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/blk-integrity.h>
  10 #include <linux/compat.h>
  11 #include <linux/delay.h>
  12 #include <linux/errno.h>
  13 #include <linux/hdreg.h>
  14 #include <linux/kernel.h>
  15 #include <linux/module.h>
  16 #include <linux/backing-dev.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 unsigned long apst_primary_timeout_ms = 100;
  61 module_param(apst_primary_timeout_ms, ulong, 0644);
  62 MODULE_PARM_DESC(apst_primary_timeout_ms,
  63         "primary APST timeout in ms");
  64
  65 static unsigned long apst_secondary_timeout_ms = 2000;
  66 module_param(apst_secondary_timeout_ms, ulong, 0644);
  67 MODULE_PARM_DESC(apst_secondary_timeout_ms,
  68         "secondary APST timeout in ms");
  69
  70 static unsigned long apst_primary_latency_tol_us = 15000;
  71 module_param(apst_primary_latency_tol_us, ulong, 0644);
  72 MODULE_PARM_DESC(apst_primary_latency_tol_us,
  73         "primary APST latency tolerance in us");
  74
  75 static unsigned long apst_secondary_latency_tol_us = 100000;
  76 module_param(apst_secondary_latency_tol_us, ulong, 0644);
  77 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
  78         "secondary APST latency tolerance in us");
  79
  80 static bool streams;
  81 module_param(streams, bool, 0644);
  82 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  83
  84 /*
  85  * nvme_wq - hosts nvme related works that are not reset or delete
  86  * nvme_reset_wq - hosts nvme reset works
  87  * nvme_delete_wq - hosts nvme delete works
  88  *
  89  * nvme_wq will host works such as scan, aen handling, fw activation,
  90  * keep-alive, periodic reconnects etc. nvme_reset_wq
  91  * runs reset works which also flush works hosted on nvme_wq for
  92  * serialization purposes. nvme_delete_wq host controller deletion
  93  * works which flush reset works for serialization.
  94  */
  95 struct workqueue_struct *nvme_wq;
  96 EXPORT_SYMBOL_GPL(nvme_wq);
  97
  98 struct workqueue_struct *nvme_reset_wq;
  99 EXPORT_SYMBOL_GPL(nvme_reset_wq);
 100
 101 struct workqueue_struct *nvme_delete_wq;
 102 EXPORT_SYMBOL_GPL(nvme_delete_wq);
 103
 104 static LIST_HEAD(nvme_subsystems);
 105 static DEFINE_MUTEX(nvme_subsystems_lock);
 106
 107 static DEFINE_IDA(nvme_instance_ida);
 108 static dev_t nvme_ctrl_base_chr_devt;
 109 static struct class *nvme_class;
 110 static struct class *nvme_subsys_class;
 111
 112 static DEFINE_IDA(nvme_ns_chr_minor_ida);
 113 static dev_t nvme_ns_chr_devt;
 114 static struct class *nvme_ns_chr_class;
 115
 116 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
 117 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
 118                                            unsigned nsid);
 119 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
 120                                    struct nvme_command *cmd);
 121
 122 void nvme_queue_scan(struct nvme_ctrl *ctrl)
 123 {
 124         /*
 125          * Only new queue scan work when admin and IO queues are both alive
 126          */
 127         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 128                 queue_work(nvme_wq, &ctrl->scan_work);
 129 }
 130
 131 /*
 132  * Use this function to proceed with scheduling reset_work for a controller
 133  * that had previously been set to the resetting state. This is intended for
 134  * code paths that can't be interrupted by other reset attempts. A hot removal
 135  * may prevent this from succeeding.
 136  */
 137 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 138 {
 139         if (ctrl->state != NVME_CTRL_RESETTING)
 140                 return -EBUSY;
 141         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 142                 return -EBUSY;
 143         return 0;
 144 }
 145 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 146
 147 static void nvme_failfast_work(struct work_struct *work)
 148 {
 149         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 150                         struct nvme_ctrl, failfast_work);
 151
 152         if (ctrl->state != NVME_CTRL_CONNECTING)
 153                 return;
 154
 155         set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 156         dev_info(ctrl->device, "failfast expired\n");
 157         nvme_kick_requeue_lists(ctrl);
 158 }
 159
 160 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 161 {
 162         if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 163                 return;
 164
 165         schedule_delayed_work(&ctrl->failfast_work,
 166                               ctrl->opts->fast_io_fail_tmo * HZ);
 167 }
 168
 169 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 170 {
 171         if (!ctrl->opts)
 172                 return;
 173
 174         cancel_delayed_work_sync(&ctrl->failfast_work);
 175         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 176 }
 177
 178
 179 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 180 {
 181         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 182                 return -EBUSY;
 183         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 184                 return -EBUSY;
 185         return 0;
 186 }
 187 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 188
 189 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 190 {
 191         int ret;
 192
 193         ret = nvme_reset_ctrl(ctrl);
 194         if (!ret) {
 195                 flush_work(&ctrl->reset_work);
 196                 if (ctrl->state != NVME_CTRL_LIVE)
 197                         ret = -ENETRESET;
 198         }
 199
 200         return ret;
 201 }
 202
 203 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 204 {
 205         dev_info(ctrl->device,
 206                  "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
 207
 208         flush_work(&ctrl->reset_work);
 209         nvme_stop_ctrl(ctrl);
 210         nvme_remove_namespaces(ctrl);
 211         ctrl->ops->delete_ctrl(ctrl);
 212         nvme_uninit_ctrl(ctrl);
 213 }
 214
 215 static void nvme_delete_ctrl_work(struct work_struct *work)
 216 {
 217         struct nvme_ctrl *ctrl =
 218                 container_of(work, struct nvme_ctrl, delete_work);
 219
 220         nvme_do_delete_ctrl(ctrl);
 221 }
 222
 223 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 224 {
 225         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 226                 return -EBUSY;
 227         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 228                 return -EBUSY;
 229         return 0;
 230 }
 231 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 232
 233 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 234 {
 235         /*
 236          * Keep a reference until nvme_do_delete_ctrl() complete,
 237          * since ->delete_ctrl can free the controller.
 238          */
 239         nvme_get_ctrl(ctrl);
 240         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 241                 nvme_do_delete_ctrl(ctrl);
 242         nvme_put_ctrl(ctrl);
 243 }
 244
 245 static blk_status_t nvme_error_status(u16 status)
 246 {
 247         switch (status & 0x7ff) {
 248         case NVME_SC_SUCCESS:
 249                 return BLK_STS_OK;
 250         case NVME_SC_CAP_EXCEEDED:
 251                 return BLK_STS_NOSPC;
 252         case NVME_SC_LBA_RANGE:
 253         case NVME_SC_CMD_INTERRUPTED:
 254         case NVME_SC_NS_NOT_READY:
 255                 return BLK_STS_TARGET;
 256         case NVME_SC_BAD_ATTRIBUTES:
 257         case NVME_SC_ONCS_NOT_SUPPORTED:
 258         case NVME_SC_INVALID_OPCODE:
 259         case NVME_SC_INVALID_FIELD:
 260         case NVME_SC_INVALID_NS:
 261                 return BLK_STS_NOTSUPP;
 262         case NVME_SC_WRITE_FAULT:
 263         case NVME_SC_READ_ERROR:
 264         case NVME_SC_UNWRITTEN_BLOCK:
 265         case NVME_SC_ACCESS_DENIED:
 266         case NVME_SC_READ_ONLY:
 267         case NVME_SC_COMPARE_FAILED:
 268                 return BLK_STS_MEDIUM;
 269         case NVME_SC_GUARD_CHECK:
 270         case NVME_SC_APPTAG_CHECK:
 271         case NVME_SC_REFTAG_CHECK:
 272         case NVME_SC_INVALID_PI:
 273                 return BLK_STS_PROTECTION;
 274         case NVME_SC_RESERVATION_CONFLICT:
 275                 return BLK_STS_NEXUS;
 276         case NVME_SC_HOST_PATH_ERROR:
 277                 return BLK_STS_TRANSPORT;
 278         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 279                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 280         case NVME_SC_ZONE_TOO_MANY_OPEN:
 281                 return BLK_STS_ZONE_OPEN_RESOURCE;
 282         default:
 283                 return BLK_STS_IOERR;
 284         }
 285 }
 286
 287 static void nvme_retry_req(struct request *req)
 288 {
 289         unsigned long delay = 0;
 290         u16 crd;
 291
 292         /* The mask and shift result must be <= 3 */
 293         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 294         if (crd)
 295                 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 296
 297         nvme_req(req)->retries++;
 298         blk_mq_requeue_request(req, false);
 299         blk_mq_delay_kick_requeue_list(req->q, delay);
 300 }
 301
 302 static void nvme_log_error(struct request *req)
 303 {
 304         struct nvme_ns *ns = req->q->queuedata;
 305         struct nvme_request *nr = nvme_req(req);
 306
 307         if (ns) {
 308                 pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
 309                        ns->disk ? ns->disk->disk_name : "?",
 310                        nvme_get_opcode_str(nr->cmd->common.opcode),
 311                        nr->cmd->common.opcode,
 312                        (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
 313                        (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
 314                        nvme_get_error_status_str(nr->status),
 315                        nr->status >> 8 & 7,     /* Status Code Type */
 316                        nr->status & 0xff,       /* Status Code */
 317                        nr->status & NVME_SC_MORE ? "MORE " : "",
 318                        nr->status & NVME_SC_DNR  ? "DNR "  : "");
 319                 return;
 320         }
 321
 322         pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
 323                            dev_name(nr->ctrl->device),
 324                            nvme_get_admin_opcode_str(nr->cmd->common.opcode),
 325                            nr->cmd->common.opcode,
 326                            nvme_get_error_status_str(nr->status),
 327                            nr->status >> 8 & 7, /* Status Code Type */
 328                            nr->status & 0xff,   /* Status Code */
 329                            nr->status & NVME_SC_MORE ? "MORE " : "",
 330                            nr->status & NVME_SC_DNR  ? "DNR "  : "");
 331 }
 332
 333 enum nvme_disposition {
 334         COMPLETE,
 335         RETRY,
 336         FAILOVER,
 337 };
 338
 339 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 340 {
 341         if (likely(nvme_req(req)->status == 0))
 342                 return COMPLETE;
 343
 344         if (blk_noretry_request(req) ||
 345             (nvme_req(req)->status & NVME_SC_DNR) ||
 346             nvme_req(req)->retries >= nvme_max_retries)
 347                 return COMPLETE;
 348
 349         if (req->cmd_flags & REQ_NVME_MPATH) {
 350                 if (nvme_is_path_error(nvme_req(req)->status) ||
 351                     blk_queue_dying(req->q))
 352                         return FAILOVER;
 353         } else {
 354                 if (blk_queue_dying(req->q))
 355                         return COMPLETE;
 356         }
 357
 358         return RETRY;
 359 }
 360
 361 static inline void nvme_end_req_zoned(struct request *req)
 362 {
 363         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 364             req_op(req) == REQ_OP_ZONE_APPEND)
 365                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 366                         le64_to_cpu(nvme_req(req)->result.u64));
 367 }
 368
 369 static inline void nvme_end_req(struct request *req)
 370 {
 371         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 372
 373         if (unlikely(nvme_req(req)->status != NVME_SC_SUCCESS))
 374                 nvme_log_error(req);
 375         nvme_end_req_zoned(req);
 376         nvme_trace_bio_complete(req);
 377         blk_mq_end_request(req, status);
 378 }
 379
 380 void nvme_complete_rq(struct request *req)
 381 {
 382         trace_nvme_complete_rq(req);
 383         nvme_cleanup_cmd(req);
 384
 385         if (nvme_req(req)->ctrl->kas)
 386                 nvme_req(req)->ctrl->comp_seen = true;
 387
 388         switch (nvme_decide_disposition(req)) {
 389         case COMPLETE:
 390                 nvme_end_req(req);
 391                 return;
 392         case RETRY:
 393                 nvme_retry_req(req);
 394                 return;
 395         case FAILOVER:
 396                 nvme_failover_req(req);
 397                 return;
 398         }
 399 }
 400 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 401
 402 void nvme_complete_batch_req(struct request *req)
 403 {
 404         nvme_cleanup_cmd(req);
 405         nvme_end_req_zoned(req);
 406 }
 407 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
 408
 409 /*
 410  * Called to unwind from ->queue_rq on a failed command submission so that the
 411  * multipathing code gets called to potentially failover to another path.
 412  * The caller needs to unwind all transport specific resource allocations and
 413  * must return propagate the return value.
 414  */
 415 blk_status_t nvme_host_path_error(struct request *req)
 416 {
 417         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 418         blk_mq_set_request_complete(req);
 419         nvme_complete_rq(req);
 420         return BLK_STS_OK;
 421 }
 422 EXPORT_SYMBOL_GPL(nvme_host_path_error);
 423
 424 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 425 {
 426         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 427                                 "Cancelling I/O %d", req->tag);
 428
 429         /* don't abort one completed request */
 430         if (blk_mq_request_completed(req))
 431                 return true;
 432
 433         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 434         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 435         blk_mq_complete_request(req);
 436         return true;
 437 }
 438 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 439
 440 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 441 {
 442         if (ctrl->tagset) {
 443                 blk_mq_tagset_busy_iter(ctrl->tagset,
 444                                 nvme_cancel_request, ctrl);
 445                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 446         }
 447 }
 448 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 449
 450 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 451 {
 452         if (ctrl->admin_tagset) {
 453                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 454                                 nvme_cancel_request, ctrl);
 455                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 456         }
 457 }
 458 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 459
 460 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 461                 enum nvme_ctrl_state new_state)
 462 {
 463         enum nvme_ctrl_state old_state;
 464         unsigned long flags;
 465         bool changed = false;
 466
 467         spin_lock_irqsave(&ctrl->lock, flags);
 468
 469         old_state = ctrl->state;
 470         switch (new_state) {
 471         case NVME_CTRL_LIVE:
 472                 switch (old_state) {
 473                 case NVME_CTRL_NEW:
 474                 case NVME_CTRL_RESETTING:
 475                 case NVME_CTRL_CONNECTING:
 476                         changed = true;
 477                         fallthrough;
 478                 default:
 479                         break;
 480                 }
 481                 break;
 482         case NVME_CTRL_RESETTING:
 483                 switch (old_state) {
 484                 case NVME_CTRL_NEW:
 485                 case NVME_CTRL_LIVE:
 486                         changed = true;
 487                         fallthrough;
 488                 default:
 489                         break;
 490                 }
 491                 break;
 492         case NVME_CTRL_CONNECTING:
 493                 switch (old_state) {
 494                 case NVME_CTRL_NEW:
 495                 case NVME_CTRL_RESETTING:
 496                         changed = true;
 497                         fallthrough;
 498                 default:
 499                         break;
 500                 }
 501                 break;
 502         case NVME_CTRL_DELETING:
 503                 switch (old_state) {
 504                 case NVME_CTRL_LIVE:
 505                 case NVME_CTRL_RESETTING:
 506                 case NVME_CTRL_CONNECTING:
 507                         changed = true;
 508                         fallthrough;
 509                 default:
 510                         break;
 511                 }
 512                 break;
 513         case NVME_CTRL_DELETING_NOIO:
 514                 switch (old_state) {
 515                 case NVME_CTRL_DELETING:
 516                 case NVME_CTRL_DEAD:
 517                         changed = true;
 518                         fallthrough;
 519                 default:
 520                         break;
 521                 }
 522                 break;
 523         case NVME_CTRL_DEAD:
 524                 switch (old_state) {
 525                 case NVME_CTRL_DELETING:
 526                         changed = true;
 527                         fallthrough;
 528                 default:
 529                         break;
 530                 }
 531                 break;
 532         default:
 533                 break;
 534         }
 535
 536         if (changed) {
 537                 ctrl->state = new_state;
 538                 wake_up_all(&ctrl->state_wq);
 539         }
 540
 541         spin_unlock_irqrestore(&ctrl->lock, flags);
 542         if (!changed)
 543                 return false;
 544
 545         if (ctrl->state == NVME_CTRL_LIVE) {
 546                 if (old_state == NVME_CTRL_CONNECTING)
 547                         nvme_stop_failfast_work(ctrl);
 548                 nvme_kick_requeue_lists(ctrl);
 549         } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 550                 old_state == NVME_CTRL_RESETTING) {
 551                 nvme_start_failfast_work(ctrl);
 552         }
 553         return changed;
 554 }
 555 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 556
 557 /*
 558  * Returns true for sink states that can't ever transition back to live.
 559  */
 560 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 561 {
 562         switch (ctrl->state) {
 563         case NVME_CTRL_NEW:
 564         case NVME_CTRL_LIVE:
 565         case NVME_CTRL_RESETTING:
 566         case NVME_CTRL_CONNECTING:
 567                 return false;
 568         case NVME_CTRL_DELETING:
 569         case NVME_CTRL_DELETING_NOIO:
 570         case NVME_CTRL_DEAD:
 571                 return true;
 572         default:
 573                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 574                 return true;
 575         }
 576 }
 577
 578 /*
 579  * Waits for the controller state to be resetting, or returns false if it is
 580  * not possible to ever transition to that state.
 581  */
 582 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 583 {
 584         wait_event(ctrl->state_wq,
 585                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 586                    nvme_state_terminal(ctrl));
 587         return ctrl->state == NVME_CTRL_RESETTING;
 588 }
 589 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 590
 591 static void nvme_free_ns_head(struct kref *ref)
 592 {
 593         struct nvme_ns_head *head =
 594                 container_of(ref, struct nvme_ns_head, ref);
 595
 596         nvme_mpath_remove_disk(head);
 597         ida_free(&head->subsys->ns_ida, head->instance);
 598         cleanup_srcu_struct(&head->srcu);
 599         nvme_put_subsystem(head->subsys);
 600         kfree(head);
 601 }
 602
 603 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
 604 {
 605         return kref_get_unless_zero(&head->ref);
 606 }
 607
 608 void nvme_put_ns_head(struct nvme_ns_head *head)
 609 {
 610         kref_put(&head->ref, nvme_free_ns_head);
 611 }
 612
 613 static void nvme_free_ns(struct kref *kref)
 614 {
 615         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 616
 617         put_disk(ns->disk);
 618         nvme_put_ns_head(ns->head);
 619         nvme_put_ctrl(ns->ctrl);
 620         kfree(ns);
 621 }
 622
 623 static inline bool nvme_get_ns(struct nvme_ns *ns)
 624 {
 625         return kref_get_unless_zero(&ns->kref);
 626 }
 627
 628 void nvme_put_ns(struct nvme_ns *ns)
 629 {
 630         kref_put(&ns->kref, nvme_free_ns);
 631 }
 632 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 633
 634 static inline void nvme_clear_nvme_request(struct request *req)
 635 {
 636         nvme_req(req)->status = 0;
 637         nvme_req(req)->retries = 0;
 638         nvme_req(req)->flags = 0;
 639         req->rq_flags |= RQF_DONTPREP;
 640 }
 641
 642 /* initialize a passthrough request */
 643 void nvme_init_request(struct request *req, struct nvme_command *cmd)
 644 {
 645         if (req->q->queuedata)
 646                 req->timeout = NVME_IO_TIMEOUT;
 647         else /* no queuedata implies admin queue */
 648                 req->timeout = NVME_ADMIN_TIMEOUT;
 649
 650         /* passthru commands should let the driver set the SGL flags */
 651         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 652
 653         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 654         if (req->mq_hctx->type == HCTX_TYPE_POLL)
 655                 req->cmd_flags |= REQ_POLLED;
 656         nvme_clear_nvme_request(req);
 657         memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
 658 }
 659 EXPORT_SYMBOL_GPL(nvme_init_request);
 660
 661 /*
 662  * For something we're not in a state to send to the device the default action
 663  * is to busy it and retry it after the controller state is recovered.  However,
 664  * if the controller is deleting or if anything is marked for failfast or
 665  * nvme multipath it is immediately failed.
 666  *
 667  * Note: commands used to initialize the controller will be marked for failfast.
 668  * Note: nvme cli/ioctl commands are marked for failfast.
 669  */
 670 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
 671                 struct request *rq)
 672 {
 673         if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
 674             ctrl->state != NVME_CTRL_DELETING &&
 675             ctrl->state != NVME_CTRL_DEAD &&
 676             !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
 677             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 678                 return BLK_STS_RESOURCE;
 679         return nvme_host_path_error(rq);
 680 }
 681 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
 682
 683 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 684                 bool queue_live)
 685 {
 686         struct nvme_request *req = nvme_req(rq);
 687
 688         /*
 689          * currently we have a problem sending passthru commands
 690          * on the admin_q if the controller is not LIVE because we can't
 691          * make sure that they are going out after the admin connect,
 692          * controller enable and/or other commands in the initialization
 693          * sequence. until the controller will be LIVE, fail with
 694          * BLK_STS_RESOURCE so that they will be rescheduled.
 695          */
 696         if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
 697                 return false;
 698
 699         if (ctrl->ops->flags & NVME_F_FABRICS) {
 700                 /*
 701                  * Only allow commands on a live queue, except for the connect
 702                  * command, which is require to set the queue live in the
 703                  * appropinquate states.
 704                  */
 705                 switch (ctrl->state) {
 706                 case NVME_CTRL_CONNECTING:
 707                         if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
 708                             req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
 709                                 return true;
 710                         break;
 711                 default:
 712                         break;
 713                 case NVME_CTRL_DEAD:
 714                         return false;
 715                 }
 716         }
 717
 718         return queue_live;
 719 }
 720 EXPORT_SYMBOL_GPL(__nvme_check_ready);
 721
 722 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 723 {
 724         struct nvme_command c = { };
 725
 726         c.directive.opcode = nvme_admin_directive_send;
 727         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 728         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 729         c.directive.dtype = NVME_DIR_IDENTIFY;
 730         c.directive.tdtype = NVME_DIR_STREAMS;
 731         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 732
 733         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 734 }
 735
 736 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 737 {
 738         return nvme_toggle_streams(ctrl, false);
 739 }
 740
 741 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 742 {
 743         return nvme_toggle_streams(ctrl, true);
 744 }
 745
 746 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 747                                   struct streams_directive_params *s, u32 nsid)
 748 {
 749         struct nvme_command c = { };
 750
 751         memset(s, 0, sizeof(*s));
 752
 753         c.directive.opcode = nvme_admin_directive_recv;
 754         c.directive.nsid = cpu_to_le32(nsid);
 755         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 756         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 757         c.directive.dtype = NVME_DIR_STREAMS;
 758
 759         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 760 }
 761
 762 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 763 {
 764         struct streams_directive_params s;
 765         u16 nssa;
 766         int ret;
 767
 768         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 769                 return 0;
 770         if (!streams)
 771                 return 0;
 772
 773         ret = nvme_enable_streams(ctrl);
 774         if (ret)
 775                 return ret;
 776
 777         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 778         if (ret)
 779                 goto out_disable_stream;
 780
 781         nssa = le16_to_cpu(s.nssa);
 782         if (nssa < BLK_MAX_WRITE_HINTS - 1) {
 783                 dev_info(ctrl->device, "too few streams (%u) available\n",
 784                                         nssa);
 785                 /* this condition is not an error: streams are optional */
 786                 ret = 0;
 787                 goto out_disable_stream;
 788         }
 789
 790         ctrl->nr_streams = min_t(u16, nssa, BLK_MAX_WRITE_HINTS - 1);
 791         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 792         return 0;
 793
 794 out_disable_stream:
 795         nvme_disable_streams(ctrl);
 796         return ret;
 797 }
 798
 799 /*
 800  * Check if 'req' has a write hint associated with it. If it does, assign
 801  * a valid namespace stream to the write.
 802  */
 803 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 804                                      struct request *req, u16 *control,
 805                                      u32 *dsmgmt)
 806 {
 807         enum rw_hint streamid = req->write_hint;
 808
 809         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 810                 streamid = 0;
 811         else {
 812                 streamid--;
 813                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 814                         return;
 815
 816                 *control |= NVME_RW_DTYPE_STREAMS;
 817                 *dsmgmt |= streamid << 16;
 818         }
 819
 820         if (streamid < ARRAY_SIZE(req->q->write_hints))
 821                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 822 }
 823
 824 static inline void nvme_setup_flush(struct nvme_ns *ns,
 825                 struct nvme_command *cmnd)
 826 {
 827         memset(cmnd, 0, sizeof(*cmnd));
 828         cmnd->common.opcode = nvme_cmd_flush;
 829         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 830 }
 831
 832 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 833                 struct nvme_command *cmnd)
 834 {
 835         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 836         struct nvme_dsm_range *range;
 837         struct bio *bio;
 838
 839         /*
 840          * Some devices do not consider the DSM 'Number of Ranges' field when
 841          * determining how much data to DMA. Always allocate memory for maximum
 842          * number of segments to prevent device reading beyond end of buffer.
 843          */
 844         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 845
 846         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 847         if (!range) {
 848                 /*
 849                  * If we fail allocation our range, fallback to the controller
 850                  * discard page. If that's also busy, it's safe to return
 851                  * busy, as we know we can make progress once that's freed.
 852                  */
 853                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 854                         return BLK_STS_RESOURCE;
 855
 856                 range = page_address(ns->ctrl->discard_page);
 857         }
 858
 859         __rq_for_each_bio(bio, req) {
 860                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 861                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 862
 863                 if (n < segments) {
 864                         range[n].cattr = cpu_to_le32(0);
 865                         range[n].nlb = cpu_to_le32(nlb);
 866                         range[n].slba = cpu_to_le64(slba);
 867                 }
 868                 n++;
 869         }
 870
 871         if (WARN_ON_ONCE(n != segments)) {
 872                 if (virt_to_page(range) == ns->ctrl->discard_page)
 873                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 874                 else
 875                         kfree(range);
 876                 return BLK_STS_IOERR;
 877         }
 878
 879         memset(cmnd, 0, sizeof(*cmnd));
 880         cmnd->dsm.opcode = nvme_cmd_dsm;
 881         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 882         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 883         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 884
 885         req->special_vec.bv_page = virt_to_page(range);
 886         req->special_vec.bv_offset = offset_in_page(range);
 887         req->special_vec.bv_len = alloc_size;
 888         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 889
 890         return BLK_STS_OK;
 891 }
 892
 893 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 894                 struct request *req, struct nvme_command *cmnd)
 895 {
 896         memset(cmnd, 0, sizeof(*cmnd));
 897
 898         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 899                 return nvme_setup_discard(ns, req, cmnd);
 900
 901         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 902         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 903         cmnd->write_zeroes.slba =
 904                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 905         cmnd->write_zeroes.length =
 906                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 907
 908         if (nvme_ns_has_pi(ns)) {
 909                 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
 910
 911                 switch (ns->pi_type) {
 912                 case NVME_NS_DPS_PI_TYPE1:
 913                 case NVME_NS_DPS_PI_TYPE2:
 914                         cmnd->write_zeroes.reftag =
 915                                 cpu_to_le32(t10_pi_ref_tag(req));
 916                         break;
 917                 }
 918         }
 919
 920         return BLK_STS_OK;
 921 }
 922
 923 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 924                 struct request *req, struct nvme_command *cmnd,
 925                 enum nvme_opcode op)
 926 {
 927         struct nvme_ctrl *ctrl = ns->ctrl;
 928         u16 control = 0;
 929         u32 dsmgmt = 0;
 930
 931         if (req->cmd_flags & REQ_FUA)
 932                 control |= NVME_RW_FUA;
 933         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 934                 control |= NVME_RW_LR;
 935
 936         if (req->cmd_flags & REQ_RAHEAD)
 937                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 938
 939         cmnd->rw.opcode = op;
 940         cmnd->rw.flags = 0;
 941         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 942         cmnd->rw.rsvd2 = 0;
 943         cmnd->rw.metadata = 0;
 944         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 945         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 946         cmnd->rw.reftag = 0;
 947         cmnd->rw.apptag = 0;
 948         cmnd->rw.appmask = 0;
 949
 950         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 951                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 952
 953         if (ns->ms) {
 954                 /*
 955                  * If formated with metadata, the block layer always provides a
 956                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 957                  * we enable the PRACT bit for protection information or set the
 958                  * namespace capacity to zero to prevent any I/O.
 959                  */
 960                 if (!blk_integrity_rq(req)) {
 961                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 962                                 return BLK_STS_NOTSUPP;
 963                         control |= NVME_RW_PRINFO_PRACT;
 964                 }
 965
 966                 switch (ns->pi_type) {
 967                 case NVME_NS_DPS_PI_TYPE3:
 968                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 969                         break;
 970                 case NVME_NS_DPS_PI_TYPE1:
 971                 case NVME_NS_DPS_PI_TYPE2:
 972                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 973                                         NVME_RW_PRINFO_PRCHK_REF;
 974                         if (op == nvme_cmd_zone_append)
 975                                 control |= NVME_RW_APPEND_PIREMAP;
 976                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 977                         break;
 978                 }
 979         }
 980
 981         cmnd->rw.control = cpu_to_le16(control);
 982         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 983         return 0;
 984 }
 985
 986 void nvme_cleanup_cmd(struct request *req)
 987 {
 988         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 989                 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 990
 991                 if (req->special_vec.bv_page == ctrl->discard_page)
 992                         clear_bit_unlock(0, &ctrl->discard_page_busy);
 993                 else
 994                         kfree(bvec_virt(&req->special_vec));
 995         }
 996 }
 997 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 998
 999 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
1000 {
1001         struct nvme_command *cmd = nvme_req(req)->cmd;
1002         blk_status_t ret = BLK_STS_OK;
1003
1004         if (!(req->rq_flags & RQF_DONTPREP))
1005                 nvme_clear_nvme_request(req);
1006
1007         switch (req_op(req)) {
1008         case REQ_OP_DRV_IN:
1009         case REQ_OP_DRV_OUT:
1010                 /* these are setup prior to execution in nvme_init_request() */
1011                 break;
1012         case REQ_OP_FLUSH:
1013                 nvme_setup_flush(ns, cmd);
1014                 break;
1015         case REQ_OP_ZONE_RESET_ALL:
1016         case REQ_OP_ZONE_RESET:
1017                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1018                 break;
1019         case REQ_OP_ZONE_OPEN:
1020                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1021                 break;
1022         case REQ_OP_ZONE_CLOSE:
1023                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1024                 break;
1025         case REQ_OP_ZONE_FINISH:
1026                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1027                 break;
1028         case REQ_OP_WRITE_ZEROES:
1029                 ret = nvme_setup_write_zeroes(ns, req, cmd);
1030                 break;
1031         case REQ_OP_DISCARD:
1032                 ret = nvme_setup_discard(ns, req, cmd);
1033                 break;
1034         case REQ_OP_READ:
1035                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1036                 break;
1037         case REQ_OP_WRITE:
1038                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1039                 break;
1040         case REQ_OP_ZONE_APPEND:
1041                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1042                 break;
1043         default:
1044                 WARN_ON_ONCE(1);
1045                 return BLK_STS_IOERR;
1046         }
1047
1048         cmd->common.command_id = nvme_cid(req);
1049         trace_nvme_setup_cmd(req, cmd);
1050         return ret;
1051 }
1052 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1053
1054 /*
1055  * Return values:
1056  * 0:  success
1057  * >0: nvme controller's cqe status response
1058  * <0: kernel error in lieu of controller response
1059  */
1060 static int nvme_execute_rq(struct request *rq, bool at_head)
1061 {
1062         blk_status_t status;
1063
1064         status = blk_execute_rq(rq, at_head);
1065         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1066                 return -EINTR;
1067         if (nvme_req(rq)->status)
1068                 return nvme_req(rq)->status;
1069         return blk_status_to_errno(status);
1070 }
1071
1072 /*
1073  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1074  * if the result is positive, it's an NVM Express status code
1075  */
1076 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1077                 union nvme_result *result, void *buffer, unsigned bufflen,
1078                 unsigned timeout, int qid, int at_head,
1079                 blk_mq_req_flags_t flags)
1080 {
1081         struct request *req;
1082         int ret;
1083
1084         if (qid == NVME_QID_ANY)
1085                 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1086         else
1087                 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1088                                                 qid ? qid - 1 : 0);
1089
1090         if (IS_ERR(req))
1091                 return PTR_ERR(req);
1092         nvme_init_request(req, cmd);
1093
1094         if (timeout)
1095                 req->timeout = timeout;
1096
1097         if (buffer && bufflen) {
1098                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1099                 if (ret)
1100                         goto out;
1101         }
1102
1103         ret = nvme_execute_rq(req, at_head);
1104         if (result && ret >= 0)
1105                 *result = nvme_req(req)->result;
1106  out:
1107         blk_mq_free_request(req);
1108         return ret;
1109 }
1110 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1111
1112 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1113                 void *buffer, unsigned bufflen)
1114 {
1115         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1116                         NVME_QID_ANY, 0, 0);
1117 }
1118 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1119
1120 static u32 nvme_known_admin_effects(u8 opcode)
1121 {
1122         switch (opcode) {
1123         case nvme_admin_format_nvm:
1124                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1125                         NVME_CMD_EFFECTS_CSE_MASK;
1126         case nvme_admin_sanitize_nvm:
1127                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1128         default:
1129                 break;
1130         }
1131         return 0;
1132 }
1133
1134 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1135 {
1136         u32 effects = 0;
1137
1138         if (ns) {
1139                 if (ns->head->effects)
1140                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1141                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1142                         dev_warn_once(ctrl->device,
1143                                 "IO command:%02x has unhandled effects:%08x\n",
1144                                 opcode, effects);
1145                 return 0;
1146         }
1147
1148         if (ctrl->effects)
1149                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1150         effects |= nvme_known_admin_effects(opcode);
1151
1152         return effects;
1153 }
1154 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1155
1156 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1157                                u8 opcode)
1158 {
1159         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1160
1161         /*
1162          * For simplicity, IO to all namespaces is quiesced even if the command
1163          * effects say only one namespace is affected.
1164          */
1165         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1166                 mutex_lock(&ctrl->scan_lock);
1167                 mutex_lock(&ctrl->subsys->lock);
1168                 nvme_mpath_start_freeze(ctrl->subsys);
1169                 nvme_mpath_wait_freeze(ctrl->subsys);
1170                 nvme_start_freeze(ctrl);
1171                 nvme_wait_freeze(ctrl);
1172         }
1173         return effects;
1174 }
1175
1176 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1177                               struct nvme_command *cmd, int status)
1178 {
1179         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1180                 nvme_unfreeze(ctrl);
1181                 nvme_mpath_unfreeze(ctrl->subsys);
1182                 mutex_unlock(&ctrl->subsys->lock);
1183                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1184                 mutex_unlock(&ctrl->scan_lock);
1185         }
1186         if (effects & NVME_CMD_EFFECTS_CCC)
1187                 nvme_init_ctrl_finish(ctrl);
1188         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1189                 nvme_queue_scan(ctrl);
1190                 flush_work(&ctrl->scan_work);
1191         }
1192
1193         switch (cmd->common.opcode) {
1194         case nvme_admin_set_features:
1195                 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1196                 case NVME_FEAT_KATO:
1197                         /*
1198                          * Keep alive commands interval on the host should be
1199                          * updated when KATO is modified by Set Features
1200                          * commands.
1201                          */
1202                         if (!status)
1203                                 nvme_update_keep_alive(ctrl, cmd);
1204                         break;
1205                 default:
1206                         break;
1207                 }
1208                 break;
1209         default:
1210                 break;
1211         }
1212 }
1213
1214 int nvme_execute_passthru_rq(struct request *rq)
1215 {
1216         struct nvme_command *cmd = nvme_req(rq)->cmd;
1217         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1218         struct nvme_ns *ns = rq->q->queuedata;
1219         u32 effects;
1220         int  ret;
1221
1222         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1223         ret = nvme_execute_rq(rq, false);
1224         if (effects) /* nothing to be done for zero cmd effects */
1225                 nvme_passthru_end(ctrl, effects, cmd, ret);
1226
1227         return ret;
1228 }
1229 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1230
1231 /*
1232  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1233  *
1234  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1235  *   accounting for transport roundtrip times [..].
1236  */
1237 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1238 {
1239         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1240 }
1241
1242 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1243 {
1244         struct nvme_ctrl *ctrl = rq->end_io_data;
1245         unsigned long flags;
1246         bool startka = false;
1247
1248         blk_mq_free_request(rq);
1249
1250         if (status) {
1251                 dev_err(ctrl->device,
1252                         "failed nvme_keep_alive_end_io error=%d\n",
1253                                 status);
1254                 return;
1255         }
1256
1257         ctrl->comp_seen = false;
1258         spin_lock_irqsave(&ctrl->lock, flags);
1259         if (ctrl->state == NVME_CTRL_LIVE ||
1260             ctrl->state == NVME_CTRL_CONNECTING)
1261                 startka = true;
1262         spin_unlock_irqrestore(&ctrl->lock, flags);
1263         if (startka)
1264                 nvme_queue_keep_alive_work(ctrl);
1265 }
1266
1267 static void nvme_keep_alive_work(struct work_struct *work)
1268 {
1269         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1270                         struct nvme_ctrl, ka_work);
1271         bool comp_seen = ctrl->comp_seen;
1272         struct request *rq;
1273
1274         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1275                 dev_dbg(ctrl->device,
1276                         "reschedule traffic based keep-alive timer\n");
1277                 ctrl->comp_seen = false;
1278                 nvme_queue_keep_alive_work(ctrl);
1279                 return;
1280         }
1281
1282         rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1283                                   BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1284         if (IS_ERR(rq)) {
1285                 /* allocation failure, reset the controller */
1286                 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1287                 nvme_reset_ctrl(ctrl);
1288                 return;
1289         }
1290         nvme_init_request(rq, &ctrl->ka_cmd);
1291
1292         rq->timeout = ctrl->kato * HZ;
1293         rq->end_io_data = ctrl;
1294         blk_execute_rq_nowait(rq, false, nvme_keep_alive_end_io);
1295 }
1296
1297 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1298 {
1299         if (unlikely(ctrl->kato == 0))
1300                 return;
1301
1302         nvme_queue_keep_alive_work(ctrl);
1303 }
1304
1305 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1306 {
1307         if (unlikely(ctrl->kato == 0))
1308                 return;
1309
1310         cancel_delayed_work_sync(&ctrl->ka_work);
1311 }
1312 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1313
1314 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1315                                    struct nvme_command *cmd)
1316 {
1317         unsigned int new_kato =
1318                 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1319
1320         dev_info(ctrl->device,
1321                  "keep alive interval updated from %u ms to %u ms\n",
1322                  ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1323
1324         nvme_stop_keep_alive(ctrl);
1325         ctrl->kato = new_kato;
1326         nvme_start_keep_alive(ctrl);
1327 }
1328
1329 /*
1330  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1331  * flag, thus sending any new CNS opcodes has a big chance of not working.
1332  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1333  * (but not for any later version).
1334  */
1335 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1336 {
1337         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1338                 return ctrl->vs < NVME_VS(1, 2, 0);
1339         return ctrl->vs < NVME_VS(1, 1, 0);
1340 }
1341
1342 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1343 {
1344         struct nvme_command c = { };
1345         int error;
1346
1347         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1348         c.identify.opcode = nvme_admin_identify;
1349         c.identify.cns = NVME_ID_CNS_CTRL;
1350
1351         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1352         if (!*id)
1353                 return -ENOMEM;
1354
1355         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1356                         sizeof(struct nvme_id_ctrl));
1357         if (error)
1358                 kfree(*id);
1359         return error;
1360 }
1361
1362 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1363                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1364 {
1365         const char *warn_str = "ctrl returned bogus length:";
1366         void *data = cur;
1367
1368         switch (cur->nidt) {
1369         case NVME_NIDT_EUI64:
1370                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1371                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1372                                  warn_str, cur->nidl);
1373                         return -1;
1374                 }
1375                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1376                 return NVME_NIDT_EUI64_LEN;
1377         case NVME_NIDT_NGUID:
1378                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1379                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1380                                  warn_str, cur->nidl);
1381                         return -1;
1382                 }
1383                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1384                 return NVME_NIDT_NGUID_LEN;
1385         case NVME_NIDT_UUID:
1386                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1387                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1388                                  warn_str, cur->nidl);
1389                         return -1;
1390                 }
1391                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1392                 return NVME_NIDT_UUID_LEN;
1393         case NVME_NIDT_CSI:
1394                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1395                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1396                                  warn_str, cur->nidl);
1397                         return -1;
1398                 }
1399                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1400                 *csi_seen = true;
1401                 return NVME_NIDT_CSI_LEN;
1402         default:
1403                 /* Skip unknown types */
1404                 return cur->nidl;
1405         }
1406 }
1407
1408 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1409                 struct nvme_ns_ids *ids)
1410 {
1411         struct nvme_command c = { };
1412         bool csi_seen = false;
1413         int status, pos, len;
1414         void *data;
1415
1416         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1417                 return 0;
1418         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1419                 return 0;
1420
1421         c.identify.opcode = nvme_admin_identify;
1422         c.identify.nsid = cpu_to_le32(nsid);
1423         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1424
1425         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1426         if (!data)
1427                 return -ENOMEM;
1428
1429         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1430                                       NVME_IDENTIFY_DATA_SIZE);
1431         if (status) {
1432                 dev_warn(ctrl->device,
1433                         "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1434                         nsid, status);
1435                 goto free_data;
1436         }
1437
1438         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1439                 struct nvme_ns_id_desc *cur = data + pos;
1440
1441                 if (cur->nidl == 0)
1442                         break;
1443
1444                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1445                 if (len < 0)
1446                         break;
1447
1448                 len += sizeof(*cur);
1449         }
1450
1451         if (nvme_multi_css(ctrl) && !csi_seen) {
1452                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1453                          nsid);
1454                 status = -EINVAL;
1455         }
1456
1457 free_data:
1458         kfree(data);
1459         return status;
1460 }
1461
1462 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1463                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1464 {
1465         struct nvme_command c = { };
1466         int error;
1467
1468         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1469         c.identify.opcode = nvme_admin_identify;
1470         c.identify.nsid = cpu_to_le32(nsid);
1471         c.identify.cns = NVME_ID_CNS_NS;
1472
1473         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1474         if (!*id)
1475                 return -ENOMEM;
1476
1477         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1478         if (error) {
1479                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1480                 goto out_free_id;
1481         }
1482
1483         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1484         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1485                 goto out_free_id;
1486
1487         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1488             !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1489                 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1490         if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1491             !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1492                 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1493
1494         return 0;
1495
1496 out_free_id:
1497         kfree(*id);
1498         return error;
1499 }
1500
1501 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1502                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1503 {
1504         union nvme_result res = { 0 };
1505         struct nvme_command c = { };
1506         int ret;
1507
1508         c.features.opcode = op;
1509         c.features.fid = cpu_to_le32(fid);
1510         c.features.dword11 = cpu_to_le32(dword11);
1511
1512         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1513                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1514         if (ret >= 0 && result)
1515                 *result = le32_to_cpu(res.u32);
1516         return ret;
1517 }
1518
1519 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1520                       unsigned int dword11, void *buffer, size_t buflen,
1521                       u32 *result)
1522 {
1523         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1524                              buflen, result);
1525 }
1526 EXPORT_SYMBOL_GPL(nvme_set_features);
1527
1528 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1529                       unsigned int dword11, void *buffer, size_t buflen,
1530                       u32 *result)
1531 {
1532         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1533                              buflen, result);
1534 }
1535 EXPORT_SYMBOL_GPL(nvme_get_features);
1536
1537 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1538 {
1539         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1540         u32 result;
1541         int status, nr_io_queues;
1542
1543         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1544                         &result);
1545         if (status < 0)
1546                 return status;
1547
1548         /*
1549          * Degraded controllers might return an error when setting the queue
1550          * count.  We still want to be able to bring them online and offer
1551          * access to the admin queue, as that might be only way to fix them up.
1552          */
1553         if (status > 0) {
1554                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1555                 *count = 0;
1556         } else {
1557                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1558                 *count = min(*count, nr_io_queues);
1559         }
1560
1561         return 0;
1562 }
1563 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1564
1565 #define NVME_AEN_SUPPORTED \
1566         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1567          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1568
1569 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1570 {
1571         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1572         int status;
1573
1574         if (!supported_aens)
1575                 return;
1576
1577         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1578                         NULL, 0, &result);
1579         if (status)
1580                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1581                          supported_aens);
1582
1583         queue_work(nvme_wq, &ctrl->async_event_work);
1584 }
1585
1586 static int nvme_ns_open(struct nvme_ns *ns)
1587 {
1588
1589         /* should never be called due to GENHD_FL_HIDDEN */
1590         if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1591                 goto fail;
1592         if (!nvme_get_ns(ns))
1593                 goto fail;
1594         if (!try_module_get(ns->ctrl->ops->module))
1595                 goto fail_put_ns;
1596
1597         return 0;
1598
1599 fail_put_ns:
1600         nvme_put_ns(ns);
1601 fail:
1602         return -ENXIO;
1603 }
1604
1605 static void nvme_ns_release(struct nvme_ns *ns)
1606 {
1607
1608         module_put(ns->ctrl->ops->module);
1609         nvme_put_ns(ns);
1610 }
1611
1612 static int nvme_open(struct block_device *bdev, fmode_t mode)
1613 {
1614         return nvme_ns_open(bdev->bd_disk->private_data);
1615 }
1616
1617 static void nvme_release(struct gendisk *disk, fmode_t mode)
1618 {
1619         nvme_ns_release(disk->private_data);
1620 }
1621
1622 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1623 {
1624         /* some standard values */
1625         geo->heads = 1 << 6;
1626         geo->sectors = 1 << 5;
1627         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1628         return 0;
1629 }
1630
1631 #ifdef CONFIG_BLK_DEV_INTEGRITY
1632 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1633                                 u32 max_integrity_segments)
1634 {
1635         struct blk_integrity integrity = { };
1636
1637         switch (pi_type) {
1638         case NVME_NS_DPS_PI_TYPE3:
1639                 integrity.profile = &t10_pi_type3_crc;
1640                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1641                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1642                 break;
1643         case NVME_NS_DPS_PI_TYPE1:
1644         case NVME_NS_DPS_PI_TYPE2:
1645                 integrity.profile = &t10_pi_type1_crc;
1646                 integrity.tag_size = sizeof(u16);
1647                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1648                 break;
1649         default:
1650                 integrity.profile = NULL;
1651                 break;
1652         }
1653         integrity.tuple_size = ms;
1654         blk_integrity_register(disk, &integrity);
1655         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1656 }
1657 #else
1658 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1659                                 u32 max_integrity_segments)
1660 {
1661 }
1662 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1663
1664 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1665 {
1666         struct nvme_ctrl *ctrl = ns->ctrl;
1667         struct request_queue *queue = disk->queue;
1668         u32 size = queue_logical_block_size(queue);
1669
1670         if (ctrl->max_discard_sectors == 0) {
1671                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1672                 return;
1673         }
1674
1675         if (ctrl->nr_streams && ns->sws && ns->sgs)
1676                 size *= ns->sws * ns->sgs;
1677
1678         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1679                         NVME_DSM_MAX_RANGES);
1680
1681         queue->limits.discard_alignment = 0;
1682         queue->limits.discard_granularity = size;
1683
1684         /* If discard is already enabled, don't reset queue limits */
1685         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1686                 return;
1687
1688         blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1689         blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1690
1691         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1692                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1693 }
1694
1695 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1696 {
1697         return uuid_equal(&a->uuid, &b->uuid) &&
1698                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1699                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1700                 a->csi == b->csi;
1701 }
1702
1703 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1704                                  u32 *phys_bs, u32 *io_opt)
1705 {
1706         struct streams_directive_params s;
1707         int ret;
1708
1709         if (!ctrl->nr_streams)
1710                 return 0;
1711
1712         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1713         if (ret)
1714                 return ret;
1715
1716         ns->sws = le32_to_cpu(s.sws);
1717         ns->sgs = le16_to_cpu(s.sgs);
1718
1719         if (ns->sws) {
1720                 *phys_bs = ns->sws * (1 << ns->lba_shift);
1721                 if (ns->sgs)
1722                         *io_opt = *phys_bs * ns->sgs;
1723         }
1724
1725         return 0;
1726 }
1727
1728 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1729 {
1730         struct nvme_ctrl *ctrl = ns->ctrl;
1731
1732         /*
1733          * The PI implementation requires the metadata size to be equal to the
1734          * t10 pi tuple size.
1735          */
1736         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1737         if (ns->ms == sizeof(struct t10_pi_tuple))
1738                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1739         else
1740                 ns->pi_type = 0;
1741
1742         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1743         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1744                 return 0;
1745         if (ctrl->ops->flags & NVME_F_FABRICS) {
1746                 /*
1747                  * The NVMe over Fabrics specification only supports metadata as
1748                  * part of the extended data LBA.  We rely on HCA/HBA support to
1749                  * remap the separate metadata buffer from the block layer.
1750                  */
1751                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1752                         return -EINVAL;
1753
1754                 ns->features |= NVME_NS_EXT_LBAS;
1755
1756                 /*
1757                  * The current fabrics transport drivers support namespace
1758                  * metadata formats only if nvme_ns_has_pi() returns true.
1759                  * Suppress support for all other formats so the namespace will
1760                  * have a 0 capacity and not be usable through the block stack.
1761                  *
1762                  * Note, this check will need to be modified if any drivers
1763                  * gain the ability to use other metadata formats.
1764                  */
1765                 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1766                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1767         } else {
1768                 /*
1769                  * For PCIe controllers, we can't easily remap the separate
1770                  * metadata buffer from the block layer and thus require a
1771                  * separate metadata buffer for block layer metadata/PI support.
1772                  * We allow extended LBAs for the passthrough interface, though.
1773                  */
1774                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1775                         ns->features |= NVME_NS_EXT_LBAS;
1776                 else
1777                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1778         }
1779
1780         return 0;
1781 }
1782
1783 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1784                 struct request_queue *q)
1785 {
1786         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1787
1788         if (ctrl->max_hw_sectors) {
1789                 u32 max_segments =
1790                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1791
1792                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1793                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1794                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1795         }
1796         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1797         blk_queue_dma_alignment(q, 7);
1798         blk_queue_write_cache(q, vwc, vwc);
1799 }
1800
1801 static void nvme_update_disk_info(struct gendisk *disk,
1802                 struct nvme_ns *ns, struct nvme_id_ns *id)
1803 {
1804         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1805         unsigned short bs = 1 << ns->lba_shift;
1806         u32 atomic_bs, phys_bs, io_opt = 0;
1807
1808         /*
1809          * The block layer can't support LBA sizes larger than the page size
1810          * yet, so catch this early and don't allow block I/O.
1811          */
1812         if (ns->lba_shift > PAGE_SHIFT) {
1813                 capacity = 0;
1814                 bs = (1 << 9);
1815         }
1816
1817         blk_integrity_unregister(disk);
1818
1819         atomic_bs = phys_bs = bs;
1820         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1821         if (id->nabo == 0) {
1822                 /*
1823                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1824                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1825                  * 0 then AWUPF must be used instead.
1826                  */
1827                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1828                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1829                 else
1830                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1831         }
1832
1833         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1834                 /* NPWG = Namespace Preferred Write Granularity */
1835                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1836                 /* NOWS = Namespace Optimal Write Size */
1837                 io_opt = bs * (1 + le16_to_cpu(id->nows));
1838         }
1839
1840         blk_queue_logical_block_size(disk->queue, bs);
1841         /*
1842          * Linux filesystems assume writing a single physical block is
1843          * an atomic operation. Hence limit the physical block size to the
1844          * value of the Atomic Write Unit Power Fail parameter.
1845          */
1846         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1847         blk_queue_io_min(disk->queue, phys_bs);
1848         blk_queue_io_opt(disk->queue, io_opt);
1849
1850         /*
1851          * Register a metadata profile for PI, or the plain non-integrity NVMe
1852          * metadata masquerading as Type 0 if supported, otherwise reject block
1853          * I/O to namespaces with metadata except when the namespace supports
1854          * PI, as it can strip/insert in that case.
1855          */
1856         if (ns->ms) {
1857                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1858                     (ns->features & NVME_NS_METADATA_SUPPORTED))
1859                         nvme_init_integrity(disk, ns->ms, ns->pi_type,
1860                                             ns->ctrl->max_integrity_segments);
1861                 else if (!nvme_ns_has_pi(ns))
1862                         capacity = 0;
1863         }
1864
1865         set_capacity_and_notify(disk, capacity);
1866
1867         nvme_config_discard(disk, ns);
1868         blk_queue_max_write_zeroes_sectors(disk->queue,
1869                                            ns->ctrl->max_zeroes_sectors);
1870
1871         set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1872                 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1873 }
1874
1875 static inline bool nvme_first_scan(struct gendisk *disk)
1876 {
1877         /* nvme_alloc_ns() scans the disk prior to adding it */
1878         return !disk_live(disk);
1879 }
1880
1881 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1882 {
1883         struct nvme_ctrl *ctrl = ns->ctrl;
1884         u32 iob;
1885
1886         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1887             is_power_of_2(ctrl->max_hw_sectors))
1888                 iob = ctrl->max_hw_sectors;
1889         else
1890                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1891
1892         if (!iob)
1893                 return;
1894
1895         if (!is_power_of_2(iob)) {
1896                 if (nvme_first_scan(ns->disk))
1897                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1898                                 ns->disk->disk_name, iob);
1899                 return;
1900         }
1901
1902         if (blk_queue_is_zoned(ns->disk->queue)) {
1903                 if (nvme_first_scan(ns->disk))
1904                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
1905                                 ns->disk->disk_name);
1906                 return;
1907         }
1908
1909         blk_queue_chunk_sectors(ns->queue, iob);
1910 }
1911
1912 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1913 {
1914         unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1915         int ret;
1916
1917         blk_mq_freeze_queue(ns->disk->queue);
1918         ns->lba_shift = id->lbaf[lbaf].ds;
1919         nvme_set_queue_limits(ns->ctrl, ns->queue);
1920
1921         ret = nvme_configure_metadata(ns, id);
1922         if (ret)
1923                 goto out_unfreeze;
1924         nvme_set_chunk_sectors(ns, id);
1925         nvme_update_disk_info(ns->disk, ns, id);
1926
1927         if (ns->head->ids.csi == NVME_CSI_ZNS) {
1928                 ret = nvme_update_zone_info(ns, lbaf);
1929                 if (ret)
1930                         goto out_unfreeze;
1931         }
1932
1933         set_bit(NVME_NS_READY, &ns->flags);
1934         blk_mq_unfreeze_queue(ns->disk->queue);
1935
1936         if (blk_queue_is_zoned(ns->queue)) {
1937                 ret = nvme_revalidate_zones(ns);
1938                 if (ret && !nvme_first_scan(ns->disk))
1939                         goto out;
1940         }
1941
1942         if (nvme_ns_head_multipath(ns->head)) {
1943                 blk_mq_freeze_queue(ns->head->disk->queue);
1944                 nvme_update_disk_info(ns->head->disk, ns, id);
1945                 nvme_mpath_revalidate_paths(ns);
1946                 blk_stack_limits(&ns->head->disk->queue->limits,
1947                                  &ns->queue->limits, 0);
1948                 disk_update_readahead(ns->head->disk);
1949                 blk_mq_unfreeze_queue(ns->head->disk->queue);
1950         }
1951         return 0;
1952
1953 out_unfreeze:
1954         blk_mq_unfreeze_queue(ns->disk->queue);
1955 out:
1956         /*
1957          * If probing fails due an unsupported feature, hide the block device,
1958          * but still allow other access.
1959          */
1960         if (ret == -ENODEV) {
1961                 ns->disk->flags |= GENHD_FL_HIDDEN;
1962                 ret = 0;
1963         }
1964         return ret;
1965 }
1966
1967 static char nvme_pr_type(enum pr_type type)
1968 {
1969         switch (type) {
1970         case PR_WRITE_EXCLUSIVE:
1971                 return 1;
1972         case PR_EXCLUSIVE_ACCESS:
1973                 return 2;
1974         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1975                 return 3;
1976         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1977                 return 4;
1978         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1979                 return 5;
1980         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1981                 return 6;
1982         default:
1983                 return 0;
1984         }
1985 }
1986
1987 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1988                 struct nvme_command *c, u8 data[16])
1989 {
1990         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1991         int srcu_idx = srcu_read_lock(&head->srcu);
1992         struct nvme_ns *ns = nvme_find_path(head);
1993         int ret = -EWOULDBLOCK;
1994
1995         if (ns) {
1996                 c->common.nsid = cpu_to_le32(ns->head->ns_id);
1997                 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1998         }
1999         srcu_read_unlock(&head->srcu, srcu_idx);
2000         return ret;
2001 }
2002
2003 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2004                 u8 data[16])
2005 {
2006         c->common.nsid = cpu_to_le32(ns->head->ns_id);
2007         return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2008 }
2009
2010 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2011                                 u64 key, u64 sa_key, u8 op)
2012 {
2013         struct nvme_command c = { };
2014         u8 data[16] = { 0, };
2015
2016         put_unaligned_le64(key, &data[0]);
2017         put_unaligned_le64(sa_key, &data[8]);
2018
2019         c.common.opcode = op;
2020         c.common.cdw10 = cpu_to_le32(cdw10);
2021
2022         if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2023             bdev->bd_disk->fops == &nvme_ns_head_ops)
2024                 return nvme_send_ns_head_pr_command(bdev, &c, data);
2025         return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2026 }
2027
2028 static int nvme_pr_register(struct block_device *bdev, u64 old,
2029                 u64 new, unsigned flags)
2030 {
2031         u32 cdw10;
2032
2033         if (flags & ~PR_FL_IGNORE_KEY)
2034                 return -EOPNOTSUPP;
2035
2036         cdw10 = old ? 2 : 0;
2037         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2038         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2039         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2040 }
2041
2042 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2043                 enum pr_type type, unsigned flags)
2044 {
2045         u32 cdw10;
2046
2047         if (flags & ~PR_FL_IGNORE_KEY)
2048                 return -EOPNOTSUPP;
2049
2050         cdw10 = nvme_pr_type(type) << 8;
2051         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2052         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2053 }
2054
2055 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2056                 enum pr_type type, bool abort)
2057 {
2058         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2059
2060         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2061 }
2062
2063 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2064 {
2065         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2066
2067         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2068 }
2069
2070 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2071 {
2072         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2073
2074         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2075 }
2076
2077 const struct pr_ops nvme_pr_ops = {
2078         .pr_register    = nvme_pr_register,
2079         .pr_reserve     = nvme_pr_reserve,
2080         .pr_release     = nvme_pr_release,
2081         .pr_preempt     = nvme_pr_preempt,
2082         .pr_clear       = nvme_pr_clear,
2083 };
2084
2085 #ifdef CONFIG_BLK_SED_OPAL
2086 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2087                 bool send)
2088 {
2089         struct nvme_ctrl *ctrl = data;
2090         struct nvme_command cmd = { };
2091
2092         if (send)
2093                 cmd.common.opcode = nvme_admin_security_send;
2094         else
2095                 cmd.common.opcode = nvme_admin_security_recv;
2096         cmd.common.nsid = 0;
2097         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2098         cmd.common.cdw11 = cpu_to_le32(len);
2099
2100         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2101                         NVME_QID_ANY, 1, 0);
2102 }
2103 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2104 #endif /* CONFIG_BLK_SED_OPAL */
2105
2106 #ifdef CONFIG_BLK_DEV_ZONED
2107 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2108                 unsigned int nr_zones, report_zones_cb cb, void *data)
2109 {
2110         return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2111                         data);
2112 }
2113 #else
2114 #define nvme_report_zones       NULL
2115 #endif /* CONFIG_BLK_DEV_ZONED */
2116
2117 static const struct block_device_operations nvme_bdev_ops = {
2118         .owner          = THIS_MODULE,
2119         .ioctl          = nvme_ioctl,
2120         .open           = nvme_open,
2121         .release        = nvme_release,
2122         .getgeo         = nvme_getgeo,
2123         .report_zones   = nvme_report_zones,
2124         .pr_ops         = &nvme_pr_ops,
2125 };
2126
2127 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2128 {
2129         unsigned long timeout =
2130                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2131         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2132         int ret;
2133
2134         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2135                 if (csts == ~0)
2136                         return -ENODEV;
2137                 if ((csts & NVME_CSTS_RDY) == bit)
2138                         break;
2139
2140                 usleep_range(1000, 2000);
2141                 if (fatal_signal_pending(current))
2142                         return -EINTR;
2143                 if (time_after(jiffies, timeout)) {
2144                         dev_err(ctrl->device,
2145                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2146                                 enabled ? "initialisation" : "reset", csts);
2147                         return -ENODEV;
2148                 }
2149         }
2150
2151         return ret;
2152 }
2153
2154 /*
2155  * If the device has been passed off to us in an enabled state, just clear
2156  * the enabled bit.  The spec says we should set the 'shutdown notification
2157  * bits', but doing so may cause the device to complete commands to the
2158  * admin queue ... and we don't know what memory that might be pointing at!
2159  */
2160 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2161 {
2162         int ret;
2163
2164         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2165         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2166
2167         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2168         if (ret)
2169                 return ret;
2170
2171         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2172                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2173
2174         return nvme_wait_ready(ctrl, ctrl->cap, false);
2175 }
2176 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2177
2178 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2179 {
2180         unsigned dev_page_min;
2181         int ret;
2182
2183         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2184         if (ret) {
2185                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2186                 return ret;
2187         }
2188         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2189
2190         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2191                 dev_err(ctrl->device,
2192                         "Minimum device page size %u too large for host (%u)\n",
2193                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2194                 return -ENODEV;
2195         }
2196
2197         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2198                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2199         else
2200                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2201         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2202         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2203         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2204         ctrl->ctrl_config |= NVME_CC_ENABLE;
2205
2206         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2207         if (ret)
2208                 return ret;
2209         return nvme_wait_ready(ctrl, ctrl->cap, true);
2210 }
2211 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2212
2213 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2214 {
2215         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2216         u32 csts;
2217         int ret;
2218
2219         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2220         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2221
2222         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2223         if (ret)
2224                 return ret;
2225
2226         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2227                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2228                         break;
2229
2230                 msleep(100);
2231                 if (fatal_signal_pending(current))
2232                         return -EINTR;
2233                 if (time_after(jiffies, timeout)) {
2234                         dev_err(ctrl->device,
2235                                 "Device shutdown incomplete; abort shutdown\n");
2236                         return -ENODEV;
2237                 }
2238         }
2239
2240         return ret;
2241 }
2242 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2243
2244 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2245 {
2246         __le64 ts;
2247         int ret;
2248
2249         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2250                 return 0;
2251
2252         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2253         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2254                         NULL);
2255         if (ret)
2256                 dev_warn_once(ctrl->device,
2257                         "could not set timestamp (%d)\n", ret);
2258         return ret;
2259 }
2260
2261 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2262 {
2263         struct nvme_feat_host_behavior *host;
2264         int ret;
2265
2266         /* Don't bother enabling the feature if retry delay is not reported */
2267         if (!ctrl->crdt[0])
2268                 return 0;
2269
2270         host = kzalloc(sizeof(*host), GFP_KERNEL);
2271         if (!host)
2272                 return 0;
2273
2274         host->acre = NVME_ENABLE_ACRE;
2275         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2276                                 host, sizeof(*host), NULL);
2277         kfree(host);
2278         return ret;
2279 }
2280
2281 /*
2282  * The function checks whether the given total (exlat + enlat) latency of
2283  * a power state allows the latter to be used as an APST transition target.
2284  * It does so by comparing the latency to the primary and secondary latency
2285  * tolerances defined by module params. If there's a match, the corresponding
2286  * timeout value is returned and the matching tolerance index (1 or 2) is
2287  * reported.
2288  */
2289 static bool nvme_apst_get_transition_time(u64 total_latency,
2290                 u64 *transition_time, unsigned *last_index)
2291 {
2292         if (total_latency <= apst_primary_latency_tol_us) {
2293                 if (*last_index == 1)
2294                         return false;
2295                 *last_index = 1;
2296                 *transition_time = apst_primary_timeout_ms;
2297                 return true;
2298         }
2299         if (apst_secondary_timeout_ms &&
2300                 total_latency <= apst_secondary_latency_tol_us) {
2301                 if (*last_index <= 2)
2302                         return false;
2303                 *last_index = 2;
2304                 *transition_time = apst_secondary_timeout_ms;
2305                 return true;
2306         }
2307         return false;
2308 }
2309
2310 /*
2311  * APST (Autonomous Power State Transition) lets us program a table of power
2312  * state transitions that the controller will perform automatically.
2313  *
2314  * Depending on module params, one of the two supported techniques will be used:
2315  *
2316  * - If the parameters provide explicit timeouts and tolerances, they will be
2317  *   used to build a table with up to 2 non-operational states to transition to.
2318  *   The default parameter values were selected based on the values used by
2319  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2320  *   regeneration of the APST table in the event of switching between external
2321  *   and battery power, the timeouts and tolerances reflect a compromise
2322  *   between values used by Microsoft for AC and battery scenarios.
2323  * - If not, we'll configure the table with a simple heuristic: we are willing
2324  *   to spend at most 2% of the time transitioning between power states.
2325  *   Therefore, when running in any given state, we will enter the next
2326  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2327  *   microseconds, as long as that state's exit latency is under the requested
2328  *   maximum latency.
2329  *
2330  * We will not autonomously enter any non-operational state for which the total
2331  * latency exceeds ps_max_latency_us.
2332  *
2333  * Users can set ps_max_latency_us to zero to turn off APST.
2334  */
2335 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2336 {
2337         struct nvme_feat_auto_pst *table;
2338         unsigned apste = 0;
2339         u64 max_lat_us = 0;
2340         __le64 target = 0;
2341         int max_ps = -1;
2342         int state;
2343         int ret;
2344         unsigned last_lt_index = UINT_MAX;
2345
2346         /*
2347          * If APST isn't supported or if we haven't been initialized yet,
2348          * then don't do anything.
2349          */
2350         if (!ctrl->apsta)
2351                 return 0;
2352
2353         if (ctrl->npss > 31) {
2354                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2355                 return 0;
2356         }
2357
2358         table = kzalloc(sizeof(*table), GFP_KERNEL);
2359         if (!table)
2360                 return 0;
2361
2362         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2363                 /* Turn off APST. */
2364                 dev_dbg(ctrl->device, "APST disabled\n");
2365                 goto done;
2366         }
2367
2368         /*
2369          * Walk through all states from lowest- to highest-power.
2370          * According to the spec, lower-numbered states use more power.  NPSS,
2371          * despite the name, is the index of the lowest-power state, not the
2372          * number of states.
2373          */
2374         for (state = (int)ctrl->npss; state >= 0; state--) {
2375                 u64 total_latency_us, exit_latency_us, transition_ms;
2376
2377                 if (target)
2378                         table->entries[state] = target;
2379
2380                 /*
2381                  * Don't allow transitions to the deepest state if it's quirked
2382                  * off.
2383                  */
2384                 if (state == ctrl->npss &&
2385                     (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2386                         continue;
2387
2388                 /*
2389                  * Is this state a useful non-operational state for higher-power
2390                  * states to autonomously transition to?
2391                  */
2392                 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2393                         continue;
2394
2395                 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2396                 if (exit_latency_us > ctrl->ps_max_latency_us)
2397                         continue;
2398
2399                 total_latency_us = exit_latency_us +
2400                         le32_to_cpu(ctrl->psd[state].entry_lat);
2401
2402                 /*
2403                  * This state is good. It can be used as the APST idle target
2404                  * for higher power states.
2405                  */
2406                 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2407                         if (!nvme_apst_get_transition_time(total_latency_us,
2408                                         &transition_ms, &last_lt_index))
2409                                 continue;
2410                 } else {
2411                         transition_ms = total_latency_us + 19;
2412                         do_div(transition_ms, 20);
2413                         if (transition_ms > (1 << 24) - 1)
2414                                 transition_ms = (1 << 24) - 1;
2415                 }
2416
2417                 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2418                 if (max_ps == -1)
2419                         max_ps = state;
2420                 if (total_latency_us > max_lat_us)
2421                         max_lat_us = total_latency_us;
2422         }
2423
2424         if (max_ps == -1)
2425                 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2426         else
2427                 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2428                         max_ps, max_lat_us, (int)sizeof(*table), table);
2429         apste = 1;
2430
2431 done:
2432         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2433                                 table, sizeof(*table), NULL);
2434         if (ret)
2435                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2436         kfree(table);
2437         return ret;
2438 }
2439
2440 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2441 {
2442         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2443         u64 latency;
2444
2445         switch (val) {
2446         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2447         case PM_QOS_LATENCY_ANY:
2448                 latency = U64_MAX;
2449                 break;
2450
2451         default:
2452                 latency = val;
2453         }
2454
2455         if (ctrl->ps_max_latency_us != latency) {
2456                 ctrl->ps_max_latency_us = latency;
2457                 if (ctrl->state == NVME_CTRL_LIVE)
2458                         nvme_configure_apst(ctrl);
2459         }
2460 }
2461
2462 struct nvme_core_quirk_entry {
2463         /*
2464          * NVMe model and firmware strings are padded with spaces.  For
2465          * simplicity, strings in the quirk table are padded with NULLs
2466          * instead.
2467          */
2468         u16 vid;
2469         const char *mn;
2470         const char *fr;
2471         unsigned long quirks;
2472 };
2473
2474 static const struct nvme_core_quirk_entry core_quirks[] = {
2475         {
2476                 /*
2477                  * This Toshiba device seems to die using any APST states.  See:
2478                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2479                  */
2480                 .vid = 0x1179,
2481                 .mn = "THNSF5256GPUK TOSHIBA",
2482                 .quirks = NVME_QUIRK_NO_APST,
2483         },
2484         {
2485                 /*
2486                  * This LiteON CL1-3D*-Q11 firmware version has a race
2487                  * condition associated with actions related to suspend to idle
2488                  * LiteON has resolved the problem in future firmware
2489                  */
2490                 .vid = 0x14a4,
2491                 .fr = "22301111",
2492                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2493         },
2494         {
2495                 /*
2496                  * This Kioxia CD6-V Series / HPE PE8030 device times out and
2497                  * aborts I/O during any load, but more easily reproducible
2498                  * with discards (fstrim).
2499                  *
2500                  * The device is left in a state where it is also not possible
2501                  * to use "nvme set-feature" to disable APST, but booting with
2502                  * nvme_core.default_ps_max_latency=0 works.
2503                  */
2504                 .vid = 0x1e0f,
2505                 .mn = "KCD6XVUL6T40",
2506                 .quirks = NVME_QUIRK_NO_APST,
2507         }
2508 };
2509
2510 /* match is null-terminated but idstr is space-padded. */
2511 static bool string_matches(const char *idstr, const char *match, size_t len)
2512 {
2513         size_t matchlen;
2514
2515         if (!match)
2516                 return true;
2517
2518         matchlen = strlen(match);
2519         WARN_ON_ONCE(matchlen > len);
2520
2521         if (memcmp(idstr, match, matchlen))
2522                 return false;
2523
2524         for (; matchlen < len; matchlen++)
2525                 if (idstr[matchlen] != ' ')
2526                         return false;
2527
2528         return true;
2529 }
2530
2531 static bool quirk_matches(const struct nvme_id_ctrl *id,
2532                           const struct nvme_core_quirk_entry *q)
2533 {
2534         return q->vid == le16_to_cpu(id->vid) &&
2535                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2536                 string_matches(id->fr, q->fr, sizeof(id->fr));
2537 }
2538
2539 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2540                 struct nvme_id_ctrl *id)
2541 {
2542         size_t nqnlen;
2543         int off;
2544
2545         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2546                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2547                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2548                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2549                         return;
2550                 }
2551
2552                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2553                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2554         }
2555
2556         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2557         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2558                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2559                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2560         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2561         off += sizeof(id->sn);
2562         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2563         off += sizeof(id->mn);
2564         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2565 }
2566
2567 static void nvme_release_subsystem(struct device *dev)
2568 {
2569         struct nvme_subsystem *subsys =
2570                 container_of(dev, struct nvme_subsystem, dev);
2571
2572         if (subsys->instance >= 0)
2573                 ida_free(&nvme_instance_ida, subsys->instance);
2574         kfree(subsys);
2575 }
2576
2577 static void nvme_destroy_subsystem(struct kref *ref)
2578 {
2579         struct nvme_subsystem *subsys =
2580                         container_of(ref, struct nvme_subsystem, ref);
2581
2582         mutex_lock(&nvme_subsystems_lock);
2583         list_del(&subsys->entry);
2584         mutex_unlock(&nvme_subsystems_lock);
2585
2586         ida_destroy(&subsys->ns_ida);
2587         device_del(&subsys->dev);
2588         put_device(&subsys->dev);
2589 }
2590
2591 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2592 {
2593         kref_put(&subsys->ref, nvme_destroy_subsystem);
2594 }
2595
2596 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2597 {
2598         struct nvme_subsystem *subsys;
2599
2600         lockdep_assert_held(&nvme_subsystems_lock);
2601
2602         /*
2603          * Fail matches for discovery subsystems. This results
2604          * in each discovery controller bound to a unique subsystem.
2605          * This avoids issues with validating controller values
2606          * that can only be true when there is a single unique subsystem.
2607          * There may be multiple and completely independent entities
2608          * that provide discovery controllers.
2609          */
2610         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2611                 return NULL;
2612
2613         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2614                 if (strcmp(subsys->subnqn, subsysnqn))
2615                         continue;
2616                 if (!kref_get_unless_zero(&subsys->ref))
2617                         continue;
2618                 return subsys;
2619         }
2620
2621         return NULL;
2622 }
2623
2624 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2625         struct device_attribute subsys_attr_##_name = \
2626                 __ATTR(_name, _mode, _show, NULL)
2627
2628 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2629                                     struct device_attribute *attr,
2630                                     char *buf)
2631 {
2632         struct nvme_subsystem *subsys =
2633                 container_of(dev, struct nvme_subsystem, dev);
2634
2635         return sysfs_emit(buf, "%s\n", subsys->subnqn);
2636 }
2637 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2638
2639 static ssize_t nvme_subsys_show_type(struct device *dev,
2640                                     struct device_attribute *attr,
2641                                     char *buf)
2642 {
2643         struct nvme_subsystem *subsys =
2644                 container_of(dev, struct nvme_subsystem, dev);
2645
2646         switch (subsys->subtype) {
2647         case NVME_NQN_DISC:
2648                 return sysfs_emit(buf, "discovery\n");
2649         case NVME_NQN_NVME:
2650                 return sysfs_emit(buf, "nvm\n");
2651         default:
2652                 return sysfs_emit(buf, "reserved\n");
2653         }
2654 }
2655 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2656
2657 #define nvme_subsys_show_str_function(field)                            \
2658 static ssize_t subsys_##field##_show(struct device *dev,                \
2659                             struct device_attribute *attr, char *buf)   \
2660 {                                                                       \
2661         struct nvme_subsystem *subsys =                                 \
2662                 container_of(dev, struct nvme_subsystem, dev);          \
2663         return sysfs_emit(buf, "%.*s\n",                                \
2664                            (int)sizeof(subsys->field), subsys->field);  \
2665 }                                                                       \
2666 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2667
2668 nvme_subsys_show_str_function(model);
2669 nvme_subsys_show_str_function(serial);
2670 nvme_subsys_show_str_function(firmware_rev);
2671
2672 static struct attribute *nvme_subsys_attrs[] = {
2673         &subsys_attr_model.attr,
2674         &subsys_attr_serial.attr,
2675         &subsys_attr_firmware_rev.attr,
2676         &subsys_attr_subsysnqn.attr,
2677         &subsys_attr_subsystype.attr,
2678 #ifdef CONFIG_NVME_MULTIPATH
2679         &subsys_attr_iopolicy.attr,
2680 #endif
2681         NULL,
2682 };
2683
2684 static const struct attribute_group nvme_subsys_attrs_group = {
2685         .attrs = nvme_subsys_attrs,
2686 };
2687
2688 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2689         &nvme_subsys_attrs_group,
2690         NULL,
2691 };
2692
2693 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2694 {
2695         return ctrl->opts && ctrl->opts->discovery_nqn;
2696 }
2697
2698 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2699                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2700 {
2701         struct nvme_ctrl *tmp;
2702
2703         lockdep_assert_held(&nvme_subsystems_lock);
2704
2705         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2706                 if (nvme_state_terminal(tmp))
2707                         continue;
2708
2709                 if (tmp->cntlid == ctrl->cntlid) {
2710                         dev_err(ctrl->device,
2711                                 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2712                                 ctrl->cntlid, dev_name(tmp->device),
2713                                 subsys->subnqn);
2714                         return false;
2715                 }
2716
2717                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2718                     nvme_discovery_ctrl(ctrl))
2719                         continue;
2720
2721                 dev_err(ctrl->device,
2722                         "Subsystem does not support multiple controllers\n");
2723                 return false;
2724         }
2725
2726         return true;
2727 }
2728
2729 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2730 {
2731         struct nvme_subsystem *subsys, *found;
2732         int ret;
2733
2734         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2735         if (!subsys)
2736                 return -ENOMEM;
2737
2738         subsys->instance = -1;
2739         mutex_init(&subsys->lock);
2740         kref_init(&subsys->ref);
2741         INIT_LIST_HEAD(&subsys->ctrls);
2742         INIT_LIST_HEAD(&subsys->nsheads);
2743         nvme_init_subnqn(subsys, ctrl, id);
2744         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2745         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2746         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2747         subsys->vendor_id = le16_to_cpu(id->vid);
2748         subsys->cmic = id->cmic;
2749
2750         /* Versions prior to 1.4 don't necessarily report a valid type */
2751         if (id->cntrltype == NVME_CTRL_DISC ||
2752             !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2753                 subsys->subtype = NVME_NQN_DISC;
2754         else
2755                 subsys->subtype = NVME_NQN_NVME;
2756
2757         if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2758                 dev_err(ctrl->device,
2759                         "Subsystem %s is not a discovery controller",
2760                         subsys->subnqn);
2761                 kfree(subsys);
2762                 return -EINVAL;
2763         }
2764         subsys->awupf = le16_to_cpu(id->awupf);
2765         nvme_mpath_default_iopolicy(subsys);
2766
2767         subsys->dev.class = nvme_subsys_class;
2768         subsys->dev.release = nvme_release_subsystem;
2769         subsys->dev.groups = nvme_subsys_attrs_groups;
2770         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2771         device_initialize(&subsys->dev);
2772
2773         mutex_lock(&nvme_subsystems_lock);
2774         found = __nvme_find_get_subsystem(subsys->subnqn);
2775         if (found) {
2776                 put_device(&subsys->dev);
2777                 subsys = found;
2778
2779                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2780                         ret = -EINVAL;
2781                         goto out_put_subsystem;
2782                 }
2783         } else {
2784                 ret = device_add(&subsys->dev);
2785                 if (ret) {
2786                         dev_err(ctrl->device,
2787                                 "failed to register subsystem device.\n");
2788                         put_device(&subsys->dev);
2789                         goto out_unlock;
2790                 }
2791                 ida_init(&subsys->ns_ida);
2792                 list_add_tail(&subsys->entry, &nvme_subsystems);
2793         }
2794
2795         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2796                                 dev_name(ctrl->device));
2797         if (ret) {
2798                 dev_err(ctrl->device,
2799                         "failed to create sysfs link from subsystem.\n");
2800                 goto out_put_subsystem;
2801         }
2802
2803         if (!found)
2804                 subsys->instance = ctrl->instance;
2805         ctrl->subsys = subsys;
2806         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2807         mutex_unlock(&nvme_subsystems_lock);
2808         return 0;
2809
2810 out_put_subsystem:
2811         nvme_put_subsystem(subsys);
2812 out_unlock:
2813         mutex_unlock(&nvme_subsystems_lock);
2814         return ret;
2815 }
2816
2817 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2818                 void *log, size_t size, u64 offset)
2819 {
2820         struct nvme_command c = { };
2821         u32 dwlen = nvme_bytes_to_numd(size);
2822
2823         c.get_log_page.opcode = nvme_admin_get_log_page;
2824         c.get_log_page.nsid = cpu_to_le32(nsid);
2825         c.get_log_page.lid = log_page;
2826         c.get_log_page.lsp = lsp;
2827         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2828         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2829         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2830         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2831         c.get_log_page.csi = csi;
2832
2833         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2834 }
2835
2836 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2837                                 struct nvme_effects_log **log)
2838 {
2839         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2840         int ret;
2841
2842         if (cel)
2843                 goto out;
2844
2845         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2846         if (!cel)
2847                 return -ENOMEM;
2848
2849         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2850                         cel, sizeof(*cel), 0);
2851         if (ret) {
2852                 kfree(cel);
2853                 return ret;
2854         }
2855
2856         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2857 out:
2858         *log = cel;
2859         return 0;
2860 }
2861
2862 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2863 {
2864         u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2865
2866         if (check_shl_overflow(1U, units + page_shift - 9, &val))
2867                 return UINT_MAX;
2868         return val;
2869 }
2870
2871 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2872 {
2873         struct nvme_command c = { };
2874         struct nvme_id_ctrl_nvm *id;
2875         int ret;
2876
2877         if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2878                 ctrl->max_discard_sectors = UINT_MAX;
2879                 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2880         } else {
2881                 ctrl->max_discard_sectors = 0;
2882                 ctrl->max_discard_segments = 0;
2883         }
2884
2885         /*
2886          * Even though NVMe spec explicitly states that MDTS is not applicable
2887          * to the write-zeroes, we are cautious and limit the size to the
2888          * controllers max_hw_sectors value, which is based on the MDTS field
2889          * and possibly other limiting factors.
2890          */
2891         if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2892             !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2893                 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2894         else
2895                 ctrl->max_zeroes_sectors = 0;
2896
2897         if (nvme_ctrl_limited_cns(ctrl))
2898                 return 0;
2899
2900         id = kzalloc(sizeof(*id), GFP_KERNEL);
2901         if (!id)
2902                 return 0;
2903
2904         c.identify.opcode = nvme_admin_identify;
2905         c.identify.cns = NVME_ID_CNS_CS_CTRL;
2906         c.identify.csi = NVME_CSI_NVM;
2907
2908         ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2909         if (ret)
2910                 goto free_data;
2911
2912         if (id->dmrl)
2913                 ctrl->max_discard_segments = id->dmrl;
2914         if (id->dmrsl)
2915                 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2916         if (id->wzsl)
2917                 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2918
2919 free_data:
2920         kfree(id);
2921         return ret;
2922 }
2923
2924 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2925 {
2926         struct nvme_id_ctrl *id;
2927         u32 max_hw_sectors;
2928         bool prev_apst_enabled;
2929         int ret;
2930
2931         ret = nvme_identify_ctrl(ctrl, &id);
2932         if (ret) {
2933                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2934                 return -EIO;
2935         }
2936
2937         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2938                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2939                 if (ret < 0)
2940                         goto out_free;
2941         }
2942
2943         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2944                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2945
2946         if (!ctrl->identified) {
2947                 unsigned int i;
2948
2949                 ret = nvme_init_subsystem(ctrl, id);
2950                 if (ret)
2951                         goto out_free;
2952
2953                 /*
2954                  * Check for quirks.  Quirk can depend on firmware version,
2955                  * so, in principle, the set of quirks present can change
2956                  * across a reset.  As a possible future enhancement, we
2957                  * could re-scan for quirks every time we reinitialize
2958                  * the device, but we'd have to make sure that the driver
2959                  * behaves intelligently if the quirks change.
2960                  */
2961                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2962                         if (quirk_matches(id, &core_quirks[i]))
2963                                 ctrl->quirks |= core_quirks[i].quirks;
2964                 }
2965         }
2966
2967         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2968                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2969                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2970         }
2971
2972         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2973         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2974         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2975
2976         ctrl->oacs = le16_to_cpu(id->oacs);
2977         ctrl->oncs = le16_to_cpu(id->oncs);
2978         ctrl->mtfa = le16_to_cpu(id->mtfa);
2979         ctrl->oaes = le32_to_cpu(id->oaes);
2980         ctrl->wctemp = le16_to_cpu(id->wctemp);
2981         ctrl->cctemp = le16_to_cpu(id->cctemp);
2982
2983         atomic_set(&ctrl->abort_limit, id->acl + 1);
2984         ctrl->vwc = id->vwc;
2985         if (id->mdts)
2986                 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2987         else
2988                 max_hw_sectors = UINT_MAX;
2989         ctrl->max_hw_sectors =
2990                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2991
2992         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2993         ctrl->sgls = le32_to_cpu(id->sgls);
2994         ctrl->kas = le16_to_cpu(id->kas);
2995         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2996         ctrl->ctratt = le32_to_cpu(id->ctratt);
2997
2998         ctrl->cntrltype = id->cntrltype;
2999         ctrl->dctype = id->dctype;
3000
3001         if (id->rtd3e) {
3002                 /* us -> s */
3003                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3004
3005                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3006                                                  shutdown_timeout, 60);
3007
3008                 if (ctrl->shutdown_timeout != shutdown_timeout)
3009                         dev_info(ctrl->device,
3010                                  "Shutdown timeout set to %u seconds\n",
3011                                  ctrl->shutdown_timeout);
3012         } else
3013                 ctrl->shutdown_timeout = shutdown_timeout;
3014
3015         ctrl->npss = id->npss;
3016         ctrl->apsta = id->apsta;
3017         prev_apst_enabled = ctrl->apst_enabled;
3018         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3019                 if (force_apst && id->apsta) {
3020                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3021                         ctrl->apst_enabled = true;
3022                 } else {
3023                         ctrl->apst_enabled = false;
3024                 }
3025         } else {
3026                 ctrl->apst_enabled = id->apsta;
3027         }
3028         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3029
3030         if (ctrl->ops->flags & NVME_F_FABRICS) {
3031                 ctrl->icdoff = le16_to_cpu(id->icdoff);
3032                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3033                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3034                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3035
3036                 /*
3037                  * In fabrics we need to verify the cntlid matches the
3038                  * admin connect
3039                  */
3040                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3041                         dev_err(ctrl->device,
3042                                 "Mismatching cntlid: Connect %u vs Identify "
3043                                 "%u, rejecting\n",
3044                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
3045                         ret = -EINVAL;
3046                         goto out_free;
3047                 }
3048
3049                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3050                         dev_err(ctrl->device,
3051                                 "keep-alive support is mandatory for fabrics\n");
3052                         ret = -EINVAL;
3053                         goto out_free;
3054                 }
3055         } else {
3056                 ctrl->hmpre = le32_to_cpu(id->hmpre);
3057                 ctrl->hmmin = le32_to_cpu(id->hmmin);
3058                 ctrl->hmminds = le32_to_cpu(id->hmminds);
3059                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3060         }
3061
3062         ret = nvme_mpath_init_identify(ctrl, id);
3063         if (ret < 0)
3064                 goto out_free;
3065
3066         if (ctrl->apst_enabled && !prev_apst_enabled)
3067                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3068         else if (!ctrl->apst_enabled && prev_apst_enabled)
3069                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3070
3071 out_free:
3072         kfree(id);
3073         return ret;
3074 }
3075
3076 /*
3077  * Initialize the cached copies of the Identify data and various controller
3078  * register in our nvme_ctrl structure.  This should be called as soon as
3079  * the admin queue is fully up and running.
3080  */
3081 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3082 {
3083         int ret;
3084
3085         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3086         if (ret) {
3087                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3088                 return ret;
3089         }
3090
3091         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3092
3093         if (ctrl->vs >= NVME_VS(1, 1, 0))
3094                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3095
3096         ret = nvme_init_identify(ctrl);
3097         if (ret)
3098                 return ret;
3099
3100         ret = nvme_init_non_mdts_limits(ctrl);
3101         if (ret < 0)
3102                 return ret;
3103
3104         ret = nvme_configure_apst(ctrl);
3105         if (ret < 0)
3106                 return ret;
3107
3108         ret = nvme_configure_timestamp(ctrl);
3109         if (ret < 0)
3110                 return ret;
3111
3112         ret = nvme_configure_directives(ctrl);
3113         if (ret < 0)
3114                 return ret;
3115
3116         ret = nvme_configure_acre(ctrl);
3117         if (ret < 0)
3118                 return ret;
3119
3120         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3121                 ret = nvme_hwmon_init(ctrl);
3122                 if (ret < 0)
3123                         return ret;
3124         }
3125
3126         ctrl->identified = true;
3127
3128         return 0;
3129 }
3130 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3131
3132 static int nvme_dev_open(struct inode *inode, struct file *file)
3133 {
3134         struct nvme_ctrl *ctrl =
3135                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3136
3137         switch (ctrl->state) {
3138         case NVME_CTRL_LIVE:
3139                 break;
3140         default:
3141                 return -EWOULDBLOCK;
3142         }
3143
3144         nvme_get_ctrl(ctrl);
3145         if (!try_module_get(ctrl->ops->module)) {
3146                 nvme_put_ctrl(ctrl);
3147                 return -EINVAL;
3148         }
3149
3150         file->private_data = ctrl;
3151         return 0;
3152 }
3153
3154 static int nvme_dev_release(struct inode *inode, struct file *file)
3155 {
3156         struct nvme_ctrl *ctrl =
3157                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3158
3159         module_put(ctrl->ops->module);
3160         nvme_put_ctrl(ctrl);
3161         return 0;
3162 }
3163
3164 static const struct file_operations nvme_dev_fops = {
3165         .owner          = THIS_MODULE,
3166         .open           = nvme_dev_open,
3167         .release        = nvme_dev_release,
3168         .unlocked_ioctl = nvme_dev_ioctl,
3169         .compat_ioctl   = compat_ptr_ioctl,
3170 };
3171
3172 static ssize_t nvme_sysfs_reset(struct device *dev,
3173                                 struct device_attribute *attr, const char *buf,
3174                                 size_t count)
3175 {
3176         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3177         int ret;
3178
3179         ret = nvme_reset_ctrl_sync(ctrl);
3180         if (ret < 0)
3181                 return ret;
3182         return count;
3183 }
3184 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3185
3186 static ssize_t nvme_sysfs_rescan(struct device *dev,
3187                                 struct device_attribute *attr, const char *buf,
3188                                 size_t count)
3189 {
3190         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3191
3192         nvme_queue_scan(ctrl);
3193         return count;
3194 }
3195 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3196
3197 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3198 {
3199         struct gendisk *disk = dev_to_disk(dev);
3200
3201         if (disk->fops == &nvme_bdev_ops)
3202                 return nvme_get_ns_from_dev(dev)->head;
3203         else
3204                 return disk->private_data;
3205 }
3206
3207 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3208                 char *buf)
3209 {
3210         struct nvme_ns_head *head = dev_to_ns_head(dev);
3211         struct nvme_ns_ids *ids = &head->ids;
3212         struct nvme_subsystem *subsys = head->subsys;
3213         int serial_len = sizeof(subsys->serial);
3214         int model_len = sizeof(subsys->model);
3215
3216         if (!uuid_is_null(&ids->uuid))
3217                 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3218
3219         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3220                 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3221
3222         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3223                 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3224
3225         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3226                                   subsys->serial[serial_len - 1] == '\0'))
3227                 serial_len--;
3228         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3229                                  subsys->model[model_len - 1] == '\0'))
3230                 model_len--;
3231
3232         return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3233                 serial_len, subsys->serial, model_len, subsys->model,
3234                 head->ns_id);
3235 }
3236 static DEVICE_ATTR_RO(wwid);
3237
3238 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3239                 char *buf)
3240 {
3241         return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3242 }
3243 static DEVICE_ATTR_RO(nguid);
3244
3245 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3246                 char *buf)
3247 {
3248         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3249
3250         /* For backward compatibility expose the NGUID to userspace if
3251          * we have no UUID set
3252          */
3253         if (uuid_is_null(&ids->uuid)) {
3254                 printk_ratelimited(KERN_WARNING
3255                                    "No UUID available providing old NGUID\n");
3256                 return sysfs_emit(buf, "%pU\n", ids->nguid);
3257         }
3258         return sysfs_emit(buf, "%pU\n", &ids->uuid);
3259 }
3260 static DEVICE_ATTR_RO(uuid);
3261
3262 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3263                 char *buf)
3264 {
3265         return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3266 }
3267 static DEVICE_ATTR_RO(eui);
3268
3269 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3270                 char *buf)
3271 {
3272         return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3273 }
3274 static DEVICE_ATTR_RO(nsid);
3275
3276 static struct attribute *nvme_ns_id_attrs[] = {
3277         &dev_attr_wwid.attr,
3278         &dev_attr_uuid.attr,
3279         &dev_attr_nguid.attr,
3280         &dev_attr_eui.attr,
3281         &dev_attr_nsid.attr,
3282 #ifdef CONFIG_NVME_MULTIPATH
3283         &dev_attr_ana_grpid.attr,
3284         &dev_attr_ana_state.attr,
3285 #endif
3286         NULL,
3287 };
3288
3289 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3290                 struct attribute *a, int n)
3291 {
3292         struct device *dev = container_of(kobj, struct device, kobj);
3293         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3294
3295         if (a == &dev_attr_uuid.attr) {
3296                 if (uuid_is_null(&ids->uuid) &&
3297                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3298                         return 0;
3299         }
3300         if (a == &dev_attr_nguid.attr) {
3301                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3302                         return 0;
3303         }
3304         if (a == &dev_attr_eui.attr) {
3305                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3306                         return 0;
3307         }
3308 #ifdef CONFIG_NVME_MULTIPATH
3309         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3310                 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3311                         return 0;
3312                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3313                         return 0;
3314         }
3315 #endif
3316         return a->mode;
3317 }
3318
3319 static const struct attribute_group nvme_ns_id_attr_group = {
3320         .attrs          = nvme_ns_id_attrs,
3321         .is_visible     = nvme_ns_id_attrs_are_visible,
3322 };
3323
3324 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3325         &nvme_ns_id_attr_group,
3326         NULL,
3327 };
3328
3329 #define nvme_show_str_function(field)                                           \
3330 static ssize_t  field##_show(struct device *dev,                                \
3331                             struct device_attribute *attr, char *buf)           \
3332 {                                                                               \
3333         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3334         return sysfs_emit(buf, "%.*s\n",                                        \
3335                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3336 }                                                                               \
3337 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3338
3339 nvme_show_str_function(model);
3340 nvme_show_str_function(serial);
3341 nvme_show_str_function(firmware_rev);
3342
3343 #define nvme_show_int_function(field)                                           \
3344 static ssize_t  field##_show(struct device *dev,                                \
3345                             struct device_attribute *attr, char *buf)           \
3346 {                                                                               \
3347         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3348         return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3349 }                                                                               \
3350 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3351
3352 nvme_show_int_function(cntlid);
3353 nvme_show_int_function(numa_node);
3354 nvme_show_int_function(queue_count);
3355 nvme_show_int_function(sqsize);
3356 nvme_show_int_function(kato);
3357
3358 static ssize_t nvme_sysfs_delete(struct device *dev,
3359                                 struct device_attribute *attr, const char *buf,
3360                                 size_t count)
3361 {
3362         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3363
3364         if (device_remove_file_self(dev, attr))
3365                 nvme_delete_ctrl_sync(ctrl);
3366         return count;
3367 }
3368 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3369
3370 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3371                                          struct device_attribute *attr,
3372                                          char *buf)
3373 {
3374         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3375
3376         return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3377 }
3378 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3379
3380 static ssize_t nvme_sysfs_show_state(struct device *dev,
3381                                      struct device_attribute *attr,
3382                                      char *buf)
3383 {
3384         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3385         static const char *const state_name[] = {
3386                 [NVME_CTRL_NEW]         = "new",
3387                 [NVME_CTRL_LIVE]        = "live",
3388                 [NVME_CTRL_RESETTING]   = "resetting",
3389                 [NVME_CTRL_CONNECTING]  = "connecting",
3390                 [NVME_CTRL_DELETING]    = "deleting",
3391                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3392                 [NVME_CTRL_DEAD]        = "dead",
3393         };
3394
3395         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3396             state_name[ctrl->state])
3397                 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3398
3399         return sysfs_emit(buf, "unknown state\n");
3400 }
3401
3402 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3403
3404 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3405                                          struct device_attribute *attr,
3406                                          char *buf)
3407 {
3408         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3409
3410         return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3411 }
3412 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3413
3414 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3415                                         struct device_attribute *attr,
3416                                         char *buf)
3417 {
3418         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3419
3420         return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3421 }
3422 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3423
3424 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3425                                         struct device_attribute *attr,
3426                                         char *buf)
3427 {
3428         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3429
3430         return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3431 }
3432 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3433
3434 static ssize_t nvme_sysfs_show_address(struct device *dev,
3435                                          struct device_attribute *attr,
3436                                          char *buf)
3437 {
3438         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3439
3440         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3441 }
3442 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3443
3444 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3445                 struct device_attribute *attr, char *buf)
3446 {
3447         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3448         struct nvmf_ctrl_options *opts = ctrl->opts;
3449
3450         if (ctrl->opts->max_reconnects == -1)
3451                 return sysfs_emit(buf, "off\n");
3452         return sysfs_emit(buf, "%d\n",
3453                           opts->max_reconnects * opts->reconnect_delay);
3454 }
3455
3456 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3457                 struct device_attribute *attr, const char *buf, size_t count)
3458 {
3459         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3460         struct nvmf_ctrl_options *opts = ctrl->opts;
3461         int ctrl_loss_tmo, err;
3462
3463         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3464         if (err)
3465                 return -EINVAL;
3466
3467         if (ctrl_loss_tmo < 0)
3468                 opts->max_reconnects = -1;
3469         else
3470                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3471                                                 opts->reconnect_delay);
3472         return count;
3473 }
3474 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3475         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3476
3477 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3478                 struct device_attribute *attr, char *buf)
3479 {
3480         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3481
3482         if (ctrl->opts->reconnect_delay == -1)
3483                 return sysfs_emit(buf, "off\n");
3484         return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3485 }
3486
3487 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3488                 struct device_attribute *attr, const char *buf, size_t count)
3489 {
3490         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3491         unsigned int v;
3492         int err;
3493
3494         err = kstrtou32(buf, 10, &v);
3495         if (err)
3496                 return err;
3497
3498         ctrl->opts->reconnect_delay = v;
3499         return count;
3500 }
3501 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3502         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3503
3504 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3505                 struct device_attribute *attr, char *buf)
3506 {
3507         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3508
3509         if (ctrl->opts->fast_io_fail_tmo == -1)
3510                 return sysfs_emit(buf, "off\n");
3511         return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3512 }
3513
3514 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3515                 struct device_attribute *attr, const char *buf, size_t count)
3516 {
3517         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3518         struct nvmf_ctrl_options *opts = ctrl->opts;
3519         int fast_io_fail_tmo, err;
3520
3521         err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3522         if (err)
3523                 return -EINVAL;
3524
3525         if (fast_io_fail_tmo < 0)
3526                 opts->fast_io_fail_tmo = -1;
3527         else
3528                 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3529         return count;
3530 }
3531 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3532         nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3533
3534 static ssize_t cntrltype_show(struct device *dev,
3535                               struct device_attribute *attr, char *buf)
3536 {
3537         static const char * const type[] = {
3538                 [NVME_CTRL_IO] = "io\n",
3539                 [NVME_CTRL_DISC] = "discovery\n",
3540                 [NVME_CTRL_ADMIN] = "admin\n",
3541         };
3542         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3543
3544         if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3545                 return sysfs_emit(buf, "reserved\n");
3546
3547         return sysfs_emit(buf, type[ctrl->cntrltype]);
3548 }
3549 static DEVICE_ATTR_RO(cntrltype);
3550
3551 static ssize_t dctype_show(struct device *dev,
3552                            struct device_attribute *attr, char *buf)
3553 {
3554         static const char * const type[] = {
3555                 [NVME_DCTYPE_NOT_REPORTED] = "none\n",
3556                 [NVME_DCTYPE_DDC] = "ddc\n",
3557                 [NVME_DCTYPE_CDC] = "cdc\n",
3558         };
3559         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3560
3561         if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3562                 return sysfs_emit(buf, "reserved\n");
3563
3564         return sysfs_emit(buf, type[ctrl->dctype]);
3565 }
3566 static DEVICE_ATTR_RO(dctype);
3567
3568 static struct attribute *nvme_dev_attrs[] = {
3569         &dev_attr_reset_controller.attr,
3570         &dev_attr_rescan_controller.attr,
3571         &dev_attr_model.attr,
3572         &dev_attr_serial.attr,
3573         &dev_attr_firmware_rev.attr,
3574         &dev_attr_cntlid.attr,
3575         &dev_attr_delete_controller.attr,
3576         &dev_attr_transport.attr,
3577         &dev_attr_subsysnqn.attr,
3578         &dev_attr_address.attr,
3579         &dev_attr_state.attr,
3580         &dev_attr_numa_node.attr,
3581         &dev_attr_queue_count.attr,
3582         &dev_attr_sqsize.attr,
3583         &dev_attr_hostnqn.attr,
3584         &dev_attr_hostid.attr,
3585         &dev_attr_ctrl_loss_tmo.attr,
3586         &dev_attr_reconnect_delay.attr,
3587         &dev_attr_fast_io_fail_tmo.attr,
3588         &dev_attr_kato.attr,
3589         &dev_attr_cntrltype.attr,
3590         &dev_attr_dctype.attr,
3591         NULL
3592 };
3593
3594 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3595                 struct attribute *a, int n)
3596 {
3597         struct device *dev = container_of(kobj, struct device, kobj);
3598         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3599
3600         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3601                 return 0;
3602         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3603                 return 0;
3604         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3605                 return 0;
3606         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3607                 return 0;
3608         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3609                 return 0;
3610         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3611                 return 0;
3612         if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3613                 return 0;
3614
3615         return a->mode;
3616 }
3617
3618 static const struct attribute_group nvme_dev_attrs_group = {
3619         .attrs          = nvme_dev_attrs,
3620         .is_visible     = nvme_dev_attrs_are_visible,
3621 };
3622
3623 static const struct attribute_group *nvme_dev_attr_groups[] = {
3624         &nvme_dev_attrs_group,
3625         NULL,
3626 };
3627
3628 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3629                 unsigned nsid)
3630 {
3631         struct nvme_ns_head *h;
3632
3633         lockdep_assert_held(&subsys->lock);
3634
3635         list_for_each_entry(h, &subsys->nsheads, entry) {
3636                 if (h->ns_id != nsid)
3637                         continue;
3638                 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3639                         return h;
3640         }
3641
3642         return NULL;
3643 }
3644
3645 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3646                 struct nvme_ns_ids *ids)
3647 {
3648         bool has_uuid = !uuid_is_null(&ids->uuid);
3649         bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3650         bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3651         struct nvme_ns_head *h;
3652
3653         lockdep_assert_held(&subsys->lock);
3654
3655         list_for_each_entry(h, &subsys->nsheads, entry) {
3656                 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3657                         return -EINVAL;
3658                 if (has_nguid &&
3659                     memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3660                         return -EINVAL;
3661                 if (has_eui64 &&
3662                     memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3663                         return -EINVAL;
3664         }
3665
3666         return 0;
3667 }
3668
3669 static void nvme_cdev_rel(struct device *dev)
3670 {
3671         ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3672 }
3673
3674 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3675 {
3676         cdev_device_del(cdev, cdev_device);
3677         put_device(cdev_device);
3678 }
3679
3680 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3681                 const struct file_operations *fops, struct module *owner)
3682 {
3683         int minor, ret;
3684
3685         minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3686         if (minor < 0)
3687                 return minor;
3688         cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3689         cdev_device->class = nvme_ns_chr_class;
3690         cdev_device->release = nvme_cdev_rel;
3691         device_initialize(cdev_device);
3692         cdev_init(cdev, fops);
3693         cdev->owner = owner;
3694         ret = cdev_device_add(cdev, cdev_device);
3695         if (ret)
3696                 put_device(cdev_device);
3697
3698         return ret;
3699 }
3700
3701 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3702 {
3703         return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3704 }
3705
3706 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3707 {
3708         nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3709         return 0;
3710 }
3711
3712 static const struct file_operations nvme_ns_chr_fops = {
3713         .owner          = THIS_MODULE,
3714         .open           = nvme_ns_chr_open,
3715         .release        = nvme_ns_chr_release,
3716         .unlocked_ioctl = nvme_ns_chr_ioctl,
3717         .compat_ioctl   = compat_ptr_ioctl,
3718 };
3719
3720 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3721 {
3722         int ret;
3723
3724         ns->cdev_device.parent = ns->ctrl->device;
3725         ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3726                            ns->ctrl->instance, ns->head->instance);
3727         if (ret)
3728                 return ret;
3729
3730         return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3731                              ns->ctrl->ops->module);
3732 }
3733
3734 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3735                 unsigned nsid, struct nvme_ns_ids *ids)
3736 {
3737         struct nvme_ns_head *head;
3738         size_t size = sizeof(*head);
3739         int ret = -ENOMEM;
3740
3741 #ifdef CONFIG_NVME_MULTIPATH
3742         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3743 #endif
3744
3745         head = kzalloc(size, GFP_KERNEL);
3746         if (!head)
3747                 goto out;
3748         ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3749         if (ret < 0)
3750                 goto out_free_head;
3751         head->instance = ret;
3752         INIT_LIST_HEAD(&head->list);
3753         ret = init_srcu_struct(&head->srcu);
3754         if (ret)
3755                 goto out_ida_remove;
3756         head->subsys = ctrl->subsys;
3757         head->ns_id = nsid;
3758         head->ids = *ids;
3759         kref_init(&head->ref);
3760
3761         if (head->ids.csi) {
3762                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3763                 if (ret)
3764                         goto out_cleanup_srcu;
3765         } else
3766                 head->effects = ctrl->effects;
3767
3768         ret = nvme_mpath_alloc_disk(ctrl, head);
3769         if (ret)
3770                 goto out_cleanup_srcu;
3771
3772         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3773
3774         kref_get(&ctrl->subsys->ref);
3775
3776         return head;
3777 out_cleanup_srcu:
3778         cleanup_srcu_struct(&head->srcu);
3779 out_ida_remove:
3780         ida_free(&ctrl->subsys->ns_ida, head->instance);
3781 out_free_head:
3782         kfree(head);
3783 out:
3784         if (ret > 0)
3785                 ret = blk_status_to_errno(nvme_error_status(ret));
3786         return ERR_PTR(ret);
3787 }
3788
3789 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3790                 struct nvme_ns_ids *ids)
3791 {
3792         struct nvme_subsystem *s;
3793         int ret = 0;
3794
3795         /*
3796          * Note that this check is racy as we try to avoid holding the global
3797          * lock over the whole ns_head creation.  But it is only intended as
3798          * a sanity check anyway.
3799          */
3800         mutex_lock(&nvme_subsystems_lock);
3801         list_for_each_entry(s, &nvme_subsystems, entry) {
3802                 if (s == this)
3803                         continue;
3804                 mutex_lock(&s->lock);
3805                 ret = nvme_subsys_check_duplicate_ids(s, ids);
3806                 mutex_unlock(&s->lock);
3807                 if (ret)
3808                         break;
3809         }
3810         mutex_unlock(&nvme_subsystems_lock);
3811
3812         return ret;
3813 }
3814
3815 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3816                 struct nvme_ns_ids *ids, bool is_shared)
3817 {
3818         struct nvme_ctrl *ctrl = ns->ctrl;
3819         struct nvme_ns_head *head = NULL;
3820         int ret;
3821
3822         ret = nvme_global_check_duplicate_ids(ctrl->subsys, ids);
3823         if (ret) {
3824                 dev_err(ctrl->device,
3825                         "globally duplicate IDs for nsid %d\n", nsid);
3826                 return ret;
3827         }
3828
3829         mutex_lock(&ctrl->subsys->lock);
3830         head = nvme_find_ns_head(ctrl->subsys, nsid);
3831         if (!head) {
3832                 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, ids);
3833                 if (ret) {
3834                         dev_err(ctrl->device,
3835                                 "duplicate IDs in subsystem for nsid %d\n",
3836                                 nsid);
3837                         goto out_unlock;
3838                 }
3839                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3840                 if (IS_ERR(head)) {
3841                         ret = PTR_ERR(head);
3842                         goto out_unlock;
3843                 }
3844                 head->shared = is_shared;
3845         } else {
3846                 ret = -EINVAL;
3847                 if (!is_shared || !head->shared) {
3848                         dev_err(ctrl->device,
3849                                 "Duplicate unshared namespace %d\n", nsid);
3850                         goto out_put_ns_head;
3851                 }
3852                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3853                         dev_err(ctrl->device,
3854                                 "IDs don't match for shared namespace %d\n",
3855                                         nsid);
3856                         goto out_put_ns_head;
3857                 }
3858         }
3859
3860         list_add_tail_rcu(&ns->siblings, &head->list);
3861         ns->head = head;
3862         mutex_unlock(&ctrl->subsys->lock);
3863         return 0;
3864
3865 out_put_ns_head:
3866         nvme_put_ns_head(head);
3867 out_unlock:
3868         mutex_unlock(&ctrl->subsys->lock);
3869         return ret;
3870 }
3871
3872 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3873 {
3874         struct nvme_ns *ns, *ret = NULL;
3875
3876         down_read(&ctrl->namespaces_rwsem);
3877         list_for_each_entry(ns, &ctrl->namespaces, list) {
3878                 if (ns->head->ns_id == nsid) {
3879                         if (!nvme_get_ns(ns))
3880                                 continue;
3881                         ret = ns;
3882                         break;
3883                 }
3884                 if (ns->head->ns_id > nsid)
3885                         break;
3886         }
3887         up_read(&ctrl->namespaces_rwsem);
3888         return ret;
3889 }
3890 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3891
3892 /*
3893  * Add the namespace to the controller list while keeping the list ordered.
3894  */
3895 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3896 {
3897         struct nvme_ns *tmp;
3898
3899         list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3900                 if (tmp->head->ns_id < ns->head->ns_id) {
3901                         list_add(&ns->list, &tmp->list);
3902                         return;
3903                 }
3904         }
3905         list_add(&ns->list, &ns->ctrl->namespaces);
3906 }
3907
3908 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3909                 struct nvme_ns_ids *ids)
3910 {
3911         struct nvme_ns *ns;
3912         struct gendisk *disk;
3913         struct nvme_id_ns *id;
3914         int node = ctrl->numa_node;
3915
3916         if (nvme_identify_ns(ctrl, nsid, ids, &id))
3917                 return;
3918
3919         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3920         if (!ns)
3921                 goto out_free_id;
3922
3923         disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3924         if (IS_ERR(disk))
3925                 goto out_free_ns;
3926         disk->fops = &nvme_bdev_ops;
3927         disk->private_data = ns;
3928
3929         ns->disk = disk;
3930         ns->queue = disk->queue;
3931
3932         if (ctrl->opts && ctrl->opts->data_digest)
3933                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3934
3935         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3936         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3937                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3938
3939         ns->ctrl = ctrl;
3940         kref_init(&ns->kref);
3941
3942         if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3943                 goto out_cleanup_disk;
3944
3945         /*
3946          * If multipathing is enabled, the device name for all disks and not
3947          * just those that represent shared namespaces needs to be based on the
3948          * subsystem instance.  Using the controller instance for private
3949          * namespaces could lead to naming collisions between shared and private
3950          * namespaces if they don't use a common numbering scheme.
3951          *
3952          * If multipathing is not enabled, disk names must use the controller
3953          * instance as shared namespaces will show up as multiple block
3954          * devices.
3955          */
3956         if (ns->head->disk) {
3957                 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3958                         ctrl->instance, ns->head->instance);
3959                 disk->flags |= GENHD_FL_HIDDEN;
3960         } else if (multipath) {
3961                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3962                         ns->head->instance);
3963         } else {
3964                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3965                         ns->head->instance);
3966         }
3967
3968         if (nvme_update_ns_info(ns, id))
3969                 goto out_unlink_ns;
3970
3971         down_write(&ctrl->namespaces_rwsem);
3972         nvme_ns_add_to_ctrl_list(ns);
3973         up_write(&ctrl->namespaces_rwsem);
3974         nvme_get_ctrl(ctrl);
3975
3976         if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3977                 goto out_cleanup_ns_from_list;
3978
3979         if (!nvme_ns_head_multipath(ns->head))
3980                 nvme_add_ns_cdev(ns);
3981
3982         nvme_mpath_add_disk(ns, id);
3983         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3984         kfree(id);
3985
3986         return;
3987
3988  out_cleanup_ns_from_list:
3989         nvme_put_ctrl(ctrl);
3990         down_write(&ctrl->namespaces_rwsem);
3991         list_del_init(&ns->list);
3992         up_write(&ctrl->namespaces_rwsem);
3993  out_unlink_ns:
3994         mutex_lock(&ctrl->subsys->lock);
3995         list_del_rcu(&ns->siblings);
3996         if (list_empty(&ns->head->list))
3997                 list_del_init(&ns->head->entry);
3998         mutex_unlock(&ctrl->subsys->lock);
3999         nvme_put_ns_head(ns->head);
4000  out_cleanup_disk:
4001         blk_cleanup_disk(disk);
4002  out_free_ns:
4003         kfree(ns);
4004  out_free_id:
4005         kfree(id);
4006 }
4007
4008 static void nvme_ns_remove(struct nvme_ns *ns)
4009 {
4010         bool last_path = false;
4011
4012         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4013                 return;
4014
4015         clear_bit(NVME_NS_READY, &ns->flags);
4016         set_capacity(ns->disk, 0);
4017         nvme_fault_inject_fini(&ns->fault_inject);
4018
4019         mutex_lock(&ns->ctrl->subsys->lock);
4020         list_del_rcu(&ns->siblings);
4021         if (list_empty(&ns->head->list)) {
4022                 list_del_init(&ns->head->entry);
4023                 last_path = true;
4024         }
4025         mutex_unlock(&ns->ctrl->subsys->lock);
4026
4027         /* guarantee not available in head->list */
4028         synchronize_rcu();
4029
4030         /* wait for concurrent submissions */
4031         if (nvme_mpath_clear_current_path(ns))
4032                 synchronize_srcu(&ns->head->srcu);
4033
4034         if (!nvme_ns_head_multipath(ns->head))
4035                 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4036         del_gendisk(ns->disk);
4037         blk_cleanup_queue(ns->queue);
4038
4039         down_write(&ns->ctrl->namespaces_rwsem);
4040         list_del_init(&ns->list);
4041         up_write(&ns->ctrl->namespaces_rwsem);
4042
4043         if (last_path)
4044                 nvme_mpath_shutdown_disk(ns->head);
4045         nvme_put_ns(ns);
4046 }
4047
4048 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4049 {
4050         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4051
4052         if (ns) {
4053                 nvme_ns_remove(ns);
4054                 nvme_put_ns(ns);
4055         }
4056 }
4057
4058 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4059 {
4060         struct nvme_id_ns *id;
4061         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4062
4063         if (test_bit(NVME_NS_DEAD, &ns->flags))
4064                 goto out;
4065
4066         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4067         if (ret)
4068                 goto out;
4069
4070         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4071         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4072                 dev_err(ns->ctrl->device,
4073                         "identifiers changed for nsid %d\n", ns->head->ns_id);
4074                 goto out_free_id;
4075         }
4076
4077         ret = nvme_update_ns_info(ns, id);
4078
4079 out_free_id:
4080         kfree(id);
4081 out:
4082         /*
4083          * Only remove the namespace if we got a fatal error back from the
4084          * device, otherwise ignore the error and just move on.
4085          *
4086          * TODO: we should probably schedule a delayed retry here.
4087          */
4088         if (ret > 0 && (ret & NVME_SC_DNR))
4089                 nvme_ns_remove(ns);
4090 }
4091
4092 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4093 {
4094         struct nvme_ns_ids ids = { };
4095         struct nvme_ns *ns;
4096
4097         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4098                 return;
4099
4100         ns = nvme_find_get_ns(ctrl, nsid);
4101         if (ns) {
4102                 nvme_validate_ns(ns, &ids);
4103                 nvme_put_ns(ns);
4104                 return;
4105         }
4106
4107         switch (ids.csi) {
4108         case NVME_CSI_NVM:
4109                 nvme_alloc_ns(ctrl, nsid, &ids);
4110                 break;
4111         case NVME_CSI_ZNS:
4112                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4113                         dev_warn(ctrl->device,
4114                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4115                                 nsid);
4116                         break;
4117                 }
4118                 if (!nvme_multi_css(ctrl)) {
4119                         dev_warn(ctrl->device,
4120                                 "command set not reported for nsid: %d\n",
4121                                 nsid);
4122                         break;
4123                 }
4124                 nvme_alloc_ns(ctrl, nsid, &ids);
4125                 break;
4126         default:
4127                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4128                         ids.csi, nsid);
4129                 break;
4130         }
4131 }
4132
4133 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4134                                         unsigned nsid)
4135 {
4136         struct nvme_ns *ns, *next;
4137         LIST_HEAD(rm_list);
4138
4139         down_write(&ctrl->namespaces_rwsem);
4140         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4141                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4142                         list_move_tail(&ns->list, &rm_list);
4143         }
4144         up_write(&ctrl->namespaces_rwsem);
4145
4146         list_for_each_entry_safe(ns, next, &rm_list, list)
4147                 nvme_ns_remove(ns);
4148
4149 }
4150
4151 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4152 {
4153         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4154         __le32 *ns_list;
4155         u32 prev = 0;
4156         int ret = 0, i;
4157
4158         if (nvme_ctrl_limited_cns(ctrl))
4159                 return -EOPNOTSUPP;
4160
4161         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4162         if (!ns_list)
4163                 return -ENOMEM;
4164
4165         for (;;) {
4166                 struct nvme_command cmd = {
4167                         .identify.opcode        = nvme_admin_identify,
4168                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
4169                         .identify.nsid          = cpu_to_le32(prev),
4170                 };
4171
4172                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4173                                             NVME_IDENTIFY_DATA_SIZE);
4174                 if (ret) {
4175                         dev_warn(ctrl->device,
4176                                 "Identify NS List failed (status=0x%x)\n", ret);
4177                         goto free;
4178                 }
4179
4180                 for (i = 0; i < nr_entries; i++) {
4181                         u32 nsid = le32_to_cpu(ns_list[i]);
4182
4183                         if (!nsid)      /* end of the list? */
4184                                 goto out;
4185                         nvme_validate_or_alloc_ns(ctrl, nsid);
4186                         while (++prev < nsid)
4187                                 nvme_ns_remove_by_nsid(ctrl, prev);
4188                 }
4189         }
4190  out:
4191         nvme_remove_invalid_namespaces(ctrl, prev);
4192  free:
4193         kfree(ns_list);
4194         return ret;
4195 }
4196
4197 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4198 {
4199         struct nvme_id_ctrl *id;
4200         u32 nn, i;
4201
4202         if (nvme_identify_ctrl(ctrl, &id))
4203                 return;
4204         nn = le32_to_cpu(id->nn);
4205         kfree(id);
4206
4207         for (i = 1; i <= nn; i++)
4208                 nvme_validate_or_alloc_ns(ctrl, i);
4209
4210         nvme_remove_invalid_namespaces(ctrl, nn);
4211 }
4212
4213 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4214 {
4215         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4216         __le32 *log;
4217         int error;
4218
4219         log = kzalloc(log_size, GFP_KERNEL);
4220         if (!log)
4221                 return;
4222
4223         /*
4224          * We need to read the log to clear the AEN, but we don't want to rely
4225          * on it for the changed namespace information as userspace could have
4226          * raced with us in reading the log page, which could cause us to miss
4227          * updates.
4228          */
4229         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4230                         NVME_CSI_NVM, log, log_size, 0);
4231         if (error)
4232                 dev_warn(ctrl->device,
4233                         "reading changed ns log failed: %d\n", error);
4234
4235         kfree(log);
4236 }
4237
4238 static void nvme_scan_work(struct work_struct *work)
4239 {
4240         struct nvme_ctrl *ctrl =
4241                 container_of(work, struct nvme_ctrl, scan_work);
4242
4243         /* No tagset on a live ctrl means IO queues could not created */
4244         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4245                 return;
4246
4247         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4248                 dev_info(ctrl->device, "rescanning namespaces.\n");
4249                 nvme_clear_changed_ns_log(ctrl);
4250         }
4251
4252         mutex_lock(&ctrl->scan_lock);
4253         if (nvme_scan_ns_list(ctrl) != 0)
4254                 nvme_scan_ns_sequential(ctrl);
4255         mutex_unlock(&ctrl->scan_lock);
4256 }
4257
4258 /*
4259  * This function iterates the namespace list unlocked to allow recovery from
4260  * controller failure. It is up to the caller to ensure the namespace list is
4261  * not modified by scan work while this function is executing.
4262  */
4263 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4264 {
4265         struct nvme_ns *ns, *next;
4266         LIST_HEAD(ns_list);
4267
4268         /*
4269          * make sure to requeue I/O to all namespaces as these
4270          * might result from the scan itself and must complete
4271          * for the scan_work to make progress
4272          */
4273         nvme_mpath_clear_ctrl_paths(ctrl);
4274
4275         /* prevent racing with ns scanning */
4276         flush_work(&ctrl->scan_work);
4277
4278         /*
4279          * The dead states indicates the controller was not gracefully
4280          * disconnected. In that case, we won't be able to flush any data while
4281          * removing the namespaces' disks; fail all the queues now to avoid
4282          * potentially having to clean up the failed sync later.
4283          */
4284         if (ctrl->state == NVME_CTRL_DEAD)
4285                 nvme_kill_queues(ctrl);
4286
4287         /* this is a no-op when called from the controller reset handler */
4288         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4289
4290         down_write(&ctrl->namespaces_rwsem);
4291         list_splice_init(&ctrl->namespaces, &ns_list);
4292         up_write(&ctrl->namespaces_rwsem);
4293
4294         list_for_each_entry_safe(ns, next, &ns_list, list)
4295                 nvme_ns_remove(ns);
4296 }
4297 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4298
4299 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4300 {
4301         struct nvme_ctrl *ctrl =
4302                 container_of(dev, struct nvme_ctrl, ctrl_device);
4303         struct nvmf_ctrl_options *opts = ctrl->opts;
4304         int ret;
4305
4306         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4307         if (ret)
4308                 return ret;
4309
4310         if (opts) {
4311                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4312                 if (ret)
4313                         return ret;
4314
4315                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4316                                 opts->trsvcid ?: "none");
4317                 if (ret)
4318                         return ret;
4319
4320                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4321                                 opts->host_traddr ?: "none");
4322                 if (ret)
4323                         return ret;
4324
4325                 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4326                                 opts->host_iface ?: "none");
4327         }
4328         return ret;
4329 }
4330
4331 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4332 {
4333         char *envp[2] = { envdata, NULL };
4334
4335         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4336 }
4337
4338 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4339 {
4340         char *envp[2] = { NULL, NULL };
4341         u32 aen_result = ctrl->aen_result;
4342
4343         ctrl->aen_result = 0;
4344         if (!aen_result)
4345                 return;
4346
4347         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4348         if (!envp[0])
4349                 return;
4350         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4351         kfree(envp[0]);
4352 }
4353
4354 static void nvme_async_event_work(struct work_struct *work)
4355 {
4356         struct nvme_ctrl *ctrl =
4357                 container_of(work, struct nvme_ctrl, async_event_work);
4358
4359         nvme_aen_uevent(ctrl);
4360         ctrl->ops->submit_async_event(ctrl);
4361 }
4362
4363 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4364 {
4365
4366         u32 csts;
4367
4368         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4369                 return false;
4370
4371         if (csts == ~0)
4372                 return false;
4373
4374         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4375 }
4376
4377 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4378 {
4379         struct nvme_fw_slot_info_log *log;
4380
4381         log = kmalloc(sizeof(*log), GFP_KERNEL);
4382         if (!log)
4383                 return;
4384
4385         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4386                         log, sizeof(*log), 0))
4387                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4388         kfree(log);
4389 }
4390
4391 static void nvme_fw_act_work(struct work_struct *work)
4392 {
4393         struct nvme_ctrl *ctrl = container_of(work,
4394                                 struct nvme_ctrl, fw_act_work);
4395         unsigned long fw_act_timeout;
4396
4397         if (ctrl->mtfa)
4398                 fw_act_timeout = jiffies +
4399                                 msecs_to_jiffies(ctrl->mtfa * 100);
4400         else
4401                 fw_act_timeout = jiffies +
4402                                 msecs_to_jiffies(admin_timeout * 1000);
4403
4404         nvme_stop_queues(ctrl);
4405         while (nvme_ctrl_pp_status(ctrl)) {
4406                 if (time_after(jiffies, fw_act_timeout)) {
4407                         dev_warn(ctrl->device,
4408                                 "Fw activation timeout, reset controller\n");
4409                         nvme_try_sched_reset(ctrl);
4410                         return;
4411                 }
4412                 msleep(100);
4413         }
4414
4415         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4416                 return;
4417
4418         nvme_start_queues(ctrl);
4419         /* read FW slot information to clear the AER */
4420         nvme_get_fw_slot_info(ctrl);
4421 }
4422
4423 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4424 {
4425         u32 aer_notice_type = (result & 0xff00) >> 8;
4426
4427         trace_nvme_async_event(ctrl, aer_notice_type);
4428
4429         switch (aer_notice_type) {
4430         case NVME_AER_NOTICE_NS_CHANGED:
4431                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4432                 nvme_queue_scan(ctrl);
4433                 break;
4434         case NVME_AER_NOTICE_FW_ACT_STARTING:
4435                 /*
4436                  * We are (ab)using the RESETTING state to prevent subsequent
4437                  * recovery actions from interfering with the controller's
4438                  * firmware activation.
4439                  */
4440                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4441                         queue_work(nvme_wq, &ctrl->fw_act_work);
4442                 break;
4443 #ifdef CONFIG_NVME_MULTIPATH
4444         case NVME_AER_NOTICE_ANA:
4445                 if (!ctrl->ana_log_buf)
4446                         break;
4447                 queue_work(nvme_wq, &ctrl->ana_work);
4448                 break;
4449 #endif
4450         case NVME_AER_NOTICE_DISC_CHANGED:
4451                 ctrl->aen_result = result;
4452                 break;
4453         default:
4454                 dev_warn(ctrl->device, "async event result %08x\n", result);
4455         }
4456 }
4457
4458 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4459                 volatile union nvme_result *res)
4460 {
4461         u32 result = le32_to_cpu(res->u32);
4462         u32 aer_type = result & 0x07;
4463
4464         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4465                 return;
4466
4467         switch (aer_type) {
4468         case NVME_AER_NOTICE:
4469                 nvme_handle_aen_notice(ctrl, result);
4470                 break;
4471         case NVME_AER_ERROR:
4472         case NVME_AER_SMART:
4473         case NVME_AER_CSS:
4474         case NVME_AER_VS:
4475                 trace_nvme_async_event(ctrl, aer_type);
4476                 ctrl->aen_result = result;
4477                 break;
4478         default:
4479                 break;
4480         }
4481         queue_work(nvme_wq, &ctrl->async_event_work);
4482 }
4483 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4484
4485 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4486 {
4487         nvme_mpath_stop(ctrl);
4488         nvme_stop_keep_alive(ctrl);
4489         nvme_stop_failfast_work(ctrl);
4490         flush_work(&ctrl->async_event_work);
4491         cancel_work_sync(&ctrl->fw_act_work);
4492 }
4493 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4494
4495 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4496 {
4497         nvme_start_keep_alive(ctrl);
4498
4499         nvme_enable_aen(ctrl);
4500
4501         if (ctrl->queue_count > 1) {
4502                 nvme_queue_scan(ctrl);
4503                 nvme_start_queues(ctrl);
4504         }
4505
4506         nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4507 }
4508 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4509
4510 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4511 {
4512         nvme_hwmon_exit(ctrl);
4513         nvme_fault_inject_fini(&ctrl->fault_inject);
4514         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4515         cdev_device_del(&ctrl->cdev, ctrl->device);
4516         nvme_put_ctrl(ctrl);
4517 }
4518 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4519
4520 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4521 {
4522         struct nvme_effects_log *cel;
4523         unsigned long i;
4524
4525         xa_for_each(&ctrl->cels, i, cel) {
4526                 xa_erase(&ctrl->cels, i);
4527                 kfree(cel);
4528         }
4529
4530         xa_destroy(&ctrl->cels);
4531 }
4532
4533 static void nvme_free_ctrl(struct device *dev)
4534 {
4535         struct nvme_ctrl *ctrl =
4536                 container_of(dev, struct nvme_ctrl, ctrl_device);
4537         struct nvme_subsystem *subsys = ctrl->subsys;
4538
4539         if (!subsys || ctrl->instance != subsys->instance)
4540                 ida_free(&nvme_instance_ida, ctrl->instance);
4541
4542         nvme_free_cels(ctrl);
4543         nvme_mpath_uninit(ctrl);
4544         __free_page(ctrl->discard_page);
4545
4546         if (subsys) {
4547                 mutex_lock(&nvme_subsystems_lock);
4548                 list_del(&ctrl->subsys_entry);
4549                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4550                 mutex_unlock(&nvme_subsystems_lock);
4551         }
4552
4553         ctrl->ops->free_ctrl(ctrl);
4554
4555         if (subsys)
4556                 nvme_put_subsystem(subsys);
4557 }
4558
4559 /*
4560  * Initialize a NVMe controller structures.  This needs to be called during
4561  * earliest initialization so that we have the initialized structured around
4562  * during probing.
4563  */
4564 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4565                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4566 {
4567         int ret;
4568
4569         ctrl->state = NVME_CTRL_NEW;
4570         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4571         spin_lock_init(&ctrl->lock);
4572         mutex_init(&ctrl->scan_lock);
4573         INIT_LIST_HEAD(&ctrl->namespaces);
4574         xa_init(&ctrl->cels);
4575         init_rwsem(&ctrl->namespaces_rwsem);
4576         ctrl->dev = dev;
4577         ctrl->ops = ops;
4578         ctrl->quirks = quirks;
4579         ctrl->numa_node = NUMA_NO_NODE;
4580         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4581         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4582         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4583         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4584         init_waitqueue_head(&ctrl->state_wq);
4585
4586         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4587         INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4588         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4589         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4590
4591         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4592                         PAGE_SIZE);
4593         ctrl->discard_page = alloc_page(GFP_KERNEL);
4594         if (!ctrl->discard_page) {
4595                 ret = -ENOMEM;
4596                 goto out;
4597         }
4598
4599         ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4600         if (ret < 0)
4601                 goto out;
4602         ctrl->instance = ret;
4603
4604         device_initialize(&ctrl->ctrl_device);
4605         ctrl->device = &ctrl->ctrl_device;
4606         ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4607                         ctrl->instance);
4608         ctrl->device->class = nvme_class;
4609         ctrl->device->parent = ctrl->dev;
4610         ctrl->device->groups = nvme_dev_attr_groups;
4611         ctrl->device->release = nvme_free_ctrl;
4612         dev_set_drvdata(ctrl->device, ctrl);
4613         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4614         if (ret)
4615                 goto out_release_instance;
4616
4617         nvme_get_ctrl(ctrl);
4618         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4619         ctrl->cdev.owner = ops->module;
4620         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4621         if (ret)
4622                 goto out_free_name;
4623
4624         /*
4625          * Initialize latency tolerance controls.  The sysfs files won't
4626          * be visible to userspace unless the device actually supports APST.
4627          */
4628         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4629         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4630                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4631
4632         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4633         nvme_mpath_init_ctrl(ctrl);
4634
4635         return 0;
4636 out_free_name:
4637         nvme_put_ctrl(ctrl);
4638         kfree_const(ctrl->device->kobj.name);
4639 out_release_instance:
4640         ida_free(&nvme_instance_ida, ctrl->instance);
4641 out:
4642         if (ctrl->discard_page)
4643                 __free_page(ctrl->discard_page);
4644         return ret;
4645 }
4646 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4647
4648 static void nvme_start_ns_queue(struct nvme_ns *ns)
4649 {
4650         if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4651                 blk_mq_unquiesce_queue(ns->queue);
4652 }
4653
4654 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4655 {
4656         if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4657                 blk_mq_quiesce_queue(ns->queue);
4658         else
4659                 blk_mq_wait_quiesce_done(ns->queue);
4660 }
4661
4662 /*
4663  * Prepare a queue for teardown.
4664  *
4665  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4666  * the capacity to 0 after that to avoid blocking dispatchers that may be
4667  * holding bd_butex.  This will end buffered writers dirtying pages that can't
4668  * be synced.
4669  */
4670 static void nvme_set_queue_dying(struct nvme_ns *ns)
4671 {
4672         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4673                 return;
4674
4675         blk_set_queue_dying(ns->queue);
4676         nvme_start_ns_queue(ns);
4677
4678         set_capacity_and_notify(ns->disk, 0);
4679 }
4680
4681 /**
4682  * nvme_kill_queues(): Ends all namespace queues
4683  * @ctrl: the dead controller that needs to end
4684  *
4685  * Call this function when the driver determines it is unable to get the
4686  * controller in a state capable of servicing IO.
4687  */
4688 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4689 {
4690         struct nvme_ns *ns;
4691
4692         down_read(&ctrl->namespaces_rwsem);
4693
4694         /* Forcibly unquiesce queues to avoid blocking dispatch */
4695         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4696                 nvme_start_admin_queue(ctrl);
4697
4698         list_for_each_entry(ns, &ctrl->namespaces, list)
4699                 nvme_set_queue_dying(ns);
4700
4701         up_read(&ctrl->namespaces_rwsem);
4702 }
4703 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4704
4705 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4706 {
4707         struct nvme_ns *ns;
4708
4709         down_read(&ctrl->namespaces_rwsem);
4710         list_for_each_entry(ns, &ctrl->namespaces, list)
4711                 blk_mq_unfreeze_queue(ns->queue);
4712         up_read(&ctrl->namespaces_rwsem);
4713 }
4714 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4715
4716 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4717 {
4718         struct nvme_ns *ns;
4719
4720         down_read(&ctrl->namespaces_rwsem);
4721         list_for_each_entry(ns, &ctrl->namespaces, list) {
4722                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4723                 if (timeout <= 0)
4724                         break;
4725         }
4726         up_read(&ctrl->namespaces_rwsem);
4727         return timeout;
4728 }
4729 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4730
4731 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4732 {
4733         struct nvme_ns *ns;
4734
4735         down_read(&ctrl->namespaces_rwsem);
4736         list_for_each_entry(ns, &ctrl->namespaces, list)
4737                 blk_mq_freeze_queue_wait(ns->queue);
4738         up_read(&ctrl->namespaces_rwsem);
4739 }
4740 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4741
4742 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4743 {
4744         struct nvme_ns *ns;
4745
4746         down_read(&ctrl->namespaces_rwsem);
4747         list_for_each_entry(ns, &ctrl->namespaces, list)
4748                 blk_freeze_queue_start(ns->queue);
4749         up_read(&ctrl->namespaces_rwsem);
4750 }
4751 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4752
4753 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4754 {
4755         struct nvme_ns *ns;
4756
4757         down_read(&ctrl->namespaces_rwsem);
4758         list_for_each_entry(ns, &ctrl->namespaces, list)
4759                 nvme_stop_ns_queue(ns);
4760         up_read(&ctrl->namespaces_rwsem);
4761 }
4762 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4763
4764 void nvme_start_queues(struct nvme_ctrl *ctrl)
4765 {
4766         struct nvme_ns *ns;
4767
4768         down_read(&ctrl->namespaces_rwsem);
4769         list_for_each_entry(ns, &ctrl->namespaces, list)
4770                 nvme_start_ns_queue(ns);
4771         up_read(&ctrl->namespaces_rwsem);
4772 }
4773 EXPORT_SYMBOL_GPL(nvme_start_queues);
4774
4775 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4776 {
4777         if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4778                 blk_mq_quiesce_queue(ctrl->admin_q);
4779         else
4780                 blk_mq_wait_quiesce_done(ctrl->admin_q);
4781 }
4782 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4783
4784 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4785 {
4786         if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4787                 blk_mq_unquiesce_queue(ctrl->admin_q);
4788 }
4789 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4790
4791 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4792 {
4793         struct nvme_ns *ns;
4794
4795         down_read(&ctrl->namespaces_rwsem);
4796         list_for_each_entry(ns, &ctrl->namespaces, list)
4797                 blk_sync_queue(ns->queue);
4798         up_read(&ctrl->namespaces_rwsem);
4799 }
4800 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4801
4802 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4803 {
4804         nvme_sync_io_queues(ctrl);
4805         if (ctrl->admin_q)
4806                 blk_sync_queue(ctrl->admin_q);
4807 }
4808 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4809
4810 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4811 {
4812         if (file->f_op != &nvme_dev_fops)
4813                 return NULL;
4814         return file->private_data;
4815 }
4816 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4817
4818 /*
4819  * Check we didn't inadvertently grow the command structure sizes:
4820  */
4821 static inline void _nvme_check_size(void)
4822 {
4823         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4824         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4825         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4826         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4827         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4828         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4829         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4830         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4831         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4832         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4833         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4834         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4835         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4836         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4837         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4838         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4839         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4840         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4841         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4842         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4843 }
4844
4845
4846 static int __init nvme_core_init(void)
4847 {
4848         int result = -ENOMEM;
4849
4850         _nvme_check_size();
4851
4852         nvme_wq = alloc_workqueue("nvme-wq",
4853                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4854         if (!nvme_wq)
4855                 goto out;
4856
4857         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4858                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4859         if (!nvme_reset_wq)
4860                 goto destroy_wq;
4861
4862         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4863                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4864         if (!nvme_delete_wq)
4865                 goto destroy_reset_wq;
4866
4867         result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4868                         NVME_MINORS, "nvme");
4869         if (result < 0)
4870                 goto destroy_delete_wq;
4871
4872         nvme_class = class_create(THIS_MODULE, "nvme");
4873         if (IS_ERR(nvme_class)) {
4874                 result = PTR_ERR(nvme_class);
4875                 goto unregister_chrdev;
4876         }
4877         nvme_class->dev_uevent = nvme_class_uevent;
4878
4879         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4880         if (IS_ERR(nvme_subsys_class)) {
4881                 result = PTR_ERR(nvme_subsys_class);
4882                 goto destroy_class;
4883         }
4884
4885         result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4886                                      "nvme-generic");
4887         if (result < 0)
4888                 goto destroy_subsys_class;
4889
4890         nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4891         if (IS_ERR(nvme_ns_chr_class)) {
4892                 result = PTR_ERR(nvme_ns_chr_class);
4893                 goto unregister_generic_ns;
4894         }
4895
4896         return 0;
4897
4898 unregister_generic_ns:
4899         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4900 destroy_subsys_class:
4901         class_destroy(nvme_subsys_class);
4902 destroy_class:
4903         class_destroy(nvme_class);
4904 unregister_chrdev:
4905         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4906 destroy_delete_wq:
4907         destroy_workqueue(nvme_delete_wq);
4908 destroy_reset_wq:
4909         destroy_workqueue(nvme_reset_wq);
4910 destroy_wq:
4911         destroy_workqueue(nvme_wq);
4912 out:
4913         return result;
4914 }
4915
4916 static void __exit nvme_core_exit(void)
4917 {
4918         class_destroy(nvme_ns_chr_class);
4919         class_destroy(nvme_subsys_class);
4920         class_destroy(nvme_class);
4921         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4922         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4923         destroy_workqueue(nvme_delete_wq);
4924         destroy_workqueue(nvme_reset_wq);
4925         destroy_workqueue(nvme_wq);
4926         ida_destroy(&nvme_ns_chr_minor_ida);
4927         ida_destroy(&nvme_instance_ida);
4928 }
4929
4930 MODULE_LICENSE("GPL");
4931 MODULE_VERSION("1.0");
4932 module_init(nvme_core_init);
4933 module_exit(nvme_core_exit);