
2442         return nvme_wait_ready(ctrl, ctrl->cap, false);
2443 }
2444 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2445
2446 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2447 {
2448         unsigned dev_page_min;
2449         int ret;
2450
2451         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2452         if (ret) {
2453                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2454                 return ret;
2455         }
2456         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2457
2458         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2459                 dev_err(ctrl->device,
2460                         "Minimum device page size %u too large for host (%u)\n",
2461                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2462                 return -ENODEV;
2463         }
2464
2465         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2466                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2467         else
2468                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2469         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2470         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2471         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2472         ctrl->ctrl_config |= NVME_CC_ENABLE;
2473
2474         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2475         if (ret)
2476                 return ret;
2477         return nvme_wait_ready(ctrl, ctrl->cap, true);
2478 }
2479 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2480
2481 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2482 {
2483         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2484         u32 csts;
2485         int ret;
2486
2487         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2488         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2489
2490         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2491         if (ret)
2492                 return ret;
2493
2494         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2495                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2496                         break;
2497
2498                 msleep(100);
2499                 if (fatal_signal_pending(current))
2500                         return -EINTR;
2501                 if (time_after(jiffies, timeout)) {
2502                         dev_err(ctrl->device,
2503                                 "Device shutdown incomplete; abort shutdown\n");
2504                         return -ENODEV;
2505                 }
2506         }
2507
2508         return ret;
2509 }
2510 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2511
2512 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2513 {
2514         __le64 ts;
2515         int ret;
2516
2517         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2518                 return 0;
2519
2520         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2521         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2522                         NULL);
2523         if (ret)
2524                 dev_warn_once(ctrl->device,
2525                         "could not set timestamp (%d)\n", ret);
2526         return ret;
2527 }
2528
2529 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2530 {
2531         struct nvme_feat_host_behavior *host;
2532         int ret;
2533
2534         /* Don't bother enabling the feature if retry delay is not reported */
2535         if (!ctrl->crdt[0])
2536                 return 0;
2537
2538         host = kzalloc(sizeof(*host), GFP_KERNEL);
2539         if (!host)
2540                 return 0;
2541
2542         host->acre = NVME_ENABLE_ACRE;
2543         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2544                                 host, sizeof(*host), NULL);
2545         kfree(host);
2546         return ret;
2547 }
2548
2549 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2550 {
2551         /*
2552          * APST (Autonomous Power State Transition) lets us program a
2553          * table of power state transitions that the controller will
2554          * perform automatically.  We configure it with a simple
2555          * heuristic: we are willing to spend at most 2% of the time
2556          * transitioning between power states.  Therefore, when running
2557          * in any given state, we will enter the next lower-power
2558          * non-operational state after waiting 50 * (enlat + exlat)
2559          * microseconds, as long as that state's exit latency is under
2560          * the requested maximum latency.
2561          *
2562          * We will not autonomously enter any non-operational state for
2563          * which the total latency exceeds ps_max_latency_us.  Users
2564          * can set ps_max_latency_us to zero to turn off APST.
2565          */
2566
2567         unsigned apste;
2568         struct nvme_feat_auto_pst *table;
2569         u64 max_lat_us = 0;
2570         int max_ps = -1;
2571         int ret;
2572
2573         /*
2574          * If APST isn't supported or if we haven't been initialized yet,
2575          * then don't do anything.
2576          */
2577         if (!ctrl->apsta)
2578                 return 0;
2579
2580         if (ctrl->npss > 31) {
2581                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2582                 return 0;
2583         }
2584
2585         table = kzalloc(sizeof(*table), GFP_KERNEL);
2586         if (!table)
2587                 return 0;
2588
2589         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2590                 /* Turn off APST. */
2591                 apste = 0;
2592                 dev_dbg(ctrl->device, "APST disabled\n");
2593         } else {
2594                 __le64 target = cpu_to_le64(0);
2595                 int state;
2596
2597                 /*
2598                  * Walk through all states from lowest- to highest-power.
2599                  * According to the spec, lower-numbered states use more
2600                  * power.  NPSS, despite the name, is the index of the
2601                  * lowest-power state, not the number of states.
2602                  */
2603                 for (state = (int)ctrl->npss; state >= 0; state--) {
2604                         u64 total_latency_us, exit_latency_us, transition_ms;
2605
2606                         if (target)
2607                                 table->entries[state] = target;
2608
2609                         /*
2610                          * Don't allow transitions to the deepest state
2611                          * if it's quirked off.
2612                          */
2613                         if (state == ctrl->npss &&
2614                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2615                                 continue;
2616
2617                         /*
2618                          * Is this state a useful non-operational state for
2619                          * higher-power states to autonomously transition to?
2620                          */
2621                         if (!(ctrl->psd[state].flags &
2622                               NVME_PS_FLAGS_NON_OP_STATE))
2623                                 continue;
2624
2625                         exit_latency_us =
2626                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2627                         if (exit_latency_us > ctrl->ps_max_latency_us)
2628                                 continue;
2629
2630                         total_latency_us =
2631                                 exit_latency_us +
2632                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2633
2634                         /*
2635                          * This state is good.  Use it as the APST idle
2636                          * target for higher power states.
2637                          */
2638                         transition_ms = total_latency_us + 19;
2639                         do_div(transition_ms, 20);
2640                         if (transition_ms > (1 << 24) - 1)
2641                                 transition_ms = (1 << 24) - 1;
2642
2643                         target = cpu_to_le64((state << 3) |
2644                                              (transition_ms << 8));
2645
2646                         if (max_ps == -1)
2647                                 max_ps = state;
2648
2649                         if (total_latency_us > max_lat_us)
2650                                 max_lat_us = total_latency_us;
2651                 }
2652
2653                 apste = 1;
2654
2655                 if (max_ps == -1) {
2656                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2657                 } else {
2658                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2659                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2660                 }
2661         }
2662
2663         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2664                                 table, sizeof(*table), NULL);
2665         if (ret)
2666                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2667
2668         kfree(table);
2669         return ret;
2670 }
2671
2672 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2673 {
2674         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2675         u64 latency;
2676
2677         switch (val) {
2678         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2679         case PM_QOS_LATENCY_ANY:
2680                 latency = U64_MAX;
2681                 break;
2682
2683         default:
2684                 latency = val;
2685         }
2686
2687         if (ctrl->ps_max_latency_us != latency) {
2688                 ctrl->ps_max_latency_us = latency;
2689                 nvme_configure_apst(ctrl);
2690         }
2691 }
2692
2693 struct nvme_core_quirk_entry {
2694         /*
2695          * NVMe model and firmware strings are padded with spaces.  For
2696          * simplicity, strings in the quirk table are padded with NULLs
2697          * instead.
2698          */
2699         u16 vid;
2700         const char *mn;
2701         const char *fr;
2702         unsigned long quirks;
2703 };
2704
2705 static const struct nvme_core_quirk_entry core_quirks[] = {
2706         {
2707                 /*
2708                  * This Toshiba device seems to die using any APST states.  See:
2709                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2710                  */
2711                 .vid = 0x1179,
2712                 .mn = "THNSF5256GPUK TOSHIBA",
2713                 .quirks = NVME_QUIRK_NO_APST,
2714         },
2715         {
2716                 /*
2717                  * This LiteON CL1-3D*-Q11 firmware version has a race
2718                  * condition associated with actions related to suspend to idle
2719                  * LiteON has resolved the problem in future firmware
2720                  */
2721                 .vid = 0x14a4,
2722                 .fr = "22301111",
2723                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2724         }
2725 };
2726
2727 /* match is null-terminated but idstr is space-padded. */
2728 static bool string_matches(const char *idstr, const char *match, size_t len)
2729 {
2730         size_t matchlen;
2731
2732         if (!match)
2733                 return true;
2734
2735         matchlen = strlen(match);
2736         WARN_ON_ONCE(matchlen > len);
2737
2738         if (memcmp(idstr, match, matchlen))
2739                 return false;
2740
2741         for (; matchlen < len; matchlen++)
2742                 if (idstr[matchlen] != ' ')
2743                         return false;
2744
2745         return true;
2746 }
2747
2748 static bool quirk_matches(const struct nvme_id_ctrl *id,
2749                           const struct nvme_core_quirk_entry *q)
2750 {
2751         return q->vid == le16_to_cpu(id->vid) &&
2752                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2753                 string_matches(id->fr, q->fr, sizeof(id->fr));
2754 }
2755
2756 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2757                 struct nvme_id_ctrl *id)
2758 {
2759         size_t nqnlen;
2760         int off;
2761
2762         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2763                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2764                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2765                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2766                         return;
2767                 }
2768
2769                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2770                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2771         }
2772
2773         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2774         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2775                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2776                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2777         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2778         off += sizeof(id->sn);
2779         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2780         off += sizeof(id->mn);
2781         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2782 }
2783
2784 static void nvme_release_subsystem(struct device *dev)
2785 {
2786         struct nvme_subsystem *subsys =
2787                 container_of(dev, struct nvme_subsystem, dev);
2788
2789         if (subsys->instance >= 0)
2790                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2791         kfree(subsys);
2792 }
2793
2794 static void nvme_destroy_subsystem(struct kref *ref)
2795 {
2796         struct nvme_subsystem *subsys =
2797                         container_of(ref, struct nvme_subsystem, ref);
2798
2799         mutex_lock(&nvme_subsystems_lock);
2800         list_del(&subsys->entry);
2801         mutex_unlock(&nvme_subsystems_lock);
2802
2803         ida_destroy(&subsys->ns_ida);
2804         device_del(&subsys->dev);
2805         put_device(&subsys->dev);
2806 }
2807
2808 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2809 {
2810         kref_put(&subsys->ref, nvme_destroy_subsystem);
2811 }
2812
2813 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2814 {
2815         struct nvme_subsystem *subsys;
2816
2817         lockdep_assert_held(&nvme_subsystems_lock);
2818
2819         /*
2820          * Fail matches for discovery subsystems. This results
2821          * in each discovery controller bound to a unique subsystem.
2822          * This avoids issues with validating controller values
2823          * that can only be true when there is a single unique subsystem.
2824          * There may be multiple and completely independent entities
2825          * that provide discovery controllers.
2826          */
2827         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2828                 return NULL;
2829
2830         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2831                 if (strcmp(subsys->subnqn, subsysnqn))
2832                         continue;
2833                 if (!kref_get_unless_zero(&subsys->ref))
2834                         continue;
2835                 return subsys;
2836         }
2837
2838         return NULL;
2839 }
2840
2841 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2842         struct device_attribute subsys_attr_##_name = \
2843                 __ATTR(_name, _mode, _show, NULL)
2844
2845 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2846                                     struct device_attribute *attr,
2847                                     char *buf)
2848 {
2849         struct nvme_subsystem *subsys =
2850                 container_of(dev, struct nvme_subsystem, dev);
2851
2852         return sysfs_emit(buf, "%s\n", subsys->subnqn);
2853 }
2854 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2855
2856 #define nvme_subsys_show_str_function(field)                            \
2857 static ssize_t subsys_##field##_show(struct device *dev,                \
2858                             struct device_attribute *attr, char *buf)   \
2859 {                                                                       \
2860         struct nvme_subsystem *subsys =                                 \
2861                 container_of(dev, struct nvme_subsystem, dev);          \
2862         return sysfs_emit(buf, "%.*s\n",                                \
2863                            (int)sizeof(subsys->field), subsys->field);  \
2864 }                                                                       \
2865 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2866
2867 nvme_subsys_show_str_function(model);
2868 nvme_subsys_show_str_function(serial);
2869 nvme_subsys_show_str_function(firmware_rev);
2870
2871 static struct attribute *nvme_subsys_attrs[] = {
2872         &subsys_attr_model.attr,
2873         &subsys_attr_serial.attr,
2874         &subsys_attr_firmware_rev.attr,
2875         &subsys_attr_subsysnqn.attr,
2876 #ifdef CONFIG_NVME_MULTIPATH
2877         &subsys_attr_iopolicy.attr,
2878 #endif
2879         NULL,
2880 };
2881
2882 static const struct attribute_group nvme_subsys_attrs_group = {
2883         .attrs = nvme_subsys_attrs,
2884 };
2885
2886 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2887         &nvme_subsys_attrs_group,
2888         NULL,
2889 };
2890
2891 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2892 {
2893         return ctrl->opts && ctrl->opts->discovery_nqn;
2894 }
2895
2896 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2897                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2898 {
2899         struct nvme_ctrl *tmp;
2900
2901         lockdep_assert_held(&nvme_subsystems_lock);
2902
2903         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2904                 if (nvme_state_terminal(tmp))
2905                         continue;
2906
2907                 if (tmp->cntlid == ctrl->cntlid) {
2908                         dev_err(ctrl->device,
2909                                 "Duplicate cntlid %u with %s, rejecting\n",
2910                                 ctrl->cntlid, dev_name(tmp->device));
2911                         return false;
2912                 }
2913
2914                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2915                     nvme_discovery_ctrl(ctrl))
2916                         continue;
2917
2918                 dev_err(ctrl->device,
2919                         "Subsystem does not support multiple controllers\n");
2920                 return false;
2921         }
2922
2923         return true;
2924 }
2925
2926 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2927 {
2928         struct nvme_subsystem *subsys, *found;
2929         int ret;
2930
2931         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2932         if (!subsys)
2933                 return -ENOMEM;
2934
2935         subsys->instance = -1;
2936         mutex_init(&subsys->lock);
2937         kref_init(&subsys->ref);
2938         INIT_LIST_HEAD(&subsys->ctrls);
2939         INIT_LIST_HEAD(&subsys->nsheads);
2940         nvme_init_subnqn(subsys, ctrl, id);
2941         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2942         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2943         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2944         subsys->vendor_id = le16_to_cpu(id->vid);
2945         subsys->cmic = id->cmic;
2946         subsys->awupf = le16_to_cpu(id->awupf);
2947 #ifdef CONFIG_NVME_MULTIPATH
2948         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2949 #endif
2950
2951         subsys->dev.class = nvme_subsys_class;
2952         subsys->dev.release = nvme_release_subsystem;
2953         subsys->dev.groups = nvme_subsys_attrs_groups;
2954         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2955         device_initialize(&subsys->dev);
2956
2957         mutex_lock(&nvme_subsystems_lock);
2958         found = __nvme_find_get_subsystem(subsys->subnqn);
2959         if (found) {
2960                 put_device(&subsys->dev);
2961                 subsys = found;
2962
2963                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2964                         ret = -EINVAL;
2965                         goto out_put_subsystem;
2966                 }
2967         } else {
2968                 ret = device_add(&subsys->dev);
2969                 if (ret) {
2970                         dev_err(ctrl->device,
2971                                 "failed to register subsystem device.\n");
2972                         put_device(&subsys->dev);
2973                         goto out_unlock;
2974                 }
2975                 ida_init(&subsys->ns_ida);
2976                 list_add_tail(&subsys->entry, &nvme_subsystems);
2977         }
2978
2979         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2980                                 dev_name(ctrl->device));
2981         if (ret) {
2982                 dev_err(ctrl->device,
2983                         "failed to create sysfs link from subsystem.\n");
2984                 goto out_put_subsystem;
2985         }
2986
2987         if (!found)
2988                 subsys->instance = ctrl->instance;
2989         ctrl->subsys = subsys;
2990         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2991         mutex_unlock(&nvme_subsystems_lock);
2992         return 0;
2993
2994 out_put_subsystem:
2995         nvme_put_subsystem(subsys);
2996 out_unlock:
2997         mutex_unlock(&nvme_subsystems_lock);
2998         return ret;
2999 }
3000
3001 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3002                 void *log, size_t size, u64 offset)
3003 {
3004         struct nvme_command c = { };
3005         u32 dwlen = nvme_bytes_to_numd(size);
3006
3007         c.get_log_page.opcode = nvme_admin_get_log_page;
3008         c.get_log_page.nsid = cpu_to_le32(nsid);
3009         c.get_log_page.lid = log_page;
3010         c.get_log_page.lsp = lsp;
3011         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3012         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3013         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3014         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3015         c.get_log_page.csi = csi;
3016
3017         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3018 }
3019
3020 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3021                                 struct nvme_effects_log **log)
3022 {
3023         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3024         int ret;
3025
3026         if (cel)
3027                 goto out;
3028
3029         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3030         if (!cel)
3031                 return -ENOMEM;
3032
3033         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3034                         cel, sizeof(*cel), 0);
3035         if (ret) {
3036                 kfree(cel);
3037                 return ret;
3038         }
3039
3040         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3041 out:
3042         *log = cel;
3043         return 0;
3044 }
3045
3046 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3047 {
3048         u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3049
3050         if (check_shl_overflow(1U, units + page_shift - 9, &val))
3051                 return UINT_MAX;
3052         return val;
3053 }
3054
3055 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3056 {
3057         struct nvme_command c = { };
3058         struct nvme_id_ctrl_nvm *id;
3059         int ret;
3060
3061         if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
3062                 ctrl->max_discard_sectors = UINT_MAX;
3063                 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
3064         } else {
3065                 ctrl->max_discard_sectors = 0;
3066                 ctrl->max_discard_segments = 0;
3067         }
3068
3069         /*
3070          * Even though NVMe spec explicitly states that MDTS is not applicable
3071          * to the write-zeroes, we are cautious and limit the size to the
3072          * controllers max_hw_sectors value, which is based on the MDTS field
3073          * and possibly other limiting factors.
3074          */
3075         if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3076             !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3077                 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3078         else
3079                 ctrl->max_zeroes_sectors = 0;
3080
3081         if (nvme_ctrl_limited_cns(ctrl))
3082                 return 0;
3083
3084         id = kzalloc(sizeof(*id), GFP_KERNEL);
3085         if (!id)
3086                 return 0;
3087
3088         c.identify.opcode = nvme_admin_identify;
3089         c.identify.cns = NVME_ID_CNS_CS_CTRL;
3090         c.identify.csi = NVME_CSI_NVM;
3091
3092         ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3093         if (ret)
3094                 goto free_data;
3095
3096         if (id->dmrl)
3097                 ctrl->max_discard_segments = id->dmrl;
3098         if (id->dmrsl)
3099                 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
3100         if (id->wzsl)
3101                 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3102
3103 free_data:
3104         kfree(id);
3105         return ret;
3106 }
3107
3108 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3109 {
3110         struct nvme_id_ctrl *id;
3111         u32 max_hw_sectors;
3112         bool prev_apst_enabled;
3113         int ret;
3114
3115         ret = nvme_identify_ctrl(ctrl, &id);
3116         if (ret) {
3117                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3118                 return -EIO;
3119         }
3120
3121         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3122                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3123                 if (ret < 0)
3124                         goto out_free;
3125         }
3126
3127         if (!(ctrl->ops->flags & NVME_F_FABRICS))
3128                 ctrl->cntlid = le16_to_cpu(id->cntlid);
3129
3130         if (!ctrl->identified) {
3131                 unsigned int i;
3132
3133                 ret = nvme_init_subsystem(ctrl, id);
3134                 if (ret)
3135                         goto out_free;
3136
3137                 /*
3138                  * Check for quirks.  Quirk can depend on firmware version,
3139                  * so, in principle, the set of quirks present can change
3140                  * across a reset.  As a possible future enhancement, we
3141                  * could re-scan for quirks every time we reinitialize
3142                  * the device, but we'd have to make sure that the driver
3143                  * behaves intelligently if the quirks change.
3144                  */
3145                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3146                         if (quirk_matches(id, &core_quirks[i]))
3147                                 ctrl->quirks |= core_quirks[i].quirks;
3148                 }
3149         }
3150
3151         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3152                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3153                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3154         }
3155
3156         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3157         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3158         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3159
3160         ctrl->oacs = le16_to_cpu(id->oacs);
3161         ctrl->oncs = le16_to_cpu(id->oncs);
3162         ctrl->mtfa = le16_to_cpu(id->mtfa);
3163         ctrl->oaes = le32_to_cpu(id->oaes);
3164         ctrl->wctemp = le16_to_cpu(id->wctemp);
3165         ctrl->cctemp = le16_to_cpu(id->cctemp);
3166
3167         atomic_set(&ctrl->abort_limit, id->acl + 1);
3168         ctrl->vwc = id->vwc;
3169         if (id->mdts)
3170                 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3171         else
3172                 max_hw_sectors = UINT_MAX;
3173         ctrl->max_hw_sectors =
3174                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3175
3176         nvme_set_queue_limits(ctrl, ctrl->admin_q);
3177         ctrl->sgls = le32_to_cpu(id->sgls);
3178         ctrl->kas = le16_to_cpu(id->kas);
3179         ctrl->max_namespaces = le32_to_cpu(id->mnan);
3180         ctrl->ctratt = le32_to_cpu(id->ctratt);
3181
3182         if (id->rtd3e) {
3183                 /* us -> s */
3184                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3185
3186                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3187                                                  shutdown_timeout, 60);
3188
3189                 if (ctrl->shutdown_timeout != shutdown_timeout)
3190                         dev_info(ctrl->device,
3191                                  "Shutdown timeout set to %u seconds\n",
3192                                  ctrl->shutdown_timeout);
3193         } else
3194                 ctrl->shutdown_timeout = shutdown_timeout;
3195
3196         ctrl->npss = id->npss;
3197         ctrl->apsta = id->apsta;
3198         prev_apst_enabled = ctrl->apst_enabled;
3199         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3200                 if (force_apst && id->apsta) {
3201                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3202                         ctrl->apst_enabled = true;
3203                 } else {
3204                         ctrl->apst_enabled = false;
3205                 }
3206         } else {
3207                 ctrl->apst_enabled = id->apsta;
3208         }
3209         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3210
3211         if (ctrl->ops->flags & NVME_F_FABRICS) {
3212                 ctrl->icdoff = le16_to_cpu(id->icdoff);
3213                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3214                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3215                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3216
3217                 /*
3218                  * In fabrics we need to verify the cntlid matches the
3219                  * admin connect
3220                  */
3221                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3222                         dev_err(ctrl->device,
3223                                 "Mismatching cntlid: Connect %u vs Identify "
3224                                 "%u, rejecting\n",
3225                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
3226                         ret = -EINVAL;
3227                         goto out_free;
3228                 }
3229
3230                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3231                         dev_err(ctrl->device,
3232                                 "keep-alive support is mandatory for fabrics\n");
3233                         ret = -EINVAL;
3234                         goto out_free;
3235                 }
3236         } else {
3237                 ctrl->hmpre = le32_to_cpu(id->hmpre);
3238                 ctrl->hmmin = le32_to_cpu(id->hmmin);
3239                 ctrl->hmminds = le32_to_cpu(id->hmminds);
3240                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3241         }
3242
3243         ret = nvme_mpath_init(ctrl, id);
3244         if (ret < 0)
3245                 goto out_free;
3246
3247         if (ctrl->apst_enabled && !prev_apst_enabled)
3248                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3249         else if (!ctrl->apst_enabled && prev_apst_enabled)
3250                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3251
3252 out_free:
3253         kfree(id);
3254         return ret;
3255 }
3256
3257 /*
3258  * Initialize the cached copies of the Identify data and various controller
3259  * register in our nvme_ctrl structure.  This should be called as soon as
3260  * the admin queue is fully up and running.
3261  */
3262 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3263 {
3264         int ret;
3265
3266         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3267         if (ret) {
3268                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3269                 return ret;
3270         }
3271
3272         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3273
3274         if (ctrl->vs >= NVME_VS(1, 1, 0))
3275                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3276
3277         ret = nvme_init_identify(ctrl);
3278         if (ret)
3279                 return ret;
3280
3281         ret = nvme_init_non_mdts_limits(ctrl);
3282         if (ret < 0)
3283                 return ret;
3284
3285         ret = nvme_configure_apst(ctrl);
3286         if (ret < 0)
3287                 return ret;
3288
3289         ret = nvme_configure_timestamp(ctrl);
3290         if (ret < 0)
3291                 return ret;
3292
3293         ret = nvme_configure_directives(ctrl);
3294         if (ret < 0)
3295                 return ret;
3296
3297         ret = nvme_configure_acre(ctrl);
3298         if (ret < 0)
3299                 return ret;
3300
3301         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3302                 ret = nvme_hwmon_init(ctrl);
3303                 if (ret < 0)
3304                         return ret;
3305         }
3306
3307         ctrl->identified = true;
3308
3309         return 0;
3310 }
3311 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3312
3313 static int nvme_dev_open(struct inode *inode, struct file *file)
3314 {
3315         struct nvme_ctrl *ctrl =
3316                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3317
3318         switch (ctrl->state) {
3319         case NVME_CTRL_LIVE:
3320                 break;
3321         default:
3322                 return -EWOULDBLOCK;
3323         }
3324
3325         nvme_get_ctrl(ctrl);
3326         if (!try_module_get(ctrl->ops->module)) {
3327                 nvme_put_ctrl(ctrl);
3328                 return -EINVAL;
3329         }
3330
3331         file->private_data = ctrl;
3332         return 0;
3333 }
3334
3335 static int nvme_dev_release(struct inode *inode, struct file *file)
3336 {
3337         struct nvme_ctrl *ctrl =
3338                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3339
3340         module_put(ctrl->ops->module);
3341         nvme_put_ctrl(ctrl);
3342         return 0;
3343 }
3344
3345 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3346 {
3347         struct nvme_ns *ns;
3348         int ret;
3349
3350         down_read(&ctrl->namespaces_rwsem);
3351         if (list_empty(&ctrl->namespaces)) {
3352                 ret = -ENOTTY;
3353                 goto out_unlock;
3354         }
3355
3356         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3357         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3358                 dev_warn(ctrl->device,
3359                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3360                 ret = -EINVAL;
3361                 goto out_unlock;
3362         }
3363
3364         dev_warn(ctrl->device,
3365                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3366         kref_get(&ns->kref);
3367         up_read(&ctrl->namespaces_rwsem);
3368
3369         ret = nvme_user_cmd(ctrl, ns, argp);
3370         nvme_put_ns(ns);
3371         return ret;
3372
3373 out_unlock:
3374         up_read(&ctrl->namespaces_rwsem);
3375         return ret;
3376 }
3377
3378 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3379                 unsigned long arg)
3380 {
3381         struct nvme_ctrl *ctrl = file->private_data;
3382         void __user *argp = (void __user *)arg;
3383
3384         switch (cmd) {
3385         case NVME_IOCTL_ADMIN_CMD:
3386                 return nvme_user_cmd(ctrl, NULL, argp);
3387         case NVME_IOCTL_ADMIN64_CMD:
3388                 return nvme_user_cmd64(ctrl, NULL, argp);
3389         case NVME_IOCTL_IO_CMD:
3390                 return nvme_dev_user_cmd(ctrl, argp);
3391         case NVME_IOCTL_RESET:
3392                 dev_warn(ctrl->device, "resetting controller\n");
3393                 return nvme_reset_ctrl_sync(ctrl);
3394         case NVME_IOCTL_SUBSYS_RESET:
3395                 return nvme_reset_subsystem(ctrl);
3396         case NVME_IOCTL_RESCAN:
3397                 nvme_queue_scan(ctrl);
3398                 return 0;
3399         default:
3400                 return -ENOTTY;
3401         }
3402 }
3403
3404 static const struct file_operations nvme_dev_fops = {
3405         .owner          = THIS_MODULE,
3406         .open           = nvme_dev_open,
3407         .release        = nvme_dev_release,
3408         .unlocked_ioctl = nvme_dev_ioctl,
3409         .compat_ioctl   = compat_ptr_ioctl,
3410 };
3411
3412 static ssize_t nvme_sysfs_reset(struct device *dev,
3413                                 struct device_attribute *attr, const char *buf,
3414                                 size_t count)
3415 {
3416         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3417         int ret;
3418
3419         ret = nvme_reset_ctrl_sync(ctrl);
3420         if (ret < 0)
3421                 return ret;
3422         return count;
3423 }
3424 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3425
3426 static ssize_t nvme_sysfs_rescan(struct device *dev,
3427                                 struct device_attribute *attr, const char *buf,
3428                                 size_t count)
3429 {
3430         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3431
3432         nvme_queue_scan(ctrl);
3433         return count;
3434 }
3435 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3436
3437 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3438 {
3439         struct gendisk *disk = dev_to_disk(dev);
3440
3441         if (disk->fops == &nvme_bdev_ops)
3442                 return nvme_get_ns_from_dev(dev)->head;
3443         else
3444                 return disk->private_data;
3445 }
3446
3447 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3448                 char *buf)
3449 {
3450         struct nvme_ns_head *head = dev_to_ns_head(dev);
3451         struct nvme_ns_ids *ids = &head->ids;
3452         struct nvme_subsystem *subsys = head->subsys;
3453         int serial_len = sizeof(subsys->serial);
3454         int model_len = sizeof(subsys->model);
3455
3456         if (!uuid_is_null(&ids->uuid))
3457                 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3458
3459         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3460                 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3461
3462         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3463                 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3464
3465         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3466                                   subsys->serial[serial_len - 1] == '\0'))
3467                 serial_len--;
3468         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3469                                  subsys->model[model_len - 1] == '\0'))
3470                 model_len--;
3471
3472         return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3473                 serial_len, subsys->serial, model_len, subsys->model,
3474                 head->ns_id);
3475 }
3476 static DEVICE_ATTR_RO(wwid);
3477
3478 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3479                 char *buf)
3480 {
3481         return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3482 }
3483 static DEVICE_ATTR_RO(nguid);
3484
3485 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3486                 char *buf)
3487 {
3488         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3489
3490         /* For backward compatibility expose the NGUID to userspace if
3491          * we have no UUID set
3492          */
3493         if (uuid_is_null(&ids->uuid)) {
3494                 printk_ratelimited(KERN_WARNING
3495                                    "No UUID available providing old NGUID\n");
3496                 return sysfs_emit(buf, "%pU\n", ids->nguid);
3497         }
3498         return sysfs_emit(buf, "%pU\n", &ids->uuid);
3499 }
3500 static DEVICE_ATTR_RO(uuid);
3501
3502 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3503                 char *buf)
3504 {
3505         return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3506 }
3507 static DEVICE_ATTR_RO(eui);
3508
3509 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3510                 char *buf)
3511 {
3512         return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3513 }
3514 static DEVICE_ATTR_RO(nsid);
3515
3516 static struct attribute *nvme_ns_id_attrs[] = {
3517         &dev_attr_wwid.attr,
3518         &dev_attr_uuid.attr,
3519         &dev_attr_nguid.attr,
3520         &dev_attr_eui.attr,
3521         &dev_attr_nsid.attr,
3522 #ifdef CONFIG_NVME_MULTIPATH
3523         &dev_attr_ana_grpid.attr,
3524         &dev_attr_ana_state.attr,
3525 #endif
3526         NULL,
3527 };
3528
3529 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3530                 struct attribute *a, int n)
3531 {
3532         struct device *dev = container_of(kobj, struct device, kobj);
3533         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3534
3535         if (a == &dev_attr_uuid.attr) {
3536                 if (uuid_is_null(&ids->uuid) &&
3537                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3538                         return 0;
3539         }
3540         if (a == &dev_attr_nguid.attr) {
3541                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3542                         return 0;
3543         }
3544         if (a == &dev_attr_eui.attr) {
3545                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3546                         return 0;
3547         }
3548 #ifdef CONFIG_NVME_MULTIPATH
3549         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3550                 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3551                         return 0;
3552                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3553                         return 0;
3554         }
3555 #endif
3556         return a->mode;
3557 }
3558
3559 static const struct attribute_group nvme_ns_id_attr_group = {
3560         .attrs          = nvme_ns_id_attrs,
3561         .is_visible     = nvme_ns_id_attrs_are_visible,
3562 };
3563
3564 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3565         &nvme_ns_id_attr_group,
3566 #ifdef CONFIG_NVM
3567         &nvme_nvm_attr_group,
3568 #endif
3569         NULL,
3570 };
3571
3572 #define nvme_show_str_function(field)                                           \
3573 static ssize_t  field##_show(struct device *dev,                                \
3574                             struct device_attribute *attr, char *buf)           \
3575 {                                                                               \
3576         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3577         return sysfs_emit(buf, "%.*s\n",                                        \
3578                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3579 }                                                                               \
3580 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3581
3582 nvme_show_str_function(model);
3583 nvme_show_str_function(serial);
3584 nvme_show_str_function(firmware_rev);
3585
3586 #define nvme_show_int_function(field)                                           \
3587 static ssize_t  field##_show(struct device *dev,                                \
3588                             struct device_attribute *attr, char *buf)           \
3589 {                                                                               \
3590         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3591         return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3592 }                                                                               \
3593 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3594
3595 nvme_show_int_function(cntlid);
3596 nvme_show_int_function(numa_node);
3597 nvme_show_int_function(queue_count);
3598 nvme_show_int_function(sqsize);
3599
3600 static ssize_t nvme_sysfs_delete(struct device *dev,
3601                                 struct device_attribute *attr, const char *buf,
3602                                 size_t count)
3603 {
3604         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3605
3606         if (device_remove_file_self(dev, attr))
3607                 nvme_delete_ctrl_sync(ctrl);
3608         return count;
3609 }
3610 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3611
3612 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3613                                          struct device_attribute *attr,
3614                                          char *buf)
3615 {
3616         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3617
3618         return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3619 }
3620 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3621
3622 static ssize_t nvme_sysfs_show_state(struct device *dev,
3623                                      struct device_attribute *attr,
3624                                      char *buf)
3625 {
3626         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3627         static const char *const state_name[] = {
3628                 [NVME_CTRL_NEW]         = "new",
3629                 [NVME_CTRL_LIVE]        = "live",
3630                 [NVME_CTRL_RESETTING]   = "resetting",
3631                 [NVME_CTRL_CONNECTING]  = "connecting",
3632                 [NVME_CTRL_DELETING]    = "deleting",
3633                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3634                 [NVME_CTRL_DEAD]        = "dead",
3635         };
3636
3637         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3638             state_name[ctrl->state])
3639                 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3640
3641         return sysfs_emit(buf, "unknown state\n");
3642 }
3643
3644 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3645
3646 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3647                                          struct device_attribute *attr,
3648                                          char *buf)
3649 {
3650         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3651
3652         return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3653 }
3654 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3655
3656 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3657                                         struct device_attribute *attr,
3658                                         char *buf)
3659 {
3660         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3661
3662         return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3663 }
3664 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3665
3666 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3667                                         struct device_attribute *attr,
3668                                         char *buf)
3669 {
3670         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3671
3672         return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3673 }
3674 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3675
3676 static ssize_t nvme_sysfs_show_address(struct device *dev,
3677                                          struct device_attribute *attr,
3678                                          char *buf)
3679 {
3680         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3681
3682         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3683 }
3684 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3685
3686 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3687                 struct device_attribute *attr, char *buf)
3688 {
3689         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3690         struct nvmf_ctrl_options *opts = ctrl->opts;
3691
3692         if (ctrl->opts->max_reconnects == -1)
3693                 return sysfs_emit(buf, "off\n");
3694         return sysfs_emit(buf, "%d\n",
3695                           opts->max_reconnects * opts->reconnect_delay);
3696 }
3697
3698 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3699                 struct device_attribute *attr, const char *buf, size_t count)
3700 {
3701         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3702         struct nvmf_ctrl_options *opts = ctrl->opts;
3703         int ctrl_loss_tmo, err;
3704
3705         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3706         if (err)
3707                 return -EINVAL;
3708
3709         if (ctrl_loss_tmo < 0)
3710                 opts->max_reconnects = -1;
3711         else
3712                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3713                                                 opts->reconnect_delay);
3714         return count;
3715 }
3716 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3717         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3718
3719 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3720                 struct device_attribute *attr, char *buf)
3721 {
3722         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3723
3724         if (ctrl->opts->reconnect_delay == -1)
3725                 return sysfs_emit(buf, "off\n");
3726         return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3727 }
3728
3729 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3730                 struct device_attribute *attr, const char *buf, size_t count)
3731 {
3732         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3733         unsigned int v;
3734         int err;
3735
3736         err = kstrtou32(buf, 10, &v);
3737         if (err)
3738                 return err;
3739
3740         ctrl->opts->reconnect_delay = v;
3741         return count;
3742 }
3743 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3744         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3745
3746 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3747                 struct device_attribute *attr, char *buf)
3748 {
3749         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3750
3751         if (ctrl->opts->fast_io_fail_tmo == -1)
3752                 return sysfs_emit(buf, "off\n");
3753         return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3754 }
3755
3756 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3757                 struct device_attribute *attr, const char *buf, size_t count)
3758 {
3759         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3760         struct nvmf_ctrl_options *opts = ctrl->opts;
3761         int fast_io_fail_tmo, err;
3762
3763         err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3764         if (err)
3765                 return -EINVAL;
3766
3767         if (fast_io_fail_tmo < 0)
3768                 opts->fast_io_fail_tmo = -1;
3769         else
3770                 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3771         return count;
3772 }
3773 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3774         nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3775
3776 static struct attribute *nvme_dev_attrs[] = {
3777         &dev_attr_reset_controller.attr,
3778         &dev_attr_rescan_controller.attr,
3779         &dev_attr_model.attr,
3780         &dev_attr_serial.attr,
3781         &dev_attr_firmware_rev.attr,
3782         &dev_attr_cntlid.attr,
3783         &dev_attr_delete_controller.attr,
3784         &dev_attr_transport.attr,
3785         &dev_attr_subsysnqn.attr,
3786         &dev_attr_address.attr,
3787         &dev_attr_state.attr,
3788         &dev_attr_numa_node.attr,
3789         &dev_attr_queue_count.attr,
3790         &dev_attr_sqsize.attr,
3791         &dev_attr_hostnqn.attr,
3792         &dev_attr_hostid.attr,
3793         &dev_attr_ctrl_loss_tmo.attr,
3794         &dev_attr_reconnect_delay.attr,
3795         &dev_attr_fast_io_fail_tmo.attr,
3796         NULL
3797 };
3798
3799 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3800                 struct attribute *a, int n)
3801 {
3802         struct device *dev = container_of(kobj, struct device, kobj);
3803         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3804
3805         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3806                 return 0;
3807         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3808                 return 0;
3809         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3810                 return 0;
3811         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3812                 return 0;
3813         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3814                 return 0;
3815         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3816                 return 0;
3817
3818         return a->mode;
3819 }
3820
3821 static const struct attribute_group nvme_dev_attrs_group = {
3822         .attrs          = nvme_dev_attrs,
3823         .is_visible     = nvme_dev_attrs_are_visible,
3824 };
3825
3826 static const struct attribute_group *nvme_dev_attr_groups[] = {
3827         &nvme_dev_attrs_group,
3828         NULL,
3829 };
3830
3831 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3832                 unsigned nsid)
3833 {
3834         struct nvme_ns_head *h;
3835
3836         lockdep_assert_held(&subsys->lock);
3837
3838         list_for_each_entry(h, &subsys->nsheads, entry) {
3839                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3840                         return h;
3841         }
3842
3843         return NULL;
3844 }
3845
3846 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3847                 struct nvme_ns_head *new)
3848 {
3849         struct nvme_ns_head *h;
3850
3851         lockdep_assert_held(&subsys->lock);
3852
3853         list_for_each_entry(h, &subsys->nsheads, entry) {
3854                 if (nvme_ns_ids_valid(&new->ids) &&
3855                     nvme_ns_ids_equal(&new->ids, &h->ids))
3856                         return -EINVAL;
3857         }
3858
3859         return 0;
3860 }
3861
3862 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3863                 unsigned nsid, struct nvme_ns_ids *ids)
3864 {
3865         struct nvme_ns_head *head;
3866         size_t size = sizeof(*head);
3867         int ret = -ENOMEM;
3868
3869 #ifdef CONFIG_NVME_MULTIPATH
3870         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3871 #endif
3872
3873         head = kzalloc(size, GFP_KERNEL);
3874         if (!head)
3875                 goto out;
3876         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3877         if (ret < 0)
3878                 goto out_free_head;
3879         head->instance = ret;
3880         INIT_LIST_HEAD(&head->list);
3881         ret = init_srcu_struct(&head->srcu);
3882         if (ret)
3883                 goto out_ida_remove;
3884         head->subsys = ctrl->subsys;
3885         head->ns_id = nsid;
3886         head->ids = *ids;
3887         kref_init(&head->ref);
3888
3889         ret = __nvme_check_ids(ctrl->subsys, head);
3890         if (ret) {
3891                 dev_err(ctrl->device,
3892                         "duplicate IDs for nsid %d\n", nsid);
3893                 goto out_cleanup_srcu;
3894         }
3895
3896         if (head->ids.csi) {
3897                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3898                 if (ret)
3899                         goto out_cleanup_srcu;
3900         } else
3901                 head->effects = ctrl->effects;
3902
3903         ret = nvme_mpath_alloc_disk(ctrl, head);
3904         if (ret)
3905                 goto out_cleanup_srcu;
3906
3907         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3908
3909         kref_get(&ctrl->subsys->ref);
3910
3911         return head;
3912 out_cleanup_srcu:
3913         cleanup_srcu_struct(&head->srcu);
3914 out_ida_remove:
3915         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3916 out_free_head:
3917         kfree(head);
3918 out:
3919         if (ret > 0)
3920                 ret = blk_status_to_errno(nvme_error_status(ret));
3921         return ERR_PTR(ret);
3922 }
3923
3924 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3925                 struct nvme_ns_ids *ids, bool is_shared)
3926 {
3927         struct nvme_ctrl *ctrl = ns->ctrl;
3928         struct nvme_ns_head *head = NULL;
3929         int ret = 0;
3930
3931         mutex_lock(&ctrl->subsys->lock);
3932         head = nvme_find_ns_head(ctrl->subsys, nsid);
3933         if (!head) {
3934                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3935                 if (IS_ERR(head)) {
3936                         ret = PTR_ERR(head);
3937                         goto out_unlock;
3938                 }
3939                 head->shared = is_shared;
3940         } else {
3941                 ret = -EINVAL;
3942                 if (!is_shared || !head->shared) {
3943                         dev_err(ctrl->device,
3944                                 "Duplicate unshared namespace %d\n", nsid);
3945                         goto out_put_ns_head;
3946                 }
3947                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3948                         dev_err(ctrl->device,
3949                                 "IDs don't match for shared namespace %d\n",
3950                                         nsid);
3951                         goto out_put_ns_head;
3952                 }
3953         }
3954
3955         list_add_tail_rcu(&ns->siblings, &head->list);
3956         ns->head = head;
3957         mutex_unlock(&ctrl->subsys->lock);
3958         return 0;
3959
3960 out_put_ns_head:
3961         nvme_put_ns_head(head);
3962 out_unlock:
3963         mutex_unlock(&ctrl->subsys->lock);
3964         return ret;
3965 }
3966
3967 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3968 {
3969         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3970         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3971
3972         return nsa->head->ns_id - nsb->head->ns_id;
3973 }
3974
3975 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3976 {
3977         struct nvme_ns *ns, *ret = NULL;
3978
3979         down_read(&ctrl->namespaces_rwsem);
3980         list_for_each_entry(ns, &ctrl->namespaces, list) {
3981                 if (ns->head->ns_id == nsid) {
3982                         if (!kref_get_unless_zero(&ns->kref))
3983                                 continue;
3984                         ret = ns;
3985                         break;
3986                 }
3987                 if (ns->head->ns_id > nsid)
3988                         break;
3989         }
3990         up_read(&ctrl->namespaces_rwsem);
3991         return ret;
3992 }
3993 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3994
3995 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3996                 struct nvme_ns_ids *ids)
3997 {
3998         struct nvme_ns *ns;
3999         struct gendisk *disk;
4000         struct nvme_id_ns *id;
4001         int node = ctrl->numa_node;
4002
4003         if (nvme_identify_ns(ctrl, nsid, ids, &id))
4004                 return;
4005
4006         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4007         if (!ns)
4008                 goto out_free_id;
4009
4010         ns->queue = blk_mq_init_queue(ctrl->tagset);
4011         if (IS_ERR(ns->queue))
4012                 goto out_free_ns;
4013
4014         if (ctrl->opts && ctrl->opts->data_digest)
4015                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
4016
4017         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
4018         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
4019                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4020
4021         ns->queue->queuedata = ns;
4022         ns->ctrl = ctrl;
4023         kref_init(&ns->kref);
4024
4025         if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
4026                 goto out_free_queue;
4027
4028         disk = alloc_disk_node(0, node);
4029         if (!disk)
4030                 goto out_unlink_ns;
4031
4032         disk->fops = &nvme_bdev_ops;
4033         disk->private_data = ns;
4034         disk->queue = ns->queue;
4035         disk->flags = GENHD_FL_EXT_DEVT;
4036         /*
4037          * Without the multipath code enabled, multiple controller per
4038          * subsystems are visible as devices and thus we cannot use the
4039          * subsystem instance.
4040          */
4041         if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
4042                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4043                         ns->head->instance);
4044         ns->disk = disk;
4045
4046         if (nvme_update_ns_info(ns, id))
4047                 goto out_put_disk;
4048
4049         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
4050                 if (nvme_nvm_register(ns, disk->disk_name, node)) {
4051                         dev_warn(ctrl->device, "LightNVM init failure\n");
4052                         goto out_put_disk;
4053                 }
4054         }
4055
4056         down_write(&ctrl->namespaces_rwsem);
4057         list_add_tail(&ns->list, &ctrl->namespaces);
4058         up_write(&ctrl->namespaces_rwsem);
4059
4060         nvme_get_ctrl(ctrl);
4061
4062         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
4063
4064         nvme_mpath_add_disk(ns, id);
4065         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4066         kfree(id);
4067
4068         return;
4069  out_put_disk:
4070         /* prevent double queue cleanup */
4071         ns->disk->queue = NULL;
4072         put_disk(ns->disk);
4073  out_unlink_ns:
4074         mutex_lock(&ctrl->subsys->lock);
4075         list_del_rcu(&ns->siblings);
4076         if (list_empty(&ns->head->list))
4077                 list_del_init(&ns->head->entry);
4078         mutex_unlock(&ctrl->subsys->lock);
4079         nvme_put_ns_head(ns->head);
4080  out_free_queue:
4081         blk_cleanup_queue(ns->queue);
4082  out_free_ns:
4083         kfree(ns);
4084  out_free_id:
4085         kfree(id);
4086 }
4087
4088 static void nvme_ns_remove(struct nvme_ns *ns)
4089 {
4090         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4091                 return;
4092
4093         set_capacity(ns->disk, 0);
4094         nvme_fault_inject_fini(&ns->fault_inject);
4095
4096         mutex_lock(&ns->ctrl->subsys->lock);
4097         list_del_rcu(&ns->siblings);
4098         if (list_empty(&ns->head->list))
4099                 list_del_init(&ns->head->entry);
4100         mutex_unlock(&ns->ctrl->subsys->lock);
4101
4102         synchronize_rcu(); /* guarantee not available in head->list */
4103         nvme_mpath_clear_current_path(ns);
4104         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
4105
4106         if (ns->disk->flags & GENHD_FL_UP) {
4107                 del_gendisk(ns->disk);
4108                 blk_cleanup_queue(ns->queue);
4109                 if (blk_get_integrity(ns->disk))
4110                         blk_integrity_unregister(ns->disk);
4111         }
4112
4113         down_write(&ns->ctrl->namespaces_rwsem);
4114         list_del_init(&ns->list);
4115         up_write(&ns->ctrl->namespaces_rwsem);
4116
4117         nvme_mpath_check_last_path(ns);
4118         nvme_put_ns(ns);
4119 }
4120
4121 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4122 {
4123         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4124
4125         if (ns) {
4126                 nvme_ns_remove(ns);
4127                 nvme_put_ns(ns);
4128         }
4129 }
4130
4131 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4132 {
4133         struct nvme_id_ns *id;
4134         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4135
4136         if (test_bit(NVME_NS_DEAD, &ns->flags))
4137                 goto out;
4138
4139         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4140         if (ret)
4141                 goto out;
4142
4143         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4144         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4145                 dev_err(ns->ctrl->device,
4146                         "identifiers changed for nsid %d\n", ns->head->ns_id);
4147                 goto out_free_id;
4148         }
4149
4150         ret = nvme_update_ns_info(ns, id);
4151
4152 out_free_id:
4153         kfree(id);
4154 out:
4155         /*
4156          * Only remove the namespace if we got a fatal error back from the
4157          * device, otherwise ignore the error and just move on.
4158          *
4159          * TODO: we should probably schedule a delayed retry here.
4160          */
4161         if (ret > 0 && (ret & NVME_SC_DNR))
4162                 nvme_ns_remove(ns);
4163 }
4164
4165 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4166 {
4167         struct nvme_ns_ids ids = { };
4168         struct nvme_ns *ns;
4169
4170         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4171                 return;
4172
4173         ns = nvme_find_get_ns(ctrl, nsid);
4174         if (ns) {
4175                 nvme_validate_ns(ns, &ids);
4176                 nvme_put_ns(ns);
4177                 return;
4178         }
4179
4180         switch (ids.csi) {
4181         case NVME_CSI_NVM:
4182                 nvme_alloc_ns(ctrl, nsid, &ids);
4183                 break;
4184         case NVME_CSI_ZNS:
4185                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4186                         dev_warn(ctrl->device,
4187                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4188                                 nsid);
4189                         break;
4190                 }
4191                 if (!nvme_multi_css(ctrl)) {
4192                         dev_warn(ctrl->device,
4193                                 "command set not reported for nsid: %d\n",
4194                                 nsid);
4195                         break;
4196                 }
4197                 nvme_alloc_ns(ctrl, nsid, &ids);
4198                 break;
4199         default:
4200                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4201                         ids.csi, nsid);
4202                 break;
4203         }
4204 }
4205
4206 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4207                                         unsigned nsid)
4208 {
4209         struct nvme_ns *ns, *next;
4210         LIST_HEAD(rm_list);
4211
4212         down_write(&ctrl->namespaces_rwsem);
4213         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4214                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4215                         list_move_tail(&ns->list, &rm_list);
4216         }
4217         up_write(&ctrl->namespaces_rwsem);
4218
4219         list_for_each_entry_safe(ns, next, &rm_list, list)
4220                 nvme_ns_remove(ns);
4221
4222 }
4223
4224 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4225 {
4226         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4227         __le32 *ns_list;
4228         u32 prev = 0;
4229         int ret = 0, i;
4230
4231         if (nvme_ctrl_limited_cns(ctrl))
4232                 return -EOPNOTSUPP;
4233
4234         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4235         if (!ns_list)
4236                 return -ENOMEM;
4237
4238         for (;;) {
4239                 struct nvme_command cmd = {
4240                         .identify.opcode        = nvme_admin_identify,
4241                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
4242                         .identify.nsid          = cpu_to_le32(prev),
4243                 };
4244
4245                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4246                                             NVME_IDENTIFY_DATA_SIZE);
4247                 if (ret) {
4248                         dev_warn(ctrl->device,
4249                                 "Identify NS List failed (status=0x%x)\n", ret);
4250                         goto free;
4251                 }
4252
4253                 for (i = 0; i < nr_entries; i++) {
4254                         u32 nsid = le32_to_cpu(ns_list[i]);
4255
4256                         if (!nsid)      /* end of the list? */
4257                                 goto out;
4258                         nvme_validate_or_alloc_ns(ctrl, nsid);
4259                         while (++prev < nsid)
4260                                 nvme_ns_remove_by_nsid(ctrl, prev);
4261                 }
4262         }
4263  out:
4264         nvme_remove_invalid_namespaces(ctrl, prev);
4265  free:
4266         kfree(ns_list);
4267         return ret;
4268 }
4269
4270 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4271 {
4272         struct nvme_id_ctrl *id;
4273         u32 nn, i;
4274
4275         if (nvme_identify_ctrl(ctrl, &id))
4276                 return;
4277         nn = le32_to_cpu(id->nn);
4278         kfree(id);
4279
4280         for (i = 1; i <= nn; i++)
4281                 nvme_validate_or_alloc_ns(ctrl, i);
4282
4283         nvme_remove_invalid_namespaces(ctrl, nn);
4284 }
4285
4286 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4287 {
4288         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4289         __le32 *log;
4290         int error;
4291
4292         log = kzalloc(log_size, GFP_KERNEL);
4293         if (!log)
4294                 return;
4295
4296         /*
4297          * We need to read the log to clear the AEN, but we don't want to rely
4298          * on it for the changed namespace information as userspace could have
4299          * raced with us in reading the log page, which could cause us to miss
4300          * updates.
4301          */
4302         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4303                         NVME_CSI_NVM, log, log_size, 0);
4304         if (error)
4305                 dev_warn(ctrl->device,
4306                         "reading changed ns log failed: %d\n", error);
4307
4308         kfree(log);
4309 }
4310
4311 static void nvme_scan_work(struct work_struct *work)
4312 {
4313         struct nvme_ctrl *ctrl =
4314                 container_of(work, struct nvme_ctrl, scan_work);
4315
4316         /* No tagset on a live ctrl means IO queues could not created */
4317         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4318                 return;
4319
4320         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4321                 dev_info(ctrl->device, "rescanning namespaces.\n");
4322                 nvme_clear_changed_ns_log(ctrl);
4323         }
4324
4325         mutex_lock(&ctrl->scan_lock);
4326         if (nvme_scan_ns_list(ctrl) != 0)
4327                 nvme_scan_ns_sequential(ctrl);
4328         mutex_unlock(&ctrl->scan_lock);
4329
4330         down_write(&ctrl->namespaces_rwsem);
4331         list_sort(NULL, &ctrl->namespaces, ns_cmp);
4332         up_write(&ctrl->namespaces_rwsem);
4333 }
4334
4335 /*
4336  * This function iterates the namespace list unlocked to allow recovery from
4337  * controller failure. It is up to the caller to ensure the namespace list is
4338  * not modified by scan work while this function is executing.
4339  */
4340 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4341 {
4342         struct nvme_ns *ns, *next;
4343         LIST_HEAD(ns_list);
4344
4345         /*
4346          * make sure to requeue I/O to all namespaces as these
4347          * might result from the scan itself and must complete
4348          * for the scan_work to make progress
4349          */
4350         nvme_mpath_clear_ctrl_paths(ctrl);
4351
4352         /* prevent racing with ns scanning */
4353         flush_work(&ctrl->scan_work);
4354
4355         /*
4356          * The dead states indicates the controller was not gracefully
4357          * disconnected. In that case, we won't be able to flush any data while
4358          * removing the namespaces' disks; fail all the queues now to avoid
4359          * potentially having to clean up the failed sync later.
4360          */
4361         if (ctrl->state == NVME_CTRL_DEAD)
4362                 nvme_kill_queues(ctrl);
4363
4364         /* this is a no-op when called from the controller reset handler */
4365         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4366
4367         down_write(&ctrl->namespaces_rwsem);
4368         list_splice_init(&ctrl->namespaces, &ns_list);
4369         up_write(&ctrl->namespaces_rwsem);
4370
4371         list_for_each_entry_safe(ns, next, &ns_list, list)
4372                 nvme_ns_remove(ns);
4373 }
4374 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4375
4376 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4377 {
4378         struct nvme_ctrl *ctrl =
4379                 container_of(dev, struct nvme_ctrl, ctrl_device);
4380         struct nvmf_ctrl_options *opts = ctrl->opts;
4381         int ret;
4382
4383         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4384         if (ret)
4385                 return ret;
4386
4387         if (opts) {
4388                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4389                 if (ret)
4390                         return ret;
4391
4392                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4393                                 opts->trsvcid ?: "none");
4394                 if (ret)
4395                         return ret;
4396
4397                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4398                                 opts->host_traddr ?: "none");
4399         }
4400         return ret;
4401 }
4402
4403 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4404 {
4405         char *envp[2] = { NULL, NULL };
4406         u32 aen_result = ctrl->aen_result;
4407
4408         ctrl->aen_result = 0;
4409         if (!aen_result)
4410                 return;
4411
4412         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4413         if (!envp[0])
4414                 return;
4415         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4416         kfree(envp[0]);
4417 }
4418
4419 static void nvme_async_event_work(struct work_struct *work)
4420 {
4421         struct nvme_ctrl *ctrl =
4422                 container_of(work, struct nvme_ctrl, async_event_work);
4423
4424         nvme_aen_uevent(ctrl);
4425         ctrl->ops->submit_async_event(ctrl);
4426 }
4427
4428 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4429 {
4430
4431         u32 csts;
4432
4433         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4434                 return false;
4435
4436         if (csts == ~0)
4437                 return false;
4438
4439         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4440 }
4441
4442 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4443 {
4444         struct nvme_fw_slot_info_log *log;
4445
4446         log = kmalloc(sizeof(*log), GFP_KERNEL);
4447         if (!log)
4448                 return;
4449
4450         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4451                         log, sizeof(*log), 0))
4452                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4453         kfree(log);
4454 }
4455
4456 static void nvme_fw_act_work(struct work_struct *work)
4457 {
4458         struct nvme_ctrl *ctrl = container_of(work,
4459                                 struct nvme_ctrl, fw_act_work);
4460         unsigned long fw_act_timeout;
4461
4462         if (ctrl->mtfa)
4463                 fw_act_timeout = jiffies +
4464                                 msecs_to_jiffies(ctrl->mtfa * 100);
4465         else
4466                 fw_act_timeout = jiffies +
4467                                 msecs_to_jiffies(admin_timeout * 1000);
4468
4469         nvme_stop_queues(ctrl);
4470         while (nvme_ctrl_pp_status(ctrl)) {
4471                 if (time_after(jiffies, fw_act_timeout)) {
4472                         dev_warn(ctrl->device,
4473                                 "Fw activation timeout, reset controller\n");
4474                         nvme_try_sched_reset(ctrl);
4475                         return;
4476                 }
4477                 msleep(100);
4478         }
4479
4480         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4481                 return;
4482
4483         nvme_start_queues(ctrl);
4484         /* read FW slot information to clear the AER */
4485         nvme_get_fw_slot_info(ctrl);
4486 }
4487
4488 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4489 {
4490         u32 aer_notice_type = (result & 0xff00) >> 8;
4491
4492         trace_nvme_async_event(ctrl, aer_notice_type);
4493
4494         switch (aer_notice_type) {
4495         case NVME_AER_NOTICE_NS_CHANGED:
4496                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4497                 nvme_queue_scan(ctrl);
4498                 break;
4499         case NVME_AER_NOTICE_FW_ACT_STARTING:
4500                 /*
4501                  * We are (ab)using the RESETTING state to prevent subsequent
4502                  * recovery actions from interfering with the controller's
4503                  * firmware activation.
4504                  */
4505                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4506                         queue_work(nvme_wq, &ctrl->fw_act_work);
4507                 break;
4508 #ifdef CONFIG_NVME_MULTIPATH
4509         case NVME_AER_NOTICE_ANA:
4510                 if (!ctrl->ana_log_buf)
4511                         break;
4512                 queue_work(nvme_wq, &ctrl->ana_work);
4513                 break;
4514 #endif
4515         case NVME_AER_NOTICE_DISC_CHANGED:
4516                 ctrl->aen_result = result;
4517                 break;
4518         default:
4519                 dev_warn(ctrl->device, "async event result %08x\n", result);
4520         }
4521 }
4522
4523 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4524                 volatile union nvme_result *res)
4525 {
4526         u32 result = le32_to_cpu(res->u32);
4527         u32 aer_type = result & 0x07;
4528
4529         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4530                 return;
4531
4532         switch (aer_type) {
4533         case NVME_AER_NOTICE:
4534                 nvme_handle_aen_notice(ctrl, result);
4535                 break;
4536         case NVME_AER_ERROR:
4537         case NVME_AER_SMART:
4538         case NVME_AER_CSS:
4539         case NVME_AER_VS:
4540                 trace_nvme_async_event(ctrl, aer_type);
4541                 ctrl->aen_result = result;
4542                 break;
4543         default:
4544                 break;
4545         }
4546         queue_work(nvme_wq, &ctrl->async_event_work);
4547 }
4548 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4549
4550 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4551 {
4552         nvme_mpath_stop(ctrl);
4553         nvme_stop_keep_alive(ctrl);
4554         nvme_stop_failfast_work(ctrl);
4555         flush_work(&ctrl->async_event_work);
4556         cancel_work_sync(&ctrl->fw_act_work);
4557 }
4558 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4559
4560 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4561 {
4562         nvme_start_keep_alive(ctrl);
4563
4564         nvme_enable_aen(ctrl);
4565
4566         if (ctrl->queue_count > 1) {
4567                 nvme_queue_scan(ctrl);
4568                 nvme_start_queues(ctrl);
4569         }
4570 }
4571 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4572
4573 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4574 {
4575         nvme_hwmon_exit(ctrl);
4576         nvme_fault_inject_fini(&ctrl->fault_inject);
4577         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4578         cdev_device_del(&ctrl->cdev, ctrl->device);
4579         nvme_put_ctrl(ctrl);
4580 }
4581 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4582
4583 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4584 {
4585         struct nvme_effects_log *cel;
4586         unsigned long i;
4587
4588         xa_for_each(&ctrl->cels, i, cel) {
4589                 xa_erase(&ctrl->cels, i);
4590                 kfree(cel);
4591         }
4592
4593         xa_destroy(&ctrl->cels);
4594 }
4595
4596 static void nvme_free_ctrl(struct device *dev)
4597 {
4598         struct nvme_ctrl *ctrl =
4599                 container_of(dev, struct nvme_ctrl, ctrl_device);
4600         struct nvme_subsystem *subsys = ctrl->subsys;
4601
4602         if (!subsys || ctrl->instance != subsys->instance)
4603                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4604
4605         nvme_free_cels(ctrl);
4606         nvme_mpath_uninit(ctrl);
4607         __free_page(ctrl->discard_page);
4608
4609         if (subsys) {
4610                 mutex_lock(&nvme_subsystems_lock);
4611                 list_del(&ctrl->subsys_entry);
4612                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4613                 mutex_unlock(&nvme_subsystems_lock);
4614         }
4615
4616         ctrl->ops->free_ctrl(ctrl);
4617
4618         if (subsys)
4619                 nvme_put_subsystem(subsys);
4620 }
4621
4622 /*
4623  * Initialize a NVMe controller structures.  This needs to be called during
4624  * earliest initialization so that we have the initialized structured around
4625  * during probing.
4626  */
4627 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4628                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4629 {
4630         int ret;
4631
4632         ctrl->state = NVME_CTRL_NEW;
4633         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4634         spin_lock_init(&ctrl->lock);
4635         mutex_init(&ctrl->scan_lock);
4636         INIT_LIST_HEAD(&ctrl->namespaces);
4637         xa_init(&ctrl->cels);
4638         init_rwsem(&ctrl->namespaces_rwsem);
4639         ctrl->dev = dev;
4640         ctrl->ops = ops;
4641         ctrl->quirks = quirks;
4642         ctrl->numa_node = NUMA_NO_NODE;
4643         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4644         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4645         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4646         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4647         init_waitqueue_head(&ctrl->state_wq);
4648
4649         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4650         INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4651         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4652         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4653
4654         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4655                         PAGE_SIZE);
4656         ctrl->discard_page = alloc_page(GFP_KERNEL);
4657         if (!ctrl->discard_page) {
4658                 ret = -ENOMEM;
4659                 goto out;
4660         }
4661
4662         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4663         if (ret < 0)
4664                 goto out;
4665         ctrl->instance = ret;
4666
4667         device_initialize(&ctrl->ctrl_device);
4668         ctrl->device = &ctrl->ctrl_device;
4669         ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4670                         ctrl->instance);
4671         ctrl->device->class = nvme_class;
4672         ctrl->device->parent = ctrl->dev;
4673         ctrl->device->groups = nvme_dev_attr_groups;
4674         ctrl->device->release = nvme_free_ctrl;
4675         dev_set_drvdata(ctrl->device, ctrl);
4676         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4677         if (ret)
4678                 goto out_release_instance;
4679
4680         nvme_get_ctrl(ctrl);
4681         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4682         ctrl->cdev.owner = ops->module;
4683         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4684         if (ret)
4685                 goto out_free_name;
4686
4687         /*
4688          * Initialize latency tolerance controls.  The sysfs files won't
4689          * be visible to userspace unless the device actually supports APST.
4690          */
4691         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4692         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4693                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4694
4695         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4696
4697         return 0;
4698 out_free_name:
4699         nvme_put_ctrl(ctrl);
4700         kfree_const(ctrl->device->kobj.name);
4701 out_release_instance:
4702         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4703 out:
4704         if (ctrl->discard_page)
4705                 __free_page(ctrl->discard_page);
4706         return ret;
4707 }
4708 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4709
4710 /**
4711  * nvme_kill_queues(): Ends all namespace queues
4712  * @ctrl: the dead controller that needs to end
4713  *
4714  * Call this function when the driver determines it is unable to get the
4715  * controller in a state capable of servicing IO.
4716  */
4717 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4718 {
4719         struct nvme_ns *ns;
4720
4721         down_read(&ctrl->namespaces_rwsem);
4722
4723         /* Forcibly unquiesce queues to avoid blocking dispatch */
4724         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4725                 blk_mq_unquiesce_queue(ctrl->admin_q);
4726
4727         list_for_each_entry(ns, &ctrl->namespaces, list)
4728                 nvme_set_queue_dying(ns);
4729
4730         up_read(&ctrl->namespaces_rwsem);
4731 }
4732 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4733
4734 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4735 {
4736         struct nvme_ns *ns;
4737
4738         down_read(&ctrl->namespaces_rwsem);
4739         list_for_each_entry(ns, &ctrl->namespaces, list)
4740                 blk_mq_unfreeze_queue(ns->queue);
4741         up_read(&ctrl->namespaces_rwsem);
4742 }
4743 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4744
4745 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4746 {
4747         struct nvme_ns *ns;
4748
4749         down_read(&ctrl->namespaces_rwsem);
4750         list_for_each_entry(ns, &ctrl->namespaces, list) {
4751                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4752                 if (timeout <= 0)
4753                         break;
4754         }
4755         up_read(&ctrl->namespaces_rwsem);
4756         return timeout;
4757 }
4758 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4759
4760 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4761 {
4762         struct nvme_ns *ns;
4763
4764         down_read(&ctrl->namespaces_rwsem);
4765         list_for_each_entry(ns, &ctrl->namespaces, list)
4766                 blk_mq_freeze_queue_wait(ns->queue);
4767         up_read(&ctrl->namespaces_rwsem);
4768 }
4769 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4770
4771 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4772 {
4773         struct nvme_ns *ns;
4774
4775         down_read(&ctrl->namespaces_rwsem);
4776         list_for_each_entry(ns, &ctrl->namespaces, list)
4777                 blk_freeze_queue_start(ns->queue);
4778         up_read(&ctrl->namespaces_rwsem);
4779 }
4780 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4781
4782 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4783 {
4784         struct nvme_ns *ns;
4785
4786         down_read(&ctrl->namespaces_rwsem);
4787         list_for_each_entry(ns, &ctrl->namespaces, list)
4788                 blk_mq_quiesce_queue(ns->queue);
4789         up_read(&ctrl->namespaces_rwsem);
4790 }
4791 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4792
4793 void nvme_start_queues(struct nvme_ctrl *ctrl)
4794 {
4795         struct nvme_ns *ns;
4796
4797         down_read(&ctrl->namespaces_rwsem);
4798         list_for_each_entry(ns, &ctrl->namespaces, list)
4799                 blk_mq_unquiesce_queue(ns->queue);
4800         up_read(&ctrl->namespaces_rwsem);
4801 }
4802 EXPORT_SYMBOL_GPL(nvme_start_queues);
4803
4804 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4805 {
4806         struct nvme_ns *ns;
4807
4808         down_read(&ctrl->namespaces_rwsem);
4809         list_for_each_entry(ns, &ctrl->namespaces, list)
4810                 blk_sync_queue(ns->queue);
4811         up_read(&ctrl->namespaces_rwsem);
4812 }
4813 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4814
4815 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4816 {
4817         nvme_sync_io_queues(ctrl);
4818         if (ctrl->admin_q)
4819                 blk_sync_queue(ctrl->admin_q);
4820 }
4821 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4822
4823 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4824 {
4825         if (file->f_op != &nvme_dev_fops)
4826                 return NULL;
4827         return file->private_data;
4828 }
4829 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4830
4831 /*
4832  * Check we didn't inadvertently grow the command structure sizes:
4833  */
4834 static inline void _nvme_check_size(void)
4835 {
4836         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4837         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4838         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4839         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4840         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4841         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4842         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4843         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4844         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4845         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4846         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4847         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4848         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4849         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4850         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4851         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4852         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4853         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4854         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4855         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4856 }
4857
4858
4859 static int __init nvme_core_init(void)
4860 {
4861         int result = -ENOMEM;
4862
4863         _nvme_check_size();
4864
4865         nvme_wq = alloc_workqueue("nvme-wq",
4866                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4867         if (!nvme_wq)
4868                 goto out;
4869
4870         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4871                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4872         if (!nvme_reset_wq)
4873                 goto destroy_wq;
4874
4875         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4876                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4877         if (!nvme_delete_wq)
4878                 goto destroy_reset_wq;
4879
4880         result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4881                         NVME_MINORS, "nvme");
4882         if (result < 0)
4883                 goto destroy_delete_wq;
4884
4885         nvme_class = class_create(THIS_MODULE, "nvme");
4886         if (IS_ERR(nvme_class)) {
4887                 result = PTR_ERR(nvme_class);
4888                 goto unregister_chrdev;
4889         }
4890         nvme_class->dev_uevent = nvme_class_uevent;
4891
4892         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4893         if (IS_ERR(nvme_subsys_class)) {
4894                 result = PTR_ERR(nvme_subsys_class);
4895                 goto destroy_class;
4896         }
4897         return 0;
4898
4899 destroy_class:
4900         class_destroy(nvme_class);
4901 unregister_chrdev:
4902         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4903 destroy_delete_wq:
4904         destroy_workqueue(nvme_delete_wq);
4905 destroy_reset_wq:
4906         destroy_workqueue(nvme_reset_wq);
4907 destroy_wq:
4908         destroy_workqueue(nvme_wq);
4909 out:
4910         return result;
4911 }
4912
4913 static void __exit nvme_core_exit(void)
4914 {
4915         class_destroy(nvme_subsys_class);
4916         class_destroy(nvme_class);
4917         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4918         destroy_workqueue(nvme_delete_wq);
4919         destroy_workqueue(nvme_reset_wq);
4920         destroy_workqueue(nvme_wq);
4921         ida_destroy(&nvme_instance_ida);
4922 }
4923
4924 MODULE_LICENSE("GPL");
4925 MODULE_VERSION("1.0");
4926 module_init(nvme_core_init);
4927 module_exit(nvme_core_exit);