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 /*
  81  * nvme_wq - hosts nvme related works that are not reset or delete
  82  * nvme_reset_wq - hosts nvme reset works
  83  * nvme_delete_wq - hosts nvme delete works
  84  *
  85  * nvme_wq will host works such as scan, aen handling, fw activation,
  86  * keep-alive, periodic reconnects etc. nvme_reset_wq
  87  * runs reset works which also flush works hosted on nvme_wq for
  88  * serialization purposes. nvme_delete_wq host controller deletion
  89  * works which flush reset works for serialization.
  90  */
  91 struct workqueue_struct *nvme_wq;
  92 EXPORT_SYMBOL_GPL(nvme_wq);
  93
  94 struct workqueue_struct *nvme_reset_wq;
  95 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  96
  97 struct workqueue_struct *nvme_delete_wq;
  98 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  99
 100 static LIST_HEAD(nvme_subsystems);
 101 static DEFINE_MUTEX(nvme_subsystems_lock);
 102
 103 static DEFINE_IDA(nvme_instance_ida);
 104 static dev_t nvme_ctrl_base_chr_devt;
 105 static struct class *nvme_class;
 106 static struct class *nvme_subsys_class;
 107
 108 static DEFINE_IDA(nvme_ns_chr_minor_ida);
 109 static dev_t nvme_ns_chr_devt;
 110 static struct class *nvme_ns_chr_class;
 111
 112 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
 113 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
 114                                            unsigned nsid);
 115 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
 116                                    struct nvme_command *cmd);
 117
 118 void nvme_queue_scan(struct nvme_ctrl *ctrl)
 119 {
 120         /*
 121          * Only new queue scan work when admin and IO queues are both alive
 122          */
 123         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 124                 queue_work(nvme_wq, &ctrl->scan_work);
 125 }
 126
 127 /*
 128  * Use this function to proceed with scheduling reset_work for a controller
 129  * that had previously been set to the resetting state. This is intended for
 130  * code paths that can't be interrupted by other reset attempts. A hot removal
 131  * may prevent this from succeeding.
 132  */
 133 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 134 {
 135         if (ctrl->state != NVME_CTRL_RESETTING)
 136                 return -EBUSY;
 137         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 138                 return -EBUSY;
 139         return 0;
 140 }
 141 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 142
 143 static void nvme_failfast_work(struct work_struct *work)
 144 {
 145         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 146                         struct nvme_ctrl, failfast_work);
 147
 148         if (ctrl->state != NVME_CTRL_CONNECTING)
 149                 return;
 150
 151         set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 152         dev_info(ctrl->device, "failfast expired\n");
 153         nvme_kick_requeue_lists(ctrl);
 154 }
 155
 156 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 157 {
 158         if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 159                 return;
 160
 161         schedule_delayed_work(&ctrl->failfast_work,
 162                               ctrl->opts->fast_io_fail_tmo * HZ);
 163 }
 164
 165 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 166 {
 167         if (!ctrl->opts)
 168                 return;
 169
 170         cancel_delayed_work_sync(&ctrl->failfast_work);
 171         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 172 }
 173
 174
 175 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 176 {
 177         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 178                 return -EBUSY;
 179         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 180                 return -EBUSY;
 181         return 0;
 182 }
 183 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 184
 185 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 186 {
 187         int ret;
 188
 189         ret = nvme_reset_ctrl(ctrl);
 190         if (!ret) {
 191                 flush_work(&ctrl->reset_work);
 192                 if (ctrl->state != NVME_CTRL_LIVE)
 193                         ret = -ENETRESET;
 194         }
 195
 196         return ret;
 197 }
 198
 199 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 200 {
 201         dev_info(ctrl->device,
 202                  "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
 203
 204         flush_work(&ctrl->reset_work);
 205         nvme_stop_ctrl(ctrl);
 206         nvme_remove_namespaces(ctrl);
 207         ctrl->ops->delete_ctrl(ctrl);
 208         nvme_uninit_ctrl(ctrl);
 209 }
 210
 211 static void nvme_delete_ctrl_work(struct work_struct *work)
 212 {
 213         struct nvme_ctrl *ctrl =
 214                 container_of(work, struct nvme_ctrl, delete_work);
 215
 216         nvme_do_delete_ctrl(ctrl);
 217 }
 218
 219 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 220 {
 221         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 222                 return -EBUSY;
 223         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 224                 return -EBUSY;
 225         return 0;
 226 }
 227 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 228
 229 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 230 {
 231         /*
 232          * Keep a reference until nvme_do_delete_ctrl() complete,
 233          * since ->delete_ctrl can free the controller.
 234          */
 235         nvme_get_ctrl(ctrl);
 236         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 237                 nvme_do_delete_ctrl(ctrl);
 238         nvme_put_ctrl(ctrl);
 239 }
 240
 241 static blk_status_t nvme_error_status(u16 status)
 242 {
 243         switch (status & 0x7ff) {
 244         case NVME_SC_SUCCESS:
 245                 return BLK_STS_OK;
 246         case NVME_SC_CAP_EXCEEDED:
 247                 return BLK_STS_NOSPC;
 248         case NVME_SC_LBA_RANGE:
 249         case NVME_SC_CMD_INTERRUPTED:
 250         case NVME_SC_NS_NOT_READY:
 251                 return BLK_STS_TARGET;
 252         case NVME_SC_BAD_ATTRIBUTES:
 253         case NVME_SC_ONCS_NOT_SUPPORTED:
 254         case NVME_SC_INVALID_OPCODE:
 255         case NVME_SC_INVALID_FIELD:
 256         case NVME_SC_INVALID_NS:
 257                 return BLK_STS_NOTSUPP;
 258         case NVME_SC_WRITE_FAULT:
 259         case NVME_SC_READ_ERROR:
 260         case NVME_SC_UNWRITTEN_BLOCK:
 261         case NVME_SC_ACCESS_DENIED:
 262         case NVME_SC_READ_ONLY:
 263         case NVME_SC_COMPARE_FAILED:
 264                 return BLK_STS_MEDIUM;
 265         case NVME_SC_GUARD_CHECK:
 266         case NVME_SC_APPTAG_CHECK:
 267         case NVME_SC_REFTAG_CHECK:
 268         case NVME_SC_INVALID_PI:
 269                 return BLK_STS_PROTECTION;
 270         case NVME_SC_RESERVATION_CONFLICT:
 271                 return BLK_STS_NEXUS;
 272         case NVME_SC_HOST_PATH_ERROR:
 273                 return BLK_STS_TRANSPORT;
 274         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 275                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 276         case NVME_SC_ZONE_TOO_MANY_OPEN:
 277                 return BLK_STS_ZONE_OPEN_RESOURCE;
 278         default:
 279                 return BLK_STS_IOERR;
 280         }
 281 }
 282
 283 static void nvme_retry_req(struct request *req)
 284 {
 285         unsigned long delay = 0;
 286         u16 crd;
 287
 288         /* The mask and shift result must be <= 3 */
 289         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 290         if (crd)
 291                 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 292
 293         nvme_req(req)->retries++;
 294         blk_mq_requeue_request(req, false);
 295         blk_mq_delay_kick_requeue_list(req->q, delay);
 296 }
 297
 298 static void nvme_log_error(struct request *req)
 299 {
 300         struct nvme_ns *ns = req->q->queuedata;
 301         struct nvme_request *nr = nvme_req(req);
 302
 303         if (ns) {
 304                 pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
 305                        ns->disk ? ns->disk->disk_name : "?",
 306                        nvme_get_opcode_str(nr->cmd->common.opcode),
 307                        nr->cmd->common.opcode,
 308                        (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
 309                        (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
 310                        nvme_get_error_status_str(nr->status),
 311                        nr->status >> 8 & 7,     /* Status Code Type */
 312                        nr->status & 0xff,       /* Status Code */
 313                        nr->status & NVME_SC_MORE ? "MORE " : "",
 314                        nr->status & NVME_SC_DNR  ? "DNR "  : "");
 315                 return;
 316         }
 317
 318         pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
 319                            dev_name(nr->ctrl->device),
 320                            nvme_get_admin_opcode_str(nr->cmd->common.opcode),
 321                            nr->cmd->common.opcode,
 322                            nvme_get_error_status_str(nr->status),
 323                            nr->status >> 8 & 7, /* Status Code Type */
 324                            nr->status & 0xff,   /* Status Code */
 325                            nr->status & NVME_SC_MORE ? "MORE " : "",
 326                            nr->status & NVME_SC_DNR  ? "DNR "  : "");
 327 }
 328
 329 enum nvme_disposition {
 330         COMPLETE,
 331         RETRY,
 332         FAILOVER,
 333 };
 334
 335 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 336 {
 337         if (likely(nvme_req(req)->status == 0))
 338                 return COMPLETE;
 339
 340         if (blk_noretry_request(req) ||
 341             (nvme_req(req)->status & NVME_SC_DNR) ||
 342             nvme_req(req)->retries >= nvme_max_retries)
 343                 return COMPLETE;
 344
 345         if (req->cmd_flags & REQ_NVME_MPATH) {
 346                 if (nvme_is_path_error(nvme_req(req)->status) ||
 347                     blk_queue_dying(req->q))
 348                         return FAILOVER;
 349         } else {
 350                 if (blk_queue_dying(req->q))
 351                         return COMPLETE;
 352         }
 353
 354         return RETRY;
 355 }
 356
 357 static inline void nvme_end_req_zoned(struct request *req)
 358 {
 359         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 360             req_op(req) == REQ_OP_ZONE_APPEND)
 361                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 362                         le64_to_cpu(nvme_req(req)->result.u64));
 363 }
 364
 365 static inline void nvme_end_req(struct request *req)
 366 {
 367         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 368
 369         if (unlikely(nvme_req(req)->status != NVME_SC_SUCCESS))
 370                 nvme_log_error(req);
 371         nvme_end_req_zoned(req);
 372         nvme_trace_bio_complete(req);
 373         blk_mq_end_request(req, status);
 374 }
 375
 376 void nvme_complete_rq(struct request *req)
 377 {
 378         trace_nvme_complete_rq(req);
 379         nvme_cleanup_cmd(req);
 380
 381         if (nvme_req(req)->ctrl->kas)
 382                 nvme_req(req)->ctrl->comp_seen = true;
 383
 384         switch (nvme_decide_disposition(req)) {
 385         case COMPLETE:
 386                 nvme_end_req(req);
 387                 return;
 388         case RETRY:
 389                 nvme_retry_req(req);
 390                 return;
 391         case FAILOVER:
 392                 nvme_failover_req(req);
 393                 return;
 394         }
 395 }
 396 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 397
 398 void nvme_complete_batch_req(struct request *req)
 399 {
 400         trace_nvme_complete_rq(req);
 401         nvme_cleanup_cmd(req);
 402         nvme_end_req_zoned(req);
 403 }
 404 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
 405
 406 /*
 407  * Called to unwind from ->queue_rq on a failed command submission so that the
 408  * multipathing code gets called to potentially failover to another path.
 409  * The caller needs to unwind all transport specific resource allocations and
 410  * must return propagate the return value.
 411  */
 412 blk_status_t nvme_host_path_error(struct request *req)
 413 {
 414         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 415         blk_mq_set_request_complete(req);
 416         nvme_complete_rq(req);
 417         return BLK_STS_OK;
 418 }
 419 EXPORT_SYMBOL_GPL(nvme_host_path_error);
 420
 421 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 422 {
 423         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 424                                 "Cancelling I/O %d", req->tag);
 425
 426         /* don't abort one completed request */
 427         if (blk_mq_request_completed(req))
 428                 return true;
 429
 430         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 431         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 432         blk_mq_complete_request(req);
 433         return true;
 434 }
 435 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 436
 437 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 438 {
 439         if (ctrl->tagset) {
 440                 blk_mq_tagset_busy_iter(ctrl->tagset,
 441                                 nvme_cancel_request, ctrl);
 442                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 443         }
 444 }
 445 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 446
 447 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 448 {
 449         if (ctrl->admin_tagset) {
 450                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 451                                 nvme_cancel_request, ctrl);
 452                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 453         }
 454 }
 455 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 456
 457 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 458                 enum nvme_ctrl_state new_state)
 459 {
 460         enum nvme_ctrl_state old_state;
 461         unsigned long flags;
 462         bool changed = false;
 463
 464         spin_lock_irqsave(&ctrl->lock, flags);
 465
 466         old_state = ctrl->state;
 467         switch (new_state) {
 468         case NVME_CTRL_LIVE:
 469                 switch (old_state) {
 470                 case NVME_CTRL_NEW:
 471                 case NVME_CTRL_RESETTING:
 472                 case NVME_CTRL_CONNECTING:
 473                         changed = true;
 474                         fallthrough;
 475                 default:
 476                         break;
 477                 }
 478                 break;
 479         case NVME_CTRL_RESETTING:
 480                 switch (old_state) {
 481                 case NVME_CTRL_NEW:
 482                 case NVME_CTRL_LIVE:
 483                         changed = true;
 484                         fallthrough;
 485                 default:
 486                         break;
 487                 }
 488                 break;
 489         case NVME_CTRL_CONNECTING:
 490                 switch (old_state) {
 491                 case NVME_CTRL_NEW:
 492                 case NVME_CTRL_RESETTING:
 493                         changed = true;
 494                         fallthrough;
 495                 default:
 496                         break;
 497                 }
 498                 break;
 499         case NVME_CTRL_DELETING:
 500                 switch (old_state) {
 501                 case NVME_CTRL_LIVE:
 502                 case NVME_CTRL_RESETTING:
 503                 case NVME_CTRL_CONNECTING:
 504                         changed = true;
 505                         fallthrough;
 506                 default:
 507                         break;
 508                 }
 509                 break;
 510         case NVME_CTRL_DELETING_NOIO:
 511                 switch (old_state) {
 512                 case NVME_CTRL_DELETING:
 513                 case NVME_CTRL_DEAD:
 514                         changed = true;
 515                         fallthrough;
 516                 default:
 517                         break;
 518                 }
 519                 break;
 520         case NVME_CTRL_DEAD:
 521                 switch (old_state) {
 522                 case NVME_CTRL_DELETING:
 523                         changed = true;
 524                         fallthrough;
 525                 default:
 526                         break;
 527                 }
 528                 break;
 529         default:
 530                 break;
 531         }
 532
 533         if (changed) {
 534                 ctrl->state = new_state;
 535                 wake_up_all(&ctrl->state_wq);
 536         }
 537
 538         spin_unlock_irqrestore(&ctrl->lock, flags);
 539         if (!changed)
 540                 return false;
 541
 542         if (ctrl->state == NVME_CTRL_LIVE) {
 543                 if (old_state == NVME_CTRL_CONNECTING)
 544                         nvme_stop_failfast_work(ctrl);
 545                 nvme_kick_requeue_lists(ctrl);
 546         } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 547                 old_state == NVME_CTRL_RESETTING) {
 548                 nvme_start_failfast_work(ctrl);
 549         }
 550         return changed;
 551 }
 552 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 553
 554 /*
 555  * Returns true for sink states that can't ever transition back to live.
 556  */
 557 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 558 {
 559         switch (ctrl->state) {
 560         case NVME_CTRL_NEW:
 561         case NVME_CTRL_LIVE:
 562         case NVME_CTRL_RESETTING:
 563         case NVME_CTRL_CONNECTING:
 564                 return false;
 565         case NVME_CTRL_DELETING:
 566         case NVME_CTRL_DELETING_NOIO:
 567         case NVME_CTRL_DEAD:
 568                 return true;
 569         default:
 570                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 571                 return true;
 572         }
 573 }
 574
 575 /*
 576  * Waits for the controller state to be resetting, or returns false if it is
 577  * not possible to ever transition to that state.
 578  */
 579 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 580 {
 581         wait_event(ctrl->state_wq,
 582                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 583                    nvme_state_terminal(ctrl));
 584         return ctrl->state == NVME_CTRL_RESETTING;
 585 }
 586 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 587
 588 static void nvme_free_ns_head(struct kref *ref)
 589 {
 590         struct nvme_ns_head *head =
 591                 container_of(ref, struct nvme_ns_head, ref);
 592
 593         nvme_mpath_remove_disk(head);
 594         ida_free(&head->subsys->ns_ida, head->instance);
 595         cleanup_srcu_struct(&head->srcu);
 596         nvme_put_subsystem(head->subsys);
 597         kfree(head);
 598 }
 599
 600 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
 601 {
 602         return kref_get_unless_zero(&head->ref);
 603 }
 604
 605 void nvme_put_ns_head(struct nvme_ns_head *head)
 606 {
 607         kref_put(&head->ref, nvme_free_ns_head);
 608 }
 609
 610 static void nvme_free_ns(struct kref *kref)
 611 {
 612         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 613
 614         put_disk(ns->disk);
 615         nvme_put_ns_head(ns->head);
 616         nvme_put_ctrl(ns->ctrl);
 617         kfree(ns);
 618 }
 619
 620 static inline bool nvme_get_ns(struct nvme_ns *ns)
 621 {
 622         return kref_get_unless_zero(&ns->kref);
 623 }
 624
 625 void nvme_put_ns(struct nvme_ns *ns)
 626 {
 627         kref_put(&ns->kref, nvme_free_ns);
 628 }
 629 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 630
 631 static inline void nvme_clear_nvme_request(struct request *req)
 632 {
 633         nvme_req(req)->status = 0;
 634         nvme_req(req)->retries = 0;
 635         nvme_req(req)->flags = 0;
 636         req->rq_flags |= RQF_DONTPREP;
 637 }
 638
 639 /* initialize a passthrough request */
 640 void nvme_init_request(struct request *req, struct nvme_command *cmd)
 641 {
 642         if (req->q->queuedata)
 643                 req->timeout = NVME_IO_TIMEOUT;
 644         else /* no queuedata implies admin queue */
 645                 req->timeout = NVME_ADMIN_TIMEOUT;
 646
 647         /* passthru commands should let the driver set the SGL flags */
 648         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 649
 650         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 651         if (req->mq_hctx->type == HCTX_TYPE_POLL)
 652                 req->cmd_flags |= REQ_POLLED;
 653         nvme_clear_nvme_request(req);
 654         memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
 655 }
 656 EXPORT_SYMBOL_GPL(nvme_init_request);
 657
 658 /*
 659  * For something we're not in a state to send to the device the default action
 660  * is to busy it and retry it after the controller state is recovered.  However,
 661  * if the controller is deleting or if anything is marked for failfast or
 662  * nvme multipath it is immediately failed.
 663  *
 664  * Note: commands used to initialize the controller will be marked for failfast.
 665  * Note: nvme cli/ioctl commands are marked for failfast.
 666  */
 667 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
 668                 struct request *rq)
 669 {
 670         if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
 671             ctrl->state != NVME_CTRL_DELETING &&
 672             ctrl->state != NVME_CTRL_DEAD &&
 673             !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
 674             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 675                 return BLK_STS_RESOURCE;
 676         return nvme_host_path_error(rq);
 677 }
 678 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
 679
 680 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 681                 bool queue_live)
 682 {
 683         struct nvme_request *req = nvme_req(rq);
 684
 685         /*
 686          * currently we have a problem sending passthru commands
 687          * on the admin_q if the controller is not LIVE because we can't
 688          * make sure that they are going out after the admin connect,
 689          * controller enable and/or other commands in the initialization
 690          * sequence. until the controller will be LIVE, fail with
 691          * BLK_STS_RESOURCE so that they will be rescheduled.
 692          */
 693         if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
 694                 return false;
 695
 696         if (ctrl->ops->flags & NVME_F_FABRICS) {
 697                 /*
 698                  * Only allow commands on a live queue, except for the connect
 699                  * command, which is require to set the queue live in the
 700                  * appropinquate states.
 701                  */
 702                 switch (ctrl->state) {
 703                 case NVME_CTRL_CONNECTING:
 704                         if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
 705                             req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
 706                                 return true;
 707                         break;
 708                 default:
 709                         break;
 710                 case NVME_CTRL_DEAD:
 711                         return false;
 712                 }
 713         }
 714
 715         return queue_live;
 716 }
 717 EXPORT_SYMBOL_GPL(__nvme_check_ready);
 718
 719 static inline void nvme_setup_flush(struct nvme_ns *ns,
 720                 struct nvme_command *cmnd)
 721 {
 722         memset(cmnd, 0, sizeof(*cmnd));
 723         cmnd->common.opcode = nvme_cmd_flush;
 724         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 725 }
 726
 727 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 728                 struct nvme_command *cmnd)
 729 {
 730         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 731         struct nvme_dsm_range *range;
 732         struct bio *bio;
 733
 734         /*
 735          * Some devices do not consider the DSM 'Number of Ranges' field when
 736          * determining how much data to DMA. Always allocate memory for maximum
 737          * number of segments to prevent device reading beyond end of buffer.
 738          */
 739         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 740
 741         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 742         if (!range) {
 743                 /*
 744                  * If we fail allocation our range, fallback to the controller
 745                  * discard page. If that's also busy, it's safe to return
 746                  * busy, as we know we can make progress once that's freed.
 747                  */
 748                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 749                         return BLK_STS_RESOURCE;
 750
 751                 range = page_address(ns->ctrl->discard_page);
 752         }
 753
 754         __rq_for_each_bio(bio, req) {
 755                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 756                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 757
 758                 if (n < segments) {
 759                         range[n].cattr = cpu_to_le32(0);
 760                         range[n].nlb = cpu_to_le32(nlb);
 761                         range[n].slba = cpu_to_le64(slba);
 762                 }
 763                 n++;
 764         }
 765
 766         if (WARN_ON_ONCE(n != segments)) {
 767                 if (virt_to_page(range) == ns->ctrl->discard_page)
 768                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 769                 else
 770                         kfree(range);
 771                 return BLK_STS_IOERR;
 772         }
 773
 774         memset(cmnd, 0, sizeof(*cmnd));
 775         cmnd->dsm.opcode = nvme_cmd_dsm;
 776         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 777         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 778         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 779
 780         req->special_vec.bv_page = virt_to_page(range);
 781         req->special_vec.bv_offset = offset_in_page(range);
 782         req->special_vec.bv_len = alloc_size;
 783         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 784
 785         return BLK_STS_OK;
 786 }
 787
 788 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
 789                               struct request *req)
 790 {
 791         u32 upper, lower;
 792         u64 ref48;
 793
 794         /* both rw and write zeroes share the same reftag format */
 795         switch (ns->guard_type) {
 796         case NVME_NVM_NS_16B_GUARD:
 797                 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 798                 break;
 799         case NVME_NVM_NS_64B_GUARD:
 800                 ref48 = ext_pi_ref_tag(req);
 801                 lower = lower_32_bits(ref48);
 802                 upper = upper_32_bits(ref48);
 803
 804                 cmnd->rw.reftag = cpu_to_le32(lower);
 805                 cmnd->rw.cdw3 = cpu_to_le32(upper);
 806                 break;
 807         default:
 808                 break;
 809         }
 810 }
 811
 812 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 813                 struct request *req, struct nvme_command *cmnd)
 814 {
 815         memset(cmnd, 0, sizeof(*cmnd));
 816
 817         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 818                 return nvme_setup_discard(ns, req, cmnd);
 819
 820         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 821         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 822         cmnd->write_zeroes.slba =
 823                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 824         cmnd->write_zeroes.length =
 825                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 826
 827         if (nvme_ns_has_pi(ns)) {
 828                 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
 829
 830                 switch (ns->pi_type) {
 831                 case NVME_NS_DPS_PI_TYPE1:
 832                 case NVME_NS_DPS_PI_TYPE2:
 833                         nvme_set_ref_tag(ns, cmnd, req);
 834                         break;
 835                 }
 836         }
 837
 838         return BLK_STS_OK;
 839 }
 840
 841 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 842                 struct request *req, struct nvme_command *cmnd,
 843                 enum nvme_opcode op)
 844 {
 845         u16 control = 0;
 846         u32 dsmgmt = 0;
 847
 848         if (req->cmd_flags & REQ_FUA)
 849                 control |= NVME_RW_FUA;
 850         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 851                 control |= NVME_RW_LR;
 852
 853         if (req->cmd_flags & REQ_RAHEAD)
 854                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 855
 856         cmnd->rw.opcode = op;
 857         cmnd->rw.flags = 0;
 858         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 859         cmnd->rw.cdw2 = 0;
 860         cmnd->rw.cdw3 = 0;
 861         cmnd->rw.metadata = 0;
 862         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 863         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 864         cmnd->rw.reftag = 0;
 865         cmnd->rw.apptag = 0;
 866         cmnd->rw.appmask = 0;
 867
 868         if (ns->ms) {
 869                 /*
 870                  * If formated with metadata, the block layer always provides a
 871                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 872                  * we enable the PRACT bit for protection information or set the
 873                  * namespace capacity to zero to prevent any I/O.
 874                  */
 875                 if (!blk_integrity_rq(req)) {
 876                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 877                                 return BLK_STS_NOTSUPP;
 878                         control |= NVME_RW_PRINFO_PRACT;
 879                 }
 880
 881                 switch (ns->pi_type) {
 882                 case NVME_NS_DPS_PI_TYPE3:
 883                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 884                         break;
 885                 case NVME_NS_DPS_PI_TYPE1:
 886                 case NVME_NS_DPS_PI_TYPE2:
 887                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 888                                         NVME_RW_PRINFO_PRCHK_REF;
 889                         if (op == nvme_cmd_zone_append)
 890                                 control |= NVME_RW_APPEND_PIREMAP;
 891                         nvme_set_ref_tag(ns, cmnd, req);
 892                         break;
 893                 }
 894         }
 895
 896         cmnd->rw.control = cpu_to_le16(control);
 897         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 898         return 0;
 899 }
 900
 901 void nvme_cleanup_cmd(struct request *req)
 902 {
 903         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 904                 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 905
 906                 if (req->special_vec.bv_page == ctrl->discard_page)
 907                         clear_bit_unlock(0, &ctrl->discard_page_busy);
 908                 else
 909                         kfree(bvec_virt(&req->special_vec));
 910         }
 911 }
 912 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 913
 914 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
 915 {
 916         struct nvme_command *cmd = nvme_req(req)->cmd;
 917         blk_status_t ret = BLK_STS_OK;
 918
 919         if (!(req->rq_flags & RQF_DONTPREP))
 920                 nvme_clear_nvme_request(req);
 921
 922         switch (req_op(req)) {
 923         case REQ_OP_DRV_IN:
 924         case REQ_OP_DRV_OUT:
 925                 /* these are setup prior to execution in nvme_init_request() */
 926                 break;
 927         case REQ_OP_FLUSH:
 928                 nvme_setup_flush(ns, cmd);
 929                 break;
 930         case REQ_OP_ZONE_RESET_ALL:
 931         case REQ_OP_ZONE_RESET:
 932                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 933                 break;
 934         case REQ_OP_ZONE_OPEN:
 935                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
 936                 break;
 937         case REQ_OP_ZONE_CLOSE:
 938                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
 939                 break;
 940         case REQ_OP_ZONE_FINISH:
 941                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
 942                 break;
 943         case REQ_OP_WRITE_ZEROES:
 944                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 945                 break;
 946         case REQ_OP_DISCARD:
 947                 ret = nvme_setup_discard(ns, req, cmd);
 948                 break;
 949         case REQ_OP_READ:
 950                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
 951                 break;
 952         case REQ_OP_WRITE:
 953                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
 954                 break;
 955         case REQ_OP_ZONE_APPEND:
 956                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
 957                 break;
 958         default:
 959                 WARN_ON_ONCE(1);
 960                 return BLK_STS_IOERR;
 961         }
 962
 963         cmd->common.command_id = nvme_cid(req);
 964         trace_nvme_setup_cmd(req, cmd);
 965         return ret;
 966 }
 967 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 968
 969 /*
 970  * Return values:
 971  * 0:  success
 972  * >0: nvme controller's cqe status response
 973  * <0: kernel error in lieu of controller response
 974  */
 975 static int nvme_execute_rq(struct request *rq, bool at_head)
 976 {
 977         blk_status_t status;
 978
 979         status = blk_execute_rq(rq, at_head);
 980         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
 981                 return -EINTR;
 982         if (nvme_req(rq)->status)
 983                 return nvme_req(rq)->status;
 984         return blk_status_to_errno(status);
 985 }
 986
 987 /*
 988  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 989  * if the result is positive, it's an NVM Express status code
 990  */
 991 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 992                 union nvme_result *result, void *buffer, unsigned bufflen,
 993                 unsigned timeout, int qid, int at_head,
 994                 blk_mq_req_flags_t flags)
 995 {
 996         struct request *req;
 997         int ret;
 998
 999         if (qid == NVME_QID_ANY)
1000                 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1001         else
1002                 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1003                                                 qid ? qid - 1 : 0);
1004
1005         if (IS_ERR(req))
1006                 return PTR_ERR(req);
1007         nvme_init_request(req, cmd);
1008
1009         if (timeout)
1010                 req->timeout = timeout;
1011
1012         if (buffer && bufflen) {
1013                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1014                 if (ret)
1015                         goto out;
1016         }
1017
1018         ret = nvme_execute_rq(req, at_head);
1019         if (result && ret >= 0)
1020                 *result = nvme_req(req)->result;
1021  out:
1022         blk_mq_free_request(req);
1023         return ret;
1024 }
1025 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1026
1027 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1028                 void *buffer, unsigned bufflen)
1029 {
1030         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1031                         NVME_QID_ANY, 0, 0);
1032 }
1033 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1034
1035 static u32 nvme_known_admin_effects(u8 opcode)
1036 {
1037         switch (opcode) {
1038         case nvme_admin_format_nvm:
1039                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1040                         NVME_CMD_EFFECTS_CSE_MASK;
1041         case nvme_admin_sanitize_nvm:
1042                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1043         default:
1044                 break;
1045         }
1046         return 0;
1047 }
1048
1049 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1050 {
1051         u32 effects = 0;
1052
1053         if (ns) {
1054                 if (ns->head->effects)
1055                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1056                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1057                         dev_warn_once(ctrl->device,
1058                                 "IO command:%02x has unhandled effects:%08x\n",
1059                                 opcode, effects);
1060                 return 0;
1061         }
1062
1063         if (ctrl->effects)
1064                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1065         effects |= nvme_known_admin_effects(opcode);
1066
1067         return effects;
1068 }
1069 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1070
1071 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1072                                u8 opcode)
1073 {
1074         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1075
1076         /*
1077          * For simplicity, IO to all namespaces is quiesced even if the command
1078          * effects say only one namespace is affected.
1079          */
1080         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1081                 mutex_lock(&ctrl->scan_lock);
1082                 mutex_lock(&ctrl->subsys->lock);
1083                 nvme_mpath_start_freeze(ctrl->subsys);
1084                 nvme_mpath_wait_freeze(ctrl->subsys);
1085                 nvme_start_freeze(ctrl);
1086                 nvme_wait_freeze(ctrl);
1087         }
1088         return effects;
1089 }
1090
1091 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1092                               struct nvme_command *cmd, int status)
1093 {
1094         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1095                 nvme_unfreeze(ctrl);
1096                 nvme_mpath_unfreeze(ctrl->subsys);
1097                 mutex_unlock(&ctrl->subsys->lock);
1098                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1099                 mutex_unlock(&ctrl->scan_lock);
1100         }
1101         if (effects & NVME_CMD_EFFECTS_CCC)
1102                 nvme_init_ctrl_finish(ctrl);
1103         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1104                 nvme_queue_scan(ctrl);
1105                 flush_work(&ctrl->scan_work);
1106         }
1107
1108         switch (cmd->common.opcode) {
1109         case nvme_admin_set_features:
1110                 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1111                 case NVME_FEAT_KATO:
1112                         /*
1113                          * Keep alive commands interval on the host should be
1114                          * updated when KATO is modified by Set Features
1115                          * commands.
1116                          */
1117                         if (!status)
1118                                 nvme_update_keep_alive(ctrl, cmd);
1119                         break;
1120                 default:
1121                         break;
1122                 }
1123                 break;
1124         default:
1125                 break;
1126         }
1127 }
1128
1129 int nvme_execute_passthru_rq(struct request *rq)
1130 {
1131         struct nvme_command *cmd = nvme_req(rq)->cmd;
1132         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1133         struct nvme_ns *ns = rq->q->queuedata;
1134         u32 effects;
1135         int  ret;
1136
1137         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1138         ret = nvme_execute_rq(rq, false);
1139         if (effects) /* nothing to be done for zero cmd effects */
1140                 nvme_passthru_end(ctrl, effects, cmd, ret);
1141
1142         return ret;
1143 }
1144 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1145
1146 /*
1147  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1148  *
1149  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1150  *   accounting for transport roundtrip times [..].
1151  */
1152 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1153 {
1154         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1155 }
1156
1157 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1158 {
1159         struct nvme_ctrl *ctrl = rq->end_io_data;
1160         unsigned long flags;
1161         bool startka = false;
1162
1163         blk_mq_free_request(rq);
1164
1165         if (status) {
1166                 dev_err(ctrl->device,
1167                         "failed nvme_keep_alive_end_io error=%d\n",
1168                                 status);
1169                 return;
1170         }
1171
1172         ctrl->comp_seen = false;
1173         spin_lock_irqsave(&ctrl->lock, flags);
1174         if (ctrl->state == NVME_CTRL_LIVE ||
1175             ctrl->state == NVME_CTRL_CONNECTING)
1176                 startka = true;
1177         spin_unlock_irqrestore(&ctrl->lock, flags);
1178         if (startka)
1179                 nvme_queue_keep_alive_work(ctrl);
1180 }
1181
1182 static void nvme_keep_alive_work(struct work_struct *work)
1183 {
1184         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1185                         struct nvme_ctrl, ka_work);
1186         bool comp_seen = ctrl->comp_seen;
1187         struct request *rq;
1188
1189         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1190                 dev_dbg(ctrl->device,
1191                         "reschedule traffic based keep-alive timer\n");
1192                 ctrl->comp_seen = false;
1193                 nvme_queue_keep_alive_work(ctrl);
1194                 return;
1195         }
1196
1197         rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1198                                   BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1199         if (IS_ERR(rq)) {
1200                 /* allocation failure, reset the controller */
1201                 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1202                 nvme_reset_ctrl(ctrl);
1203                 return;
1204         }
1205         nvme_init_request(rq, &ctrl->ka_cmd);
1206
1207         rq->timeout = ctrl->kato * HZ;
1208         rq->end_io_data = ctrl;
1209         blk_execute_rq_nowait(rq, false, nvme_keep_alive_end_io);
1210 }
1211
1212 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1213 {
1214         if (unlikely(ctrl->kato == 0))
1215                 return;
1216
1217         nvme_queue_keep_alive_work(ctrl);
1218 }
1219
1220 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1221 {
1222         if (unlikely(ctrl->kato == 0))
1223                 return;
1224
1225         cancel_delayed_work_sync(&ctrl->ka_work);
1226 }
1227 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1228
1229 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1230                                    struct nvme_command *cmd)
1231 {
1232         unsigned int new_kato =
1233                 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1234
1235         dev_info(ctrl->device,
1236                  "keep alive interval updated from %u ms to %u ms\n",
1237                  ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1238
1239         nvme_stop_keep_alive(ctrl);
1240         ctrl->kato = new_kato;
1241         nvme_start_keep_alive(ctrl);
1242 }
1243
1244 /*
1245  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1246  * flag, thus sending any new CNS opcodes has a big chance of not working.
1247  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1248  * (but not for any later version).
1249  */
1250 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1251 {
1252         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1253                 return ctrl->vs < NVME_VS(1, 2, 0);
1254         return ctrl->vs < NVME_VS(1, 1, 0);
1255 }
1256
1257 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1258 {
1259         struct nvme_command c = { };
1260         int error;
1261
1262         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1263         c.identify.opcode = nvme_admin_identify;
1264         c.identify.cns = NVME_ID_CNS_CTRL;
1265
1266         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1267         if (!*id)
1268                 return -ENOMEM;
1269
1270         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1271                         sizeof(struct nvme_id_ctrl));
1272         if (error)
1273                 kfree(*id);
1274         return error;
1275 }
1276
1277 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1278                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1279 {
1280         const char *warn_str = "ctrl returned bogus length:";
1281         void *data = cur;
1282
1283         switch (cur->nidt) {
1284         case NVME_NIDT_EUI64:
1285                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1286                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1287                                  warn_str, cur->nidl);
1288                         return -1;
1289                 }
1290                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1291                 return NVME_NIDT_EUI64_LEN;
1292         case NVME_NIDT_NGUID:
1293                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1294                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1295                                  warn_str, cur->nidl);
1296                         return -1;
1297                 }
1298                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1299                 return NVME_NIDT_NGUID_LEN;
1300         case NVME_NIDT_UUID:
1301                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1302                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1303                                  warn_str, cur->nidl);
1304                         return -1;
1305                 }
1306                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1307                 return NVME_NIDT_UUID_LEN;
1308         case NVME_NIDT_CSI:
1309                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1310                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1311                                  warn_str, cur->nidl);
1312                         return -1;
1313                 }
1314                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1315                 *csi_seen = true;
1316                 return NVME_NIDT_CSI_LEN;
1317         default:
1318                 /* Skip unknown types */
1319                 return cur->nidl;
1320         }
1321 }
1322
1323 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1324                 struct nvme_ns_ids *ids)
1325 {
1326         struct nvme_command c = { };
1327         bool csi_seen = false;
1328         int status, pos, len;
1329         void *data;
1330
1331         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1332                 return 0;
1333         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1334                 return 0;
1335
1336         c.identify.opcode = nvme_admin_identify;
1337         c.identify.nsid = cpu_to_le32(nsid);
1338         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1339
1340         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1341         if (!data)
1342                 return -ENOMEM;
1343
1344         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1345                                       NVME_IDENTIFY_DATA_SIZE);
1346         if (status) {
1347                 dev_warn(ctrl->device,
1348                         "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1349                         nsid, status);
1350                 goto free_data;
1351         }
1352
1353         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1354                 struct nvme_ns_id_desc *cur = data + pos;
1355
1356                 if (cur->nidl == 0)
1357                         break;
1358
1359                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1360                 if (len < 0)
1361                         break;
1362
1363                 len += sizeof(*cur);
1364         }
1365
1366         if (nvme_multi_css(ctrl) && !csi_seen) {
1367                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1368                          nsid);
1369                 status = -EINVAL;
1370         }
1371
1372 free_data:
1373         kfree(data);
1374         return status;
1375 }
1376
1377 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1378                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1379 {
1380         struct nvme_command c = { };
1381         int error;
1382
1383         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1384         c.identify.opcode = nvme_admin_identify;
1385         c.identify.nsid = cpu_to_le32(nsid);
1386         c.identify.cns = NVME_ID_CNS_NS;
1387
1388         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1389         if (!*id)
1390                 return -ENOMEM;
1391
1392         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1393         if (error) {
1394                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1395                 goto out_free_id;
1396         }
1397
1398         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1399         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1400                 goto out_free_id;
1401
1402         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1403             !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1404                 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1405         if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1406             !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1407                 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1408
1409         return 0;
1410
1411 out_free_id:
1412         kfree(*id);
1413         return error;
1414 }
1415
1416 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1417                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1418 {
1419         union nvme_result res = { 0 };
1420         struct nvme_command c = { };
1421         int ret;
1422
1423         c.features.opcode = op;
1424         c.features.fid = cpu_to_le32(fid);
1425         c.features.dword11 = cpu_to_le32(dword11);
1426
1427         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1428                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1429         if (ret >= 0 && result)
1430                 *result = le32_to_cpu(res.u32);
1431         return ret;
1432 }
1433
1434 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1435                       unsigned int dword11, void *buffer, size_t buflen,
1436                       u32 *result)
1437 {
1438         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1439                              buflen, result);
1440 }
1441 EXPORT_SYMBOL_GPL(nvme_set_features);
1442
1443 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1444                       unsigned int dword11, void *buffer, size_t buflen,
1445                       u32 *result)
1446 {
1447         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1448                              buflen, result);
1449 }
1450 EXPORT_SYMBOL_GPL(nvme_get_features);
1451
1452 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1453 {
1454         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1455         u32 result;
1456         int status, nr_io_queues;
1457
1458         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1459                         &result);
1460         if (status < 0)
1461                 return status;
1462
1463         /*
1464          * Degraded controllers might return an error when setting the queue
1465          * count.  We still want to be able to bring them online and offer
1466          * access to the admin queue, as that might be only way to fix them up.
1467          */
1468         if (status > 0) {
1469                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1470                 *count = 0;
1471         } else {
1472                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1473                 *count = min(*count, nr_io_queues);
1474         }
1475
1476         return 0;
1477 }
1478 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1479
1480 #define NVME_AEN_SUPPORTED \
1481         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1482          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1483
1484 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1485 {
1486         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1487         int status;
1488
1489         if (!supported_aens)
1490                 return;
1491
1492         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1493                         NULL, 0, &result);
1494         if (status)
1495                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1496                          supported_aens);
1497
1498         queue_work(nvme_wq, &ctrl->async_event_work);
1499 }
1500
1501 static int nvme_ns_open(struct nvme_ns *ns)
1502 {
1503
1504         /* should never be called due to GENHD_FL_HIDDEN */
1505         if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1506                 goto fail;
1507         if (!nvme_get_ns(ns))
1508                 goto fail;
1509         if (!try_module_get(ns->ctrl->ops->module))
1510                 goto fail_put_ns;
1511
1512         return 0;
1513
1514 fail_put_ns:
1515         nvme_put_ns(ns);
1516 fail:
1517         return -ENXIO;
1518 }
1519
1520 static void nvme_ns_release(struct nvme_ns *ns)
1521 {
1522
1523         module_put(ns->ctrl->ops->module);
1524         nvme_put_ns(ns);
1525 }
1526
1527 static int nvme_open(struct block_device *bdev, fmode_t mode)
1528 {
1529         return nvme_ns_open(bdev->bd_disk->private_data);
1530 }
1531
1532 static void nvme_release(struct gendisk *disk, fmode_t mode)
1533 {
1534         nvme_ns_release(disk->private_data);
1535 }
1536
1537 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1538 {
1539         /* some standard values */
1540         geo->heads = 1 << 6;
1541         geo->sectors = 1 << 5;
1542         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1543         return 0;
1544 }
1545
1546 #ifdef CONFIG_BLK_DEV_INTEGRITY
1547 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1548                                 u32 max_integrity_segments)
1549 {
1550         struct blk_integrity integrity = { };
1551
1552         switch (ns->pi_type) {
1553         case NVME_NS_DPS_PI_TYPE3:
1554                 switch (ns->guard_type) {
1555                 case NVME_NVM_NS_16B_GUARD:
1556                         integrity.profile = &t10_pi_type3_crc;
1557                         integrity.tag_size = sizeof(u16) + sizeof(u32);
1558                         integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1559                         break;
1560                 case NVME_NVM_NS_64B_GUARD:
1561                         integrity.profile = &ext_pi_type3_crc64;
1562                         integrity.tag_size = sizeof(u16) + 6;
1563                         integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1564                         break;
1565                 default:
1566                         integrity.profile = NULL;
1567                         break;
1568                 }
1569                 break;
1570         case NVME_NS_DPS_PI_TYPE1:
1571         case NVME_NS_DPS_PI_TYPE2:
1572                 switch (ns->guard_type) {
1573                 case NVME_NVM_NS_16B_GUARD:
1574                         integrity.profile = &t10_pi_type1_crc;
1575                         integrity.tag_size = sizeof(u16);
1576                         integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1577                         break;
1578                 case NVME_NVM_NS_64B_GUARD:
1579                         integrity.profile = &ext_pi_type1_crc64;
1580                         integrity.tag_size = sizeof(u16);
1581                         integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1582                         break;
1583                 default:
1584                         integrity.profile = NULL;
1585                         break;
1586                 }
1587                 break;
1588         default:
1589                 integrity.profile = NULL;
1590                 break;
1591         }
1592
1593         integrity.tuple_size = ns->ms;
1594         blk_integrity_register(disk, &integrity);
1595         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1596 }
1597 #else
1598 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1599                                 u32 max_integrity_segments)
1600 {
1601 }
1602 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1603
1604 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1605 {
1606         struct nvme_ctrl *ctrl = ns->ctrl;
1607         struct request_queue *queue = disk->queue;
1608         u32 size = queue_logical_block_size(queue);
1609
1610         if (ctrl->max_discard_sectors == 0) {
1611                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1612                 return;
1613         }
1614
1615         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1616                         NVME_DSM_MAX_RANGES);
1617
1618         queue->limits.discard_alignment = 0;
1619         queue->limits.discard_granularity = size;
1620
1621         /* If discard is already enabled, don't reset queue limits */
1622         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1623                 return;
1624
1625         blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1626         blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1627
1628         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1629                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1630 }
1631
1632 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1633 {
1634         return uuid_equal(&a->uuid, &b->uuid) &&
1635                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1636                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1637                 a->csi == b->csi;
1638 }
1639
1640 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1641 {
1642         bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1643         unsigned lbaf = nvme_lbaf_index(id->flbas);
1644         struct nvme_ctrl *ctrl = ns->ctrl;
1645         struct nvme_command c = { };
1646         struct nvme_id_ns_nvm *nvm;
1647         int ret = 0;
1648         u32 elbaf;
1649
1650         ns->pi_size = 0;
1651         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1652         if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1653                 ns->pi_size = sizeof(struct t10_pi_tuple);
1654                 ns->guard_type = NVME_NVM_NS_16B_GUARD;
1655                 goto set_pi;
1656         }
1657
1658         nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1659         if (!nvm)
1660                 return -ENOMEM;
1661
1662         c.identify.opcode = nvme_admin_identify;
1663         c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1664         c.identify.cns = NVME_ID_CNS_CS_NS;
1665         c.identify.csi = NVME_CSI_NVM;
1666
1667         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1668         if (ret)
1669                 goto free_data;
1670
1671         elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1672
1673         /* no support for storage tag formats right now */
1674         if (nvme_elbaf_sts(elbaf))
1675                 goto free_data;
1676
1677         ns->guard_type = nvme_elbaf_guard_type(elbaf);
1678         switch (ns->guard_type) {
1679         case NVME_NVM_NS_64B_GUARD:
1680                 ns->pi_size = sizeof(struct crc64_pi_tuple);
1681                 break;
1682         case NVME_NVM_NS_16B_GUARD:
1683                 ns->pi_size = sizeof(struct t10_pi_tuple);
1684                 break;
1685         default:
1686                 break;
1687         }
1688
1689 free_data:
1690         kfree(nvm);
1691 set_pi:
1692         if (ns->pi_size && (first || ns->ms == ns->pi_size))
1693                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1694         else
1695                 ns->pi_type = 0;
1696
1697         return ret;
1698 }
1699
1700 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1701 {
1702         struct nvme_ctrl *ctrl = ns->ctrl;
1703
1704         if (nvme_init_ms(ns, id))
1705                 return;
1706
1707         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1708         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1709                 return;
1710
1711         if (ctrl->ops->flags & NVME_F_FABRICS) {
1712                 /*
1713                  * The NVMe over Fabrics specification only supports metadata as
1714                  * part of the extended data LBA.  We rely on HCA/HBA support to
1715                  * remap the separate metadata buffer from the block layer.
1716                  */
1717                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1718                         return;
1719
1720                 ns->features |= NVME_NS_EXT_LBAS;
1721
1722                 /*
1723                  * The current fabrics transport drivers support namespace
1724                  * metadata formats only if nvme_ns_has_pi() returns true.
1725                  * Suppress support for all other formats so the namespace will
1726                  * have a 0 capacity and not be usable through the block stack.
1727                  *
1728                  * Note, this check will need to be modified if any drivers
1729                  * gain the ability to use other metadata formats.
1730                  */
1731                 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1732                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1733         } else {
1734                 /*
1735                  * For PCIe controllers, we can't easily remap the separate
1736                  * metadata buffer from the block layer and thus require a
1737                  * separate metadata buffer for block layer metadata/PI support.
1738                  * We allow extended LBAs for the passthrough interface, though.
1739                  */
1740                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1741                         ns->features |= NVME_NS_EXT_LBAS;
1742                 else
1743                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1744         }
1745 }
1746
1747 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1748                 struct request_queue *q)
1749 {
1750         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1751
1752         if (ctrl->max_hw_sectors) {
1753                 u32 max_segments =
1754                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1755
1756                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1757                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1758                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1759         }
1760         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1761         blk_queue_dma_alignment(q, 7);
1762         blk_queue_write_cache(q, vwc, vwc);
1763 }
1764
1765 static void nvme_update_disk_info(struct gendisk *disk,
1766                 struct nvme_ns *ns, struct nvme_id_ns *id)
1767 {
1768         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1769         unsigned short bs = 1 << ns->lba_shift;
1770         u32 atomic_bs, phys_bs, io_opt = 0;
1771
1772         /*
1773          * The block layer can't support LBA sizes larger than the page size
1774          * yet, so catch this early and don't allow block I/O.
1775          */
1776         if (ns->lba_shift > PAGE_SHIFT) {
1777                 capacity = 0;
1778                 bs = (1 << 9);
1779         }
1780
1781         blk_integrity_unregister(disk);
1782
1783         atomic_bs = phys_bs = bs;
1784         if (id->nabo == 0) {
1785                 /*
1786                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1787                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1788                  * 0 then AWUPF must be used instead.
1789                  */
1790                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1791                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1792                 else
1793                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1794         }
1795
1796         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1797                 /* NPWG = Namespace Preferred Write Granularity */
1798                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1799                 /* NOWS = Namespace Optimal Write Size */
1800                 io_opt = bs * (1 + le16_to_cpu(id->nows));
1801         }
1802
1803         blk_queue_logical_block_size(disk->queue, bs);
1804         /*
1805          * Linux filesystems assume writing a single physical block is
1806          * an atomic operation. Hence limit the physical block size to the
1807          * value of the Atomic Write Unit Power Fail parameter.
1808          */
1809         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1810         blk_queue_io_min(disk->queue, phys_bs);
1811         blk_queue_io_opt(disk->queue, io_opt);
1812
1813         /*
1814          * Register a metadata profile for PI, or the plain non-integrity NVMe
1815          * metadata masquerading as Type 0 if supported, otherwise reject block
1816          * I/O to namespaces with metadata except when the namespace supports
1817          * PI, as it can strip/insert in that case.
1818          */
1819         if (ns->ms) {
1820                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1821                     (ns->features & NVME_NS_METADATA_SUPPORTED))
1822                         nvme_init_integrity(disk, ns,
1823                                             ns->ctrl->max_integrity_segments);
1824                 else if (!nvme_ns_has_pi(ns))
1825                         capacity = 0;
1826         }
1827
1828         set_capacity_and_notify(disk, capacity);
1829
1830         nvme_config_discard(disk, ns);
1831         blk_queue_max_write_zeroes_sectors(disk->queue,
1832                                            ns->ctrl->max_zeroes_sectors);
1833
1834         set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1835                 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1836 }
1837
1838 static inline bool nvme_first_scan(struct gendisk *disk)
1839 {
1840         /* nvme_alloc_ns() scans the disk prior to adding it */
1841         return !disk_live(disk);
1842 }
1843
1844 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1845 {
1846         struct nvme_ctrl *ctrl = ns->ctrl;
1847         u32 iob;
1848
1849         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1850             is_power_of_2(ctrl->max_hw_sectors))
1851                 iob = ctrl->max_hw_sectors;
1852         else
1853                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1854
1855         if (!iob)
1856                 return;
1857
1858         if (!is_power_of_2(iob)) {
1859                 if (nvme_first_scan(ns->disk))
1860                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1861                                 ns->disk->disk_name, iob);
1862                 return;
1863         }
1864
1865         if (blk_queue_is_zoned(ns->disk->queue)) {
1866                 if (nvme_first_scan(ns->disk))
1867                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
1868                                 ns->disk->disk_name);
1869                 return;
1870         }
1871
1872         blk_queue_chunk_sectors(ns->queue, iob);
1873 }
1874
1875 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1876 {
1877         unsigned lbaf = nvme_lbaf_index(id->flbas);
1878         int ret;
1879
1880         blk_mq_freeze_queue(ns->disk->queue);
1881         ns->lba_shift = id->lbaf[lbaf].ds;
1882         nvme_set_queue_limits(ns->ctrl, ns->queue);
1883
1884         nvme_configure_metadata(ns, id);
1885         nvme_set_chunk_sectors(ns, id);
1886         nvme_update_disk_info(ns->disk, ns, id);
1887
1888         if (ns->head->ids.csi == NVME_CSI_ZNS) {
1889                 ret = nvme_update_zone_info(ns, lbaf);
1890                 if (ret)
1891                         goto out_unfreeze;
1892         }
1893
1894         set_bit(NVME_NS_READY, &ns->flags);
1895         blk_mq_unfreeze_queue(ns->disk->queue);
1896
1897         if (blk_queue_is_zoned(ns->queue)) {
1898                 ret = nvme_revalidate_zones(ns);
1899                 if (ret && !nvme_first_scan(ns->disk))
1900                         return ret;
1901         }
1902
1903         if (nvme_ns_head_multipath(ns->head)) {
1904                 blk_mq_freeze_queue(ns->head->disk->queue);
1905                 nvme_update_disk_info(ns->head->disk, ns, id);
1906                 nvme_mpath_revalidate_paths(ns);
1907                 blk_stack_limits(&ns->head->disk->queue->limits,
1908                                  &ns->queue->limits, 0);
1909                 disk_update_readahead(ns->head->disk);
1910                 blk_mq_unfreeze_queue(ns->head->disk->queue);
1911         }
1912         return 0;
1913
1914 out_unfreeze:
1915         /*
1916          * If probing fails due an unsupported feature, hide the block device,
1917          * but still allow other access.
1918          */
1919         if (ret == -ENODEV) {
1920                 ns->disk->flags |= GENHD_FL_HIDDEN;
1921                 set_bit(NVME_NS_READY, &ns->flags);
1922                 ret = 0;
1923         }
1924         blk_mq_unfreeze_queue(ns->disk->queue);
1925         return ret;
1926 }
1927
1928 static char nvme_pr_type(enum pr_type type)
1929 {
1930         switch (type) {
1931         case PR_WRITE_EXCLUSIVE:
1932                 return 1;
1933         case PR_EXCLUSIVE_ACCESS:
1934                 return 2;
1935         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1936                 return 3;
1937         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1938                 return 4;
1939         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1940                 return 5;
1941         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1942                 return 6;
1943         default:
1944                 return 0;
1945         }
1946 }
1947
1948 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1949                 struct nvme_command *c, u8 data[16])
1950 {
1951         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1952         int srcu_idx = srcu_read_lock(&head->srcu);
1953         struct nvme_ns *ns = nvme_find_path(head);
1954         int ret = -EWOULDBLOCK;
1955
1956         if (ns) {
1957                 c->common.nsid = cpu_to_le32(ns->head->ns_id);
1958                 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1959         }
1960         srcu_read_unlock(&head->srcu, srcu_idx);
1961         return ret;
1962 }
1963
1964 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
1965                 u8 data[16])
1966 {
1967         c->common.nsid = cpu_to_le32(ns->head->ns_id);
1968         return nvme_submit_sync_cmd(ns->queue, c, data, 16);
1969 }
1970
1971 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1972                                 u64 key, u64 sa_key, u8 op)
1973 {
1974         struct nvme_command c = { };
1975         u8 data[16] = { 0, };
1976
1977         put_unaligned_le64(key, &data[0]);
1978         put_unaligned_le64(sa_key, &data[8]);
1979
1980         c.common.opcode = op;
1981         c.common.cdw10 = cpu_to_le32(cdw10);
1982
1983         if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
1984             bdev->bd_disk->fops == &nvme_ns_head_ops)
1985                 return nvme_send_ns_head_pr_command(bdev, &c, data);
1986         return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
1987 }
1988
1989 static int nvme_pr_register(struct block_device *bdev, u64 old,
1990                 u64 new, unsigned flags)
1991 {
1992         u32 cdw10;
1993
1994         if (flags & ~PR_FL_IGNORE_KEY)
1995                 return -EOPNOTSUPP;
1996
1997         cdw10 = old ? 2 : 0;
1998         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1999         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2000         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2001 }
2002
2003 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2004                 enum pr_type type, unsigned flags)
2005 {
2006         u32 cdw10;
2007
2008         if (flags & ~PR_FL_IGNORE_KEY)
2009                 return -EOPNOTSUPP;
2010
2011         cdw10 = nvme_pr_type(type) << 8;
2012         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2013         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2014 }
2015
2016 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2017                 enum pr_type type, bool abort)
2018 {
2019         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2020
2021         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2022 }
2023
2024 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2025 {
2026         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2027
2028         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2029 }
2030
2031 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2032 {
2033         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2034
2035         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2036 }
2037
2038 const struct pr_ops nvme_pr_ops = {
2039         .pr_register    = nvme_pr_register,
2040         .pr_reserve     = nvme_pr_reserve,
2041         .pr_release     = nvme_pr_release,
2042         .pr_preempt     = nvme_pr_preempt,
2043         .pr_clear       = nvme_pr_clear,
2044 };
2045
2046 #ifdef CONFIG_BLK_SED_OPAL
2047 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2048                 bool send)
2049 {
2050         struct nvme_ctrl *ctrl = data;
2051         struct nvme_command cmd = { };
2052
2053         if (send)
2054                 cmd.common.opcode = nvme_admin_security_send;
2055         else
2056                 cmd.common.opcode = nvme_admin_security_recv;
2057         cmd.common.nsid = 0;
2058         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2059         cmd.common.cdw11 = cpu_to_le32(len);
2060
2061         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2062                         NVME_QID_ANY, 1, 0);
2063 }
2064 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2065 #endif /* CONFIG_BLK_SED_OPAL */
2066
2067 #ifdef CONFIG_BLK_DEV_ZONED
2068 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2069                 unsigned int nr_zones, report_zones_cb cb, void *data)
2070 {
2071         return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2072                         data);
2073 }
2074 #else
2075 #define nvme_report_zones       NULL
2076 #endif /* CONFIG_BLK_DEV_ZONED */
2077
2078 static const struct block_device_operations nvme_bdev_ops = {
2079         .owner          = THIS_MODULE,
2080         .ioctl          = nvme_ioctl,
2081         .open           = nvme_open,
2082         .release        = nvme_release,
2083         .getgeo         = nvme_getgeo,
2084         .report_zones   = nvme_report_zones,
2085         .pr_ops         = &nvme_pr_ops,
2086 };
2087
2088 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2089 {
2090         unsigned long timeout =
2091                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2092         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2093         int ret;
2094
2095         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2096                 if (csts == ~0)
2097                         return -ENODEV;
2098                 if ((csts & NVME_CSTS_RDY) == bit)
2099                         break;
2100
2101                 usleep_range(1000, 2000);
2102                 if (fatal_signal_pending(current))
2103                         return -EINTR;
2104                 if (time_after(jiffies, timeout)) {
2105                         dev_err(ctrl->device,
2106                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2107                                 enabled ? "initialisation" : "reset", csts);
2108                         return -ENODEV;
2109                 }
2110         }
2111
2112         return ret;
2113 }
2114
2115 /*
2116  * If the device has been passed off to us in an enabled state, just clear
2117  * the enabled bit.  The spec says we should set the 'shutdown notification
2118  * bits', but doing so may cause the device to complete commands to the
2119  * admin queue ... and we don't know what memory that might be pointing at!
2120  */
2121 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2122 {
2123         int ret;
2124
2125         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2126         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2127
2128         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2129         if (ret)
2130                 return ret;
2131
2132         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2133                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2134
2135         return nvme_wait_ready(ctrl, ctrl->cap, false);
2136 }
2137 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2138
2139 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2140 {
2141         unsigned dev_page_min;
2142         int ret;
2143
2144         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2145         if (ret) {
2146                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2147                 return ret;
2148         }
2149         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2150
2151         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2152                 dev_err(ctrl->device,
2153                         "Minimum device page size %u too large for host (%u)\n",
2154                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2155                 return -ENODEV;
2156         }
2157
2158         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2159                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2160         else
2161                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2162         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2163         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2164         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
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         return nvme_wait_ready(ctrl, ctrl->cap, true);
2171 }
2172 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2173
2174 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2175 {
2176         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2177         u32 csts;
2178         int ret;
2179
2180         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2181         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2182
2183         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2184         if (ret)
2185                 return ret;
2186
2187         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2188                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2189                         break;
2190
2191                 msleep(100);
2192                 if (fatal_signal_pending(current))
2193                         return -EINTR;
2194                 if (time_after(jiffies, timeout)) {
2195                         dev_err(ctrl->device,
2196                                 "Device shutdown incomplete; abort shutdown\n");
2197                         return -ENODEV;
2198                 }
2199         }
2200
2201         return ret;
2202 }
2203 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2204
2205 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2206 {
2207         __le64 ts;
2208         int ret;
2209
2210         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2211                 return 0;
2212
2213         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2214         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2215                         NULL);
2216         if (ret)
2217                 dev_warn_once(ctrl->device,
2218                         "could not set timestamp (%d)\n", ret);
2219         return ret;
2220 }
2221
2222 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2223 {
2224         struct nvme_feat_host_behavior *host;
2225         u8 acre = 0, lbafee = 0;
2226         int ret;
2227
2228         /* Don't bother enabling the feature if retry delay is not reported */
2229         if (ctrl->crdt[0])
2230                 acre = NVME_ENABLE_ACRE;
2231         if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2232                 lbafee = NVME_ENABLE_LBAFEE;
2233
2234         if (!acre && !lbafee)
2235                 return 0;
2236
2237         host = kzalloc(sizeof(*host), GFP_KERNEL);
2238         if (!host)
2239                 return 0;
2240
2241         host->acre = acre;
2242         host->lbafee = lbafee;
2243         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2244                                 host, sizeof(*host), NULL);
2245         kfree(host);
2246         return ret;
2247 }
2248
2249 /*
2250  * The function checks whether the given total (exlat + enlat) latency of
2251  * a power state allows the latter to be used as an APST transition target.
2252  * It does so by comparing the latency to the primary and secondary latency
2253  * tolerances defined by module params. If there's a match, the corresponding
2254  * timeout value is returned and the matching tolerance index (1 or 2) is
2255  * reported.
2256  */
2257 static bool nvme_apst_get_transition_time(u64 total_latency,
2258                 u64 *transition_time, unsigned *last_index)
2259 {
2260         if (total_latency <= apst_primary_latency_tol_us) {
2261                 if (*last_index == 1)
2262                         return false;
2263                 *last_index = 1;
2264                 *transition_time = apst_primary_timeout_ms;
2265                 return true;
2266         }
2267         if (apst_secondary_timeout_ms &&
2268                 total_latency <= apst_secondary_latency_tol_us) {
2269                 if (*last_index <= 2)
2270                         return false;
2271                 *last_index = 2;
2272                 *transition_time = apst_secondary_timeout_ms;
2273                 return true;
2274         }
2275         return false;
2276 }
2277
2278 /*
2279  * APST (Autonomous Power State Transition) lets us program a table of power
2280  * state transitions that the controller will perform automatically.
2281  *
2282  * Depending on module params, one of the two supported techniques will be used:
2283  *
2284  * - If the parameters provide explicit timeouts and tolerances, they will be
2285  *   used to build a table with up to 2 non-operational states to transition to.
2286  *   The default parameter values were selected based on the values used by
2287  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2288  *   regeneration of the APST table in the event of switching between external
2289  *   and battery power, the timeouts and tolerances reflect a compromise
2290  *   between values used by Microsoft for AC and battery scenarios.
2291  * - If not, we'll configure the table with a simple heuristic: we are willing
2292  *   to spend at most 2% of the time transitioning between power states.
2293  *   Therefore, when running in any given state, we will enter the next
2294  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2295  *   microseconds, as long as that state's exit latency is under the requested
2296  *   maximum latency.
2297  *
2298  * We will not autonomously enter any non-operational state for which the total
2299  * latency exceeds ps_max_latency_us.
2300  *
2301  * Users can set ps_max_latency_us to zero to turn off APST.
2302  */
2303 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2304 {
2305         struct nvme_feat_auto_pst *table;
2306         unsigned apste = 0;
2307         u64 max_lat_us = 0;
2308         __le64 target = 0;
2309         int max_ps = -1;
2310         int state;
2311         int ret;
2312         unsigned last_lt_index = UINT_MAX;
2313
2314         /*
2315          * If APST isn't supported or if we haven't been initialized yet,
2316          * then don't do anything.
2317          */
2318         if (!ctrl->apsta)
2319                 return 0;
2320
2321         if (ctrl->npss > 31) {
2322                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2323                 return 0;
2324         }
2325
2326         table = kzalloc(sizeof(*table), GFP_KERNEL);
2327         if (!table)
2328                 return 0;
2329
2330         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2331                 /* Turn off APST. */
2332                 dev_dbg(ctrl->device, "APST disabled\n");
2333                 goto done;
2334         }
2335
2336         /*
2337          * Walk through all states from lowest- to highest-power.
2338          * According to the spec, lower-numbered states use more power.  NPSS,
2339          * despite the name, is the index of the lowest-power state, not the
2340          * number of states.
2341          */
2342         for (state = (int)ctrl->npss; state >= 0; state--) {
2343                 u64 total_latency_us, exit_latency_us, transition_ms;
2344
2345                 if (target)
2346                         table->entries[state] = target;
2347
2348                 /*
2349                  * Don't allow transitions to the deepest state if it's quirked
2350                  * off.
2351                  */
2352                 if (state == ctrl->npss &&
2353                     (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2354                         continue;
2355
2356                 /*
2357                  * Is this state a useful non-operational state for higher-power
2358                  * states to autonomously transition to?
2359                  */
2360                 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2361                         continue;
2362
2363                 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2364                 if (exit_latency_us > ctrl->ps_max_latency_us)
2365                         continue;
2366
2367                 total_latency_us = exit_latency_us +
2368                         le32_to_cpu(ctrl->psd[state].entry_lat);
2369
2370                 /*
2371                  * This state is good. It can be used as the APST idle target
2372                  * for higher power states.
2373                  */
2374                 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2375                         if (!nvme_apst_get_transition_time(total_latency_us,
2376                                         &transition_ms, &last_lt_index))
2377                                 continue;
2378                 } else {
2379                         transition_ms = total_latency_us + 19;
2380                         do_div(transition_ms, 20);
2381                         if (transition_ms > (1 << 24) - 1)
2382                                 transition_ms = (1 << 24) - 1;
2383                 }
2384
2385                 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2386                 if (max_ps == -1)
2387                         max_ps = state;
2388                 if (total_latency_us > max_lat_us)
2389                         max_lat_us = total_latency_us;
2390         }
2391
2392         if (max_ps == -1)
2393                 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2394         else
2395                 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2396                         max_ps, max_lat_us, (int)sizeof(*table), table);
2397         apste = 1;
2398
2399 done:
2400         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2401                                 table, sizeof(*table), NULL);
2402         if (ret)
2403                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2404         kfree(table);
2405         return ret;
2406 }
2407
2408 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2409 {
2410         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2411         u64 latency;
2412
2413         switch (val) {
2414         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2415         case PM_QOS_LATENCY_ANY:
2416                 latency = U64_MAX;
2417                 break;
2418
2419         default:
2420                 latency = val;
2421         }
2422
2423         if (ctrl->ps_max_latency_us != latency) {
2424                 ctrl->ps_max_latency_us = latency;
2425                 if (ctrl->state == NVME_CTRL_LIVE)
2426                         nvme_configure_apst(ctrl);
2427         }
2428 }
2429
2430 struct nvme_core_quirk_entry {
2431         /*
2432          * NVMe model and firmware strings are padded with spaces.  For
2433          * simplicity, strings in the quirk table are padded with NULLs
2434          * instead.
2435          */
2436         u16 vid;
2437         const char *mn;
2438         const char *fr;
2439         unsigned long quirks;
2440 };
2441
2442 static const struct nvme_core_quirk_entry core_quirks[] = {
2443         {
2444                 /*
2445                  * This Toshiba device seems to die using any APST states.  See:
2446                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2447                  */
2448                 .vid = 0x1179,
2449                 .mn = "THNSF5256GPUK TOSHIBA",
2450                 .quirks = NVME_QUIRK_NO_APST,
2451         },
2452         {
2453                 /*
2454                  * This LiteON CL1-3D*-Q11 firmware version has a race
2455                  * condition associated with actions related to suspend to idle
2456                  * LiteON has resolved the problem in future firmware
2457                  */
2458                 .vid = 0x14a4,
2459                 .fr = "22301111",
2460                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2461         },
2462         {
2463                 /*
2464                  * This Kioxia CD6-V Series / HPE PE8030 device times out and
2465                  * aborts I/O during any load, but more easily reproducible
2466                  * with discards (fstrim).
2467                  *
2468                  * The device is left in a state where it is also not possible
2469                  * to use "nvme set-feature" to disable APST, but booting with
2470                  * nvme_core.default_ps_max_latency=0 works.
2471                  */
2472                 .vid = 0x1e0f,
2473                 .mn = "KCD6XVUL6T40",
2474                 .quirks = NVME_QUIRK_NO_APST,
2475         }
2476 };
2477
2478 /* match is null-terminated but idstr is space-padded. */
2479 static bool string_matches(const char *idstr, const char *match, size_t len)
2480 {
2481         size_t matchlen;
2482
2483         if (!match)
2484                 return true;
2485
2486         matchlen = strlen(match);
2487         WARN_ON_ONCE(matchlen > len);
2488
2489         if (memcmp(idstr, match, matchlen))
2490                 return false;
2491
2492         for (; matchlen < len; matchlen++)
2493                 if (idstr[matchlen] != ' ')
2494                         return false;
2495
2496         return true;
2497 }
2498
2499 static bool quirk_matches(const struct nvme_id_ctrl *id,
2500                           const struct nvme_core_quirk_entry *q)
2501 {
2502         return q->vid == le16_to_cpu(id->vid) &&
2503                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2504                 string_matches(id->fr, q->fr, sizeof(id->fr));
2505 }
2506
2507 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2508                 struct nvme_id_ctrl *id)
2509 {
2510         size_t nqnlen;
2511         int off;
2512
2513         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2514                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2515                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2516                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2517                         return;
2518                 }
2519
2520                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2521                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2522         }
2523
2524         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2525         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2526                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2527                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2528         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2529         off += sizeof(id->sn);
2530         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2531         off += sizeof(id->mn);
2532         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2533 }
2534
2535 static void nvme_release_subsystem(struct device *dev)
2536 {
2537         struct nvme_subsystem *subsys =
2538                 container_of(dev, struct nvme_subsystem, dev);
2539
2540         if (subsys->instance >= 0)
2541                 ida_free(&nvme_instance_ida, subsys->instance);
2542         kfree(subsys);
2543 }
2544
2545 static void nvme_destroy_subsystem(struct kref *ref)
2546 {
2547         struct nvme_subsystem *subsys =
2548                         container_of(ref, struct nvme_subsystem, ref);
2549
2550         mutex_lock(&nvme_subsystems_lock);
2551         list_del(&subsys->entry);
2552         mutex_unlock(&nvme_subsystems_lock);
2553
2554         ida_destroy(&subsys->ns_ida);
2555         device_del(&subsys->dev);
2556         put_device(&subsys->dev);
2557 }
2558
2559 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2560 {
2561         kref_put(&subsys->ref, nvme_destroy_subsystem);
2562 }
2563
2564 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2565 {
2566         struct nvme_subsystem *subsys;
2567
2568         lockdep_assert_held(&nvme_subsystems_lock);
2569
2570         /*
2571          * Fail matches for discovery subsystems. This results
2572          * in each discovery controller bound to a unique subsystem.
2573          * This avoids issues with validating controller values
2574          * that can only be true when there is a single unique subsystem.
2575          * There may be multiple and completely independent entities
2576          * that provide discovery controllers.
2577          */
2578         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2579                 return NULL;
2580
2581         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2582                 if (strcmp(subsys->subnqn, subsysnqn))
2583                         continue;
2584                 if (!kref_get_unless_zero(&subsys->ref))
2585                         continue;
2586                 return subsys;
2587         }
2588
2589         return NULL;
2590 }
2591
2592 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2593         struct device_attribute subsys_attr_##_name = \
2594                 __ATTR(_name, _mode, _show, NULL)
2595
2596 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2597                                     struct device_attribute *attr,
2598                                     char *buf)
2599 {
2600         struct nvme_subsystem *subsys =
2601                 container_of(dev, struct nvme_subsystem, dev);
2602
2603         return sysfs_emit(buf, "%s\n", subsys->subnqn);
2604 }
2605 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2606
2607 static ssize_t nvme_subsys_show_type(struct device *dev,
2608                                     struct device_attribute *attr,
2609                                     char *buf)
2610 {
2611         struct nvme_subsystem *subsys =
2612                 container_of(dev, struct nvme_subsystem, dev);
2613
2614         switch (subsys->subtype) {
2615         case NVME_NQN_DISC:
2616                 return sysfs_emit(buf, "discovery\n");
2617         case NVME_NQN_NVME:
2618                 return sysfs_emit(buf, "nvm\n");
2619         default:
2620                 return sysfs_emit(buf, "reserved\n");
2621         }
2622 }
2623 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2624
2625 #define nvme_subsys_show_str_function(field)                            \
2626 static ssize_t subsys_##field##_show(struct device *dev,                \
2627                             struct device_attribute *attr, char *buf)   \
2628 {                                                                       \
2629         struct nvme_subsystem *subsys =                                 \
2630                 container_of(dev, struct nvme_subsystem, dev);          \
2631         return sysfs_emit(buf, "%.*s\n",                                \
2632                            (int)sizeof(subsys->field), subsys->field);  \
2633 }                                                                       \
2634 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2635
2636 nvme_subsys_show_str_function(model);
2637 nvme_subsys_show_str_function(serial);
2638 nvme_subsys_show_str_function(firmware_rev);
2639
2640 static struct attribute *nvme_subsys_attrs[] = {
2641         &subsys_attr_model.attr,
2642         &subsys_attr_serial.attr,
2643         &subsys_attr_firmware_rev.attr,
2644         &subsys_attr_subsysnqn.attr,
2645         &subsys_attr_subsystype.attr,
2646 #ifdef CONFIG_NVME_MULTIPATH
2647         &subsys_attr_iopolicy.attr,
2648 #endif
2649         NULL,
2650 };
2651
2652 static const struct attribute_group nvme_subsys_attrs_group = {
2653         .attrs = nvme_subsys_attrs,
2654 };
2655
2656 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2657         &nvme_subsys_attrs_group,
2658         NULL,
2659 };
2660
2661 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2662 {
2663         return ctrl->opts && ctrl->opts->discovery_nqn;
2664 }
2665
2666 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2667                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2668 {
2669         struct nvme_ctrl *tmp;
2670
2671         lockdep_assert_held(&nvme_subsystems_lock);
2672
2673         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2674                 if (nvme_state_terminal(tmp))
2675                         continue;
2676
2677                 if (tmp->cntlid == ctrl->cntlid) {
2678                         dev_err(ctrl->device,
2679                                 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2680                                 ctrl->cntlid, dev_name(tmp->device),
2681                                 subsys->subnqn);
2682                         return false;
2683                 }
2684
2685                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2686                     nvme_discovery_ctrl(ctrl))
2687                         continue;
2688
2689                 dev_err(ctrl->device,
2690                         "Subsystem does not support multiple controllers\n");
2691                 return false;
2692         }
2693
2694         return true;
2695 }
2696
2697 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2698 {
2699         struct nvme_subsystem *subsys, *found;
2700         int ret;
2701
2702         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2703         if (!subsys)
2704                 return -ENOMEM;
2705
2706         subsys->instance = -1;
2707         mutex_init(&subsys->lock);
2708         kref_init(&subsys->ref);
2709         INIT_LIST_HEAD(&subsys->ctrls);
2710         INIT_LIST_HEAD(&subsys->nsheads);
2711         nvme_init_subnqn(subsys, ctrl, id);
2712         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2713         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2714         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2715         subsys->vendor_id = le16_to_cpu(id->vid);
2716         subsys->cmic = id->cmic;
2717
2718         /* Versions prior to 1.4 don't necessarily report a valid type */
2719         if (id->cntrltype == NVME_CTRL_DISC ||
2720             !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2721                 subsys->subtype = NVME_NQN_DISC;
2722         else
2723                 subsys->subtype = NVME_NQN_NVME;
2724
2725         if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2726                 dev_err(ctrl->device,
2727                         "Subsystem %s is not a discovery controller",
2728                         subsys->subnqn);
2729                 kfree(subsys);
2730                 return -EINVAL;
2731         }
2732         subsys->awupf = le16_to_cpu(id->awupf);
2733         nvme_mpath_default_iopolicy(subsys);
2734
2735         subsys->dev.class = nvme_subsys_class;
2736         subsys->dev.release = nvme_release_subsystem;
2737         subsys->dev.groups = nvme_subsys_attrs_groups;
2738         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2739         device_initialize(&subsys->dev);
2740
2741         mutex_lock(&nvme_subsystems_lock);
2742         found = __nvme_find_get_subsystem(subsys->subnqn);
2743         if (found) {
2744                 put_device(&subsys->dev);
2745                 subsys = found;
2746
2747                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2748                         ret = -EINVAL;
2749                         goto out_put_subsystem;
2750                 }
2751         } else {
2752                 ret = device_add(&subsys->dev);
2753                 if (ret) {
2754                         dev_err(ctrl->device,
2755                                 "failed to register subsystem device.\n");
2756                         put_device(&subsys->dev);
2757                         goto out_unlock;
2758                 }
2759                 ida_init(&subsys->ns_ida);
2760                 list_add_tail(&subsys->entry, &nvme_subsystems);
2761         }
2762
2763         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2764                                 dev_name(ctrl->device));
2765         if (ret) {
2766                 dev_err(ctrl->device,
2767                         "failed to create sysfs link from subsystem.\n");
2768                 goto out_put_subsystem;
2769         }
2770
2771         if (!found)
2772                 subsys->instance = ctrl->instance;
2773         ctrl->subsys = subsys;
2774         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2775         mutex_unlock(&nvme_subsystems_lock);
2776         return 0;
2777
2778 out_put_subsystem:
2779         nvme_put_subsystem(subsys);
2780 out_unlock:
2781         mutex_unlock(&nvme_subsystems_lock);
2782         return ret;
2783 }
2784
2785 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2786                 void *log, size_t size, u64 offset)
2787 {
2788         struct nvme_command c = { };
2789         u32 dwlen = nvme_bytes_to_numd(size);
2790
2791         c.get_log_page.opcode = nvme_admin_get_log_page;
2792         c.get_log_page.nsid = cpu_to_le32(nsid);
2793         c.get_log_page.lid = log_page;
2794         c.get_log_page.lsp = lsp;
2795         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2796         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2797         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2798         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2799         c.get_log_page.csi = csi;
2800
2801         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2802 }
2803
2804 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2805                                 struct nvme_effects_log **log)
2806 {
2807         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2808         int ret;
2809
2810         if (cel)
2811                 goto out;
2812
2813         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2814         if (!cel)
2815                 return -ENOMEM;
2816
2817         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2818                         cel, sizeof(*cel), 0);
2819         if (ret) {
2820                 kfree(cel);
2821                 return ret;
2822         }
2823
2824         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2825 out:
2826         *log = cel;
2827         return 0;
2828 }
2829
2830 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2831 {
2832         u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2833
2834         if (check_shl_overflow(1U, units + page_shift - 9, &val))
2835                 return UINT_MAX;
2836         return val;
2837 }
2838
2839 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2840 {
2841         struct nvme_command c = { };
2842         struct nvme_id_ctrl_nvm *id;
2843         int ret;
2844
2845         if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2846                 ctrl->max_discard_sectors = UINT_MAX;
2847                 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2848         } else {
2849                 ctrl->max_discard_sectors = 0;
2850                 ctrl->max_discard_segments = 0;
2851         }
2852
2853         /*
2854          * Even though NVMe spec explicitly states that MDTS is not applicable
2855          * to the write-zeroes, we are cautious and limit the size to the
2856          * controllers max_hw_sectors value, which is based on the MDTS field
2857          * and possibly other limiting factors.
2858          */
2859         if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2860             !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2861                 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2862         else
2863                 ctrl->max_zeroes_sectors = 0;
2864
2865         if (nvme_ctrl_limited_cns(ctrl))
2866                 return 0;
2867
2868         id = kzalloc(sizeof(*id), GFP_KERNEL);
2869         if (!id)
2870                 return 0;
2871
2872         c.identify.opcode = nvme_admin_identify;
2873         c.identify.cns = NVME_ID_CNS_CS_CTRL;
2874         c.identify.csi = NVME_CSI_NVM;
2875
2876         ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2877         if (ret)
2878                 goto free_data;
2879
2880         if (id->dmrl)
2881                 ctrl->max_discard_segments = id->dmrl;
2882         if (id->dmrsl)
2883                 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2884         if (id->wzsl)
2885                 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2886
2887 free_data:
2888         kfree(id);
2889         return ret;
2890 }
2891
2892 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2893 {
2894         struct nvme_id_ctrl *id;
2895         u32 max_hw_sectors;
2896         bool prev_apst_enabled;
2897         int ret;
2898
2899         ret = nvme_identify_ctrl(ctrl, &id);
2900         if (ret) {
2901                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2902                 return -EIO;
2903         }
2904
2905         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2906                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2907                 if (ret < 0)
2908                         goto out_free;
2909         }
2910
2911         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2912                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2913
2914         if (!ctrl->identified) {
2915                 unsigned int i;
2916
2917                 ret = nvme_init_subsystem(ctrl, id);
2918                 if (ret)
2919                         goto out_free;
2920
2921                 /*
2922                  * Check for quirks.  Quirk can depend on firmware version,
2923                  * so, in principle, the set of quirks present can change
2924                  * across a reset.  As a possible future enhancement, we
2925                  * could re-scan for quirks every time we reinitialize
2926                  * the device, but we'd have to make sure that the driver
2927                  * behaves intelligently if the quirks change.
2928                  */
2929                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2930                         if (quirk_matches(id, &core_quirks[i]))
2931                                 ctrl->quirks |= core_quirks[i].quirks;
2932                 }
2933         }
2934
2935         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2936                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2937                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2938         }
2939
2940         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2941         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2942         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2943
2944         ctrl->oacs = le16_to_cpu(id->oacs);
2945         ctrl->oncs = le16_to_cpu(id->oncs);
2946         ctrl->mtfa = le16_to_cpu(id->mtfa);
2947         ctrl->oaes = le32_to_cpu(id->oaes);
2948         ctrl->wctemp = le16_to_cpu(id->wctemp);
2949         ctrl->cctemp = le16_to_cpu(id->cctemp);
2950
2951         atomic_set(&ctrl->abort_limit, id->acl + 1);
2952         ctrl->vwc = id->vwc;
2953         if (id->mdts)
2954                 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2955         else
2956                 max_hw_sectors = UINT_MAX;
2957         ctrl->max_hw_sectors =
2958                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2959
2960         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2961         ctrl->sgls = le32_to_cpu(id->sgls);
2962         ctrl->kas = le16_to_cpu(id->kas);
2963         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2964         ctrl->ctratt = le32_to_cpu(id->ctratt);
2965
2966         ctrl->cntrltype = id->cntrltype;
2967         ctrl->dctype = id->dctype;
2968
2969         if (id->rtd3e) {
2970                 /* us -> s */
2971                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2972
2973                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2974                                                  shutdown_timeout, 60);
2975
2976                 if (ctrl->shutdown_timeout != shutdown_timeout)
2977                         dev_info(ctrl->device,
2978                                  "Shutdown timeout set to %u seconds\n",
2979                                  ctrl->shutdown_timeout);
2980         } else
2981                 ctrl->shutdown_timeout = shutdown_timeout;
2982
2983         ctrl->npss = id->npss;
2984         ctrl->apsta = id->apsta;
2985         prev_apst_enabled = ctrl->apst_enabled;
2986         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2987                 if (force_apst && id->apsta) {
2988                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2989                         ctrl->apst_enabled = true;
2990                 } else {
2991                         ctrl->apst_enabled = false;
2992                 }
2993         } else {
2994                 ctrl->apst_enabled = id->apsta;
2995         }
2996         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2997
2998         if (ctrl->ops->flags & NVME_F_FABRICS) {
2999                 ctrl->icdoff = le16_to_cpu(id->icdoff);
3000                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3001                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3002                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3003
3004                 /*
3005                  * In fabrics we need to verify the cntlid matches the
3006                  * admin connect
3007                  */
3008                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3009                         dev_err(ctrl->device,
3010                                 "Mismatching cntlid: Connect %u vs Identify "
3011                                 "%u, rejecting\n",
3012                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
3013                         ret = -EINVAL;
3014                         goto out_free;
3015                 }
3016
3017                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3018                         dev_err(ctrl->device,
3019                                 "keep-alive support is mandatory for fabrics\n");
3020                         ret = -EINVAL;
3021                         goto out_free;
3022                 }
3023         } else {
3024                 ctrl->hmpre = le32_to_cpu(id->hmpre);
3025                 ctrl->hmmin = le32_to_cpu(id->hmmin);
3026                 ctrl->hmminds = le32_to_cpu(id->hmminds);
3027                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3028         }
3029
3030         ret = nvme_mpath_init_identify(ctrl, id);
3031         if (ret < 0)
3032                 goto out_free;
3033
3034         if (ctrl->apst_enabled && !prev_apst_enabled)
3035                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3036         else if (!ctrl->apst_enabled && prev_apst_enabled)
3037                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3038
3039 out_free:
3040         kfree(id);
3041         return ret;
3042 }
3043
3044 /*
3045  * Initialize the cached copies of the Identify data and various controller
3046  * register in our nvme_ctrl structure.  This should be called as soon as
3047  * the admin queue is fully up and running.
3048  */
3049 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3050 {
3051         int ret;
3052
3053         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3054         if (ret) {
3055                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3056                 return ret;
3057         }
3058
3059         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3060
3061         if (ctrl->vs >= NVME_VS(1, 1, 0))
3062                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3063
3064         ret = nvme_init_identify(ctrl);
3065         if (ret)
3066                 return ret;
3067
3068         ret = nvme_init_non_mdts_limits(ctrl);
3069         if (ret < 0)
3070                 return ret;
3071
3072         ret = nvme_configure_apst(ctrl);
3073         if (ret < 0)
3074                 return ret;
3075
3076         ret = nvme_configure_timestamp(ctrl);
3077         if (ret < 0)
3078                 return ret;
3079
3080         ret = nvme_configure_host_options(ctrl);
3081         if (ret < 0)
3082                 return ret;
3083
3084         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3085                 ret = nvme_hwmon_init(ctrl);
3086                 if (ret < 0)
3087                         return ret;
3088         }
3089
3090         ctrl->identified = true;
3091
3092         return 0;
3093 }
3094 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3095
3096 static int nvme_dev_open(struct inode *inode, struct file *file)
3097 {
3098         struct nvme_ctrl *ctrl =
3099                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3100
3101         switch (ctrl->state) {
3102         case NVME_CTRL_LIVE:
3103                 break;
3104         default:
3105                 return -EWOULDBLOCK;
3106         }
3107
3108         nvme_get_ctrl(ctrl);
3109         if (!try_module_get(ctrl->ops->module)) {
3110                 nvme_put_ctrl(ctrl);
3111                 return -EINVAL;
3112         }
3113
3114         file->private_data = ctrl;
3115         return 0;
3116 }
3117
3118 static int nvme_dev_release(struct inode *inode, struct file *file)
3119 {
3120         struct nvme_ctrl *ctrl =
3121                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3122
3123         module_put(ctrl->ops->module);
3124         nvme_put_ctrl(ctrl);
3125         return 0;
3126 }
3127
3128 static const struct file_operations nvme_dev_fops = {
3129         .owner          = THIS_MODULE,
3130         .open           = nvme_dev_open,
3131         .release        = nvme_dev_release,
3132         .unlocked_ioctl = nvme_dev_ioctl,
3133         .compat_ioctl   = compat_ptr_ioctl,
3134 };
3135
3136 static ssize_t nvme_sysfs_reset(struct device *dev,
3137                                 struct device_attribute *attr, const char *buf,
3138                                 size_t count)
3139 {
3140         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3141         int ret;
3142
3143         ret = nvme_reset_ctrl_sync(ctrl);
3144         if (ret < 0)
3145                 return ret;
3146         return count;
3147 }
3148 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3149
3150 static ssize_t nvme_sysfs_rescan(struct device *dev,
3151                                 struct device_attribute *attr, const char *buf,
3152                                 size_t count)
3153 {
3154         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3155
3156         nvme_queue_scan(ctrl);
3157         return count;
3158 }
3159 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3160
3161 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3162 {
3163         struct gendisk *disk = dev_to_disk(dev);
3164
3165         if (disk->fops == &nvme_bdev_ops)
3166                 return nvme_get_ns_from_dev(dev)->head;
3167         else
3168                 return disk->private_data;
3169 }
3170
3171 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3172                 char *buf)
3173 {
3174         struct nvme_ns_head *head = dev_to_ns_head(dev);
3175         struct nvme_ns_ids *ids = &head->ids;
3176         struct nvme_subsystem *subsys = head->subsys;
3177         int serial_len = sizeof(subsys->serial);
3178         int model_len = sizeof(subsys->model);
3179
3180         if (!uuid_is_null(&ids->uuid))
3181                 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3182
3183         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3184                 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3185
3186         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3187                 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3188
3189         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3190                                   subsys->serial[serial_len - 1] == '\0'))
3191                 serial_len--;
3192         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3193                                  subsys->model[model_len - 1] == '\0'))
3194                 model_len--;
3195
3196         return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3197                 serial_len, subsys->serial, model_len, subsys->model,
3198                 head->ns_id);
3199 }
3200 static DEVICE_ATTR_RO(wwid);
3201
3202 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3203                 char *buf)
3204 {
3205         return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3206 }
3207 static DEVICE_ATTR_RO(nguid);
3208
3209 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3210                 char *buf)
3211 {
3212         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3213
3214         /* For backward compatibility expose the NGUID to userspace if
3215          * we have no UUID set
3216          */
3217         if (uuid_is_null(&ids->uuid)) {
3218                 printk_ratelimited(KERN_WARNING
3219                                    "No UUID available providing old NGUID\n");
3220                 return sysfs_emit(buf, "%pU\n", ids->nguid);
3221         }
3222         return sysfs_emit(buf, "%pU\n", &ids->uuid);
3223 }
3224 static DEVICE_ATTR_RO(uuid);
3225
3226 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3227                 char *buf)
3228 {
3229         return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3230 }
3231 static DEVICE_ATTR_RO(eui);
3232
3233 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3234                 char *buf)
3235 {
3236         return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3237 }
3238 static DEVICE_ATTR_RO(nsid);
3239
3240 static struct attribute *nvme_ns_id_attrs[] = {
3241         &dev_attr_wwid.attr,
3242         &dev_attr_uuid.attr,
3243         &dev_attr_nguid.attr,
3244         &dev_attr_eui.attr,
3245         &dev_attr_nsid.attr,
3246 #ifdef CONFIG_NVME_MULTIPATH
3247         &dev_attr_ana_grpid.attr,
3248         &dev_attr_ana_state.attr,
3249 #endif
3250         NULL,
3251 };
3252
3253 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3254                 struct attribute *a, int n)
3255 {
3256         struct device *dev = container_of(kobj, struct device, kobj);
3257         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3258
3259         if (a == &dev_attr_uuid.attr) {
3260                 if (uuid_is_null(&ids->uuid) &&
3261                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3262                         return 0;
3263         }
3264         if (a == &dev_attr_nguid.attr) {
3265                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3266                         return 0;
3267         }
3268         if (a == &dev_attr_eui.attr) {
3269                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3270                         return 0;
3271         }
3272 #ifdef CONFIG_NVME_MULTIPATH
3273         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3274                 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3275                         return 0;
3276                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3277                         return 0;
3278         }
3279 #endif
3280         return a->mode;
3281 }
3282
3283 static const struct attribute_group nvme_ns_id_attr_group = {
3284         .attrs          = nvme_ns_id_attrs,
3285         .is_visible     = nvme_ns_id_attrs_are_visible,
3286 };
3287
3288 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3289         &nvme_ns_id_attr_group,
3290         NULL,
3291 };
3292
3293 #define nvme_show_str_function(field)                                           \
3294 static ssize_t  field##_show(struct device *dev,                                \
3295                             struct device_attribute *attr, char *buf)           \
3296 {                                                                               \
3297         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3298         return sysfs_emit(buf, "%.*s\n",                                        \
3299                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3300 }                                                                               \
3301 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3302
3303 nvme_show_str_function(model);
3304 nvme_show_str_function(serial);
3305 nvme_show_str_function(firmware_rev);
3306
3307 #define nvme_show_int_function(field)                                           \
3308 static ssize_t  field##_show(struct device *dev,                                \
3309                             struct device_attribute *attr, char *buf)           \
3310 {                                                                               \
3311         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3312         return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3313 }                                                                               \
3314 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3315
3316 nvme_show_int_function(cntlid);
3317 nvme_show_int_function(numa_node);
3318 nvme_show_int_function(queue_count);
3319 nvme_show_int_function(sqsize);
3320 nvme_show_int_function(kato);
3321
3322 static ssize_t nvme_sysfs_delete(struct device *dev,
3323                                 struct device_attribute *attr, const char *buf,
3324                                 size_t count)
3325 {
3326         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3327
3328         if (device_remove_file_self(dev, attr))
3329                 nvme_delete_ctrl_sync(ctrl);
3330         return count;
3331 }
3332 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3333
3334 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3335                                          struct device_attribute *attr,
3336                                          char *buf)
3337 {
3338         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3339
3340         return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3341 }
3342 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3343
3344 static ssize_t nvme_sysfs_show_state(struct device *dev,
3345                                      struct device_attribute *attr,
3346                                      char *buf)
3347 {
3348         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3349         static const char *const state_name[] = {
3350                 [NVME_CTRL_NEW]         = "new",
3351                 [NVME_CTRL_LIVE]        = "live",
3352                 [NVME_CTRL_RESETTING]   = "resetting",
3353                 [NVME_CTRL_CONNECTING]  = "connecting",
3354                 [NVME_CTRL_DELETING]    = "deleting",
3355                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3356                 [NVME_CTRL_DEAD]        = "dead",
3357         };
3358
3359         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3360             state_name[ctrl->state])
3361                 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3362
3363         return sysfs_emit(buf, "unknown state\n");
3364 }
3365
3366 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3367
3368 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3369                                          struct device_attribute *attr,
3370                                          char *buf)
3371 {
3372         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3373
3374         return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3375 }
3376 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3377
3378 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3379                                         struct device_attribute *attr,
3380                                         char *buf)
3381 {
3382         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3383
3384         return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3385 }
3386 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3387
3388 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3389                                         struct device_attribute *attr,
3390                                         char *buf)
3391 {
3392         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3393
3394         return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3395 }
3396 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3397
3398 static ssize_t nvme_sysfs_show_address(struct device *dev,
3399                                          struct device_attribute *attr,
3400                                          char *buf)
3401 {
3402         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3403
3404         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3405 }
3406 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3407
3408 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3409                 struct device_attribute *attr, char *buf)
3410 {
3411         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3412         struct nvmf_ctrl_options *opts = ctrl->opts;
3413
3414         if (ctrl->opts->max_reconnects == -1)
3415                 return sysfs_emit(buf, "off\n");
3416         return sysfs_emit(buf, "%d\n",
3417                           opts->max_reconnects * opts->reconnect_delay);
3418 }
3419
3420 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3421                 struct device_attribute *attr, const char *buf, size_t count)
3422 {
3423         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3424         struct nvmf_ctrl_options *opts = ctrl->opts;
3425         int ctrl_loss_tmo, err;
3426
3427         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3428         if (err)
3429                 return -EINVAL;
3430
3431         if (ctrl_loss_tmo < 0)
3432                 opts->max_reconnects = -1;
3433         else
3434                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3435                                                 opts->reconnect_delay);
3436         return count;
3437 }
3438 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3439         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3440
3441 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3442                 struct device_attribute *attr, char *buf)
3443 {
3444         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3445
3446         if (ctrl->opts->reconnect_delay == -1)
3447                 return sysfs_emit(buf, "off\n");
3448         return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3449 }
3450
3451 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3452                 struct device_attribute *attr, const char *buf, size_t count)
3453 {
3454         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3455         unsigned int v;
3456         int err;
3457
3458         err = kstrtou32(buf, 10, &v);
3459         if (err)
3460                 return err;
3461
3462         ctrl->opts->reconnect_delay = v;
3463         return count;
3464 }
3465 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3466         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3467
3468 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3469                 struct device_attribute *attr, char *buf)
3470 {
3471         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3472
3473         if (ctrl->opts->fast_io_fail_tmo == -1)
3474                 return sysfs_emit(buf, "off\n");
3475         return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3476 }
3477
3478 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3479                 struct device_attribute *attr, const char *buf, size_t count)
3480 {
3481         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3482         struct nvmf_ctrl_options *opts = ctrl->opts;
3483         int fast_io_fail_tmo, err;
3484
3485         err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3486         if (err)
3487                 return -EINVAL;
3488
3489         if (fast_io_fail_tmo < 0)
3490                 opts->fast_io_fail_tmo = -1;
3491         else
3492                 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3493         return count;
3494 }
3495 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3496         nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3497
3498 static ssize_t cntrltype_show(struct device *dev,
3499                               struct device_attribute *attr, char *buf)
3500 {
3501         static const char * const type[] = {
3502                 [NVME_CTRL_IO] = "io\n",
3503                 [NVME_CTRL_DISC] = "discovery\n",
3504                 [NVME_CTRL_ADMIN] = "admin\n",
3505         };
3506         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3507
3508         if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3509                 return sysfs_emit(buf, "reserved\n");
3510
3511         return sysfs_emit(buf, type[ctrl->cntrltype]);
3512 }
3513 static DEVICE_ATTR_RO(cntrltype);
3514
3515 static ssize_t dctype_show(struct device *dev,
3516                            struct device_attribute *attr, char *buf)
3517 {
3518         static const char * const type[] = {
3519                 [NVME_DCTYPE_NOT_REPORTED] = "none\n",
3520                 [NVME_DCTYPE_DDC] = "ddc\n",
3521                 [NVME_DCTYPE_CDC] = "cdc\n",
3522         };
3523         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3524
3525         if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3526                 return sysfs_emit(buf, "reserved\n");
3527
3528         return sysfs_emit(buf, type[ctrl->dctype]);
3529 }
3530 static DEVICE_ATTR_RO(dctype);
3531
3532 static struct attribute *nvme_dev_attrs[] = {
3533         &dev_attr_reset_controller.attr,
3534         &dev_attr_rescan_controller.attr,
3535         &dev_attr_model.attr,
3536         &dev_attr_serial.attr,
3537         &dev_attr_firmware_rev.attr,
3538         &dev_attr_cntlid.attr,
3539         &dev_attr_delete_controller.attr,
3540         &dev_attr_transport.attr,
3541         &dev_attr_subsysnqn.attr,
3542         &dev_attr_address.attr,
3543         &dev_attr_state.attr,
3544         &dev_attr_numa_node.attr,
3545         &dev_attr_queue_count.attr,
3546         &dev_attr_sqsize.attr,
3547         &dev_attr_hostnqn.attr,
3548         &dev_attr_hostid.attr,
3549         &dev_attr_ctrl_loss_tmo.attr,
3550         &dev_attr_reconnect_delay.attr,
3551         &dev_attr_fast_io_fail_tmo.attr,
3552         &dev_attr_kato.attr,
3553         &dev_attr_cntrltype.attr,
3554         &dev_attr_dctype.attr,
3555         NULL
3556 };
3557
3558 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3559                 struct attribute *a, int n)
3560 {
3561         struct device *dev = container_of(kobj, struct device, kobj);
3562         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3563
3564         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3565                 return 0;
3566         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3567                 return 0;
3568         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3569                 return 0;
3570         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3571                 return 0;
3572         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3573                 return 0;
3574         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3575                 return 0;
3576         if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3577                 return 0;
3578
3579         return a->mode;
3580 }
3581
3582 static const struct attribute_group nvme_dev_attrs_group = {
3583         .attrs          = nvme_dev_attrs,
3584         .is_visible     = nvme_dev_attrs_are_visible,
3585 };
3586
3587 static const struct attribute_group *nvme_dev_attr_groups[] = {
3588         &nvme_dev_attrs_group,
3589         NULL,
3590 };
3591
3592 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3593                 unsigned nsid)
3594 {
3595         struct nvme_ns_head *h;
3596
3597         lockdep_assert_held(&subsys->lock);
3598
3599         list_for_each_entry(h, &subsys->nsheads, entry) {
3600                 if (h->ns_id != nsid)
3601                         continue;
3602                 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3603                         return h;
3604         }
3605
3606         return NULL;
3607 }
3608
3609 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3610                 struct nvme_ns_ids *ids)
3611 {
3612         bool has_uuid = !uuid_is_null(&ids->uuid);
3613         bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3614         bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3615         struct nvme_ns_head *h;
3616
3617         lockdep_assert_held(&subsys->lock);
3618
3619         list_for_each_entry(h, &subsys->nsheads, entry) {
3620                 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3621                         return -EINVAL;
3622                 if (has_nguid &&
3623                     memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3624                         return -EINVAL;
3625                 if (has_eui64 &&
3626                     memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3627                         return -EINVAL;
3628         }
3629
3630         return 0;
3631 }
3632
3633 static void nvme_cdev_rel(struct device *dev)
3634 {
3635         ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3636 }
3637
3638 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3639 {
3640         cdev_device_del(cdev, cdev_device);
3641         put_device(cdev_device);
3642 }
3643
3644 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3645                 const struct file_operations *fops, struct module *owner)
3646 {
3647         int minor, ret;
3648
3649         minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3650         if (minor < 0)
3651                 return minor;
3652         cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3653         cdev_device->class = nvme_ns_chr_class;
3654         cdev_device->release = nvme_cdev_rel;
3655         device_initialize(cdev_device);
3656         cdev_init(cdev, fops);
3657         cdev->owner = owner;
3658         ret = cdev_device_add(cdev, cdev_device);
3659         if (ret)
3660                 put_device(cdev_device);
3661
3662         return ret;
3663 }
3664
3665 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3666 {
3667         return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3668 }
3669
3670 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3671 {
3672         nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3673         return 0;
3674 }
3675
3676 static const struct file_operations nvme_ns_chr_fops = {
3677         .owner          = THIS_MODULE,
3678         .open           = nvme_ns_chr_open,
3679         .release        = nvme_ns_chr_release,
3680         .unlocked_ioctl = nvme_ns_chr_ioctl,
3681         .compat_ioctl   = compat_ptr_ioctl,
3682 };
3683
3684 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3685 {
3686         int ret;
3687
3688         ns->cdev_device.parent = ns->ctrl->device;
3689         ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3690                            ns->ctrl->instance, ns->head->instance);
3691         if (ret)
3692                 return ret;
3693
3694         return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3695                              ns->ctrl->ops->module);
3696 }
3697
3698 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3699                 unsigned nsid, struct nvme_ns_ids *ids)
3700 {
3701         struct nvme_ns_head *head;
3702         size_t size = sizeof(*head);
3703         int ret = -ENOMEM;
3704
3705 #ifdef CONFIG_NVME_MULTIPATH
3706         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3707 #endif
3708
3709         head = kzalloc(size, GFP_KERNEL);
3710         if (!head)
3711                 goto out;
3712         ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3713         if (ret < 0)
3714                 goto out_free_head;
3715         head->instance = ret;
3716         INIT_LIST_HEAD(&head->list);
3717         ret = init_srcu_struct(&head->srcu);
3718         if (ret)
3719                 goto out_ida_remove;
3720         head->subsys = ctrl->subsys;
3721         head->ns_id = nsid;
3722         head->ids = *ids;
3723         kref_init(&head->ref);
3724
3725         if (head->ids.csi) {
3726                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3727                 if (ret)
3728                         goto out_cleanup_srcu;
3729         } else
3730                 head->effects = ctrl->effects;
3731
3732         ret = nvme_mpath_alloc_disk(ctrl, head);
3733         if (ret)
3734                 goto out_cleanup_srcu;
3735
3736         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3737
3738         kref_get(&ctrl->subsys->ref);
3739
3740         return head;
3741 out_cleanup_srcu:
3742         cleanup_srcu_struct(&head->srcu);
3743 out_ida_remove:
3744         ida_free(&ctrl->subsys->ns_ida, head->instance);
3745 out_free_head:
3746         kfree(head);
3747 out:
3748         if (ret > 0)
3749                 ret = blk_status_to_errno(nvme_error_status(ret));
3750         return ERR_PTR(ret);
3751 }
3752
3753 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3754                 struct nvme_ns_ids *ids)
3755 {
3756         struct nvme_subsystem *s;
3757         int ret = 0;
3758
3759         /*
3760          * Note that this check is racy as we try to avoid holding the global
3761          * lock over the whole ns_head creation.  But it is only intended as
3762          * a sanity check anyway.
3763          */
3764         mutex_lock(&nvme_subsystems_lock);
3765         list_for_each_entry(s, &nvme_subsystems, entry) {
3766                 if (s == this)
3767                         continue;
3768                 mutex_lock(&s->lock);
3769                 ret = nvme_subsys_check_duplicate_ids(s, ids);
3770                 mutex_unlock(&s->lock);
3771                 if (ret)
3772                         break;
3773         }
3774         mutex_unlock(&nvme_subsystems_lock);
3775
3776         return ret;
3777 }
3778
3779 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3780                 struct nvme_ns_ids *ids, bool is_shared)
3781 {
3782         struct nvme_ctrl *ctrl = ns->ctrl;
3783         struct nvme_ns_head *head = NULL;
3784         int ret;
3785
3786         ret = nvme_global_check_duplicate_ids(ctrl->subsys, ids);
3787         if (ret) {
3788                 dev_err(ctrl->device,
3789                         "globally duplicate IDs for nsid %d\n", nsid);
3790                 return ret;
3791         }
3792
3793         mutex_lock(&ctrl->subsys->lock);
3794         head = nvme_find_ns_head(ctrl->subsys, nsid);
3795         if (!head) {
3796                 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, ids);
3797                 if (ret) {
3798                         dev_err(ctrl->device,
3799                                 "duplicate IDs in subsystem for nsid %d\n",
3800                                 nsid);
3801                         goto out_unlock;
3802                 }
3803                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3804                 if (IS_ERR(head)) {
3805                         ret = PTR_ERR(head);
3806                         goto out_unlock;
3807                 }
3808                 head->shared = is_shared;
3809         } else {
3810                 ret = -EINVAL;
3811                 if (!is_shared || !head->shared) {
3812                         dev_err(ctrl->device,
3813                                 "Duplicate unshared namespace %d\n", nsid);
3814                         goto out_put_ns_head;
3815                 }
3816                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3817                         dev_err(ctrl->device,
3818                                 "IDs don't match for shared namespace %d\n",
3819                                         nsid);
3820                         goto out_put_ns_head;
3821                 }
3822
3823                 if (!multipath && !list_empty(&head->list)) {
3824                         dev_warn(ctrl->device,
3825                                 "Found shared namespace %d, but multipathing not supported.\n",
3826                                 nsid);
3827                         dev_warn_once(ctrl->device,
3828                                 "Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3829                 }
3830         }
3831
3832         list_add_tail_rcu(&ns->siblings, &head->list);
3833         ns->head = head;
3834         mutex_unlock(&ctrl->subsys->lock);
3835         return 0;
3836
3837 out_put_ns_head:
3838         nvme_put_ns_head(head);
3839 out_unlock:
3840         mutex_unlock(&ctrl->subsys->lock);
3841         return ret;
3842 }
3843
3844 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3845 {
3846         struct nvme_ns *ns, *ret = NULL;
3847
3848         down_read(&ctrl->namespaces_rwsem);
3849         list_for_each_entry(ns, &ctrl->namespaces, list) {
3850                 if (ns->head->ns_id == nsid) {
3851                         if (!nvme_get_ns(ns))
3852                                 continue;
3853                         ret = ns;
3854                         break;
3855                 }
3856                 if (ns->head->ns_id > nsid)
3857                         break;
3858         }
3859         up_read(&ctrl->namespaces_rwsem);
3860         return ret;
3861 }
3862 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3863
3864 /*
3865  * Add the namespace to the controller list while keeping the list ordered.
3866  */
3867 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3868 {
3869         struct nvme_ns *tmp;
3870
3871         list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3872                 if (tmp->head->ns_id < ns->head->ns_id) {
3873                         list_add(&ns->list, &tmp->list);
3874                         return;
3875                 }
3876         }
3877         list_add(&ns->list, &ns->ctrl->namespaces);
3878 }
3879
3880 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3881                 struct nvme_ns_ids *ids)
3882 {
3883         struct nvme_ns *ns;
3884         struct gendisk *disk;
3885         struct nvme_id_ns *id;
3886         int node = ctrl->numa_node;
3887
3888         if (nvme_identify_ns(ctrl, nsid, ids, &id))
3889                 return;
3890
3891         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3892         if (!ns)
3893                 goto out_free_id;
3894
3895         disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3896         if (IS_ERR(disk))
3897                 goto out_free_ns;
3898         disk->fops = &nvme_bdev_ops;
3899         disk->private_data = ns;
3900
3901         ns->disk = disk;
3902         ns->queue = disk->queue;
3903
3904         if (ctrl->opts && ctrl->opts->data_digest)
3905                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3906
3907         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3908         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3909                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3910
3911         ns->ctrl = ctrl;
3912         kref_init(&ns->kref);
3913
3914         if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3915                 goto out_cleanup_disk;
3916
3917         /*
3918          * If multipathing is enabled, the device name for all disks and not
3919          * just those that represent shared namespaces needs to be based on the
3920          * subsystem instance.  Using the controller instance for private
3921          * namespaces could lead to naming collisions between shared and private
3922          * namespaces if they don't use a common numbering scheme.
3923          *
3924          * If multipathing is not enabled, disk names must use the controller
3925          * instance as shared namespaces will show up as multiple block
3926          * devices.
3927          */
3928         if (ns->head->disk) {
3929                 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3930                         ctrl->instance, ns->head->instance);
3931                 disk->flags |= GENHD_FL_HIDDEN;
3932         } else if (multipath) {
3933                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3934                         ns->head->instance);
3935         } else {
3936                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3937                         ns->head->instance);
3938         }
3939
3940         if (nvme_update_ns_info(ns, id))
3941                 goto out_unlink_ns;
3942
3943         down_write(&ctrl->namespaces_rwsem);
3944         nvme_ns_add_to_ctrl_list(ns);
3945         up_write(&ctrl->namespaces_rwsem);
3946         nvme_get_ctrl(ctrl);
3947
3948         if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3949                 goto out_cleanup_ns_from_list;
3950
3951         if (!nvme_ns_head_multipath(ns->head))
3952                 nvme_add_ns_cdev(ns);
3953
3954         nvme_mpath_add_disk(ns, id);
3955         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3956         kfree(id);
3957
3958         return;
3959
3960  out_cleanup_ns_from_list:
3961         nvme_put_ctrl(ctrl);
3962         down_write(&ctrl->namespaces_rwsem);
3963         list_del_init(&ns->list);
3964         up_write(&ctrl->namespaces_rwsem);
3965  out_unlink_ns:
3966         mutex_lock(&ctrl->subsys->lock);
3967         list_del_rcu(&ns->siblings);
3968         if (list_empty(&ns->head->list))
3969                 list_del_init(&ns->head->entry);
3970         mutex_unlock(&ctrl->subsys->lock);
3971         nvme_put_ns_head(ns->head);
3972  out_cleanup_disk:
3973         blk_cleanup_disk(disk);
3974  out_free_ns:
3975         kfree(ns);
3976  out_free_id:
3977         kfree(id);
3978 }
3979
3980 static void nvme_ns_remove(struct nvme_ns *ns)
3981 {
3982         bool last_path = false;
3983
3984         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3985                 return;
3986
3987         clear_bit(NVME_NS_READY, &ns->flags);
3988         set_capacity(ns->disk, 0);
3989         nvme_fault_inject_fini(&ns->fault_inject);
3990
3991         mutex_lock(&ns->ctrl->subsys->lock);
3992         list_del_rcu(&ns->siblings);
3993         if (list_empty(&ns->head->list)) {
3994                 list_del_init(&ns->head->entry);
3995                 last_path = true;
3996         }
3997         mutex_unlock(&ns->ctrl->subsys->lock);
3998
3999         /* guarantee not available in head->list */
4000         synchronize_rcu();
4001
4002         /* wait for concurrent submissions */
4003         if (nvme_mpath_clear_current_path(ns))
4004                 synchronize_srcu(&ns->head->srcu);
4005
4006         if (!nvme_ns_head_multipath(ns->head))
4007                 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4008         del_gendisk(ns->disk);
4009         blk_cleanup_queue(ns->queue);
4010
4011         down_write(&ns->ctrl->namespaces_rwsem);
4012         list_del_init(&ns->list);
4013         up_write(&ns->ctrl->namespaces_rwsem);
4014
4015         if (last_path)
4016                 nvme_mpath_shutdown_disk(ns->head);
4017         nvme_put_ns(ns);
4018 }
4019
4020 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4021 {
4022         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4023
4024         if (ns) {
4025                 nvme_ns_remove(ns);
4026                 nvme_put_ns(ns);
4027         }
4028 }
4029
4030 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4031 {
4032         struct nvme_id_ns *id;
4033         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4034
4035         if (test_bit(NVME_NS_DEAD, &ns->flags))
4036                 goto out;
4037
4038         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4039         if (ret)
4040                 goto out;
4041
4042         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4043         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4044                 dev_err(ns->ctrl->device,
4045                         "identifiers changed for nsid %d\n", ns->head->ns_id);
4046                 goto out_free_id;
4047         }
4048
4049         ret = nvme_update_ns_info(ns, id);
4050
4051 out_free_id:
4052         kfree(id);
4053 out:
4054         /*
4055          * Only remove the namespace if we got a fatal error back from the
4056          * device, otherwise ignore the error and just move on.
4057          *
4058          * TODO: we should probably schedule a delayed retry here.
4059          */
4060         if (ret > 0 && (ret & NVME_SC_DNR))
4061                 nvme_ns_remove(ns);
4062 }
4063
4064 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4065 {
4066         struct nvme_ns_ids ids = { };
4067         struct nvme_ns *ns;
4068
4069         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4070                 return;
4071
4072         ns = nvme_find_get_ns(ctrl, nsid);
4073         if (ns) {
4074                 nvme_validate_ns(ns, &ids);
4075                 nvme_put_ns(ns);
4076                 return;
4077         }
4078
4079         switch (ids.csi) {
4080         case NVME_CSI_NVM:
4081                 nvme_alloc_ns(ctrl, nsid, &ids);
4082                 break;
4083         case NVME_CSI_ZNS:
4084                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4085                         dev_warn(ctrl->device,
4086                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4087                                 nsid);
4088                         break;
4089                 }
4090                 if (!nvme_multi_css(ctrl)) {
4091                         dev_warn(ctrl->device,
4092                                 "command set not reported for nsid: %d\n",
4093                                 nsid);
4094                         break;
4095                 }
4096                 nvme_alloc_ns(ctrl, nsid, &ids);
4097                 break;
4098         default:
4099                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4100                         ids.csi, nsid);
4101                 break;
4102         }
4103 }
4104
4105 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4106                                         unsigned nsid)
4107 {
4108         struct nvme_ns *ns, *next;
4109         LIST_HEAD(rm_list);
4110
4111         down_write(&ctrl->namespaces_rwsem);
4112         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4113                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4114                         list_move_tail(&ns->list, &rm_list);
4115         }
4116         up_write(&ctrl->namespaces_rwsem);
4117
4118         list_for_each_entry_safe(ns, next, &rm_list, list)
4119                 nvme_ns_remove(ns);
4120
4121 }
4122
4123 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4124 {
4125         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4126         __le32 *ns_list;
4127         u32 prev = 0;
4128         int ret = 0, i;
4129
4130         if (nvme_ctrl_limited_cns(ctrl))
4131                 return -EOPNOTSUPP;
4132
4133         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4134         if (!ns_list)
4135                 return -ENOMEM;
4136
4137         for (;;) {
4138                 struct nvme_command cmd = {
4139                         .identify.opcode        = nvme_admin_identify,
4140                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
4141                         .identify.nsid          = cpu_to_le32(prev),
4142                 };
4143
4144                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4145                                             NVME_IDENTIFY_DATA_SIZE);
4146                 if (ret) {
4147                         dev_warn(ctrl->device,
4148                                 "Identify NS List failed (status=0x%x)\n", ret);
4149                         goto free;
4150                 }
4151
4152                 for (i = 0; i < nr_entries; i++) {
4153                         u32 nsid = le32_to_cpu(ns_list[i]);
4154
4155                         if (!nsid)      /* end of the list? */
4156                                 goto out;
4157                         nvme_validate_or_alloc_ns(ctrl, nsid);
4158                         while (++prev < nsid)
4159                                 nvme_ns_remove_by_nsid(ctrl, prev);
4160                 }
4161         }
4162  out:
4163         nvme_remove_invalid_namespaces(ctrl, prev);
4164  free:
4165         kfree(ns_list);
4166         return ret;
4167 }
4168
4169 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4170 {
4171         struct nvme_id_ctrl *id;
4172         u32 nn, i;
4173
4174         if (nvme_identify_ctrl(ctrl, &id))
4175                 return;
4176         nn = le32_to_cpu(id->nn);
4177         kfree(id);
4178
4179         for (i = 1; i <= nn; i++)
4180                 nvme_validate_or_alloc_ns(ctrl, i);
4181
4182         nvme_remove_invalid_namespaces(ctrl, nn);
4183 }
4184
4185 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4186 {
4187         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4188         __le32 *log;
4189         int error;
4190
4191         log = kzalloc(log_size, GFP_KERNEL);
4192         if (!log)
4193                 return;
4194
4195         /*
4196          * We need to read the log to clear the AEN, but we don't want to rely
4197          * on it for the changed namespace information as userspace could have
4198          * raced with us in reading the log page, which could cause us to miss
4199          * updates.
4200          */
4201         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4202                         NVME_CSI_NVM, log, log_size, 0);
4203         if (error)
4204                 dev_warn(ctrl->device,
4205                         "reading changed ns log failed: %d\n", error);
4206
4207         kfree(log);
4208 }
4209
4210 static void nvme_scan_work(struct work_struct *work)
4211 {
4212         struct nvme_ctrl *ctrl =
4213                 container_of(work, struct nvme_ctrl, scan_work);
4214
4215         /* No tagset on a live ctrl means IO queues could not created */
4216         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4217                 return;
4218
4219         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4220                 dev_info(ctrl->device, "rescanning namespaces.\n");
4221                 nvme_clear_changed_ns_log(ctrl);
4222         }
4223
4224         mutex_lock(&ctrl->scan_lock);
4225         if (nvme_scan_ns_list(ctrl) != 0)
4226                 nvme_scan_ns_sequential(ctrl);
4227         mutex_unlock(&ctrl->scan_lock);
4228 }
4229
4230 /*
4231  * This function iterates the namespace list unlocked to allow recovery from
4232  * controller failure. It is up to the caller to ensure the namespace list is
4233  * not modified by scan work while this function is executing.
4234  */
4235 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4236 {
4237         struct nvme_ns *ns, *next;
4238         LIST_HEAD(ns_list);
4239
4240         /*
4241          * make sure to requeue I/O to all namespaces as these
4242          * might result from the scan itself and must complete
4243          * for the scan_work to make progress
4244          */
4245         nvme_mpath_clear_ctrl_paths(ctrl);
4246
4247         /* prevent racing with ns scanning */
4248         flush_work(&ctrl->scan_work);
4249
4250         /*
4251          * The dead states indicates the controller was not gracefully
4252          * disconnected. In that case, we won't be able to flush any data while
4253          * removing the namespaces' disks; fail all the queues now to avoid
4254          * potentially having to clean up the failed sync later.
4255          */
4256         if (ctrl->state == NVME_CTRL_DEAD)
4257                 nvme_kill_queues(ctrl);
4258
4259         /* this is a no-op when called from the controller reset handler */
4260         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4261
4262         down_write(&ctrl->namespaces_rwsem);
4263         list_splice_init(&ctrl->namespaces, &ns_list);
4264         up_write(&ctrl->namespaces_rwsem);
4265
4266         list_for_each_entry_safe(ns, next, &ns_list, list)
4267                 nvme_ns_remove(ns);
4268 }
4269 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4270
4271 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4272 {
4273         struct nvme_ctrl *ctrl =
4274                 container_of(dev, struct nvme_ctrl, ctrl_device);
4275         struct nvmf_ctrl_options *opts = ctrl->opts;
4276         int ret;
4277
4278         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4279         if (ret)
4280                 return ret;
4281
4282         if (opts) {
4283                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4284                 if (ret)
4285                         return ret;
4286
4287                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4288                                 opts->trsvcid ?: "none");
4289                 if (ret)
4290                         return ret;
4291
4292                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4293                                 opts->host_traddr ?: "none");
4294                 if (ret)
4295                         return ret;
4296
4297                 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4298                                 opts->host_iface ?: "none");
4299         }
4300         return ret;
4301 }
4302
4303 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4304 {
4305         char *envp[2] = { envdata, NULL };
4306
4307         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4308 }
4309
4310 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4311 {
4312         char *envp[2] = { NULL, NULL };
4313         u32 aen_result = ctrl->aen_result;
4314
4315         ctrl->aen_result = 0;
4316         if (!aen_result)
4317                 return;
4318
4319         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4320         if (!envp[0])
4321                 return;
4322         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4323         kfree(envp[0]);
4324 }
4325
4326 static void nvme_async_event_work(struct work_struct *work)
4327 {
4328         struct nvme_ctrl *ctrl =
4329                 container_of(work, struct nvme_ctrl, async_event_work);
4330
4331         nvme_aen_uevent(ctrl);
4332
4333         /*
4334          * The transport drivers must guarantee AER submission here is safe by
4335          * flushing ctrl async_event_work after changing the controller state
4336          * from LIVE and before freeing the admin queue.
4337         */
4338         if (ctrl->state == NVME_CTRL_LIVE)
4339                 ctrl->ops->submit_async_event(ctrl);
4340 }
4341
4342 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4343 {
4344
4345         u32 csts;
4346
4347         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4348                 return false;
4349
4350         if (csts == ~0)
4351                 return false;
4352
4353         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4354 }
4355
4356 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4357 {
4358         struct nvme_fw_slot_info_log *log;
4359
4360         log = kmalloc(sizeof(*log), GFP_KERNEL);
4361         if (!log)
4362                 return;
4363
4364         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4365                         log, sizeof(*log), 0))
4366                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4367         kfree(log);
4368 }
4369
4370 static void nvme_fw_act_work(struct work_struct *work)
4371 {
4372         struct nvme_ctrl *ctrl = container_of(work,
4373                                 struct nvme_ctrl, fw_act_work);
4374         unsigned long fw_act_timeout;
4375
4376         if (ctrl->mtfa)
4377                 fw_act_timeout = jiffies +
4378                                 msecs_to_jiffies(ctrl->mtfa * 100);
4379         else
4380                 fw_act_timeout = jiffies +
4381                                 msecs_to_jiffies(admin_timeout * 1000);
4382
4383         nvme_stop_queues(ctrl);
4384         while (nvme_ctrl_pp_status(ctrl)) {
4385                 if (time_after(jiffies, fw_act_timeout)) {
4386                         dev_warn(ctrl->device,
4387                                 "Fw activation timeout, reset controller\n");
4388                         nvme_try_sched_reset(ctrl);
4389                         return;
4390                 }
4391                 msleep(100);
4392         }
4393
4394         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4395                 return;
4396
4397         nvme_start_queues(ctrl);
4398         /* read FW slot information to clear the AER */
4399         nvme_get_fw_slot_info(ctrl);
4400 }
4401
4402 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4403 {
4404         u32 aer_notice_type = (result & 0xff00) >> 8;
4405
4406         trace_nvme_async_event(ctrl, aer_notice_type);
4407
4408         switch (aer_notice_type) {
4409         case NVME_AER_NOTICE_NS_CHANGED:
4410                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4411                 nvme_queue_scan(ctrl);
4412                 break;
4413         case NVME_AER_NOTICE_FW_ACT_STARTING:
4414                 /*
4415                  * We are (ab)using the RESETTING state to prevent subsequent
4416                  * recovery actions from interfering with the controller's
4417                  * firmware activation.
4418                  */
4419                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4420                         queue_work(nvme_wq, &ctrl->fw_act_work);
4421                 break;
4422 #ifdef CONFIG_NVME_MULTIPATH
4423         case NVME_AER_NOTICE_ANA:
4424                 if (!ctrl->ana_log_buf)
4425                         break;
4426                 queue_work(nvme_wq, &ctrl->ana_work);
4427                 break;
4428 #endif
4429         case NVME_AER_NOTICE_DISC_CHANGED:
4430                 ctrl->aen_result = result;
4431                 break;
4432         default:
4433                 dev_warn(ctrl->device, "async event result %08x\n", result);
4434         }
4435 }
4436
4437 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4438                 volatile union nvme_result *res)
4439 {
4440         u32 result = le32_to_cpu(res->u32);
4441         u32 aer_type = result & 0x07;
4442
4443         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4444                 return;
4445
4446         switch (aer_type) {
4447         case NVME_AER_NOTICE:
4448                 nvme_handle_aen_notice(ctrl, result);
4449                 break;
4450         case NVME_AER_ERROR:
4451         case NVME_AER_SMART:
4452         case NVME_AER_CSS:
4453         case NVME_AER_VS:
4454                 trace_nvme_async_event(ctrl, aer_type);
4455                 ctrl->aen_result = result;
4456                 break;
4457         default:
4458                 break;
4459         }
4460         queue_work(nvme_wq, &ctrl->async_event_work);
4461 }
4462 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4463
4464 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4465 {
4466         nvme_mpath_stop(ctrl);
4467         nvme_stop_keep_alive(ctrl);
4468         nvme_stop_failfast_work(ctrl);
4469         flush_work(&ctrl->async_event_work);
4470         cancel_work_sync(&ctrl->fw_act_work);
4471 }
4472 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4473
4474 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4475 {
4476         nvme_start_keep_alive(ctrl);
4477
4478         nvme_enable_aen(ctrl);
4479
4480         if (ctrl->queue_count > 1) {
4481                 nvme_queue_scan(ctrl);
4482                 nvme_start_queues(ctrl);
4483         }
4484
4485         nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4486 }
4487 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4488
4489 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4490 {
4491         nvme_hwmon_exit(ctrl);
4492         nvme_fault_inject_fini(&ctrl->fault_inject);
4493         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4494         cdev_device_del(&ctrl->cdev, ctrl->device);
4495         nvme_put_ctrl(ctrl);
4496 }
4497 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4498
4499 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4500 {
4501         struct nvme_effects_log *cel;
4502         unsigned long i;
4503
4504         xa_for_each(&ctrl->cels, i, cel) {
4505                 xa_erase(&ctrl->cels, i);
4506                 kfree(cel);
4507         }
4508
4509         xa_destroy(&ctrl->cels);
4510 }
4511
4512 static void nvme_free_ctrl(struct device *dev)
4513 {
4514         struct nvme_ctrl *ctrl =
4515                 container_of(dev, struct nvme_ctrl, ctrl_device);
4516         struct nvme_subsystem *subsys = ctrl->subsys;
4517
4518         if (!subsys || ctrl->instance != subsys->instance)
4519                 ida_free(&nvme_instance_ida, ctrl->instance);
4520
4521         nvme_free_cels(ctrl);
4522         nvme_mpath_uninit(ctrl);
4523         __free_page(ctrl->discard_page);
4524
4525         if (subsys) {
4526                 mutex_lock(&nvme_subsystems_lock);
4527                 list_del(&ctrl->subsys_entry);
4528                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4529                 mutex_unlock(&nvme_subsystems_lock);
4530         }
4531
4532         ctrl->ops->free_ctrl(ctrl);
4533
4534         if (subsys)
4535                 nvme_put_subsystem(subsys);
4536 }
4537
4538 /*
4539  * Initialize a NVMe controller structures.  This needs to be called during
4540  * earliest initialization so that we have the initialized structured around
4541  * during probing.
4542  */
4543 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4544                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4545 {
4546         int ret;
4547
4548         ctrl->state = NVME_CTRL_NEW;
4549         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4550         spin_lock_init(&ctrl->lock);
4551         mutex_init(&ctrl->scan_lock);
4552         INIT_LIST_HEAD(&ctrl->namespaces);
4553         xa_init(&ctrl->cels);
4554         init_rwsem(&ctrl->namespaces_rwsem);
4555         ctrl->dev = dev;
4556         ctrl->ops = ops;
4557         ctrl->quirks = quirks;
4558         ctrl->numa_node = NUMA_NO_NODE;
4559         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4560         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4561         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4562         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4563         init_waitqueue_head(&ctrl->state_wq);
4564
4565         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4566         INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4567         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4568         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4569
4570         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4571                         PAGE_SIZE);
4572         ctrl->discard_page = alloc_page(GFP_KERNEL);
4573         if (!ctrl->discard_page) {
4574                 ret = -ENOMEM;
4575                 goto out;
4576         }
4577
4578         ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4579         if (ret < 0)
4580                 goto out;
4581         ctrl->instance = ret;
4582
4583         device_initialize(&ctrl->ctrl_device);
4584         ctrl->device = &ctrl->ctrl_device;
4585         ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4586                         ctrl->instance);
4587         ctrl->device->class = nvme_class;
4588         ctrl->device->parent = ctrl->dev;
4589         ctrl->device->groups = nvme_dev_attr_groups;
4590         ctrl->device->release = nvme_free_ctrl;
4591         dev_set_drvdata(ctrl->device, ctrl);
4592         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4593         if (ret)
4594                 goto out_release_instance;
4595
4596         nvme_get_ctrl(ctrl);
4597         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4598         ctrl->cdev.owner = ops->module;
4599         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4600         if (ret)
4601                 goto out_free_name;
4602
4603         /*
4604          * Initialize latency tolerance controls.  The sysfs files won't
4605          * be visible to userspace unless the device actually supports APST.
4606          */
4607         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4608         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4609                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4610
4611         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4612         nvme_mpath_init_ctrl(ctrl);
4613
4614         return 0;
4615 out_free_name:
4616         nvme_put_ctrl(ctrl);
4617         kfree_const(ctrl->device->kobj.name);
4618 out_release_instance:
4619         ida_free(&nvme_instance_ida, ctrl->instance);
4620 out:
4621         if (ctrl->discard_page)
4622                 __free_page(ctrl->discard_page);
4623         return ret;
4624 }
4625 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4626
4627 static void nvme_start_ns_queue(struct nvme_ns *ns)
4628 {
4629         if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4630                 blk_mq_unquiesce_queue(ns->queue);
4631 }
4632
4633 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4634 {
4635         if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4636                 blk_mq_quiesce_queue(ns->queue);
4637         else
4638                 blk_mq_wait_quiesce_done(ns->queue);
4639 }
4640
4641 /*
4642  * Prepare a queue for teardown.
4643  *
4644  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4645  * the capacity to 0 after that to avoid blocking dispatchers that may be
4646  * holding bd_butex.  This will end buffered writers dirtying pages that can't
4647  * be synced.
4648  */
4649 static void nvme_set_queue_dying(struct nvme_ns *ns)
4650 {
4651         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4652                 return;
4653
4654         blk_mark_disk_dead(ns->disk);
4655         nvme_start_ns_queue(ns);
4656
4657         set_capacity_and_notify(ns->disk, 0);
4658 }
4659
4660 /**
4661  * nvme_kill_queues(): Ends all namespace queues
4662  * @ctrl: the dead controller that needs to end
4663  *
4664  * Call this function when the driver determines it is unable to get the
4665  * controller in a state capable of servicing IO.
4666  */
4667 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4668 {
4669         struct nvme_ns *ns;
4670
4671         down_read(&ctrl->namespaces_rwsem);
4672
4673         /* Forcibly unquiesce queues to avoid blocking dispatch */
4674         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4675                 nvme_start_admin_queue(ctrl);
4676
4677         list_for_each_entry(ns, &ctrl->namespaces, list)
4678                 nvme_set_queue_dying(ns);
4679
4680         up_read(&ctrl->namespaces_rwsem);
4681 }
4682 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4683
4684 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4685 {
4686         struct nvme_ns *ns;
4687
4688         down_read(&ctrl->namespaces_rwsem);
4689         list_for_each_entry(ns, &ctrl->namespaces, list)
4690                 blk_mq_unfreeze_queue(ns->queue);
4691         up_read(&ctrl->namespaces_rwsem);
4692 }
4693 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4694
4695 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4696 {
4697         struct nvme_ns *ns;
4698
4699         down_read(&ctrl->namespaces_rwsem);
4700         list_for_each_entry(ns, &ctrl->namespaces, list) {
4701                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4702                 if (timeout <= 0)
4703                         break;
4704         }
4705         up_read(&ctrl->namespaces_rwsem);
4706         return timeout;
4707 }
4708 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4709
4710 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4711 {
4712         struct nvme_ns *ns;
4713
4714         down_read(&ctrl->namespaces_rwsem);
4715         list_for_each_entry(ns, &ctrl->namespaces, list)
4716                 blk_mq_freeze_queue_wait(ns->queue);
4717         up_read(&ctrl->namespaces_rwsem);
4718 }
4719 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4720
4721 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4722 {
4723         struct nvme_ns *ns;
4724
4725         down_read(&ctrl->namespaces_rwsem);
4726         list_for_each_entry(ns, &ctrl->namespaces, list)
4727                 blk_freeze_queue_start(ns->queue);
4728         up_read(&ctrl->namespaces_rwsem);
4729 }
4730 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4731
4732 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4733 {
4734         struct nvme_ns *ns;
4735
4736         down_read(&ctrl->namespaces_rwsem);
4737         list_for_each_entry(ns, &ctrl->namespaces, list)
4738                 nvme_stop_ns_queue(ns);
4739         up_read(&ctrl->namespaces_rwsem);
4740 }
4741 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4742
4743 void nvme_start_queues(struct nvme_ctrl *ctrl)
4744 {
4745         struct nvme_ns *ns;
4746
4747         down_read(&ctrl->namespaces_rwsem);
4748         list_for_each_entry(ns, &ctrl->namespaces, list)
4749                 nvme_start_ns_queue(ns);
4750         up_read(&ctrl->namespaces_rwsem);
4751 }
4752 EXPORT_SYMBOL_GPL(nvme_start_queues);
4753
4754 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4755 {
4756         if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4757                 blk_mq_quiesce_queue(ctrl->admin_q);
4758         else
4759                 blk_mq_wait_quiesce_done(ctrl->admin_q);
4760 }
4761 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4762
4763 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4764 {
4765         if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4766                 blk_mq_unquiesce_queue(ctrl->admin_q);
4767 }
4768 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4769
4770 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4771 {
4772         struct nvme_ns *ns;
4773
4774         down_read(&ctrl->namespaces_rwsem);
4775         list_for_each_entry(ns, &ctrl->namespaces, list)
4776                 blk_sync_queue(ns->queue);
4777         up_read(&ctrl->namespaces_rwsem);
4778 }
4779 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4780
4781 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4782 {
4783         nvme_sync_io_queues(ctrl);
4784         if (ctrl->admin_q)
4785                 blk_sync_queue(ctrl->admin_q);
4786 }
4787 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4788
4789 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4790 {
4791         if (file->f_op != &nvme_dev_fops)
4792                 return NULL;
4793         return file->private_data;
4794 }
4795 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4796
4797 /*
4798  * Check we didn't inadvertently grow the command structure sizes:
4799  */
4800 static inline void _nvme_check_size(void)
4801 {
4802         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4803         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4804         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4805         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4806         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4807         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4808         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4809         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4810         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4811         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4812         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4813         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4814         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4815         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4816         BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4817         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4818         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4819         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4820         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4821         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4822         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4823         BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4824 }
4825
4826
4827 static int __init nvme_core_init(void)
4828 {
4829         int result = -ENOMEM;
4830
4831         _nvme_check_size();
4832
4833         nvme_wq = alloc_workqueue("nvme-wq",
4834                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4835         if (!nvme_wq)
4836                 goto out;
4837
4838         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4839                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4840         if (!nvme_reset_wq)
4841                 goto destroy_wq;
4842
4843         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4844                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4845         if (!nvme_delete_wq)
4846                 goto destroy_reset_wq;
4847
4848         result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4849                         NVME_MINORS, "nvme");
4850         if (result < 0)
4851                 goto destroy_delete_wq;
4852
4853         nvme_class = class_create(THIS_MODULE, "nvme");
4854         if (IS_ERR(nvme_class)) {
4855                 result = PTR_ERR(nvme_class);
4856                 goto unregister_chrdev;
4857         }
4858         nvme_class->dev_uevent = nvme_class_uevent;
4859
4860         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4861         if (IS_ERR(nvme_subsys_class)) {
4862                 result = PTR_ERR(nvme_subsys_class);
4863                 goto destroy_class;
4864         }
4865
4866         result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4867                                      "nvme-generic");
4868         if (result < 0)
4869                 goto destroy_subsys_class;
4870
4871         nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4872         if (IS_ERR(nvme_ns_chr_class)) {
4873                 result = PTR_ERR(nvme_ns_chr_class);
4874                 goto unregister_generic_ns;
4875         }
4876
4877         return 0;
4878
4879 unregister_generic_ns:
4880         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4881 destroy_subsys_class:
4882         class_destroy(nvme_subsys_class);
4883 destroy_class:
4884         class_destroy(nvme_class);
4885 unregister_chrdev:
4886         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4887 destroy_delete_wq:
4888         destroy_workqueue(nvme_delete_wq);
4889 destroy_reset_wq:
4890         destroy_workqueue(nvme_reset_wq);
4891 destroy_wq:
4892         destroy_workqueue(nvme_wq);
4893 out:
4894         return result;
4895 }
4896
4897 static void __exit nvme_core_exit(void)
4898 {
4899         class_destroy(nvme_ns_chr_class);
4900         class_destroy(nvme_subsys_class);
4901         class_destroy(nvme_class);
4902         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4903         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4904         destroy_workqueue(nvme_delete_wq);
4905         destroy_workqueue(nvme_reset_wq);
4906         destroy_workqueue(nvme_wq);
4907         ida_destroy(&nvme_ns_chr_minor_ida);
4908         ida_destroy(&nvme_instance_ida);
4909 }
4910
4911 MODULE_LICENSE("GPL");
4912 MODULE_VERSION("1.0");
4913 module_init(nvme_core_init);
4914 module_exit(nvme_core_exit);