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 #include <linux/nvme-auth.h>
  28
  29 #define CREATE_TRACE_POINTS
  30 #include "trace.h"
  31
  32 #define NVME_MINORS             (1U << MINORBITS)
  33
  34 struct nvme_ns_info {
  35         struct nvme_ns_ids ids;
  36         u32 nsid;
  37         __le32 anagrpid;
  38         bool is_shared;
  39         bool is_readonly;
  40         bool is_ready;
  41 };
  42
  43 unsigned int admin_timeout = 60;
  44 module_param(admin_timeout, uint, 0644);
  45 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  46 EXPORT_SYMBOL_GPL(admin_timeout);
  47
  48 unsigned int nvme_io_timeout = 30;
  49 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  50 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  51 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  52
  53 static unsigned char shutdown_timeout = 5;
  54 module_param(shutdown_timeout, byte, 0644);
  55 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  56
  57 static u8 nvme_max_retries = 5;
  58 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  59 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  60
  61 static unsigned long default_ps_max_latency_us = 100000;
  62 module_param(default_ps_max_latency_us, ulong, 0644);
  63 MODULE_PARM_DESC(default_ps_max_latency_us,
  64                  "max power saving latency for new devices; use PM QOS to change per device");
  65
  66 static bool force_apst;
  67 module_param(force_apst, bool, 0644);
  68 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  69
  70 static unsigned long apst_primary_timeout_ms = 100;
  71 module_param(apst_primary_timeout_ms, ulong, 0644);
  72 MODULE_PARM_DESC(apst_primary_timeout_ms,
  73         "primary APST timeout in ms");
  74
  75 static unsigned long apst_secondary_timeout_ms = 2000;
  76 module_param(apst_secondary_timeout_ms, ulong, 0644);
  77 MODULE_PARM_DESC(apst_secondary_timeout_ms,
  78         "secondary APST timeout in ms");
  79
  80 static unsigned long apst_primary_latency_tol_us = 15000;
  81 module_param(apst_primary_latency_tol_us, ulong, 0644);
  82 MODULE_PARM_DESC(apst_primary_latency_tol_us,
  83         "primary APST latency tolerance in us");
  84
  85 static unsigned long apst_secondary_latency_tol_us = 100000;
  86 module_param(apst_secondary_latency_tol_us, ulong, 0644);
  87 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
  88         "secondary APST latency tolerance in us");
  89
  90 /*
  91  * nvme_wq - hosts nvme related works that are not reset or delete
  92  * nvme_reset_wq - hosts nvme reset works
  93  * nvme_delete_wq - hosts nvme delete works
  94  *
  95  * nvme_wq will host works such as scan, aen handling, fw activation,
  96  * keep-alive, periodic reconnects etc. nvme_reset_wq
  97  * runs reset works which also flush works hosted on nvme_wq for
  98  * serialization purposes. nvme_delete_wq host controller deletion
  99  * works which flush reset works for serialization.
 100  */
 101 struct workqueue_struct *nvme_wq;
 102 EXPORT_SYMBOL_GPL(nvme_wq);
 103
 104 struct workqueue_struct *nvme_reset_wq;
 105 EXPORT_SYMBOL_GPL(nvme_reset_wq);
 106
 107 struct workqueue_struct *nvme_delete_wq;
 108 EXPORT_SYMBOL_GPL(nvme_delete_wq);
 109
 110 static LIST_HEAD(nvme_subsystems);
 111 static DEFINE_MUTEX(nvme_subsystems_lock);
 112
 113 static DEFINE_IDA(nvme_instance_ida);
 114 static dev_t nvme_ctrl_base_chr_devt;
 115 static struct class *nvme_class;
 116 static struct class *nvme_subsys_class;
 117
 118 static DEFINE_IDA(nvme_ns_chr_minor_ida);
 119 static dev_t nvme_ns_chr_devt;
 120 static struct class *nvme_ns_chr_class;
 121
 122 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
 123 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
 124                                            unsigned nsid);
 125 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
 126                                    struct nvme_command *cmd);
 127
 128 void nvme_queue_scan(struct nvme_ctrl *ctrl)
 129 {
 130         /*
 131          * Only new queue scan work when admin and IO queues are both alive
 132          */
 133         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 134                 queue_work(nvme_wq, &ctrl->scan_work);
 135 }
 136
 137 /*
 138  * Use this function to proceed with scheduling reset_work for a controller
 139  * that had previously been set to the resetting state. This is intended for
 140  * code paths that can't be interrupted by other reset attempts. A hot removal
 141  * may prevent this from succeeding.
 142  */
 143 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 144 {
 145         if (ctrl->state != NVME_CTRL_RESETTING)
 146                 return -EBUSY;
 147         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 148                 return -EBUSY;
 149         return 0;
 150 }
 151 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 152
 153 static void nvme_failfast_work(struct work_struct *work)
 154 {
 155         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 156                         struct nvme_ctrl, failfast_work);
 157
 158         if (ctrl->state != NVME_CTRL_CONNECTING)
 159                 return;
 160
 161         set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 162         dev_info(ctrl->device, "failfast expired\n");
 163         nvme_kick_requeue_lists(ctrl);
 164 }
 165
 166 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 167 {
 168         if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 169                 return;
 170
 171         schedule_delayed_work(&ctrl->failfast_work,
 172                               ctrl->opts->fast_io_fail_tmo * HZ);
 173 }
 174
 175 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 176 {
 177         if (!ctrl->opts)
 178                 return;
 179
 180         cancel_delayed_work_sync(&ctrl->failfast_work);
 181         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 182 }
 183
 184
 185 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 186 {
 187         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 188                 return -EBUSY;
 189         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 190                 return -EBUSY;
 191         return 0;
 192 }
 193 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 194
 195 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 196 {
 197         int ret;
 198
 199         ret = nvme_reset_ctrl(ctrl);
 200         if (!ret) {
 201                 flush_work(&ctrl->reset_work);
 202                 if (ctrl->state != NVME_CTRL_LIVE)
 203                         ret = -ENETRESET;
 204         }
 205
 206         return ret;
 207 }
 208
 209 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 210 {
 211         dev_info(ctrl->device,
 212                  "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
 213
 214         flush_work(&ctrl->reset_work);
 215         nvme_stop_ctrl(ctrl);
 216         nvme_remove_namespaces(ctrl);
 217         ctrl->ops->delete_ctrl(ctrl);
 218         nvme_uninit_ctrl(ctrl);
 219 }
 220
 221 static void nvme_delete_ctrl_work(struct work_struct *work)
 222 {
 223         struct nvme_ctrl *ctrl =
 224                 container_of(work, struct nvme_ctrl, delete_work);
 225
 226         nvme_do_delete_ctrl(ctrl);
 227 }
 228
 229 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 230 {
 231         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 232                 return -EBUSY;
 233         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 234                 return -EBUSY;
 235         return 0;
 236 }
 237 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 238
 239 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 240 {
 241         /*
 242          * Keep a reference until nvme_do_delete_ctrl() complete,
 243          * since ->delete_ctrl can free the controller.
 244          */
 245         nvme_get_ctrl(ctrl);
 246         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 247                 nvme_do_delete_ctrl(ctrl);
 248         nvme_put_ctrl(ctrl);
 249 }
 250
 251 static blk_status_t nvme_error_status(u16 status)
 252 {
 253         switch (status & 0x7ff) {
 254         case NVME_SC_SUCCESS:
 255                 return BLK_STS_OK;
 256         case NVME_SC_CAP_EXCEEDED:
 257                 return BLK_STS_NOSPC;
 258         case NVME_SC_LBA_RANGE:
 259         case NVME_SC_CMD_INTERRUPTED:
 260         case NVME_SC_NS_NOT_READY:
 261                 return BLK_STS_TARGET;
 262         case NVME_SC_BAD_ATTRIBUTES:
 263         case NVME_SC_ONCS_NOT_SUPPORTED:
 264         case NVME_SC_INVALID_OPCODE:
 265         case NVME_SC_INVALID_FIELD:
 266         case NVME_SC_INVALID_NS:
 267                 return BLK_STS_NOTSUPP;
 268         case NVME_SC_WRITE_FAULT:
 269         case NVME_SC_READ_ERROR:
 270         case NVME_SC_UNWRITTEN_BLOCK:
 271         case NVME_SC_ACCESS_DENIED:
 272         case NVME_SC_READ_ONLY:
 273         case NVME_SC_COMPARE_FAILED:
 274                 return BLK_STS_MEDIUM;
 275         case NVME_SC_GUARD_CHECK:
 276         case NVME_SC_APPTAG_CHECK:
 277         case NVME_SC_REFTAG_CHECK:
 278         case NVME_SC_INVALID_PI:
 279                 return BLK_STS_PROTECTION;
 280         case NVME_SC_RESERVATION_CONFLICT:
 281                 return BLK_STS_NEXUS;
 282         case NVME_SC_HOST_PATH_ERROR:
 283                 return BLK_STS_TRANSPORT;
 284         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 285                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 286         case NVME_SC_ZONE_TOO_MANY_OPEN:
 287                 return BLK_STS_ZONE_OPEN_RESOURCE;
 288         default:
 289                 return BLK_STS_IOERR;
 290         }
 291 }
 292
 293 static void nvme_retry_req(struct request *req)
 294 {
 295         unsigned long delay = 0;
 296         u16 crd;
 297
 298         /* The mask and shift result must be <= 3 */
 299         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 300         if (crd)
 301                 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 302
 303         nvme_req(req)->retries++;
 304         blk_mq_requeue_request(req, false);
 305         blk_mq_delay_kick_requeue_list(req->q, delay);
 306 }
 307
 308 static void nvme_log_error(struct request *req)
 309 {
 310         struct nvme_ns *ns = req->q->queuedata;
 311         struct nvme_request *nr = nvme_req(req);
 312
 313         if (ns) {
 314                 pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
 315                        ns->disk ? ns->disk->disk_name : "?",
 316                        nvme_get_opcode_str(nr->cmd->common.opcode),
 317                        nr->cmd->common.opcode,
 318                        (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
 319                        (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
 320                        nvme_get_error_status_str(nr->status),
 321                        nr->status >> 8 & 7,     /* Status Code Type */
 322                        nr->status & 0xff,       /* Status Code */
 323                        nr->status & NVME_SC_MORE ? "MORE " : "",
 324                        nr->status & NVME_SC_DNR  ? "DNR "  : "");
 325                 return;
 326         }
 327
 328         pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
 329                            dev_name(nr->ctrl->device),
 330                            nvme_get_admin_opcode_str(nr->cmd->common.opcode),
 331                            nr->cmd->common.opcode,
 332                            nvme_get_error_status_str(nr->status),
 333                            nr->status >> 8 & 7, /* Status Code Type */
 334                            nr->status & 0xff,   /* Status Code */
 335                            nr->status & NVME_SC_MORE ? "MORE " : "",
 336                            nr->status & NVME_SC_DNR  ? "DNR "  : "");
 337 }
 338
 339 enum nvme_disposition {
 340         COMPLETE,
 341         RETRY,
 342         FAILOVER,
 343         AUTHENTICATE,
 344 };
 345
 346 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 347 {
 348         if (likely(nvme_req(req)->status == 0))
 349                 return COMPLETE;
 350
 351         if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
 352                 return AUTHENTICATE;
 353
 354         if (blk_noretry_request(req) ||
 355             (nvme_req(req)->status & NVME_SC_DNR) ||
 356             nvme_req(req)->retries >= nvme_max_retries)
 357                 return COMPLETE;
 358
 359         if (req->cmd_flags & REQ_NVME_MPATH) {
 360                 if (nvme_is_path_error(nvme_req(req)->status) ||
 361                     blk_queue_dying(req->q))
 362                         return FAILOVER;
 363         } else {
 364                 if (blk_queue_dying(req->q))
 365                         return COMPLETE;
 366         }
 367
 368         return RETRY;
 369 }
 370
 371 static inline void nvme_end_req_zoned(struct request *req)
 372 {
 373         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 374             req_op(req) == REQ_OP_ZONE_APPEND)
 375                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 376                         le64_to_cpu(nvme_req(req)->result.u64));
 377 }
 378
 379 static inline void nvme_end_req(struct request *req)
 380 {
 381         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 382
 383         if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET)))
 384                 nvme_log_error(req);
 385         nvme_end_req_zoned(req);
 386         nvme_trace_bio_complete(req);
 387         blk_mq_end_request(req, status);
 388 }
 389
 390 void nvme_complete_rq(struct request *req)
 391 {
 392         struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 393
 394         trace_nvme_complete_rq(req);
 395         nvme_cleanup_cmd(req);
 396
 397         if (ctrl->kas)
 398                 ctrl->comp_seen = true;
 399
 400         switch (nvme_decide_disposition(req)) {
 401         case COMPLETE:
 402                 nvme_end_req(req);
 403                 return;
 404         case RETRY:
 405                 nvme_retry_req(req);
 406                 return;
 407         case FAILOVER:
 408                 nvme_failover_req(req);
 409                 return;
 410         case AUTHENTICATE:
 411 #ifdef CONFIG_NVME_AUTH
 412                 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
 413                 nvme_retry_req(req);
 414 #else
 415                 nvme_end_req(req);
 416 #endif
 417                 return;
 418         }
 419 }
 420 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 421
 422 void nvme_complete_batch_req(struct request *req)
 423 {
 424         trace_nvme_complete_rq(req);
 425         nvme_cleanup_cmd(req);
 426         nvme_end_req_zoned(req);
 427 }
 428 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
 429
 430 /*
 431  * Called to unwind from ->queue_rq on a failed command submission so that the
 432  * multipathing code gets called to potentially failover to another path.
 433  * The caller needs to unwind all transport specific resource allocations and
 434  * must return propagate the return value.
 435  */
 436 blk_status_t nvme_host_path_error(struct request *req)
 437 {
 438         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 439         blk_mq_set_request_complete(req);
 440         nvme_complete_rq(req);
 441         return BLK_STS_OK;
 442 }
 443 EXPORT_SYMBOL_GPL(nvme_host_path_error);
 444
 445 bool nvme_cancel_request(struct request *req, void *data)
 446 {
 447         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 448                                 "Cancelling I/O %d", req->tag);
 449
 450         /* don't abort one completed request */
 451         if (blk_mq_request_completed(req))
 452                 return true;
 453
 454         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 455         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 456         blk_mq_complete_request(req);
 457         return true;
 458 }
 459 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 460
 461 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 462 {
 463         if (ctrl->tagset) {
 464                 blk_mq_tagset_busy_iter(ctrl->tagset,
 465                                 nvme_cancel_request, ctrl);
 466                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 467         }
 468 }
 469 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 470
 471 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 472 {
 473         if (ctrl->admin_tagset) {
 474                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 475                                 nvme_cancel_request, ctrl);
 476                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 477         }
 478 }
 479 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 480
 481 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 482                 enum nvme_ctrl_state new_state)
 483 {
 484         enum nvme_ctrl_state old_state;
 485         unsigned long flags;
 486         bool changed = false;
 487
 488         spin_lock_irqsave(&ctrl->lock, flags);
 489
 490         old_state = ctrl->state;
 491         switch (new_state) {
 492         case NVME_CTRL_LIVE:
 493                 switch (old_state) {
 494                 case NVME_CTRL_NEW:
 495                 case NVME_CTRL_RESETTING:
 496                 case NVME_CTRL_CONNECTING:
 497                         changed = true;
 498                         fallthrough;
 499                 default:
 500                         break;
 501                 }
 502                 break;
 503         case NVME_CTRL_RESETTING:
 504                 switch (old_state) {
 505                 case NVME_CTRL_NEW:
 506                 case NVME_CTRL_LIVE:
 507                         changed = true;
 508                         fallthrough;
 509                 default:
 510                         break;
 511                 }
 512                 break;
 513         case NVME_CTRL_CONNECTING:
 514                 switch (old_state) {
 515                 case NVME_CTRL_NEW:
 516                 case NVME_CTRL_RESETTING:
 517                         changed = true;
 518                         fallthrough;
 519                 default:
 520                         break;
 521                 }
 522                 break;
 523         case NVME_CTRL_DELETING:
 524                 switch (old_state) {
 525                 case NVME_CTRL_LIVE:
 526                 case NVME_CTRL_RESETTING:
 527                 case NVME_CTRL_CONNECTING:
 528                         changed = true;
 529                         fallthrough;
 530                 default:
 531                         break;
 532                 }
 533                 break;
 534         case NVME_CTRL_DELETING_NOIO:
 535                 switch (old_state) {
 536                 case NVME_CTRL_DELETING:
 537                 case NVME_CTRL_DEAD:
 538                         changed = true;
 539                         fallthrough;
 540                 default:
 541                         break;
 542                 }
 543                 break;
 544         case NVME_CTRL_DEAD:
 545                 switch (old_state) {
 546                 case NVME_CTRL_DELETING:
 547                         changed = true;
 548                         fallthrough;
 549                 default:
 550                         break;
 551                 }
 552                 break;
 553         default:
 554                 break;
 555         }
 556
 557         if (changed) {
 558                 ctrl->state = new_state;
 559                 wake_up_all(&ctrl->state_wq);
 560         }
 561
 562         spin_unlock_irqrestore(&ctrl->lock, flags);
 563         if (!changed)
 564                 return false;
 565
 566         if (ctrl->state == NVME_CTRL_LIVE) {
 567                 if (old_state == NVME_CTRL_CONNECTING)
 568                         nvme_stop_failfast_work(ctrl);
 569                 nvme_kick_requeue_lists(ctrl);
 570         } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 571                 old_state == NVME_CTRL_RESETTING) {
 572                 nvme_start_failfast_work(ctrl);
 573         }
 574         return changed;
 575 }
 576 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 577
 578 /*
 579  * Returns true for sink states that can't ever transition back to live.
 580  */
 581 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 582 {
 583         switch (ctrl->state) {
 584         case NVME_CTRL_NEW:
 585         case NVME_CTRL_LIVE:
 586         case NVME_CTRL_RESETTING:
 587         case NVME_CTRL_CONNECTING:
 588                 return false;
 589         case NVME_CTRL_DELETING:
 590         case NVME_CTRL_DELETING_NOIO:
 591         case NVME_CTRL_DEAD:
 592                 return true;
 593         default:
 594                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 595                 return true;
 596         }
 597 }
 598
 599 /*
 600  * Waits for the controller state to be resetting, or returns false if it is
 601  * not possible to ever transition to that state.
 602  */
 603 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 604 {
 605         wait_event(ctrl->state_wq,
 606                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 607                    nvme_state_terminal(ctrl));
 608         return ctrl->state == NVME_CTRL_RESETTING;
 609 }
 610 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 611
 612 static void nvme_free_ns_head(struct kref *ref)
 613 {
 614         struct nvme_ns_head *head =
 615                 container_of(ref, struct nvme_ns_head, ref);
 616
 617         nvme_mpath_remove_disk(head);
 618         ida_free(&head->subsys->ns_ida, head->instance);
 619         cleanup_srcu_struct(&head->srcu);
 620         nvme_put_subsystem(head->subsys);
 621         kfree(head);
 622 }
 623
 624 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
 625 {
 626         return kref_get_unless_zero(&head->ref);
 627 }
 628
 629 void nvme_put_ns_head(struct nvme_ns_head *head)
 630 {
 631         kref_put(&head->ref, nvme_free_ns_head);
 632 }
 633
 634 static void nvme_free_ns(struct kref *kref)
 635 {
 636         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 637
 638         put_disk(ns->disk);
 639         nvme_put_ns_head(ns->head);
 640         nvme_put_ctrl(ns->ctrl);
 641         kfree(ns);
 642 }
 643
 644 static inline bool nvme_get_ns(struct nvme_ns *ns)
 645 {
 646         return kref_get_unless_zero(&ns->kref);
 647 }
 648
 649 void nvme_put_ns(struct nvme_ns *ns)
 650 {
 651         kref_put(&ns->kref, nvme_free_ns);
 652 }
 653 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 654
 655 static inline void nvme_clear_nvme_request(struct request *req)
 656 {
 657         nvme_req(req)->status = 0;
 658         nvme_req(req)->retries = 0;
 659         nvme_req(req)->flags = 0;
 660         req->rq_flags |= RQF_DONTPREP;
 661 }
 662
 663 /* initialize a passthrough request */
 664 void nvme_init_request(struct request *req, struct nvme_command *cmd)
 665 {
 666         if (req->q->queuedata)
 667                 req->timeout = NVME_IO_TIMEOUT;
 668         else /* no queuedata implies admin queue */
 669                 req->timeout = NVME_ADMIN_TIMEOUT;
 670
 671         /* passthru commands should let the driver set the SGL flags */
 672         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 673
 674         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 675         if (req->mq_hctx->type == HCTX_TYPE_POLL)
 676                 req->cmd_flags |= REQ_POLLED;
 677         nvme_clear_nvme_request(req);
 678         memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
 679 }
 680 EXPORT_SYMBOL_GPL(nvme_init_request);
 681
 682 /*
 683  * For something we're not in a state to send to the device the default action
 684  * is to busy it and retry it after the controller state is recovered.  However,
 685  * if the controller is deleting or if anything is marked for failfast or
 686  * nvme multipath it is immediately failed.
 687  *
 688  * Note: commands used to initialize the controller will be marked for failfast.
 689  * Note: nvme cli/ioctl commands are marked for failfast.
 690  */
 691 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
 692                 struct request *rq)
 693 {
 694         if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
 695             ctrl->state != NVME_CTRL_DELETING &&
 696             ctrl->state != NVME_CTRL_DEAD &&
 697             !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
 698             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 699                 return BLK_STS_RESOURCE;
 700         return nvme_host_path_error(rq);
 701 }
 702 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
 703
 704 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 705                 bool queue_live)
 706 {
 707         struct nvme_request *req = nvme_req(rq);
 708
 709         /*
 710          * currently we have a problem sending passthru commands
 711          * on the admin_q if the controller is not LIVE because we can't
 712          * make sure that they are going out after the admin connect,
 713          * controller enable and/or other commands in the initialization
 714          * sequence. until the controller will be LIVE, fail with
 715          * BLK_STS_RESOURCE so that they will be rescheduled.
 716          */
 717         if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
 718                 return false;
 719
 720         if (ctrl->ops->flags & NVME_F_FABRICS) {
 721                 /*
 722                  * Only allow commands on a live queue, except for the connect
 723                  * command, which is require to set the queue live in the
 724                  * appropinquate states.
 725                  */
 726                 switch (ctrl->state) {
 727                 case NVME_CTRL_CONNECTING:
 728                         if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
 729                             (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
 730                              req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
 731                              req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
 732                                 return true;
 733                         break;
 734                 default:
 735                         break;
 736                 case NVME_CTRL_DEAD:
 737                         return false;
 738                 }
 739         }
 740
 741         return queue_live;
 742 }
 743 EXPORT_SYMBOL_GPL(__nvme_check_ready);
 744
 745 static inline void nvme_setup_flush(struct nvme_ns *ns,
 746                 struct nvme_command *cmnd)
 747 {
 748         memset(cmnd, 0, sizeof(*cmnd));
 749         cmnd->common.opcode = nvme_cmd_flush;
 750         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 751 }
 752
 753 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 754                 struct nvme_command *cmnd)
 755 {
 756         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 757         struct nvme_dsm_range *range;
 758         struct bio *bio;
 759
 760         /*
 761          * Some devices do not consider the DSM 'Number of Ranges' field when
 762          * determining how much data to DMA. Always allocate memory for maximum
 763          * number of segments to prevent device reading beyond end of buffer.
 764          */
 765         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 766
 767         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 768         if (!range) {
 769                 /*
 770                  * If we fail allocation our range, fallback to the controller
 771                  * discard page. If that's also busy, it's safe to return
 772                  * busy, as we know we can make progress once that's freed.
 773                  */
 774                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 775                         return BLK_STS_RESOURCE;
 776
 777                 range = page_address(ns->ctrl->discard_page);
 778         }
 779
 780         __rq_for_each_bio(bio, req) {
 781                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 782                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 783
 784                 if (n < segments) {
 785                         range[n].cattr = cpu_to_le32(0);
 786                         range[n].nlb = cpu_to_le32(nlb);
 787                         range[n].slba = cpu_to_le64(slba);
 788                 }
 789                 n++;
 790         }
 791
 792         if (WARN_ON_ONCE(n != segments)) {
 793                 if (virt_to_page(range) == ns->ctrl->discard_page)
 794                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 795                 else
 796                         kfree(range);
 797                 return BLK_STS_IOERR;
 798         }
 799
 800         memset(cmnd, 0, sizeof(*cmnd));
 801         cmnd->dsm.opcode = nvme_cmd_dsm;
 802         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 803         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 804         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 805
 806         req->special_vec.bv_page = virt_to_page(range);
 807         req->special_vec.bv_offset = offset_in_page(range);
 808         req->special_vec.bv_len = alloc_size;
 809         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 810
 811         return BLK_STS_OK;
 812 }
 813
 814 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
 815                               struct request *req)
 816 {
 817         u32 upper, lower;
 818         u64 ref48;
 819
 820         /* both rw and write zeroes share the same reftag format */
 821         switch (ns->guard_type) {
 822         case NVME_NVM_NS_16B_GUARD:
 823                 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 824                 break;
 825         case NVME_NVM_NS_64B_GUARD:
 826                 ref48 = ext_pi_ref_tag(req);
 827                 lower = lower_32_bits(ref48);
 828                 upper = upper_32_bits(ref48);
 829
 830                 cmnd->rw.reftag = cpu_to_le32(lower);
 831                 cmnd->rw.cdw3 = cpu_to_le32(upper);
 832                 break;
 833         default:
 834                 break;
 835         }
 836 }
 837
 838 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 839                 struct request *req, struct nvme_command *cmnd)
 840 {
 841         memset(cmnd, 0, sizeof(*cmnd));
 842
 843         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 844                 return nvme_setup_discard(ns, req, cmnd);
 845
 846         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 847         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 848         cmnd->write_zeroes.slba =
 849                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 850         cmnd->write_zeroes.length =
 851                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 852
 853         if (nvme_ns_has_pi(ns)) {
 854                 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
 855
 856                 switch (ns->pi_type) {
 857                 case NVME_NS_DPS_PI_TYPE1:
 858                 case NVME_NS_DPS_PI_TYPE2:
 859                         nvme_set_ref_tag(ns, cmnd, req);
 860                         break;
 861                 }
 862         }
 863
 864         return BLK_STS_OK;
 865 }
 866
 867 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 868                 struct request *req, struct nvme_command *cmnd,
 869                 enum nvme_opcode op)
 870 {
 871         u16 control = 0;
 872         u32 dsmgmt = 0;
 873
 874         if (req->cmd_flags & REQ_FUA)
 875                 control |= NVME_RW_FUA;
 876         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 877                 control |= NVME_RW_LR;
 878
 879         if (req->cmd_flags & REQ_RAHEAD)
 880                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 881
 882         cmnd->rw.opcode = op;
 883         cmnd->rw.flags = 0;
 884         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 885         cmnd->rw.cdw2 = 0;
 886         cmnd->rw.cdw3 = 0;
 887         cmnd->rw.metadata = 0;
 888         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 889         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 890         cmnd->rw.reftag = 0;
 891         cmnd->rw.apptag = 0;
 892         cmnd->rw.appmask = 0;
 893
 894         if (ns->ms) {
 895                 /*
 896                  * If formated with metadata, the block layer always provides a
 897                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 898                  * we enable the PRACT bit for protection information or set the
 899                  * namespace capacity to zero to prevent any I/O.
 900                  */
 901                 if (!blk_integrity_rq(req)) {
 902                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 903                                 return BLK_STS_NOTSUPP;
 904                         control |= NVME_RW_PRINFO_PRACT;
 905                 }
 906
 907                 switch (ns->pi_type) {
 908                 case NVME_NS_DPS_PI_TYPE3:
 909                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 910                         break;
 911                 case NVME_NS_DPS_PI_TYPE1:
 912                 case NVME_NS_DPS_PI_TYPE2:
 913                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 914                                         NVME_RW_PRINFO_PRCHK_REF;
 915                         if (op == nvme_cmd_zone_append)
 916                                 control |= NVME_RW_APPEND_PIREMAP;
 917                         nvme_set_ref_tag(ns, cmnd, req);
 918                         break;
 919                 }
 920         }
 921
 922         cmnd->rw.control = cpu_to_le16(control);
 923         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 924         return 0;
 925 }
 926
 927 void nvme_cleanup_cmd(struct request *req)
 928 {
 929         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 930                 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 931
 932                 if (req->special_vec.bv_page == ctrl->discard_page)
 933                         clear_bit_unlock(0, &ctrl->discard_page_busy);
 934                 else
 935                         kfree(bvec_virt(&req->special_vec));
 936         }
 937 }
 938 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 939
 940 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
 941 {
 942         struct nvme_command *cmd = nvme_req(req)->cmd;
 943         blk_status_t ret = BLK_STS_OK;
 944
 945         if (!(req->rq_flags & RQF_DONTPREP))
 946                 nvme_clear_nvme_request(req);
 947
 948         switch (req_op(req)) {
 949         case REQ_OP_DRV_IN:
 950         case REQ_OP_DRV_OUT:
 951                 /* these are setup prior to execution in nvme_init_request() */
 952                 break;
 953         case REQ_OP_FLUSH:
 954                 nvme_setup_flush(ns, cmd);
 955                 break;
 956         case REQ_OP_ZONE_RESET_ALL:
 957         case REQ_OP_ZONE_RESET:
 958                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 959                 break;
 960         case REQ_OP_ZONE_OPEN:
 961                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
 962                 break;
 963         case REQ_OP_ZONE_CLOSE:
 964                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
 965                 break;
 966         case REQ_OP_ZONE_FINISH:
 967                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
 968                 break;
 969         case REQ_OP_WRITE_ZEROES:
 970                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 971                 break;
 972         case REQ_OP_DISCARD:
 973                 ret = nvme_setup_discard(ns, req, cmd);
 974                 break;
 975         case REQ_OP_READ:
 976                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
 977                 break;
 978         case REQ_OP_WRITE:
 979                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
 980                 break;
 981         case REQ_OP_ZONE_APPEND:
 982                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
 983                 break;
 984         default:
 985                 WARN_ON_ONCE(1);
 986                 return BLK_STS_IOERR;
 987         }
 988
 989         cmd->common.command_id = nvme_cid(req);
 990         trace_nvme_setup_cmd(req, cmd);
 991         return ret;
 992 }
 993 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 994
 995 /*
 996  * Return values:
 997  * 0:  success
 998  * >0: nvme controller's cqe status response
 999  * <0: kernel error in lieu of controller response
1000  */
1001 static int nvme_execute_rq(struct request *rq, bool at_head)
1002 {
1003         blk_status_t status;
1004
1005         status = blk_execute_rq(rq, at_head);
1006         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1007                 return -EINTR;
1008         if (nvme_req(rq)->status)
1009                 return nvme_req(rq)->status;
1010         return blk_status_to_errno(status);
1011 }
1012
1013 /*
1014  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1015  * if the result is positive, it's an NVM Express status code
1016  */
1017 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1018                 union nvme_result *result, void *buffer, unsigned bufflen,
1019                 int qid, int at_head, blk_mq_req_flags_t flags)
1020 {
1021         struct request *req;
1022         int ret;
1023
1024         if (qid == NVME_QID_ANY)
1025                 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1026         else
1027                 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1028                                                 qid - 1);
1029
1030         if (IS_ERR(req))
1031                 return PTR_ERR(req);
1032         nvme_init_request(req, cmd);
1033
1034         if (buffer && bufflen) {
1035                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1036                 if (ret)
1037                         goto out;
1038         }
1039
1040         req->rq_flags |= RQF_QUIET;
1041         ret = nvme_execute_rq(req, at_head);
1042         if (result && ret >= 0)
1043                 *result = nvme_req(req)->result;
1044  out:
1045         blk_mq_free_request(req);
1046         return ret;
1047 }
1048 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1049
1050 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1051                 void *buffer, unsigned bufflen)
1052 {
1053         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1054                         NVME_QID_ANY, 0, 0);
1055 }
1056 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1057
1058 static u32 nvme_known_admin_effects(u8 opcode)
1059 {
1060         switch (opcode) {
1061         case nvme_admin_format_nvm:
1062                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1063                         NVME_CMD_EFFECTS_CSE_MASK;
1064         case nvme_admin_sanitize_nvm:
1065                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1066         default:
1067                 break;
1068         }
1069         return 0;
1070 }
1071
1072 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1073 {
1074         u32 effects = 0;
1075
1076         if (ns) {
1077                 if (ns->head->effects)
1078                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1079                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1080                         dev_warn_once(ctrl->device,
1081                                 "IO command:%02x has unhandled effects:%08x\n",
1082                                 opcode, effects);
1083                 return 0;
1084         }
1085
1086         if (ctrl->effects)
1087                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1088         effects |= nvme_known_admin_effects(opcode);
1089
1090         return effects;
1091 }
1092 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1093
1094 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1095                                u8 opcode)
1096 {
1097         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1098
1099         /*
1100          * For simplicity, IO to all namespaces is quiesced even if the command
1101          * effects say only one namespace is affected.
1102          */
1103         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1104                 mutex_lock(&ctrl->scan_lock);
1105                 mutex_lock(&ctrl->subsys->lock);
1106                 nvme_mpath_start_freeze(ctrl->subsys);
1107                 nvme_mpath_wait_freeze(ctrl->subsys);
1108                 nvme_start_freeze(ctrl);
1109                 nvme_wait_freeze(ctrl);
1110         }
1111         return effects;
1112 }
1113
1114 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1115                               struct nvme_command *cmd, int status)
1116 {
1117         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1118                 nvme_unfreeze(ctrl);
1119                 nvme_mpath_unfreeze(ctrl->subsys);
1120                 mutex_unlock(&ctrl->subsys->lock);
1121                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1122                 mutex_unlock(&ctrl->scan_lock);
1123         }
1124         if (effects & NVME_CMD_EFFECTS_CCC)
1125                 nvme_init_ctrl_finish(ctrl);
1126         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1127                 nvme_queue_scan(ctrl);
1128                 flush_work(&ctrl->scan_work);
1129         }
1130
1131         switch (cmd->common.opcode) {
1132         case nvme_admin_set_features:
1133                 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1134                 case NVME_FEAT_KATO:
1135                         /*
1136                          * Keep alive commands interval on the host should be
1137                          * updated when KATO is modified by Set Features
1138                          * commands.
1139                          */
1140                         if (!status)
1141                                 nvme_update_keep_alive(ctrl, cmd);
1142                         break;
1143                 default:
1144                         break;
1145                 }
1146                 break;
1147         default:
1148                 break;
1149         }
1150 }
1151
1152 int nvme_execute_passthru_rq(struct request *rq)
1153 {
1154         struct nvme_command *cmd = nvme_req(rq)->cmd;
1155         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1156         struct nvme_ns *ns = rq->q->queuedata;
1157         u32 effects;
1158         int  ret;
1159
1160         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1161         ret = nvme_execute_rq(rq, false);
1162         if (effects) /* nothing to be done for zero cmd effects */
1163                 nvme_passthru_end(ctrl, effects, cmd, ret);
1164
1165         return ret;
1166 }
1167 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1168
1169 /*
1170  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1171  *
1172  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1173  *   accounting for transport roundtrip times [..].
1174  */
1175 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1176 {
1177         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1178 }
1179
1180 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1181 {
1182         struct nvme_ctrl *ctrl = rq->end_io_data;
1183         unsigned long flags;
1184         bool startka = false;
1185
1186         blk_mq_free_request(rq);
1187
1188         if (status) {
1189                 dev_err(ctrl->device,
1190                         "failed nvme_keep_alive_end_io error=%d\n",
1191                                 status);
1192                 return;
1193         }
1194
1195         ctrl->comp_seen = false;
1196         spin_lock_irqsave(&ctrl->lock, flags);
1197         if (ctrl->state == NVME_CTRL_LIVE ||
1198             ctrl->state == NVME_CTRL_CONNECTING)
1199                 startka = true;
1200         spin_unlock_irqrestore(&ctrl->lock, flags);
1201         if (startka)
1202                 nvme_queue_keep_alive_work(ctrl);
1203 }
1204
1205 static void nvme_keep_alive_work(struct work_struct *work)
1206 {
1207         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1208                         struct nvme_ctrl, ka_work);
1209         bool comp_seen = ctrl->comp_seen;
1210         struct request *rq;
1211
1212         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1213                 dev_dbg(ctrl->device,
1214                         "reschedule traffic based keep-alive timer\n");
1215                 ctrl->comp_seen = false;
1216                 nvme_queue_keep_alive_work(ctrl);
1217                 return;
1218         }
1219
1220         rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1221                                   BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1222         if (IS_ERR(rq)) {
1223                 /* allocation failure, reset the controller */
1224                 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1225                 nvme_reset_ctrl(ctrl);
1226                 return;
1227         }
1228         nvme_init_request(rq, &ctrl->ka_cmd);
1229
1230         rq->timeout = ctrl->kato * HZ;
1231         rq->end_io = nvme_keep_alive_end_io;
1232         rq->end_io_data = ctrl;
1233         rq->rq_flags |= RQF_QUIET;
1234         blk_execute_rq_nowait(rq, false);
1235 }
1236
1237 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1238 {
1239         if (unlikely(ctrl->kato == 0))
1240                 return;
1241
1242         nvme_queue_keep_alive_work(ctrl);
1243 }
1244
1245 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1246 {
1247         if (unlikely(ctrl->kato == 0))
1248                 return;
1249
1250         cancel_delayed_work_sync(&ctrl->ka_work);
1251 }
1252 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1253
1254 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1255                                    struct nvme_command *cmd)
1256 {
1257         unsigned int new_kato =
1258                 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1259
1260         dev_info(ctrl->device,
1261                  "keep alive interval updated from %u ms to %u ms\n",
1262                  ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1263
1264         nvme_stop_keep_alive(ctrl);
1265         ctrl->kato = new_kato;
1266         nvme_start_keep_alive(ctrl);
1267 }
1268
1269 /*
1270  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1271  * flag, thus sending any new CNS opcodes has a big chance of not working.
1272  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1273  * (but not for any later version).
1274  */
1275 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1276 {
1277         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1278                 return ctrl->vs < NVME_VS(1, 2, 0);
1279         return ctrl->vs < NVME_VS(1, 1, 0);
1280 }
1281
1282 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1283 {
1284         struct nvme_command c = { };
1285         int error;
1286
1287         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1288         c.identify.opcode = nvme_admin_identify;
1289         c.identify.cns = NVME_ID_CNS_CTRL;
1290
1291         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1292         if (!*id)
1293                 return -ENOMEM;
1294
1295         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1296                         sizeof(struct nvme_id_ctrl));
1297         if (error)
1298                 kfree(*id);
1299         return error;
1300 }
1301
1302 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1303                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1304 {
1305         const char *warn_str = "ctrl returned bogus length:";
1306         void *data = cur;
1307
1308         switch (cur->nidt) {
1309         case NVME_NIDT_EUI64:
1310                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1311                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1312                                  warn_str, cur->nidl);
1313                         return -1;
1314                 }
1315                 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1316                         return NVME_NIDT_EUI64_LEN;
1317                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1318                 return NVME_NIDT_EUI64_LEN;
1319         case NVME_NIDT_NGUID:
1320                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1321                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1322                                  warn_str, cur->nidl);
1323                         return -1;
1324                 }
1325                 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1326                         return NVME_NIDT_NGUID_LEN;
1327                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1328                 return NVME_NIDT_NGUID_LEN;
1329         case NVME_NIDT_UUID:
1330                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1331                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1332                                  warn_str, cur->nidl);
1333                         return -1;
1334                 }
1335                 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1336                         return NVME_NIDT_UUID_LEN;
1337                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1338                 return NVME_NIDT_UUID_LEN;
1339         case NVME_NIDT_CSI:
1340                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1341                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1342                                  warn_str, cur->nidl);
1343                         return -1;
1344                 }
1345                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1346                 *csi_seen = true;
1347                 return NVME_NIDT_CSI_LEN;
1348         default:
1349                 /* Skip unknown types */
1350                 return cur->nidl;
1351         }
1352 }
1353
1354 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1355                 struct nvme_ns_info *info)
1356 {
1357         struct nvme_command c = { };
1358         bool csi_seen = false;
1359         int status, pos, len;
1360         void *data;
1361
1362         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1363                 return 0;
1364         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1365                 return 0;
1366
1367         c.identify.opcode = nvme_admin_identify;
1368         c.identify.nsid = cpu_to_le32(info->nsid);
1369         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1370
1371         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1372         if (!data)
1373                 return -ENOMEM;
1374
1375         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1376                                       NVME_IDENTIFY_DATA_SIZE);
1377         if (status) {
1378                 dev_warn(ctrl->device,
1379                         "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1380                         info->nsid, status);
1381                 goto free_data;
1382         }
1383
1384         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1385                 struct nvme_ns_id_desc *cur = data + pos;
1386
1387                 if (cur->nidl == 0)
1388                         break;
1389
1390                 len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1391                 if (len < 0)
1392                         break;
1393
1394                 len += sizeof(*cur);
1395         }
1396
1397         if (nvme_multi_css(ctrl) && !csi_seen) {
1398                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1399                          info->nsid);
1400                 status = -EINVAL;
1401         }
1402
1403 free_data:
1404         kfree(data);
1405         return status;
1406 }
1407
1408 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1409                         struct nvme_id_ns **id)
1410 {
1411         struct nvme_command c = { };
1412         int error;
1413
1414         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1415         c.identify.opcode = nvme_admin_identify;
1416         c.identify.nsid = cpu_to_le32(nsid);
1417         c.identify.cns = NVME_ID_CNS_NS;
1418
1419         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1420         if (!*id)
1421                 return -ENOMEM;
1422
1423         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1424         if (error) {
1425                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1426                 goto out_free_id;
1427         }
1428
1429         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1430         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1431                 goto out_free_id;
1432         return 0;
1433
1434 out_free_id:
1435         kfree(*id);
1436         return error;
1437 }
1438
1439 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1440                 struct nvme_ns_info *info)
1441 {
1442         struct nvme_ns_ids *ids = &info->ids;
1443         struct nvme_id_ns *id;
1444         int ret;
1445
1446         ret = nvme_identify_ns(ctrl, info->nsid, &id);
1447         if (ret)
1448                 return ret;
1449         info->anagrpid = id->anagrpid;
1450         info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1451         info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1452         info->is_ready = true;
1453         if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1454                 dev_info(ctrl->device,
1455                          "Ignoring bogus Namespace Identifiers\n");
1456         } else {
1457                 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1458                     !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1459                         memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1460                 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1461                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1462                         memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1463         }
1464         kfree(id);
1465         return 0;
1466 }
1467
1468 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1469                 struct nvme_ns_info *info)
1470 {
1471         struct nvme_id_ns_cs_indep *id;
1472         struct nvme_command c = {
1473                 .identify.opcode        = nvme_admin_identify,
1474                 .identify.nsid          = cpu_to_le32(info->nsid),
1475                 .identify.cns           = NVME_ID_CNS_NS_CS_INDEP,
1476         };
1477         int ret;
1478
1479         id = kmalloc(sizeof(*id), GFP_KERNEL);
1480         if (!id)
1481                 return -ENOMEM;
1482
1483         ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1484         if (!ret) {
1485                 info->anagrpid = id->anagrpid;
1486                 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1487                 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1488                 info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1489         }
1490         kfree(id);
1491         return ret;
1492 }
1493
1494 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1495                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1496 {
1497         union nvme_result res = { 0 };
1498         struct nvme_command c = { };
1499         int ret;
1500
1501         c.features.opcode = op;
1502         c.features.fid = cpu_to_le32(fid);
1503         c.features.dword11 = cpu_to_le32(dword11);
1504
1505         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1506                         buffer, buflen, NVME_QID_ANY, 0, 0);
1507         if (ret >= 0 && result)
1508                 *result = le32_to_cpu(res.u32);
1509         return ret;
1510 }
1511
1512 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1513                       unsigned int dword11, void *buffer, size_t buflen,
1514                       u32 *result)
1515 {
1516         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1517                              buflen, result);
1518 }
1519 EXPORT_SYMBOL_GPL(nvme_set_features);
1520
1521 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1522                       unsigned int dword11, void *buffer, size_t buflen,
1523                       u32 *result)
1524 {
1525         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1526                              buflen, result);
1527 }
1528 EXPORT_SYMBOL_GPL(nvme_get_features);
1529
1530 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1531 {
1532         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1533         u32 result;
1534         int status, nr_io_queues;
1535
1536         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1537                         &result);
1538         if (status < 0)
1539                 return status;
1540
1541         /*
1542          * Degraded controllers might return an error when setting the queue
1543          * count.  We still want to be able to bring them online and offer
1544          * access to the admin queue, as that might be only way to fix them up.
1545          */
1546         if (status > 0) {
1547                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1548                 *count = 0;
1549         } else {
1550                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1551                 *count = min(*count, nr_io_queues);
1552         }
1553
1554         return 0;
1555 }
1556 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1557
1558 #define NVME_AEN_SUPPORTED \
1559         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1560          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1561
1562 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1563 {
1564         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1565         int status;
1566
1567         if (!supported_aens)
1568                 return;
1569
1570         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1571                         NULL, 0, &result);
1572         if (status)
1573                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1574                          supported_aens);
1575
1576         queue_work(nvme_wq, &ctrl->async_event_work);
1577 }
1578
1579 static int nvme_ns_open(struct nvme_ns *ns)
1580 {
1581
1582         /* should never be called due to GENHD_FL_HIDDEN */
1583         if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1584                 goto fail;
1585         if (!nvme_get_ns(ns))
1586                 goto fail;
1587         if (!try_module_get(ns->ctrl->ops->module))
1588                 goto fail_put_ns;
1589
1590         return 0;
1591
1592 fail_put_ns:
1593         nvme_put_ns(ns);
1594 fail:
1595         return -ENXIO;
1596 }
1597
1598 static void nvme_ns_release(struct nvme_ns *ns)
1599 {
1600
1601         module_put(ns->ctrl->ops->module);
1602         nvme_put_ns(ns);
1603 }
1604
1605 static int nvme_open(struct block_device *bdev, fmode_t mode)
1606 {
1607         return nvme_ns_open(bdev->bd_disk->private_data);
1608 }
1609
1610 static void nvme_release(struct gendisk *disk, fmode_t mode)
1611 {
1612         nvme_ns_release(disk->private_data);
1613 }
1614
1615 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1616 {
1617         /* some standard values */
1618         geo->heads = 1 << 6;
1619         geo->sectors = 1 << 5;
1620         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1621         return 0;
1622 }
1623
1624 #ifdef CONFIG_BLK_DEV_INTEGRITY
1625 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1626                                 u32 max_integrity_segments)
1627 {
1628         struct blk_integrity integrity = { };
1629
1630         switch (ns->pi_type) {
1631         case NVME_NS_DPS_PI_TYPE3:
1632                 switch (ns->guard_type) {
1633                 case NVME_NVM_NS_16B_GUARD:
1634                         integrity.profile = &t10_pi_type3_crc;
1635                         integrity.tag_size = sizeof(u16) + sizeof(u32);
1636                         integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1637                         break;
1638                 case NVME_NVM_NS_64B_GUARD:
1639                         integrity.profile = &ext_pi_type3_crc64;
1640                         integrity.tag_size = sizeof(u16) + 6;
1641                         integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1642                         break;
1643                 default:
1644                         integrity.profile = NULL;
1645                         break;
1646                 }
1647                 break;
1648         case NVME_NS_DPS_PI_TYPE1:
1649         case NVME_NS_DPS_PI_TYPE2:
1650                 switch (ns->guard_type) {
1651                 case NVME_NVM_NS_16B_GUARD:
1652                         integrity.profile = &t10_pi_type1_crc;
1653                         integrity.tag_size = sizeof(u16);
1654                         integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1655                         break;
1656                 case NVME_NVM_NS_64B_GUARD:
1657                         integrity.profile = &ext_pi_type1_crc64;
1658                         integrity.tag_size = sizeof(u16);
1659                         integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1660                         break;
1661                 default:
1662                         integrity.profile = NULL;
1663                         break;
1664                 }
1665                 break;
1666         default:
1667                 integrity.profile = NULL;
1668                 break;
1669         }
1670
1671         integrity.tuple_size = ns->ms;
1672         blk_integrity_register(disk, &integrity);
1673         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1674 }
1675 #else
1676 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1677                                 u32 max_integrity_segments)
1678 {
1679 }
1680 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1681
1682 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1683 {
1684         struct nvme_ctrl *ctrl = ns->ctrl;
1685         struct request_queue *queue = disk->queue;
1686         u32 size = queue_logical_block_size(queue);
1687
1688         if (ctrl->max_discard_sectors == 0) {
1689                 blk_queue_max_discard_sectors(queue, 0);
1690                 return;
1691         }
1692
1693         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1694                         NVME_DSM_MAX_RANGES);
1695
1696         queue->limits.discard_granularity = size;
1697
1698         /* If discard is already enabled, don't reset queue limits */
1699         if (queue->limits.max_discard_sectors)
1700                 return;
1701
1702         if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1703                 ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1704
1705         blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1706         blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1707
1708         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1709                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1710 }
1711
1712 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1713 {
1714         return uuid_equal(&a->uuid, &b->uuid) &&
1715                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1716                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1717                 a->csi == b->csi;
1718 }
1719
1720 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1721 {
1722         bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1723         unsigned lbaf = nvme_lbaf_index(id->flbas);
1724         struct nvme_ctrl *ctrl = ns->ctrl;
1725         struct nvme_command c = { };
1726         struct nvme_id_ns_nvm *nvm;
1727         int ret = 0;
1728         u32 elbaf;
1729
1730         ns->pi_size = 0;
1731         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1732         if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1733                 ns->pi_size = sizeof(struct t10_pi_tuple);
1734                 ns->guard_type = NVME_NVM_NS_16B_GUARD;
1735                 goto set_pi;
1736         }
1737
1738         nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1739         if (!nvm)
1740                 return -ENOMEM;
1741
1742         c.identify.opcode = nvme_admin_identify;
1743         c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1744         c.identify.cns = NVME_ID_CNS_CS_NS;
1745         c.identify.csi = NVME_CSI_NVM;
1746
1747         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1748         if (ret)
1749                 goto free_data;
1750
1751         elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1752
1753         /* no support for storage tag formats right now */
1754         if (nvme_elbaf_sts(elbaf))
1755                 goto free_data;
1756
1757         ns->guard_type = nvme_elbaf_guard_type(elbaf);
1758         switch (ns->guard_type) {
1759         case NVME_NVM_NS_64B_GUARD:
1760                 ns->pi_size = sizeof(struct crc64_pi_tuple);
1761                 break;
1762         case NVME_NVM_NS_16B_GUARD:
1763                 ns->pi_size = sizeof(struct t10_pi_tuple);
1764                 break;
1765         default:
1766                 break;
1767         }
1768
1769 free_data:
1770         kfree(nvm);
1771 set_pi:
1772         if (ns->pi_size && (first || ns->ms == ns->pi_size))
1773                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1774         else
1775                 ns->pi_type = 0;
1776
1777         return ret;
1778 }
1779
1780 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1781 {
1782         struct nvme_ctrl *ctrl = ns->ctrl;
1783
1784         if (nvme_init_ms(ns, id))
1785                 return;
1786
1787         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1788         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1789                 return;
1790
1791         if (ctrl->ops->flags & NVME_F_FABRICS) {
1792                 /*
1793                  * The NVMe over Fabrics specification only supports metadata as
1794                  * part of the extended data LBA.  We rely on HCA/HBA support to
1795                  * remap the separate metadata buffer from the block layer.
1796                  */
1797                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1798                         return;
1799
1800                 ns->features |= NVME_NS_EXT_LBAS;
1801
1802                 /*
1803                  * The current fabrics transport drivers support namespace
1804                  * metadata formats only if nvme_ns_has_pi() returns true.
1805                  * Suppress support for all other formats so the namespace will
1806                  * have a 0 capacity and not be usable through the block stack.
1807                  *
1808                  * Note, this check will need to be modified if any drivers
1809                  * gain the ability to use other metadata formats.
1810                  */
1811                 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1812                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1813         } else {
1814                 /*
1815                  * For PCIe controllers, we can't easily remap the separate
1816                  * metadata buffer from the block layer and thus require a
1817                  * separate metadata buffer for block layer metadata/PI support.
1818                  * We allow extended LBAs for the passthrough interface, though.
1819                  */
1820                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1821                         ns->features |= NVME_NS_EXT_LBAS;
1822                 else
1823                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1824         }
1825 }
1826
1827 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1828                 struct request_queue *q)
1829 {
1830         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1831
1832         if (ctrl->max_hw_sectors) {
1833                 u32 max_segments =
1834                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1835
1836                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1837                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1838                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1839         }
1840         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1841         blk_queue_dma_alignment(q, 3);
1842         blk_queue_write_cache(q, vwc, vwc);
1843 }
1844
1845 static void nvme_update_disk_info(struct gendisk *disk,
1846                 struct nvme_ns *ns, struct nvme_id_ns *id)
1847 {
1848         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1849         unsigned short bs = 1 << ns->lba_shift;
1850         u32 atomic_bs, phys_bs, io_opt = 0;
1851
1852         /*
1853          * The block layer can't support LBA sizes larger than the page size
1854          * yet, so catch this early and don't allow block I/O.
1855          */
1856         if (ns->lba_shift > PAGE_SHIFT) {
1857                 capacity = 0;
1858                 bs = (1 << 9);
1859         }
1860
1861         blk_integrity_unregister(disk);
1862
1863         atomic_bs = phys_bs = bs;
1864         if (id->nabo == 0) {
1865                 /*
1866                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1867                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1868                  * 0 then AWUPF must be used instead.
1869                  */
1870                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1871                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1872                 else
1873                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1874         }
1875
1876         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1877                 /* NPWG = Namespace Preferred Write Granularity */
1878                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1879                 /* NOWS = Namespace Optimal Write Size */
1880                 io_opt = bs * (1 + le16_to_cpu(id->nows));
1881         }
1882
1883         blk_queue_logical_block_size(disk->queue, bs);
1884         /*
1885          * Linux filesystems assume writing a single physical block is
1886          * an atomic operation. Hence limit the physical block size to the
1887          * value of the Atomic Write Unit Power Fail parameter.
1888          */
1889         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1890         blk_queue_io_min(disk->queue, phys_bs);
1891         blk_queue_io_opt(disk->queue, io_opt);
1892
1893         /*
1894          * Register a metadata profile for PI, or the plain non-integrity NVMe
1895          * metadata masquerading as Type 0 if supported, otherwise reject block
1896          * I/O to namespaces with metadata except when the namespace supports
1897          * PI, as it can strip/insert in that case.
1898          */
1899         if (ns->ms) {
1900                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1901                     (ns->features & NVME_NS_METADATA_SUPPORTED))
1902                         nvme_init_integrity(disk, ns,
1903                                             ns->ctrl->max_integrity_segments);
1904                 else if (!nvme_ns_has_pi(ns))
1905                         capacity = 0;
1906         }
1907
1908         set_capacity_and_notify(disk, capacity);
1909
1910         nvme_config_discard(disk, ns);
1911         blk_queue_max_write_zeroes_sectors(disk->queue,
1912                                            ns->ctrl->max_zeroes_sectors);
1913 }
1914
1915 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1916 {
1917         return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1918 }
1919
1920 static inline bool nvme_first_scan(struct gendisk *disk)
1921 {
1922         /* nvme_alloc_ns() scans the disk prior to adding it */
1923         return !disk_live(disk);
1924 }
1925
1926 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1927 {
1928         struct nvme_ctrl *ctrl = ns->ctrl;
1929         u32 iob;
1930
1931         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1932             is_power_of_2(ctrl->max_hw_sectors))
1933                 iob = ctrl->max_hw_sectors;
1934         else
1935                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1936
1937         if (!iob)
1938                 return;
1939
1940         if (!is_power_of_2(iob)) {
1941                 if (nvme_first_scan(ns->disk))
1942                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1943                                 ns->disk->disk_name, iob);
1944                 return;
1945         }
1946
1947         if (blk_queue_is_zoned(ns->disk->queue)) {
1948                 if (nvme_first_scan(ns->disk))
1949                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
1950                                 ns->disk->disk_name);
1951                 return;
1952         }
1953
1954         blk_queue_chunk_sectors(ns->queue, iob);
1955 }
1956
1957 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1958                 struct nvme_ns_info *info)
1959 {
1960         blk_mq_freeze_queue(ns->disk->queue);
1961         nvme_set_queue_limits(ns->ctrl, ns->queue);
1962         set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
1963         blk_mq_unfreeze_queue(ns->disk->queue);
1964
1965         if (nvme_ns_head_multipath(ns->head)) {
1966                 blk_mq_freeze_queue(ns->head->disk->queue);
1967                 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
1968                 nvme_mpath_revalidate_paths(ns);
1969                 blk_stack_limits(&ns->head->disk->queue->limits,
1970                                  &ns->queue->limits, 0);
1971                 ns->head->disk->flags |= GENHD_FL_HIDDEN;
1972                 blk_mq_unfreeze_queue(ns->head->disk->queue);
1973         }
1974
1975         /* Hide the block-interface for these devices */
1976         ns->disk->flags |= GENHD_FL_HIDDEN;
1977         set_bit(NVME_NS_READY, &ns->flags);
1978
1979         return 0;
1980 }
1981
1982 static int nvme_update_ns_info_block(struct nvme_ns *ns,
1983                 struct nvme_ns_info *info)
1984 {
1985         struct nvme_id_ns *id;
1986         unsigned lbaf;
1987         int ret;
1988
1989         ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
1990         if (ret)
1991                 return ret;
1992
1993         blk_mq_freeze_queue(ns->disk->queue);
1994         lbaf = nvme_lbaf_index(id->flbas);
1995         ns->lba_shift = id->lbaf[lbaf].ds;
1996         nvme_set_queue_limits(ns->ctrl, ns->queue);
1997
1998         nvme_configure_metadata(ns, id);
1999         nvme_set_chunk_sectors(ns, id);
2000         nvme_update_disk_info(ns->disk, ns, id);
2001
2002         if (ns->head->ids.csi == NVME_CSI_ZNS) {
2003                 ret = nvme_update_zone_info(ns, lbaf);
2004                 if (ret) {
2005                         blk_mq_unfreeze_queue(ns->disk->queue);
2006                         goto out;
2007                 }
2008         }
2009
2010         set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2011         set_bit(NVME_NS_READY, &ns->flags);
2012         blk_mq_unfreeze_queue(ns->disk->queue);
2013
2014         if (blk_queue_is_zoned(ns->queue)) {
2015                 ret = nvme_revalidate_zones(ns);
2016                 if (ret && !nvme_first_scan(ns->disk))
2017                         goto out;
2018         }
2019
2020         if (nvme_ns_head_multipath(ns->head)) {
2021                 blk_mq_freeze_queue(ns->head->disk->queue);
2022                 nvme_update_disk_info(ns->head->disk, ns, id);
2023                 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2024                 nvme_mpath_revalidate_paths(ns);
2025                 blk_stack_limits(&ns->head->disk->queue->limits,
2026                                  &ns->queue->limits, 0);
2027                 disk_update_readahead(ns->head->disk);
2028                 blk_mq_unfreeze_queue(ns->head->disk->queue);
2029         }
2030
2031         ret = 0;
2032 out:
2033         /*
2034          * If probing fails due an unsupported feature, hide the block device,
2035          * but still allow other access.
2036          */
2037         if (ret == -ENODEV) {
2038                 ns->disk->flags |= GENHD_FL_HIDDEN;
2039                 set_bit(NVME_NS_READY, &ns->flags);
2040                 ret = 0;
2041         }
2042         kfree(id);
2043         return ret;
2044 }
2045
2046 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2047 {
2048         switch (info->ids.csi) {
2049         case NVME_CSI_ZNS:
2050                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2051                         dev_info(ns->ctrl->device,
2052         "block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2053                                 info->nsid);
2054                         return nvme_update_ns_info_generic(ns, info);
2055                 }
2056                 return nvme_update_ns_info_block(ns, info);
2057         case NVME_CSI_NVM:
2058                 return nvme_update_ns_info_block(ns, info);
2059         default:
2060                 dev_info(ns->ctrl->device,
2061                         "block device for nsid %u not supported (csi %u)\n",
2062                         info->nsid, info->ids.csi);
2063                 return nvme_update_ns_info_generic(ns, info);
2064         }
2065 }
2066
2067 static char nvme_pr_type(enum pr_type type)
2068 {
2069         switch (type) {
2070         case PR_WRITE_EXCLUSIVE:
2071                 return 1;
2072         case PR_EXCLUSIVE_ACCESS:
2073                 return 2;
2074         case PR_WRITE_EXCLUSIVE_REG_ONLY:
2075                 return 3;
2076         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2077                 return 4;
2078         case PR_WRITE_EXCLUSIVE_ALL_REGS:
2079                 return 5;
2080         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2081                 return 6;
2082         default:
2083                 return 0;
2084         }
2085 }
2086
2087 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
2088                 struct nvme_command *c, u8 data[16])
2089 {
2090         struct nvme_ns_head *head = bdev->bd_disk->private_data;
2091         int srcu_idx = srcu_read_lock(&head->srcu);
2092         struct nvme_ns *ns = nvme_find_path(head);
2093         int ret = -EWOULDBLOCK;
2094
2095         if (ns) {
2096                 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2097                 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
2098         }
2099         srcu_read_unlock(&head->srcu, srcu_idx);
2100         return ret;
2101 }
2102
2103 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2104                 u8 data[16])
2105 {
2106         c->common.nsid = cpu_to_le32(ns->head->ns_id);
2107         return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2108 }
2109
2110 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2111                                 u64 key, u64 sa_key, u8 op)
2112 {
2113         struct nvme_command c = { };
2114         u8 data[16] = { 0, };
2115
2116         put_unaligned_le64(key, &data[0]);
2117         put_unaligned_le64(sa_key, &data[8]);
2118
2119         c.common.opcode = op;
2120         c.common.cdw10 = cpu_to_le32(cdw10);
2121
2122         if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2123             bdev->bd_disk->fops == &nvme_ns_head_ops)
2124                 return nvme_send_ns_head_pr_command(bdev, &c, data);
2125         return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2126 }
2127
2128 static int nvme_pr_register(struct block_device *bdev, u64 old,
2129                 u64 new, unsigned flags)
2130 {
2131         u32 cdw10;
2132
2133         if (flags & ~PR_FL_IGNORE_KEY)
2134                 return -EOPNOTSUPP;
2135
2136         cdw10 = old ? 2 : 0;
2137         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2138         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2139         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2140 }
2141
2142 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2143                 enum pr_type type, unsigned flags)
2144 {
2145         u32 cdw10;
2146
2147         if (flags & ~PR_FL_IGNORE_KEY)
2148                 return -EOPNOTSUPP;
2149
2150         cdw10 = nvme_pr_type(type) << 8;
2151         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2152         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2153 }
2154
2155 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2156                 enum pr_type type, bool abort)
2157 {
2158         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2159
2160         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2161 }
2162
2163 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2164 {
2165         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2166
2167         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2168 }
2169
2170 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2171 {
2172         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2173
2174         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2175 }
2176
2177 const struct pr_ops nvme_pr_ops = {
2178         .pr_register    = nvme_pr_register,
2179         .pr_reserve     = nvme_pr_reserve,
2180         .pr_release     = nvme_pr_release,
2181         .pr_preempt     = nvme_pr_preempt,
2182         .pr_clear       = nvme_pr_clear,
2183 };
2184
2185 #ifdef CONFIG_BLK_SED_OPAL
2186 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2187                 bool send)
2188 {
2189         struct nvme_ctrl *ctrl = data;
2190         struct nvme_command cmd = { };
2191
2192         if (send)
2193                 cmd.common.opcode = nvme_admin_security_send;
2194         else
2195                 cmd.common.opcode = nvme_admin_security_recv;
2196         cmd.common.nsid = 0;
2197         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2198         cmd.common.cdw11 = cpu_to_le32(len);
2199
2200         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2201                         NVME_QID_ANY, 1, 0);
2202 }
2203 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2204 #endif /* CONFIG_BLK_SED_OPAL */
2205
2206 #ifdef CONFIG_BLK_DEV_ZONED
2207 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2208                 unsigned int nr_zones, report_zones_cb cb, void *data)
2209 {
2210         return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2211                         data);
2212 }
2213 #else
2214 #define nvme_report_zones       NULL
2215 #endif /* CONFIG_BLK_DEV_ZONED */
2216
2217 static const struct block_device_operations nvme_bdev_ops = {
2218         .owner          = THIS_MODULE,
2219         .ioctl          = nvme_ioctl,
2220         .compat_ioctl   = blkdev_compat_ptr_ioctl,
2221         .open           = nvme_open,
2222         .release        = nvme_release,
2223         .getgeo         = nvme_getgeo,
2224         .report_zones   = nvme_report_zones,
2225         .pr_ops         = &nvme_pr_ops,
2226 };
2227
2228 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 timeout, bool enabled)
2229 {
2230         unsigned long timeout_jiffies = ((timeout + 1) * HZ / 2) + jiffies;
2231         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2232         int ret;
2233
2234         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2235                 if (csts == ~0)
2236                         return -ENODEV;
2237                 if ((csts & NVME_CSTS_RDY) == bit)
2238                         break;
2239
2240                 usleep_range(1000, 2000);
2241                 if (fatal_signal_pending(current))
2242                         return -EINTR;
2243                 if (time_after(jiffies, timeout_jiffies)) {
2244                         dev_err(ctrl->device,
2245                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2246                                 enabled ? "initialisation" : "reset", csts);
2247                         return -ENODEV;
2248                 }
2249         }
2250
2251         return ret;
2252 }
2253
2254 /*
2255  * If the device has been passed off to us in an enabled state, just clear
2256  * the enabled bit.  The spec says we should set the 'shutdown notification
2257  * bits', but doing so may cause the device to complete commands to the
2258  * admin queue ... and we don't know what memory that might be pointing at!
2259  */
2260 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2261 {
2262         int ret;
2263
2264         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2265         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2266
2267         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2268         if (ret)
2269                 return ret;
2270
2271         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2272                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2273
2274         return nvme_wait_ready(ctrl, NVME_CAP_TIMEOUT(ctrl->cap), false);
2275 }
2276 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2277
2278 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2279 {
2280         unsigned dev_page_min;
2281         u32 timeout;
2282         int ret;
2283
2284         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2285         if (ret) {
2286                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2287                 return ret;
2288         }
2289         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2290
2291         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2292                 dev_err(ctrl->device,
2293                         "Minimum device page size %u too large for host (%u)\n",
2294                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2295                 return -ENODEV;
2296         }
2297
2298         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2299                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2300         else
2301                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2302
2303         if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2304                 u32 crto;
2305
2306                 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2307                 if (ret) {
2308                         dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2309                                 ret);
2310                         return ret;
2311                 }
2312
2313                 if (ctrl->cap & NVME_CAP_CRMS_CRIMS) {
2314                         ctrl->ctrl_config |= NVME_CC_CRIME;
2315                         timeout = NVME_CRTO_CRIMT(crto);
2316                 } else {
2317                         timeout = NVME_CRTO_CRWMT(crto);
2318                 }
2319         } else {
2320                 timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2321         }
2322
2323         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2324         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2325         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2326         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2327         if (ret)
2328                 return ret;
2329
2330         /* Flush write to device (required if transport is PCI) */
2331         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2332         if (ret)
2333                 return ret;
2334
2335         ctrl->ctrl_config |= NVME_CC_ENABLE;
2336         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2337         if (ret)
2338                 return ret;
2339         return nvme_wait_ready(ctrl, timeout, true);
2340 }
2341 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2342
2343 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2344 {
2345         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2346         u32 csts;
2347         int ret;
2348
2349         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2350         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2351
2352         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2353         if (ret)
2354                 return ret;
2355
2356         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2357                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2358                         break;
2359
2360                 msleep(100);
2361                 if (fatal_signal_pending(current))
2362                         return -EINTR;
2363                 if (time_after(jiffies, timeout)) {
2364                         dev_err(ctrl->device,
2365                                 "Device shutdown incomplete; abort shutdown\n");
2366                         return -ENODEV;
2367                 }
2368         }
2369
2370         return ret;
2371 }
2372 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2373
2374 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2375 {
2376         __le64 ts;
2377         int ret;
2378
2379         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2380                 return 0;
2381
2382         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2383         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2384                         NULL);
2385         if (ret)
2386                 dev_warn_once(ctrl->device,
2387                         "could not set timestamp (%d)\n", ret);
2388         return ret;
2389 }
2390
2391 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2392 {
2393         struct nvme_feat_host_behavior *host;
2394         u8 acre = 0, lbafee = 0;
2395         int ret;
2396
2397         /* Don't bother enabling the feature if retry delay is not reported */
2398         if (ctrl->crdt[0])
2399                 acre = NVME_ENABLE_ACRE;
2400         if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2401                 lbafee = NVME_ENABLE_LBAFEE;
2402
2403         if (!acre && !lbafee)
2404                 return 0;
2405
2406         host = kzalloc(sizeof(*host), GFP_KERNEL);
2407         if (!host)
2408                 return 0;
2409
2410         host->acre = acre;
2411         host->lbafee = lbafee;
2412         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2413                                 host, sizeof(*host), NULL);
2414         kfree(host);
2415         return ret;
2416 }
2417
2418 /*
2419  * The function checks whether the given total (exlat + enlat) latency of
2420  * a power state allows the latter to be used as an APST transition target.
2421  * It does so by comparing the latency to the primary and secondary latency
2422  * tolerances defined by module params. If there's a match, the corresponding
2423  * timeout value is returned and the matching tolerance index (1 or 2) is
2424  * reported.
2425  */
2426 static bool nvme_apst_get_transition_time(u64 total_latency,
2427                 u64 *transition_time, unsigned *last_index)
2428 {
2429         if (total_latency <= apst_primary_latency_tol_us) {
2430                 if (*last_index == 1)
2431                         return false;
2432                 *last_index = 1;
2433                 *transition_time = apst_primary_timeout_ms;
2434                 return true;
2435         }
2436         if (apst_secondary_timeout_ms &&
2437                 total_latency <= apst_secondary_latency_tol_us) {
2438                 if (*last_index <= 2)
2439                         return false;
2440                 *last_index = 2;
2441                 *transition_time = apst_secondary_timeout_ms;
2442                 return true;
2443         }
2444         return false;
2445 }
2446
2447 /*
2448  * APST (Autonomous Power State Transition) lets us program a table of power
2449  * state transitions that the controller will perform automatically.
2450  *
2451  * Depending on module params, one of the two supported techniques will be used:
2452  *
2453  * - If the parameters provide explicit timeouts and tolerances, they will be
2454  *   used to build a table with up to 2 non-operational states to transition to.
2455  *   The default parameter values were selected based on the values used by
2456  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2457  *   regeneration of the APST table in the event of switching between external
2458  *   and battery power, the timeouts and tolerances reflect a compromise
2459  *   between values used by Microsoft for AC and battery scenarios.
2460  * - If not, we'll configure the table with a simple heuristic: we are willing
2461  *   to spend at most 2% of the time transitioning between power states.
2462  *   Therefore, when running in any given state, we will enter the next
2463  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2464  *   microseconds, as long as that state's exit latency is under the requested
2465  *   maximum latency.
2466  *
2467  * We will not autonomously enter any non-operational state for which the total
2468  * latency exceeds ps_max_latency_us.
2469  *
2470  * Users can set ps_max_latency_us to zero to turn off APST.
2471  */
2472 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2473 {
2474         struct nvme_feat_auto_pst *table;
2475         unsigned apste = 0;
2476         u64 max_lat_us = 0;
2477         __le64 target = 0;
2478         int max_ps = -1;
2479         int state;
2480         int ret;
2481         unsigned last_lt_index = UINT_MAX;
2482
2483         /*
2484          * If APST isn't supported or if we haven't been initialized yet,
2485          * then don't do anything.
2486          */
2487         if (!ctrl->apsta)
2488                 return 0;
2489
2490         if (ctrl->npss > 31) {
2491                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2492                 return 0;
2493         }
2494
2495         table = kzalloc(sizeof(*table), GFP_KERNEL);
2496         if (!table)
2497                 return 0;
2498
2499         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2500                 /* Turn off APST. */
2501                 dev_dbg(ctrl->device, "APST disabled\n");
2502                 goto done;
2503         }
2504
2505         /*
2506          * Walk through all states from lowest- to highest-power.
2507          * According to the spec, lower-numbered states use more power.  NPSS,
2508          * despite the name, is the index of the lowest-power state, not the
2509          * number of states.
2510          */
2511         for (state = (int)ctrl->npss; state >= 0; state--) {
2512                 u64 total_latency_us, exit_latency_us, transition_ms;
2513
2514                 if (target)
2515                         table->entries[state] = target;
2516
2517                 /*
2518                  * Don't allow transitions to the deepest state if it's quirked
2519                  * off.
2520                  */
2521                 if (state == ctrl->npss &&
2522                     (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2523                         continue;
2524
2525                 /*
2526                  * Is this state a useful non-operational state for higher-power
2527                  * states to autonomously transition to?
2528                  */
2529                 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2530                         continue;
2531
2532                 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2533                 if (exit_latency_us > ctrl->ps_max_latency_us)
2534                         continue;
2535
2536                 total_latency_us = exit_latency_us +
2537                         le32_to_cpu(ctrl->psd[state].entry_lat);
2538
2539                 /*
2540                  * This state is good. It can be used as the APST idle target
2541                  * for higher power states.
2542                  */
2543                 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2544                         if (!nvme_apst_get_transition_time(total_latency_us,
2545                                         &transition_ms, &last_lt_index))
2546                                 continue;
2547                 } else {
2548                         transition_ms = total_latency_us + 19;
2549                         do_div(transition_ms, 20);
2550                         if (transition_ms > (1 << 24) - 1)
2551                                 transition_ms = (1 << 24) - 1;
2552                 }
2553
2554                 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2555                 if (max_ps == -1)
2556                         max_ps = state;
2557                 if (total_latency_us > max_lat_us)
2558                         max_lat_us = total_latency_us;
2559         }
2560
2561         if (max_ps == -1)
2562                 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2563         else
2564                 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2565                         max_ps, max_lat_us, (int)sizeof(*table), table);
2566         apste = 1;
2567
2568 done:
2569         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2570                                 table, sizeof(*table), NULL);
2571         if (ret)
2572                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2573         kfree(table);
2574         return ret;
2575 }
2576
2577 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2578 {
2579         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2580         u64 latency;
2581
2582         switch (val) {
2583         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2584         case PM_QOS_LATENCY_ANY:
2585                 latency = U64_MAX;
2586                 break;
2587
2588         default:
2589                 latency = val;
2590         }
2591
2592         if (ctrl->ps_max_latency_us != latency) {
2593                 ctrl->ps_max_latency_us = latency;
2594                 if (ctrl->state == NVME_CTRL_LIVE)
2595                         nvme_configure_apst(ctrl);
2596         }
2597 }
2598
2599 struct nvme_core_quirk_entry {
2600         /*
2601          * NVMe model and firmware strings are padded with spaces.  For
2602          * simplicity, strings in the quirk table are padded with NULLs
2603          * instead.
2604          */
2605         u16 vid;
2606         const char *mn;
2607         const char *fr;
2608         unsigned long quirks;
2609 };
2610
2611 static const struct nvme_core_quirk_entry core_quirks[] = {
2612         {
2613                 /*
2614                  * This Toshiba device seems to die using any APST states.  See:
2615                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2616                  */
2617                 .vid = 0x1179,
2618                 .mn = "THNSF5256GPUK TOSHIBA",
2619                 .quirks = NVME_QUIRK_NO_APST,
2620         },
2621         {
2622                 /*
2623                  * This LiteON CL1-3D*-Q11 firmware version has a race
2624                  * condition associated with actions related to suspend to idle
2625                  * LiteON has resolved the problem in future firmware
2626                  */
2627                 .vid = 0x14a4,
2628                 .fr = "22301111",
2629                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2630         },
2631         {
2632                 /*
2633                  * This Kioxia CD6-V Series / HPE PE8030 device times out and
2634                  * aborts I/O during any load, but more easily reproducible
2635                  * with discards (fstrim).
2636                  *
2637                  * The device is left in a state where it is also not possible
2638                  * to use "nvme set-feature" to disable APST, but booting with
2639                  * nvme_core.default_ps_max_latency=0 works.
2640                  */
2641                 .vid = 0x1e0f,
2642                 .mn = "KCD6XVUL6T40",
2643                 .quirks = NVME_QUIRK_NO_APST,
2644         },
2645         {
2646                 /*
2647                  * The external Samsung X5 SSD fails initialization without a
2648                  * delay before checking if it is ready and has a whole set of
2649                  * other problems.  To make this even more interesting, it
2650                  * shares the PCI ID with internal Samsung 970 Evo Plus that
2651                  * does not need or want these quirks.
2652                  */
2653                 .vid = 0x144d,
2654                 .mn = "Samsung Portable SSD X5",
2655                 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2656                           NVME_QUIRK_NO_DEEPEST_PS |
2657                           NVME_QUIRK_IGNORE_DEV_SUBNQN,
2658         }
2659 };
2660
2661 /* match is null-terminated but idstr is space-padded. */
2662 static bool string_matches(const char *idstr, const char *match, size_t len)
2663 {
2664         size_t matchlen;
2665
2666         if (!match)
2667                 return true;
2668
2669         matchlen = strlen(match);
2670         WARN_ON_ONCE(matchlen > len);
2671
2672         if (memcmp(idstr, match, matchlen))
2673                 return false;
2674
2675         for (; matchlen < len; matchlen++)
2676                 if (idstr[matchlen] != ' ')
2677                         return false;
2678
2679         return true;
2680 }
2681
2682 static bool quirk_matches(const struct nvme_id_ctrl *id,
2683                           const struct nvme_core_quirk_entry *q)
2684 {
2685         return q->vid == le16_to_cpu(id->vid) &&
2686                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2687                 string_matches(id->fr, q->fr, sizeof(id->fr));
2688 }
2689
2690 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2691                 struct nvme_id_ctrl *id)
2692 {
2693         size_t nqnlen;
2694         int off;
2695
2696         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2697                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2698                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2699                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2700                         return;
2701                 }
2702
2703                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2704                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2705         }
2706
2707         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2708         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2709                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2710                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2711         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2712         off += sizeof(id->sn);
2713         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2714         off += sizeof(id->mn);
2715         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2716 }
2717
2718 static void nvme_release_subsystem(struct device *dev)
2719 {
2720         struct nvme_subsystem *subsys =
2721                 container_of(dev, struct nvme_subsystem, dev);
2722
2723         if (subsys->instance >= 0)
2724                 ida_free(&nvme_instance_ida, subsys->instance);
2725         kfree(subsys);
2726 }
2727
2728 static void nvme_destroy_subsystem(struct kref *ref)
2729 {
2730         struct nvme_subsystem *subsys =
2731                         container_of(ref, struct nvme_subsystem, ref);
2732
2733         mutex_lock(&nvme_subsystems_lock);
2734         list_del(&subsys->entry);
2735         mutex_unlock(&nvme_subsystems_lock);
2736
2737         ida_destroy(&subsys->ns_ida);
2738         device_del(&subsys->dev);
2739         put_device(&subsys->dev);
2740 }
2741
2742 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2743 {
2744         kref_put(&subsys->ref, nvme_destroy_subsystem);
2745 }
2746
2747 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2748 {
2749         struct nvme_subsystem *subsys;
2750
2751         lockdep_assert_held(&nvme_subsystems_lock);
2752
2753         /*
2754          * Fail matches for discovery subsystems. This results
2755          * in each discovery controller bound to a unique subsystem.
2756          * This avoids issues with validating controller values
2757          * that can only be true when there is a single unique subsystem.
2758          * There may be multiple and completely independent entities
2759          * that provide discovery controllers.
2760          */
2761         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2762                 return NULL;
2763
2764         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2765                 if (strcmp(subsys->subnqn, subsysnqn))
2766                         continue;
2767                 if (!kref_get_unless_zero(&subsys->ref))
2768                         continue;
2769                 return subsys;
2770         }
2771
2772         return NULL;
2773 }
2774
2775 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2776         struct device_attribute subsys_attr_##_name = \
2777                 __ATTR(_name, _mode, _show, NULL)
2778
2779 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2780                                     struct device_attribute *attr,
2781                                     char *buf)
2782 {
2783         struct nvme_subsystem *subsys =
2784                 container_of(dev, struct nvme_subsystem, dev);
2785
2786         return sysfs_emit(buf, "%s\n", subsys->subnqn);
2787 }
2788 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2789
2790 static ssize_t nvme_subsys_show_type(struct device *dev,
2791                                     struct device_attribute *attr,
2792                                     char *buf)
2793 {
2794         struct nvme_subsystem *subsys =
2795                 container_of(dev, struct nvme_subsystem, dev);
2796
2797         switch (subsys->subtype) {
2798         case NVME_NQN_DISC:
2799                 return sysfs_emit(buf, "discovery\n");
2800         case NVME_NQN_NVME:
2801                 return sysfs_emit(buf, "nvm\n");
2802         default:
2803                 return sysfs_emit(buf, "reserved\n");
2804         }
2805 }
2806 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2807
2808 #define nvme_subsys_show_str_function(field)                            \
2809 static ssize_t subsys_##field##_show(struct device *dev,                \
2810                             struct device_attribute *attr, char *buf)   \
2811 {                                                                       \
2812         struct nvme_subsystem *subsys =                                 \
2813                 container_of(dev, struct nvme_subsystem, dev);          \
2814         return sysfs_emit(buf, "%.*s\n",                                \
2815                            (int)sizeof(subsys->field), subsys->field);  \
2816 }                                                                       \
2817 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2818
2819 nvme_subsys_show_str_function(model);
2820 nvme_subsys_show_str_function(serial);
2821 nvme_subsys_show_str_function(firmware_rev);
2822
2823 static struct attribute *nvme_subsys_attrs[] = {
2824         &subsys_attr_model.attr,
2825         &subsys_attr_serial.attr,
2826         &subsys_attr_firmware_rev.attr,
2827         &subsys_attr_subsysnqn.attr,
2828         &subsys_attr_subsystype.attr,
2829 #ifdef CONFIG_NVME_MULTIPATH
2830         &subsys_attr_iopolicy.attr,
2831 #endif
2832         NULL,
2833 };
2834
2835 static const struct attribute_group nvme_subsys_attrs_group = {
2836         .attrs = nvme_subsys_attrs,
2837 };
2838
2839 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2840         &nvme_subsys_attrs_group,
2841         NULL,
2842 };
2843
2844 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2845 {
2846         return ctrl->opts && ctrl->opts->discovery_nqn;
2847 }
2848
2849 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2850                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2851 {
2852         struct nvme_ctrl *tmp;
2853
2854         lockdep_assert_held(&nvme_subsystems_lock);
2855
2856         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2857                 if (nvme_state_terminal(tmp))
2858                         continue;
2859
2860                 if (tmp->cntlid == ctrl->cntlid) {
2861                         dev_err(ctrl->device,
2862                                 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2863                                 ctrl->cntlid, dev_name(tmp->device),
2864                                 subsys->subnqn);
2865                         return false;
2866                 }
2867
2868                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2869                     nvme_discovery_ctrl(ctrl))
2870                         continue;
2871
2872                 dev_err(ctrl->device,
2873                         "Subsystem does not support multiple controllers\n");
2874                 return false;
2875         }
2876
2877         return true;
2878 }
2879
2880 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2881 {
2882         struct nvme_subsystem *subsys, *found;
2883         int ret;
2884
2885         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2886         if (!subsys)
2887                 return -ENOMEM;
2888
2889         subsys->instance = -1;
2890         mutex_init(&subsys->lock);
2891         kref_init(&subsys->ref);
2892         INIT_LIST_HEAD(&subsys->ctrls);
2893         INIT_LIST_HEAD(&subsys->nsheads);
2894         nvme_init_subnqn(subsys, ctrl, id);
2895         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2896         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2897         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2898         subsys->vendor_id = le16_to_cpu(id->vid);
2899         subsys->cmic = id->cmic;
2900
2901         /* Versions prior to 1.4 don't necessarily report a valid type */
2902         if (id->cntrltype == NVME_CTRL_DISC ||
2903             !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2904                 subsys->subtype = NVME_NQN_DISC;
2905         else
2906                 subsys->subtype = NVME_NQN_NVME;
2907
2908         if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2909                 dev_err(ctrl->device,
2910                         "Subsystem %s is not a discovery controller",
2911                         subsys->subnqn);
2912                 kfree(subsys);
2913                 return -EINVAL;
2914         }
2915         subsys->awupf = le16_to_cpu(id->awupf);
2916         nvme_mpath_default_iopolicy(subsys);
2917
2918         subsys->dev.class = nvme_subsys_class;
2919         subsys->dev.release = nvme_release_subsystem;
2920         subsys->dev.groups = nvme_subsys_attrs_groups;
2921         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2922         device_initialize(&subsys->dev);
2923
2924         mutex_lock(&nvme_subsystems_lock);
2925         found = __nvme_find_get_subsystem(subsys->subnqn);
2926         if (found) {
2927                 put_device(&subsys->dev);
2928                 subsys = found;
2929
2930                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2931                         ret = -EINVAL;
2932                         goto out_put_subsystem;
2933                 }
2934         } else {
2935                 ret = device_add(&subsys->dev);
2936                 if (ret) {
2937                         dev_err(ctrl->device,
2938                                 "failed to register subsystem device.\n");
2939                         put_device(&subsys->dev);
2940                         goto out_unlock;
2941                 }
2942                 ida_init(&subsys->ns_ida);
2943                 list_add_tail(&subsys->entry, &nvme_subsystems);
2944         }
2945
2946         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2947                                 dev_name(ctrl->device));
2948         if (ret) {
2949                 dev_err(ctrl->device,
2950                         "failed to create sysfs link from subsystem.\n");
2951                 goto out_put_subsystem;
2952         }
2953
2954         if (!found)
2955                 subsys->instance = ctrl->instance;
2956         ctrl->subsys = subsys;
2957         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2958         mutex_unlock(&nvme_subsystems_lock);
2959         return 0;
2960
2961 out_put_subsystem:
2962         nvme_put_subsystem(subsys);
2963 out_unlock:
2964         mutex_unlock(&nvme_subsystems_lock);
2965         return ret;
2966 }
2967
2968 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2969                 void *log, size_t size, u64 offset)
2970 {
2971         struct nvme_command c = { };
2972         u32 dwlen = nvme_bytes_to_numd(size);
2973
2974         c.get_log_page.opcode = nvme_admin_get_log_page;
2975         c.get_log_page.nsid = cpu_to_le32(nsid);
2976         c.get_log_page.lid = log_page;
2977         c.get_log_page.lsp = lsp;
2978         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2979         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2980         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2981         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2982         c.get_log_page.csi = csi;
2983
2984         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2985 }
2986
2987 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2988                                 struct nvme_effects_log **log)
2989 {
2990         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2991         int ret;
2992
2993         if (cel)
2994                 goto out;
2995
2996         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2997         if (!cel)
2998                 return -ENOMEM;
2999
3000         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3001                         cel, sizeof(*cel), 0);
3002         if (ret) {
3003                 kfree(cel);
3004                 return ret;
3005         }
3006
3007         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3008 out:
3009         *log = cel;
3010         return 0;
3011 }
3012
3013 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3014 {
3015         u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3016
3017         if (check_shl_overflow(1U, units + page_shift - 9, &val))
3018                 return UINT_MAX;
3019         return val;
3020 }
3021
3022 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3023 {
3024         struct nvme_command c = { };
3025         struct nvme_id_ctrl_nvm *id;
3026         int ret;
3027
3028         if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
3029                 ctrl->max_discard_sectors = UINT_MAX;
3030                 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
3031         } else {
3032                 ctrl->max_discard_sectors = 0;
3033                 ctrl->max_discard_segments = 0;
3034         }
3035
3036         /*
3037          * Even though NVMe spec explicitly states that MDTS is not applicable
3038          * to the write-zeroes, we are cautious and limit the size to the
3039          * controllers max_hw_sectors value, which is based on the MDTS field
3040          * and possibly other limiting factors.
3041          */
3042         if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3043             !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3044                 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3045         else
3046                 ctrl->max_zeroes_sectors = 0;
3047
3048         if (nvme_ctrl_limited_cns(ctrl))
3049                 return 0;
3050
3051         id = kzalloc(sizeof(*id), GFP_KERNEL);
3052         if (!id)
3053                 return 0;
3054
3055         c.identify.opcode = nvme_admin_identify;
3056         c.identify.cns = NVME_ID_CNS_CS_CTRL;
3057         c.identify.csi = NVME_CSI_NVM;
3058
3059         ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3060         if (ret)
3061                 goto free_data;
3062
3063         if (id->dmrl)
3064                 ctrl->max_discard_segments = id->dmrl;
3065         ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3066         if (id->wzsl)
3067                 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3068
3069 free_data:
3070         kfree(id);
3071         return ret;
3072 }
3073
3074 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3075 {
3076         struct nvme_id_ctrl *id;
3077         u32 max_hw_sectors;
3078         bool prev_apst_enabled;
3079         int ret;
3080
3081         ret = nvme_identify_ctrl(ctrl, &id);
3082         if (ret) {
3083                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3084                 return -EIO;
3085         }
3086
3087         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3088                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3089                 if (ret < 0)
3090                         goto out_free;
3091         }
3092
3093         if (!(ctrl->ops->flags & NVME_F_FABRICS))
3094                 ctrl->cntlid = le16_to_cpu(id->cntlid);
3095
3096         if (!ctrl->identified) {
3097                 unsigned int i;
3098
3099                 ret = nvme_init_subsystem(ctrl, id);
3100                 if (ret)
3101                         goto out_free;
3102
3103                 /*
3104                  * Check for quirks.  Quirk can depend on firmware version,
3105                  * so, in principle, the set of quirks present can change
3106                  * across a reset.  As a possible future enhancement, we
3107                  * could re-scan for quirks every time we reinitialize
3108                  * the device, but we'd have to make sure that the driver
3109                  * behaves intelligently if the quirks change.
3110                  */
3111                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3112                         if (quirk_matches(id, &core_quirks[i]))
3113                                 ctrl->quirks |= core_quirks[i].quirks;
3114                 }
3115         }
3116
3117         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3118                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3119                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3120         }
3121
3122         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3123         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3124         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3125
3126         ctrl->oacs = le16_to_cpu(id->oacs);
3127         ctrl->oncs = le16_to_cpu(id->oncs);
3128         ctrl->mtfa = le16_to_cpu(id->mtfa);
3129         ctrl->oaes = le32_to_cpu(id->oaes);
3130         ctrl->wctemp = le16_to_cpu(id->wctemp);
3131         ctrl->cctemp = le16_to_cpu(id->cctemp);
3132
3133         atomic_set(&ctrl->abort_limit, id->acl + 1);
3134         ctrl->vwc = id->vwc;
3135         if (id->mdts)
3136                 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3137         else
3138                 max_hw_sectors = UINT_MAX;
3139         ctrl->max_hw_sectors =
3140                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3141
3142         nvme_set_queue_limits(ctrl, ctrl->admin_q);
3143         ctrl->sgls = le32_to_cpu(id->sgls);
3144         ctrl->kas = le16_to_cpu(id->kas);
3145         ctrl->max_namespaces = le32_to_cpu(id->mnan);
3146         ctrl->ctratt = le32_to_cpu(id->ctratt);
3147
3148         ctrl->cntrltype = id->cntrltype;
3149         ctrl->dctype = id->dctype;
3150
3151         if (id->rtd3e) {
3152                 /* us -> s */
3153                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3154
3155                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3156                                                  shutdown_timeout, 60);
3157
3158                 if (ctrl->shutdown_timeout != shutdown_timeout)
3159                         dev_info(ctrl->device,
3160                                  "Shutdown timeout set to %u seconds\n",
3161                                  ctrl->shutdown_timeout);
3162         } else
3163                 ctrl->shutdown_timeout = shutdown_timeout;
3164
3165         ctrl->npss = id->npss;
3166         ctrl->apsta = id->apsta;
3167         prev_apst_enabled = ctrl->apst_enabled;
3168         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3169                 if (force_apst && id->apsta) {
3170                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3171                         ctrl->apst_enabled = true;
3172                 } else {
3173                         ctrl->apst_enabled = false;
3174                 }
3175         } else {
3176                 ctrl->apst_enabled = id->apsta;
3177         }
3178         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3179
3180         if (ctrl->ops->flags & NVME_F_FABRICS) {
3181                 ctrl->icdoff = le16_to_cpu(id->icdoff);
3182                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3183                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3184                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3185
3186                 /*
3187                  * In fabrics we need to verify the cntlid matches the
3188                  * admin connect
3189                  */
3190                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3191                         dev_err(ctrl->device,
3192                                 "Mismatching cntlid: Connect %u vs Identify "
3193                                 "%u, rejecting\n",
3194                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
3195                         ret = -EINVAL;
3196                         goto out_free;
3197                 }
3198
3199                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3200                         dev_err(ctrl->device,
3201                                 "keep-alive support is mandatory for fabrics\n");
3202                         ret = -EINVAL;
3203                         goto out_free;
3204                 }
3205         } else {
3206                 ctrl->hmpre = le32_to_cpu(id->hmpre);
3207                 ctrl->hmmin = le32_to_cpu(id->hmmin);
3208                 ctrl->hmminds = le32_to_cpu(id->hmminds);
3209                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3210         }
3211
3212         ret = nvme_mpath_init_identify(ctrl, id);
3213         if (ret < 0)
3214                 goto out_free;
3215
3216         if (ctrl->apst_enabled && !prev_apst_enabled)
3217                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3218         else if (!ctrl->apst_enabled && prev_apst_enabled)
3219                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3220
3221 out_free:
3222         kfree(id);
3223         return ret;
3224 }
3225
3226 /*
3227  * Initialize the cached copies of the Identify data and various controller
3228  * register in our nvme_ctrl structure.  This should be called as soon as
3229  * the admin queue is fully up and running.
3230  */
3231 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3232 {
3233         int ret;
3234
3235         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3236         if (ret) {
3237                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3238                 return ret;
3239         }
3240
3241         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3242
3243         if (ctrl->vs >= NVME_VS(1, 1, 0))
3244                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3245
3246         ret = nvme_init_identify(ctrl);
3247         if (ret)
3248                 return ret;
3249
3250         ret = nvme_configure_apst(ctrl);
3251         if (ret < 0)
3252                 return ret;
3253
3254         ret = nvme_configure_timestamp(ctrl);
3255         if (ret < 0)
3256                 return ret;
3257
3258         ret = nvme_configure_host_options(ctrl);
3259         if (ret < 0)
3260                 return ret;
3261
3262         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3263                 ret = nvme_hwmon_init(ctrl);
3264                 if (ret < 0)
3265                         return ret;
3266         }
3267
3268         ctrl->identified = true;
3269
3270         return 0;
3271 }
3272 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3273
3274 static int nvme_dev_open(struct inode *inode, struct file *file)
3275 {
3276         struct nvme_ctrl *ctrl =
3277                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3278
3279         switch (ctrl->state) {
3280         case NVME_CTRL_LIVE:
3281                 break;
3282         default:
3283                 return -EWOULDBLOCK;
3284         }
3285
3286         nvme_get_ctrl(ctrl);
3287         if (!try_module_get(ctrl->ops->module)) {
3288                 nvme_put_ctrl(ctrl);
3289                 return -EINVAL;
3290         }
3291
3292         file->private_data = ctrl;
3293         return 0;
3294 }
3295
3296 static int nvme_dev_release(struct inode *inode, struct file *file)
3297 {
3298         struct nvme_ctrl *ctrl =
3299                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3300
3301         module_put(ctrl->ops->module);
3302         nvme_put_ctrl(ctrl);
3303         return 0;
3304 }
3305
3306 static const struct file_operations nvme_dev_fops = {
3307         .owner          = THIS_MODULE,
3308         .open           = nvme_dev_open,
3309         .release        = nvme_dev_release,
3310         .unlocked_ioctl = nvme_dev_ioctl,
3311         .compat_ioctl   = compat_ptr_ioctl,
3312         .uring_cmd      = nvme_dev_uring_cmd,
3313 };
3314
3315 static ssize_t nvme_sysfs_reset(struct device *dev,
3316                                 struct device_attribute *attr, const char *buf,
3317                                 size_t count)
3318 {
3319         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3320         int ret;
3321
3322         ret = nvme_reset_ctrl_sync(ctrl);
3323         if (ret < 0)
3324                 return ret;
3325         return count;
3326 }
3327 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3328
3329 static ssize_t nvme_sysfs_rescan(struct device *dev,
3330                                 struct device_attribute *attr, const char *buf,
3331                                 size_t count)
3332 {
3333         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3334
3335         nvme_queue_scan(ctrl);
3336         return count;
3337 }
3338 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3339
3340 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3341 {
3342         struct gendisk *disk = dev_to_disk(dev);
3343
3344         if (disk->fops == &nvme_bdev_ops)
3345                 return nvme_get_ns_from_dev(dev)->head;
3346         else
3347                 return disk->private_data;
3348 }
3349
3350 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3351                 char *buf)
3352 {
3353         struct nvme_ns_head *head = dev_to_ns_head(dev);
3354         struct nvme_ns_ids *ids = &head->ids;
3355         struct nvme_subsystem *subsys = head->subsys;
3356         int serial_len = sizeof(subsys->serial);
3357         int model_len = sizeof(subsys->model);
3358
3359         if (!uuid_is_null(&ids->uuid))
3360                 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3361
3362         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3363                 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3364
3365         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3366                 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3367
3368         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3369                                   subsys->serial[serial_len - 1] == '\0'))
3370                 serial_len--;
3371         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3372                                  subsys->model[model_len - 1] == '\0'))
3373                 model_len--;
3374
3375         return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3376                 serial_len, subsys->serial, model_len, subsys->model,
3377                 head->ns_id);
3378 }
3379 static DEVICE_ATTR_RO(wwid);
3380
3381 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3382                 char *buf)
3383 {
3384         return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3385 }
3386 static DEVICE_ATTR_RO(nguid);
3387
3388 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3389                 char *buf)
3390 {
3391         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3392
3393         /* For backward compatibility expose the NGUID to userspace if
3394          * we have no UUID set
3395          */
3396         if (uuid_is_null(&ids->uuid)) {
3397                 dev_warn_ratelimited(dev,
3398                         "No UUID available providing old NGUID\n");
3399                 return sysfs_emit(buf, "%pU\n", ids->nguid);
3400         }
3401         return sysfs_emit(buf, "%pU\n", &ids->uuid);
3402 }
3403 static DEVICE_ATTR_RO(uuid);
3404
3405 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3406                 char *buf)
3407 {
3408         return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3409 }
3410 static DEVICE_ATTR_RO(eui);
3411
3412 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3413                 char *buf)
3414 {
3415         return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3416 }
3417 static DEVICE_ATTR_RO(nsid);
3418
3419 static struct attribute *nvme_ns_id_attrs[] = {
3420         &dev_attr_wwid.attr,
3421         &dev_attr_uuid.attr,
3422         &dev_attr_nguid.attr,
3423         &dev_attr_eui.attr,
3424         &dev_attr_nsid.attr,
3425 #ifdef CONFIG_NVME_MULTIPATH
3426         &dev_attr_ana_grpid.attr,
3427         &dev_attr_ana_state.attr,
3428 #endif
3429         NULL,
3430 };
3431
3432 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3433                 struct attribute *a, int n)
3434 {
3435         struct device *dev = container_of(kobj, struct device, kobj);
3436         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3437
3438         if (a == &dev_attr_uuid.attr) {
3439                 if (uuid_is_null(&ids->uuid) &&
3440                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3441                         return 0;
3442         }
3443         if (a == &dev_attr_nguid.attr) {
3444                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3445                         return 0;
3446         }
3447         if (a == &dev_attr_eui.attr) {
3448                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3449                         return 0;
3450         }
3451 #ifdef CONFIG_NVME_MULTIPATH
3452         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3453                 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3454                         return 0;
3455                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3456                         return 0;
3457         }
3458 #endif
3459         return a->mode;
3460 }
3461
3462 static const struct attribute_group nvme_ns_id_attr_group = {
3463         .attrs          = nvme_ns_id_attrs,
3464         .is_visible     = nvme_ns_id_attrs_are_visible,
3465 };
3466
3467 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3468         &nvme_ns_id_attr_group,
3469         NULL,
3470 };
3471
3472 #define nvme_show_str_function(field)                                           \
3473 static ssize_t  field##_show(struct device *dev,                                \
3474                             struct device_attribute *attr, char *buf)           \
3475 {                                                                               \
3476         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3477         return sysfs_emit(buf, "%.*s\n",                                        \
3478                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3479 }                                                                               \
3480 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3481
3482 nvme_show_str_function(model);
3483 nvme_show_str_function(serial);
3484 nvme_show_str_function(firmware_rev);
3485
3486 #define nvme_show_int_function(field)                                           \
3487 static ssize_t  field##_show(struct device *dev,                                \
3488                             struct device_attribute *attr, char *buf)           \
3489 {                                                                               \
3490         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3491         return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3492 }                                                                               \
3493 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3494
3495 nvme_show_int_function(cntlid);
3496 nvme_show_int_function(numa_node);
3497 nvme_show_int_function(queue_count);
3498 nvme_show_int_function(sqsize);
3499 nvme_show_int_function(kato);
3500
3501 static ssize_t nvme_sysfs_delete(struct device *dev,
3502                                 struct device_attribute *attr, const char *buf,
3503                                 size_t count)
3504 {
3505         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3506
3507         if (device_remove_file_self(dev, attr))
3508                 nvme_delete_ctrl_sync(ctrl);
3509         return count;
3510 }
3511 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3512
3513 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3514                                          struct device_attribute *attr,
3515                                          char *buf)
3516 {
3517         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3518
3519         return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3520 }
3521 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3522
3523 static ssize_t nvme_sysfs_show_state(struct device *dev,
3524                                      struct device_attribute *attr,
3525                                      char *buf)
3526 {
3527         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3528         static const char *const state_name[] = {
3529                 [NVME_CTRL_NEW]         = "new",
3530                 [NVME_CTRL_LIVE]        = "live",
3531                 [NVME_CTRL_RESETTING]   = "resetting",
3532                 [NVME_CTRL_CONNECTING]  = "connecting",
3533                 [NVME_CTRL_DELETING]    = "deleting",
3534                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3535                 [NVME_CTRL_DEAD]        = "dead",
3536         };
3537
3538         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3539             state_name[ctrl->state])
3540                 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3541
3542         return sysfs_emit(buf, "unknown state\n");
3543 }
3544
3545 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3546
3547 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3548                                          struct device_attribute *attr,
3549                                          char *buf)
3550 {
3551         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3552
3553         return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3554 }
3555 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3556
3557 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3558                                         struct device_attribute *attr,
3559                                         char *buf)
3560 {
3561         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3562
3563         return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3564 }
3565 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3566
3567 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3568                                         struct device_attribute *attr,
3569                                         char *buf)
3570 {
3571         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3572
3573         return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3574 }
3575 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3576
3577 static ssize_t nvme_sysfs_show_address(struct device *dev,
3578                                          struct device_attribute *attr,
3579                                          char *buf)
3580 {
3581         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3582
3583         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3584 }
3585 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3586
3587 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3588                 struct device_attribute *attr, char *buf)
3589 {
3590         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3591         struct nvmf_ctrl_options *opts = ctrl->opts;
3592
3593         if (ctrl->opts->max_reconnects == -1)
3594                 return sysfs_emit(buf, "off\n");
3595         return sysfs_emit(buf, "%d\n",
3596                           opts->max_reconnects * opts->reconnect_delay);
3597 }
3598
3599 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3600                 struct device_attribute *attr, const char *buf, size_t count)
3601 {
3602         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3603         struct nvmf_ctrl_options *opts = ctrl->opts;
3604         int ctrl_loss_tmo, err;
3605
3606         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3607         if (err)
3608                 return -EINVAL;
3609
3610         if (ctrl_loss_tmo < 0)
3611                 opts->max_reconnects = -1;
3612         else
3613                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3614                                                 opts->reconnect_delay);
3615         return count;
3616 }
3617 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3618         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3619
3620 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3621                 struct device_attribute *attr, char *buf)
3622 {
3623         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3624
3625         if (ctrl->opts->reconnect_delay == -1)
3626                 return sysfs_emit(buf, "off\n");
3627         return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3628 }
3629
3630 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3631                 struct device_attribute *attr, const char *buf, size_t count)
3632 {
3633         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3634         unsigned int v;
3635         int err;
3636
3637         err = kstrtou32(buf, 10, &v);
3638         if (err)
3639                 return err;
3640
3641         ctrl->opts->reconnect_delay = v;
3642         return count;
3643 }
3644 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3645         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3646
3647 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3648                 struct device_attribute *attr, char *buf)
3649 {
3650         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3651
3652         if (ctrl->opts->fast_io_fail_tmo == -1)
3653                 return sysfs_emit(buf, "off\n");
3654         return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3655 }
3656
3657 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3658                 struct device_attribute *attr, const char *buf, size_t count)
3659 {
3660         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3661         struct nvmf_ctrl_options *opts = ctrl->opts;
3662         int fast_io_fail_tmo, err;
3663
3664         err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3665         if (err)
3666                 return -EINVAL;
3667
3668         if (fast_io_fail_tmo < 0)
3669                 opts->fast_io_fail_tmo = -1;
3670         else
3671                 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3672         return count;
3673 }
3674 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3675         nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3676
3677 static ssize_t cntrltype_show(struct device *dev,
3678                               struct device_attribute *attr, char *buf)
3679 {
3680         static const char * const type[] = {
3681                 [NVME_CTRL_IO] = "io\n",
3682                 [NVME_CTRL_DISC] = "discovery\n",
3683                 [NVME_CTRL_ADMIN] = "admin\n",
3684         };
3685         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3686
3687         if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3688                 return sysfs_emit(buf, "reserved\n");
3689
3690         return sysfs_emit(buf, type[ctrl->cntrltype]);
3691 }
3692 static DEVICE_ATTR_RO(cntrltype);
3693
3694 static ssize_t dctype_show(struct device *dev,
3695                            struct device_attribute *attr, char *buf)
3696 {
3697         static const char * const type[] = {
3698                 [NVME_DCTYPE_NOT_REPORTED] = "none\n",
3699                 [NVME_DCTYPE_DDC] = "ddc\n",
3700                 [NVME_DCTYPE_CDC] = "cdc\n",
3701         };
3702         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3703
3704         if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3705                 return sysfs_emit(buf, "reserved\n");
3706
3707         return sysfs_emit(buf, type[ctrl->dctype]);
3708 }
3709 static DEVICE_ATTR_RO(dctype);
3710
3711 #ifdef CONFIG_NVME_AUTH
3712 static ssize_t nvme_ctrl_dhchap_secret_show(struct device *dev,
3713                 struct device_attribute *attr, char *buf)
3714 {
3715         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3716         struct nvmf_ctrl_options *opts = ctrl->opts;
3717
3718         if (!opts->dhchap_secret)
3719                 return sysfs_emit(buf, "none\n");
3720         return sysfs_emit(buf, "%s\n", opts->dhchap_secret);
3721 }
3722
3723 static ssize_t nvme_ctrl_dhchap_secret_store(struct device *dev,
3724                 struct device_attribute *attr, const char *buf, size_t count)
3725 {
3726         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3727         struct nvmf_ctrl_options *opts = ctrl->opts;
3728         char *dhchap_secret;
3729
3730         if (!ctrl->opts->dhchap_secret)
3731                 return -EINVAL;
3732         if (count < 7)
3733                 return -EINVAL;
3734         if (memcmp(buf, "DHHC-1:", 7))
3735                 return -EINVAL;
3736
3737         dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3738         if (!dhchap_secret)
3739                 return -ENOMEM;
3740         memcpy(dhchap_secret, buf, count);
3741         nvme_auth_stop(ctrl);
3742         if (strcmp(dhchap_secret, opts->dhchap_secret)) {
3743                 int ret;
3744
3745                 ret = nvme_auth_generate_key(dhchap_secret, &ctrl->host_key);
3746                 if (ret)
3747                         return ret;
3748                 kfree(opts->dhchap_secret);
3749                 opts->dhchap_secret = dhchap_secret;
3750                 /* Key has changed; re-authentication with new key */
3751                 nvme_auth_reset(ctrl);
3752         }
3753         /* Start re-authentication */
3754         dev_info(ctrl->device, "re-authenticating controller\n");
3755         queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3756
3757         return count;
3758 }
3759 static DEVICE_ATTR(dhchap_secret, S_IRUGO | S_IWUSR,
3760         nvme_ctrl_dhchap_secret_show, nvme_ctrl_dhchap_secret_store);
3761
3762 static ssize_t nvme_ctrl_dhchap_ctrl_secret_show(struct device *dev,
3763                 struct device_attribute *attr, char *buf)
3764 {
3765         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3766         struct nvmf_ctrl_options *opts = ctrl->opts;
3767
3768         if (!opts->dhchap_ctrl_secret)
3769                 return sysfs_emit(buf, "none\n");
3770         return sysfs_emit(buf, "%s\n", opts->dhchap_ctrl_secret);
3771 }
3772
3773 static ssize_t nvme_ctrl_dhchap_ctrl_secret_store(struct device *dev,
3774                 struct device_attribute *attr, const char *buf, size_t count)
3775 {
3776         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3777         struct nvmf_ctrl_options *opts = ctrl->opts;
3778         char *dhchap_secret;
3779
3780         if (!ctrl->opts->dhchap_ctrl_secret)
3781                 return -EINVAL;
3782         if (count < 7)
3783                 return -EINVAL;
3784         if (memcmp(buf, "DHHC-1:", 7))
3785                 return -EINVAL;
3786
3787         dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3788         if (!dhchap_secret)
3789                 return -ENOMEM;
3790         memcpy(dhchap_secret, buf, count);
3791         nvme_auth_stop(ctrl);
3792         if (strcmp(dhchap_secret, opts->dhchap_ctrl_secret)) {
3793                 int ret;
3794
3795                 ret = nvme_auth_generate_key(dhchap_secret, &ctrl->ctrl_key);
3796                 if (ret)
3797                         return ret;
3798                 kfree(opts->dhchap_ctrl_secret);
3799                 opts->dhchap_ctrl_secret = dhchap_secret;
3800                 /* Key has changed; re-authentication with new key */
3801                 nvme_auth_reset(ctrl);
3802         }
3803         /* Start re-authentication */
3804         dev_info(ctrl->device, "re-authenticating controller\n");
3805         queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3806
3807         return count;
3808 }
3809 static DEVICE_ATTR(dhchap_ctrl_secret, S_IRUGO | S_IWUSR,
3810         nvme_ctrl_dhchap_ctrl_secret_show, nvme_ctrl_dhchap_ctrl_secret_store);
3811 #endif
3812
3813 static struct attribute *nvme_dev_attrs[] = {
3814         &dev_attr_reset_controller.attr,
3815         &dev_attr_rescan_controller.attr,
3816         &dev_attr_model.attr,
3817         &dev_attr_serial.attr,
3818         &dev_attr_firmware_rev.attr,
3819         &dev_attr_cntlid.attr,
3820         &dev_attr_delete_controller.attr,
3821         &dev_attr_transport.attr,
3822         &dev_attr_subsysnqn.attr,
3823         &dev_attr_address.attr,
3824         &dev_attr_state.attr,
3825         &dev_attr_numa_node.attr,
3826         &dev_attr_queue_count.attr,
3827         &dev_attr_sqsize.attr,
3828         &dev_attr_hostnqn.attr,
3829         &dev_attr_hostid.attr,
3830         &dev_attr_ctrl_loss_tmo.attr,
3831         &dev_attr_reconnect_delay.attr,
3832         &dev_attr_fast_io_fail_tmo.attr,
3833         &dev_attr_kato.attr,
3834         &dev_attr_cntrltype.attr,
3835         &dev_attr_dctype.attr,
3836 #ifdef CONFIG_NVME_AUTH
3837         &dev_attr_dhchap_secret.attr,
3838         &dev_attr_dhchap_ctrl_secret.attr,
3839 #endif
3840         NULL
3841 };
3842
3843 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3844                 struct attribute *a, int n)
3845 {
3846         struct device *dev = container_of(kobj, struct device, kobj);
3847         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3848
3849         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3850                 return 0;
3851         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3852                 return 0;
3853         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3854                 return 0;
3855         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3856                 return 0;
3857         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3858                 return 0;
3859         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3860                 return 0;
3861         if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3862                 return 0;
3863 #ifdef CONFIG_NVME_AUTH
3864         if (a == &dev_attr_dhchap_secret.attr && !ctrl->opts)
3865                 return 0;
3866         if (a == &dev_attr_dhchap_ctrl_secret.attr && !ctrl->opts)
3867                 return 0;
3868 #endif
3869
3870         return a->mode;
3871 }
3872
3873 static const struct attribute_group nvme_dev_attrs_group = {
3874         .attrs          = nvme_dev_attrs,
3875         .is_visible     = nvme_dev_attrs_are_visible,
3876 };
3877
3878 static const struct attribute_group *nvme_dev_attr_groups[] = {
3879         &nvme_dev_attrs_group,
3880         NULL,
3881 };
3882
3883 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3884                 unsigned nsid)
3885 {
3886         struct nvme_ns_head *h;
3887
3888         lockdep_assert_held(&ctrl->subsys->lock);
3889
3890         list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3891                 /*
3892                  * Private namespaces can share NSIDs under some conditions.
3893                  * In that case we can't use the same ns_head for namespaces
3894                  * with the same NSID.
3895                  */
3896                 if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3897                         continue;
3898                 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3899                         return h;
3900         }
3901
3902         return NULL;
3903 }
3904
3905 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3906                 struct nvme_ns_ids *ids)
3907 {
3908         bool has_uuid = !uuid_is_null(&ids->uuid);
3909         bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3910         bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3911         struct nvme_ns_head *h;
3912
3913         lockdep_assert_held(&subsys->lock);
3914
3915         list_for_each_entry(h, &subsys->nsheads, entry) {
3916                 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3917                         return -EINVAL;
3918                 if (has_nguid &&
3919                     memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3920                         return -EINVAL;
3921                 if (has_eui64 &&
3922                     memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3923                         return -EINVAL;
3924         }
3925
3926         return 0;
3927 }
3928
3929 static void nvme_cdev_rel(struct device *dev)
3930 {
3931         ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3932 }
3933
3934 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3935 {
3936         cdev_device_del(cdev, cdev_device);
3937         put_device(cdev_device);
3938 }
3939
3940 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3941                 const struct file_operations *fops, struct module *owner)
3942 {
3943         int minor, ret;
3944
3945         minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3946         if (minor < 0)
3947                 return minor;
3948         cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3949         cdev_device->class = nvme_ns_chr_class;
3950         cdev_device->release = nvme_cdev_rel;
3951         device_initialize(cdev_device);
3952         cdev_init(cdev, fops);
3953         cdev->owner = owner;
3954         ret = cdev_device_add(cdev, cdev_device);
3955         if (ret)
3956                 put_device(cdev_device);
3957
3958         return ret;
3959 }
3960
3961 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3962 {
3963         return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3964 }
3965
3966 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3967 {
3968         nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3969         return 0;
3970 }
3971
3972 static const struct file_operations nvme_ns_chr_fops = {
3973         .owner          = THIS_MODULE,
3974         .open           = nvme_ns_chr_open,
3975         .release        = nvme_ns_chr_release,
3976         .unlocked_ioctl = nvme_ns_chr_ioctl,
3977         .compat_ioctl   = compat_ptr_ioctl,
3978         .uring_cmd      = nvme_ns_chr_uring_cmd,
3979 };
3980
3981 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3982 {
3983         int ret;
3984
3985         ns->cdev_device.parent = ns->ctrl->device;
3986         ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3987                            ns->ctrl->instance, ns->head->instance);
3988         if (ret)
3989                 return ret;
3990
3991         return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3992                              ns->ctrl->ops->module);
3993 }
3994
3995 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3996                 struct nvme_ns_info *info)
3997 {
3998         struct nvme_ns_head *head;
3999         size_t size = sizeof(*head);
4000         int ret = -ENOMEM;
4001
4002 #ifdef CONFIG_NVME_MULTIPATH
4003         size += num_possible_nodes() * sizeof(struct nvme_ns *);
4004 #endif
4005
4006         head = kzalloc(size, GFP_KERNEL);
4007         if (!head)
4008                 goto out;
4009         ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
4010         if (ret < 0)
4011                 goto out_free_head;
4012         head->instance = ret;
4013         INIT_LIST_HEAD(&head->list);
4014         ret = init_srcu_struct(&head->srcu);
4015         if (ret)
4016                 goto out_ida_remove;
4017         head->subsys = ctrl->subsys;
4018         head->ns_id = info->nsid;
4019         head->ids = info->ids;
4020         head->shared = info->is_shared;
4021         kref_init(&head->ref);
4022
4023         if (head->ids.csi) {
4024                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
4025                 if (ret)
4026                         goto out_cleanup_srcu;
4027         } else
4028                 head->effects = ctrl->effects;
4029
4030         ret = nvme_mpath_alloc_disk(ctrl, head);
4031         if (ret)
4032                 goto out_cleanup_srcu;
4033
4034         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
4035
4036         kref_get(&ctrl->subsys->ref);
4037
4038         return head;
4039 out_cleanup_srcu:
4040         cleanup_srcu_struct(&head->srcu);
4041 out_ida_remove:
4042         ida_free(&ctrl->subsys->ns_ida, head->instance);
4043 out_free_head:
4044         kfree(head);
4045 out:
4046         if (ret > 0)
4047                 ret = blk_status_to_errno(nvme_error_status(ret));
4048         return ERR_PTR(ret);
4049 }
4050
4051 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
4052                 struct nvme_ns_ids *ids)
4053 {
4054         struct nvme_subsystem *s;
4055         int ret = 0;
4056
4057         /*
4058          * Note that this check is racy as we try to avoid holding the global
4059          * lock over the whole ns_head creation.  But it is only intended as
4060          * a sanity check anyway.
4061          */
4062         mutex_lock(&nvme_subsystems_lock);
4063         list_for_each_entry(s, &nvme_subsystems, entry) {
4064                 if (s == this)
4065                         continue;
4066                 mutex_lock(&s->lock);
4067                 ret = nvme_subsys_check_duplicate_ids(s, ids);
4068                 mutex_unlock(&s->lock);
4069                 if (ret)
4070                         break;
4071         }
4072         mutex_unlock(&nvme_subsystems_lock);
4073
4074         return ret;
4075 }
4076
4077 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
4078 {
4079         struct nvme_ctrl *ctrl = ns->ctrl;
4080         struct nvme_ns_head *head = NULL;
4081         int ret;
4082
4083         ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
4084         if (ret) {
4085                 dev_err(ctrl->device,
4086                         "globally duplicate IDs for nsid %d\n", info->nsid);
4087                 nvme_print_device_info(ctrl);
4088                 return ret;
4089         }
4090
4091         mutex_lock(&ctrl->subsys->lock);
4092         head = nvme_find_ns_head(ctrl, info->nsid);
4093         if (!head) {
4094                 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
4095                 if (ret) {
4096                         dev_err(ctrl->device,
4097                                 "duplicate IDs in subsystem for nsid %d\n",
4098                                 info->nsid);
4099                         goto out_unlock;
4100                 }
4101                 head = nvme_alloc_ns_head(ctrl, info);
4102                 if (IS_ERR(head)) {
4103                         ret = PTR_ERR(head);
4104                         goto out_unlock;
4105                 }
4106         } else {
4107                 ret = -EINVAL;
4108                 if (!info->is_shared || !head->shared) {
4109                         dev_err(ctrl->device,
4110                                 "Duplicate unshared namespace %d\n",
4111                                 info->nsid);
4112                         goto out_put_ns_head;
4113                 }
4114                 if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
4115                         dev_err(ctrl->device,
4116                                 "IDs don't match for shared namespace %d\n",
4117                                         info->nsid);
4118                         goto out_put_ns_head;
4119                 }
4120
4121                 if (!multipath && !list_empty(&head->list)) {
4122                         dev_warn(ctrl->device,
4123                                 "Found shared namespace %d, but multipathing not supported.\n",
4124                                 info->nsid);
4125                         dev_warn_once(ctrl->device,
4126                                 "Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
4127                 }
4128         }
4129
4130         list_add_tail_rcu(&ns->siblings, &head->list);
4131         ns->head = head;
4132         mutex_unlock(&ctrl->subsys->lock);
4133         return 0;
4134
4135 out_put_ns_head:
4136         nvme_put_ns_head(head);
4137 out_unlock:
4138         mutex_unlock(&ctrl->subsys->lock);
4139         return ret;
4140 }
4141
4142 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4143 {
4144         struct nvme_ns *ns, *ret = NULL;
4145
4146         down_read(&ctrl->namespaces_rwsem);
4147         list_for_each_entry(ns, &ctrl->namespaces, list) {
4148                 if (ns->head->ns_id == nsid) {
4149                         if (!nvme_get_ns(ns))
4150                                 continue;
4151                         ret = ns;
4152                         break;
4153                 }
4154                 if (ns->head->ns_id > nsid)
4155                         break;
4156         }
4157         up_read(&ctrl->namespaces_rwsem);
4158         return ret;
4159 }
4160 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
4161
4162 /*
4163  * Add the namespace to the controller list while keeping the list ordered.
4164  */
4165 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
4166 {
4167         struct nvme_ns *tmp;
4168
4169         list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
4170                 if (tmp->head->ns_id < ns->head->ns_id) {
4171                         list_add(&ns->list, &tmp->list);
4172                         return;
4173                 }
4174         }
4175         list_add(&ns->list, &ns->ctrl->namespaces);
4176 }
4177
4178 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
4179 {
4180         struct nvme_ns *ns;
4181         struct gendisk *disk;
4182         int node = ctrl->numa_node;
4183
4184         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4185         if (!ns)
4186                 return;
4187
4188         disk = blk_mq_alloc_disk(ctrl->tagset, ns);
4189         if (IS_ERR(disk))
4190                 goto out_free_ns;
4191         disk->fops = &nvme_bdev_ops;
4192         disk->private_data = ns;
4193
4194         ns->disk = disk;
4195         ns->queue = disk->queue;
4196
4197         if (ctrl->opts && ctrl->opts->data_digest)
4198                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
4199
4200         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
4201         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
4202                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4203
4204         ns->ctrl = ctrl;
4205         kref_init(&ns->kref);
4206
4207         if (nvme_init_ns_head(ns, info))
4208                 goto out_cleanup_disk;
4209
4210         /*
4211          * If multipathing is enabled, the device name for all disks and not
4212          * just those that represent shared namespaces needs to be based on the
4213          * subsystem instance.  Using the controller instance for private
4214          * namespaces could lead to naming collisions between shared and private
4215          * namespaces if they don't use a common numbering scheme.
4216          *
4217          * If multipathing is not enabled, disk names must use the controller
4218          * instance as shared namespaces will show up as multiple block
4219          * devices.
4220          */
4221         if (ns->head->disk) {
4222                 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
4223                         ctrl->instance, ns->head->instance);
4224                 disk->flags |= GENHD_FL_HIDDEN;
4225         } else if (multipath) {
4226                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4227                         ns->head->instance);
4228         } else {
4229                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4230                         ns->head->instance);
4231         }
4232
4233         if (nvme_update_ns_info(ns, info))
4234                 goto out_unlink_ns;
4235
4236         down_write(&ctrl->namespaces_rwsem);
4237         nvme_ns_add_to_ctrl_list(ns);
4238         up_write(&ctrl->namespaces_rwsem);
4239         nvme_get_ctrl(ctrl);
4240
4241         if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
4242                 goto out_cleanup_ns_from_list;
4243
4244         if (!nvme_ns_head_multipath(ns->head))
4245                 nvme_add_ns_cdev(ns);
4246
4247         nvme_mpath_add_disk(ns, info->anagrpid);
4248         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4249
4250         return;
4251
4252  out_cleanup_ns_from_list:
4253         nvme_put_ctrl(ctrl);
4254         down_write(&ctrl->namespaces_rwsem);
4255         list_del_init(&ns->list);
4256         up_write(&ctrl->namespaces_rwsem);
4257  out_unlink_ns:
4258         mutex_lock(&ctrl->subsys->lock);
4259         list_del_rcu(&ns->siblings);
4260         if (list_empty(&ns->head->list))
4261                 list_del_init(&ns->head->entry);
4262         mutex_unlock(&ctrl->subsys->lock);
4263         nvme_put_ns_head(ns->head);
4264  out_cleanup_disk:
4265         put_disk(disk);
4266  out_free_ns:
4267         kfree(ns);
4268 }
4269
4270 static void nvme_ns_remove(struct nvme_ns *ns)
4271 {
4272         bool last_path = false;
4273
4274         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4275                 return;
4276
4277         clear_bit(NVME_NS_READY, &ns->flags);
4278         set_capacity(ns->disk, 0);
4279         nvme_fault_inject_fini(&ns->fault_inject);
4280
4281         /*
4282          * Ensure that !NVME_NS_READY is seen by other threads to prevent
4283          * this ns going back into current_path.
4284          */
4285         synchronize_srcu(&ns->head->srcu);
4286
4287         /* wait for concurrent submissions */
4288         if (nvme_mpath_clear_current_path(ns))
4289                 synchronize_srcu(&ns->head->srcu);
4290
4291         mutex_lock(&ns->ctrl->subsys->lock);
4292         list_del_rcu(&ns->siblings);
4293         if (list_empty(&ns->head->list)) {
4294                 list_del_init(&ns->head->entry);
4295                 last_path = true;
4296         }
4297         mutex_unlock(&ns->ctrl->subsys->lock);
4298
4299         /* guarantee not available in head->list */
4300         synchronize_rcu();
4301
4302         if (!nvme_ns_head_multipath(ns->head))
4303                 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4304         del_gendisk(ns->disk);
4305
4306         down_write(&ns->ctrl->namespaces_rwsem);
4307         list_del_init(&ns->list);
4308         up_write(&ns->ctrl->namespaces_rwsem);
4309
4310         if (last_path)
4311                 nvme_mpath_shutdown_disk(ns->head);
4312         nvme_put_ns(ns);
4313 }
4314
4315 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4316 {
4317         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4318
4319         if (ns) {
4320                 nvme_ns_remove(ns);
4321                 nvme_put_ns(ns);
4322         }
4323 }
4324
4325 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
4326 {
4327         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4328
4329         if (test_bit(NVME_NS_DEAD, &ns->flags))
4330                 goto out;
4331
4332         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4333         if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
4334                 dev_err(ns->ctrl->device,
4335                         "identifiers changed for nsid %d\n", ns->head->ns_id);
4336                 goto out;
4337         }
4338
4339         ret = nvme_update_ns_info(ns, info);
4340 out:
4341         /*
4342          * Only remove the namespace if we got a fatal error back from the
4343          * device, otherwise ignore the error and just move on.
4344          *
4345          * TODO: we should probably schedule a delayed retry here.
4346          */
4347         if (ret > 0 && (ret & NVME_SC_DNR))
4348                 nvme_ns_remove(ns);
4349 }
4350
4351 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4352 {
4353         struct nvme_ns_info info = { .nsid = nsid };
4354         struct nvme_ns *ns;
4355
4356         if (nvme_identify_ns_descs(ctrl, &info))
4357                 return;
4358
4359         if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
4360                 dev_warn(ctrl->device,
4361                         "command set not reported for nsid: %d\n", nsid);
4362                 return;
4363         }
4364
4365         /*
4366          * If available try to use the Command Set Idependent Identify Namespace
4367          * data structure to find all the generic information that is needed to
4368          * set up a namespace.  If not fall back to the legacy version.
4369          */
4370         if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
4371             (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS)) {
4372                 if (nvme_ns_info_from_id_cs_indep(ctrl, &info))
4373                         return;
4374         } else {
4375                 if (nvme_ns_info_from_identify(ctrl, &info))
4376                         return;
4377         }
4378
4379         /*
4380          * Ignore the namespace if it is not ready. We will get an AEN once it
4381          * becomes ready and restart the scan.
4382          */
4383         if (!info.is_ready)
4384                 return;
4385
4386         ns = nvme_find_get_ns(ctrl, nsid);
4387         if (ns) {
4388                 nvme_validate_ns(ns, &info);
4389                 nvme_put_ns(ns);
4390         } else {
4391                 nvme_alloc_ns(ctrl, &info);
4392         }
4393 }
4394
4395 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4396                                         unsigned nsid)
4397 {
4398         struct nvme_ns *ns, *next;
4399         LIST_HEAD(rm_list);
4400
4401         down_write(&ctrl->namespaces_rwsem);
4402         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4403                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4404                         list_move_tail(&ns->list, &rm_list);
4405         }
4406         up_write(&ctrl->namespaces_rwsem);
4407
4408         list_for_each_entry_safe(ns, next, &rm_list, list)
4409                 nvme_ns_remove(ns);
4410
4411 }
4412
4413 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4414 {
4415         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4416         __le32 *ns_list;
4417         u32 prev = 0;
4418         int ret = 0, i;
4419
4420         if (nvme_ctrl_limited_cns(ctrl))
4421                 return -EOPNOTSUPP;
4422
4423         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4424         if (!ns_list)
4425                 return -ENOMEM;
4426
4427         for (;;) {
4428                 struct nvme_command cmd = {
4429                         .identify.opcode        = nvme_admin_identify,
4430                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
4431                         .identify.nsid          = cpu_to_le32(prev),
4432                 };
4433
4434                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4435                                             NVME_IDENTIFY_DATA_SIZE);
4436                 if (ret) {
4437                         dev_warn(ctrl->device,
4438                                 "Identify NS List failed (status=0x%x)\n", ret);
4439                         goto free;
4440                 }
4441
4442                 for (i = 0; i < nr_entries; i++) {
4443                         u32 nsid = le32_to_cpu(ns_list[i]);
4444
4445                         if (!nsid)      /* end of the list? */
4446                                 goto out;
4447                         nvme_scan_ns(ctrl, nsid);
4448                         while (++prev < nsid)
4449                                 nvme_ns_remove_by_nsid(ctrl, prev);
4450                 }
4451         }
4452  out:
4453         nvme_remove_invalid_namespaces(ctrl, prev);
4454  free:
4455         kfree(ns_list);
4456         return ret;
4457 }
4458
4459 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4460 {
4461         struct nvme_id_ctrl *id;
4462         u32 nn, i;
4463
4464         if (nvme_identify_ctrl(ctrl, &id))
4465                 return;
4466         nn = le32_to_cpu(id->nn);
4467         kfree(id);
4468
4469         for (i = 1; i <= nn; i++)
4470                 nvme_scan_ns(ctrl, i);
4471
4472         nvme_remove_invalid_namespaces(ctrl, nn);
4473 }
4474
4475 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4476 {
4477         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4478         __le32 *log;
4479         int error;
4480
4481         log = kzalloc(log_size, GFP_KERNEL);
4482         if (!log)
4483                 return;
4484
4485         /*
4486          * We need to read the log to clear the AEN, but we don't want to rely
4487          * on it for the changed namespace information as userspace could have
4488          * raced with us in reading the log page, which could cause us to miss
4489          * updates.
4490          */
4491         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4492                         NVME_CSI_NVM, log, log_size, 0);
4493         if (error)
4494                 dev_warn(ctrl->device,
4495                         "reading changed ns log failed: %d\n", error);
4496
4497         kfree(log);
4498 }
4499
4500 static void nvme_scan_work(struct work_struct *work)
4501 {
4502         struct nvme_ctrl *ctrl =
4503                 container_of(work, struct nvme_ctrl, scan_work);
4504         int ret;
4505
4506         /* No tagset on a live ctrl means IO queues could not created */
4507         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4508                 return;
4509
4510         /*
4511          * Identify controller limits can change at controller reset due to
4512          * new firmware download, even though it is not common we cannot ignore
4513          * such scenario. Controller's non-mdts limits are reported in the unit
4514          * of logical blocks that is dependent on the format of attached
4515          * namespace. Hence re-read the limits at the time of ns allocation.
4516          */
4517         ret = nvme_init_non_mdts_limits(ctrl);
4518         if (ret < 0) {
4519                 dev_warn(ctrl->device,
4520                         "reading non-mdts-limits failed: %d\n", ret);
4521                 return;
4522         }
4523
4524         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4525                 dev_info(ctrl->device, "rescanning namespaces.\n");
4526                 nvme_clear_changed_ns_log(ctrl);
4527         }
4528
4529         mutex_lock(&ctrl->scan_lock);
4530         if (nvme_scan_ns_list(ctrl) != 0)
4531                 nvme_scan_ns_sequential(ctrl);
4532         mutex_unlock(&ctrl->scan_lock);
4533 }
4534
4535 /*
4536  * This function iterates the namespace list unlocked to allow recovery from
4537  * controller failure. It is up to the caller to ensure the namespace list is
4538  * not modified by scan work while this function is executing.
4539  */
4540 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4541 {
4542         struct nvme_ns *ns, *next;
4543         LIST_HEAD(ns_list);
4544
4545         /*
4546          * make sure to requeue I/O to all namespaces as these
4547          * might result from the scan itself and must complete
4548          * for the scan_work to make progress
4549          */
4550         nvme_mpath_clear_ctrl_paths(ctrl);
4551
4552         /* prevent racing with ns scanning */
4553         flush_work(&ctrl->scan_work);
4554
4555         /*
4556          * The dead states indicates the controller was not gracefully
4557          * disconnected. In that case, we won't be able to flush any data while
4558          * removing the namespaces' disks; fail all the queues now to avoid
4559          * potentially having to clean up the failed sync later.
4560          */
4561         if (ctrl->state == NVME_CTRL_DEAD)
4562                 nvme_kill_queues(ctrl);
4563
4564         /* this is a no-op when called from the controller reset handler */
4565         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4566
4567         down_write(&ctrl->namespaces_rwsem);
4568         list_splice_init(&ctrl->namespaces, &ns_list);
4569         up_write(&ctrl->namespaces_rwsem);
4570
4571         list_for_each_entry_safe(ns, next, &ns_list, list)
4572                 nvme_ns_remove(ns);
4573 }
4574 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4575
4576 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4577 {
4578         struct nvme_ctrl *ctrl =
4579                 container_of(dev, struct nvme_ctrl, ctrl_device);
4580         struct nvmf_ctrl_options *opts = ctrl->opts;
4581         int ret;
4582
4583         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4584         if (ret)
4585                 return ret;
4586
4587         if (opts) {
4588                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4589                 if (ret)
4590                         return ret;
4591
4592                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4593                                 opts->trsvcid ?: "none");
4594                 if (ret)
4595                         return ret;
4596
4597                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4598                                 opts->host_traddr ?: "none");
4599                 if (ret)
4600                         return ret;
4601
4602                 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4603                                 opts->host_iface ?: "none");
4604         }
4605         return ret;
4606 }
4607
4608 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4609 {
4610         char *envp[2] = { envdata, NULL };
4611
4612         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4613 }
4614
4615 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4616 {
4617         char *envp[2] = { NULL, NULL };
4618         u32 aen_result = ctrl->aen_result;
4619
4620         ctrl->aen_result = 0;
4621         if (!aen_result)
4622                 return;
4623
4624         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4625         if (!envp[0])
4626                 return;
4627         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4628         kfree(envp[0]);
4629 }
4630
4631 static void nvme_async_event_work(struct work_struct *work)
4632 {
4633         struct nvme_ctrl *ctrl =
4634                 container_of(work, struct nvme_ctrl, async_event_work);
4635
4636         nvme_aen_uevent(ctrl);
4637
4638         /*
4639          * The transport drivers must guarantee AER submission here is safe by
4640          * flushing ctrl async_event_work after changing the controller state
4641          * from LIVE and before freeing the admin queue.
4642         */
4643         if (ctrl->state == NVME_CTRL_LIVE)
4644                 ctrl->ops->submit_async_event(ctrl);
4645 }
4646
4647 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4648 {
4649
4650         u32 csts;
4651
4652         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4653                 return false;
4654
4655         if (csts == ~0)
4656                 return false;
4657
4658         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4659 }
4660
4661 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4662 {
4663         struct nvme_fw_slot_info_log *log;
4664
4665         log = kmalloc(sizeof(*log), GFP_KERNEL);
4666         if (!log)
4667                 return;
4668
4669         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4670                         log, sizeof(*log), 0))
4671                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4672         kfree(log);
4673 }
4674
4675 static void nvme_fw_act_work(struct work_struct *work)
4676 {
4677         struct nvme_ctrl *ctrl = container_of(work,
4678                                 struct nvme_ctrl, fw_act_work);
4679         unsigned long fw_act_timeout;
4680
4681         if (ctrl->mtfa)
4682                 fw_act_timeout = jiffies +
4683                                 msecs_to_jiffies(ctrl->mtfa * 100);
4684         else
4685                 fw_act_timeout = jiffies +
4686                                 msecs_to_jiffies(admin_timeout * 1000);
4687
4688         nvme_stop_queues(ctrl);
4689         while (nvme_ctrl_pp_status(ctrl)) {
4690                 if (time_after(jiffies, fw_act_timeout)) {
4691                         dev_warn(ctrl->device,
4692                                 "Fw activation timeout, reset controller\n");
4693                         nvme_try_sched_reset(ctrl);
4694                         return;
4695                 }
4696                 msleep(100);
4697         }
4698
4699         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4700                 return;
4701
4702         nvme_start_queues(ctrl);
4703         /* read FW slot information to clear the AER */
4704         nvme_get_fw_slot_info(ctrl);
4705 }
4706
4707 static u32 nvme_aer_type(u32 result)
4708 {
4709         return result & 0x7;
4710 }
4711
4712 static u32 nvme_aer_subtype(u32 result)
4713 {
4714         return (result & 0xff00) >> 8;
4715 }
4716
4717 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4718 {
4719         u32 aer_notice_type = nvme_aer_subtype(result);
4720
4721         trace_nvme_async_event(ctrl, aer_notice_type);
4722
4723         switch (aer_notice_type) {
4724         case NVME_AER_NOTICE_NS_CHANGED:
4725                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4726                 nvme_queue_scan(ctrl);
4727                 break;
4728         case NVME_AER_NOTICE_FW_ACT_STARTING:
4729                 /*
4730                  * We are (ab)using the RESETTING state to prevent subsequent
4731                  * recovery actions from interfering with the controller's
4732                  * firmware activation.
4733                  */
4734                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4735                         nvme_auth_stop(ctrl);
4736                         queue_work(nvme_wq, &ctrl->fw_act_work);
4737                 }
4738                 break;
4739 #ifdef CONFIG_NVME_MULTIPATH
4740         case NVME_AER_NOTICE_ANA:
4741                 if (!ctrl->ana_log_buf)
4742                         break;
4743                 queue_work(nvme_wq, &ctrl->ana_work);
4744                 break;
4745 #endif
4746         case NVME_AER_NOTICE_DISC_CHANGED:
4747                 ctrl->aen_result = result;
4748                 break;
4749         default:
4750                 dev_warn(ctrl->device, "async event result %08x\n", result);
4751         }
4752 }
4753
4754 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4755 {
4756         trace_nvme_async_event(ctrl, NVME_AER_ERROR);
4757         dev_warn(ctrl->device, "resetting controller due to AER\n");
4758         nvme_reset_ctrl(ctrl);
4759 }
4760
4761 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4762                 volatile union nvme_result *res)
4763 {
4764         u32 result = le32_to_cpu(res->u32);
4765         u32 aer_type = nvme_aer_type(result);
4766         u32 aer_subtype = nvme_aer_subtype(result);
4767
4768         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4769                 return;
4770
4771         switch (aer_type) {
4772         case NVME_AER_NOTICE:
4773                 nvme_handle_aen_notice(ctrl, result);
4774                 break;
4775         case NVME_AER_ERROR:
4776                 /*
4777                  * For a persistent internal error, don't run async_event_work
4778                  * to submit a new AER. The controller reset will do it.
4779                  */
4780                 if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4781                         nvme_handle_aer_persistent_error(ctrl);
4782                         return;
4783                 }
4784                 fallthrough;
4785         case NVME_AER_SMART:
4786         case NVME_AER_CSS:
4787         case NVME_AER_VS:
4788                 trace_nvme_async_event(ctrl, aer_type);
4789                 ctrl->aen_result = result;
4790                 break;
4791         default:
4792                 break;
4793         }
4794         queue_work(nvme_wq, &ctrl->async_event_work);
4795 }
4796 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4797
4798 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4799 {
4800         nvme_mpath_stop(ctrl);
4801         nvme_auth_stop(ctrl);
4802         nvme_stop_keep_alive(ctrl);
4803         nvme_stop_failfast_work(ctrl);
4804         flush_work(&ctrl->async_event_work);
4805         cancel_work_sync(&ctrl->fw_act_work);
4806         if (ctrl->ops->stop_ctrl)
4807                 ctrl->ops->stop_ctrl(ctrl);
4808 }
4809 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4810
4811 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4812 {
4813         nvme_start_keep_alive(ctrl);
4814
4815         nvme_enable_aen(ctrl);
4816
4817         if (ctrl->queue_count > 1) {
4818                 nvme_queue_scan(ctrl);
4819                 nvme_start_queues(ctrl);
4820                 nvme_mpath_update(ctrl);
4821         }
4822
4823         nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4824 }
4825 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4826
4827 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4828 {
4829         nvme_hwmon_exit(ctrl);
4830         nvme_fault_inject_fini(&ctrl->fault_inject);
4831         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4832         cdev_device_del(&ctrl->cdev, ctrl->device);
4833         nvme_put_ctrl(ctrl);
4834 }
4835 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4836
4837 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4838 {
4839         struct nvme_effects_log *cel;
4840         unsigned long i;
4841
4842         xa_for_each(&ctrl->cels, i, cel) {
4843                 xa_erase(&ctrl->cels, i);
4844                 kfree(cel);
4845         }
4846
4847         xa_destroy(&ctrl->cels);
4848 }
4849
4850 static void nvme_free_ctrl(struct device *dev)
4851 {
4852         struct nvme_ctrl *ctrl =
4853                 container_of(dev, struct nvme_ctrl, ctrl_device);
4854         struct nvme_subsystem *subsys = ctrl->subsys;
4855
4856         if (!subsys || ctrl->instance != subsys->instance)
4857                 ida_free(&nvme_instance_ida, ctrl->instance);
4858
4859         nvme_free_cels(ctrl);
4860         nvme_mpath_uninit(ctrl);
4861         nvme_auth_stop(ctrl);
4862         nvme_auth_free(ctrl);
4863         __free_page(ctrl->discard_page);
4864
4865         if (subsys) {
4866                 mutex_lock(&nvme_subsystems_lock);
4867                 list_del(&ctrl->subsys_entry);
4868                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4869                 mutex_unlock(&nvme_subsystems_lock);
4870         }
4871
4872         ctrl->ops->free_ctrl(ctrl);
4873
4874         if (subsys)
4875                 nvme_put_subsystem(subsys);
4876 }
4877
4878 /*
4879  * Initialize a NVMe controller structures.  This needs to be called during
4880  * earliest initialization so that we have the initialized structured around
4881  * during probing.
4882  */
4883 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4884                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4885 {
4886         int ret;
4887
4888         ctrl->state = NVME_CTRL_NEW;
4889         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4890         spin_lock_init(&ctrl->lock);
4891         mutex_init(&ctrl->scan_lock);
4892         INIT_LIST_HEAD(&ctrl->namespaces);
4893         xa_init(&ctrl->cels);
4894         init_rwsem(&ctrl->namespaces_rwsem);
4895         ctrl->dev = dev;
4896         ctrl->ops = ops;
4897         ctrl->quirks = quirks;
4898         ctrl->numa_node = NUMA_NO_NODE;
4899         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4900         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4901         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4902         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4903         init_waitqueue_head(&ctrl->state_wq);
4904
4905         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4906         INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4907         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4908         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4909
4910         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4911                         PAGE_SIZE);
4912         ctrl->discard_page = alloc_page(GFP_KERNEL);
4913         if (!ctrl->discard_page) {
4914                 ret = -ENOMEM;
4915                 goto out;
4916         }
4917
4918         ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4919         if (ret < 0)
4920                 goto out;
4921         ctrl->instance = ret;
4922
4923         device_initialize(&ctrl->ctrl_device);
4924         ctrl->device = &ctrl->ctrl_device;
4925         ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4926                         ctrl->instance);
4927         ctrl->device->class = nvme_class;
4928         ctrl->device->parent = ctrl->dev;
4929         ctrl->device->groups = nvme_dev_attr_groups;
4930         ctrl->device->release = nvme_free_ctrl;
4931         dev_set_drvdata(ctrl->device, ctrl);
4932         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4933         if (ret)
4934                 goto out_release_instance;
4935
4936         nvme_get_ctrl(ctrl);
4937         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4938         ctrl->cdev.owner = ops->module;
4939         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4940         if (ret)
4941                 goto out_free_name;
4942
4943         /*
4944          * Initialize latency tolerance controls.  The sysfs files won't
4945          * be visible to userspace unless the device actually supports APST.
4946          */
4947         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4948         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4949                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4950
4951         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4952         nvme_mpath_init_ctrl(ctrl);
4953         nvme_auth_init_ctrl(ctrl);
4954
4955         return 0;
4956 out_free_name:
4957         nvme_put_ctrl(ctrl);
4958         kfree_const(ctrl->device->kobj.name);
4959 out_release_instance:
4960         ida_free(&nvme_instance_ida, ctrl->instance);
4961 out:
4962         if (ctrl->discard_page)
4963                 __free_page(ctrl->discard_page);
4964         return ret;
4965 }
4966 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4967
4968 static void nvme_start_ns_queue(struct nvme_ns *ns)
4969 {
4970         if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4971                 blk_mq_unquiesce_queue(ns->queue);
4972 }
4973
4974 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4975 {
4976         if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4977                 blk_mq_quiesce_queue(ns->queue);
4978         else
4979                 blk_mq_wait_quiesce_done(ns->queue);
4980 }
4981
4982 /*
4983  * Prepare a queue for teardown.
4984  *
4985  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4986  * the capacity to 0 after that to avoid blocking dispatchers that may be
4987  * holding bd_butex.  This will end buffered writers dirtying pages that can't
4988  * be synced.
4989  */
4990 static void nvme_set_queue_dying(struct nvme_ns *ns)
4991 {
4992         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4993                 return;
4994
4995         blk_mark_disk_dead(ns->disk);
4996         nvme_start_ns_queue(ns);
4997
4998         set_capacity_and_notify(ns->disk, 0);
4999 }
5000
5001 /**
5002  * nvme_kill_queues(): Ends all namespace queues
5003  * @ctrl: the dead controller that needs to end
5004  *
5005  * Call this function when the driver determines it is unable to get the
5006  * controller in a state capable of servicing IO.
5007  */
5008 void nvme_kill_queues(struct nvme_ctrl *ctrl)
5009 {
5010         struct nvme_ns *ns;
5011
5012         down_read(&ctrl->namespaces_rwsem);
5013
5014         /* Forcibly unquiesce queues to avoid blocking dispatch */
5015         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
5016                 nvme_start_admin_queue(ctrl);
5017
5018         list_for_each_entry(ns, &ctrl->namespaces, list)
5019                 nvme_set_queue_dying(ns);
5020
5021         up_read(&ctrl->namespaces_rwsem);
5022 }
5023 EXPORT_SYMBOL_GPL(nvme_kill_queues);
5024
5025 void nvme_unfreeze(struct nvme_ctrl *ctrl)
5026 {
5027         struct nvme_ns *ns;
5028
5029         down_read(&ctrl->namespaces_rwsem);
5030         list_for_each_entry(ns, &ctrl->namespaces, list)
5031                 blk_mq_unfreeze_queue(ns->queue);
5032         up_read(&ctrl->namespaces_rwsem);
5033 }
5034 EXPORT_SYMBOL_GPL(nvme_unfreeze);
5035
5036 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
5037 {
5038         struct nvme_ns *ns;
5039
5040         down_read(&ctrl->namespaces_rwsem);
5041         list_for_each_entry(ns, &ctrl->namespaces, list) {
5042                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
5043                 if (timeout <= 0)
5044                         break;
5045         }
5046         up_read(&ctrl->namespaces_rwsem);
5047         return timeout;
5048 }
5049 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
5050
5051 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
5052 {
5053         struct nvme_ns *ns;
5054
5055         down_read(&ctrl->namespaces_rwsem);
5056         list_for_each_entry(ns, &ctrl->namespaces, list)
5057                 blk_mq_freeze_queue_wait(ns->queue);
5058         up_read(&ctrl->namespaces_rwsem);
5059 }
5060 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
5061
5062 void nvme_start_freeze(struct nvme_ctrl *ctrl)
5063 {
5064         struct nvme_ns *ns;
5065
5066         down_read(&ctrl->namespaces_rwsem);
5067         list_for_each_entry(ns, &ctrl->namespaces, list)
5068                 blk_freeze_queue_start(ns->queue);
5069         up_read(&ctrl->namespaces_rwsem);
5070 }
5071 EXPORT_SYMBOL_GPL(nvme_start_freeze);
5072
5073 void nvme_stop_queues(struct nvme_ctrl *ctrl)
5074 {
5075         struct nvme_ns *ns;
5076
5077         down_read(&ctrl->namespaces_rwsem);
5078         list_for_each_entry(ns, &ctrl->namespaces, list)
5079                 nvme_stop_ns_queue(ns);
5080         up_read(&ctrl->namespaces_rwsem);
5081 }
5082 EXPORT_SYMBOL_GPL(nvme_stop_queues);
5083
5084 void nvme_start_queues(struct nvme_ctrl *ctrl)
5085 {
5086         struct nvme_ns *ns;
5087
5088         down_read(&ctrl->namespaces_rwsem);
5089         list_for_each_entry(ns, &ctrl->namespaces, list)
5090                 nvme_start_ns_queue(ns);
5091         up_read(&ctrl->namespaces_rwsem);
5092 }
5093 EXPORT_SYMBOL_GPL(nvme_start_queues);
5094
5095 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
5096 {
5097         if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5098                 blk_mq_quiesce_queue(ctrl->admin_q);
5099         else
5100                 blk_mq_wait_quiesce_done(ctrl->admin_q);
5101 }
5102 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
5103
5104 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
5105 {
5106         if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5107                 blk_mq_unquiesce_queue(ctrl->admin_q);
5108 }
5109 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
5110
5111 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
5112 {
5113         struct nvme_ns *ns;
5114
5115         down_read(&ctrl->namespaces_rwsem);
5116         list_for_each_entry(ns, &ctrl->namespaces, list)
5117                 blk_sync_queue(ns->queue);
5118         up_read(&ctrl->namespaces_rwsem);
5119 }
5120 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
5121
5122 void nvme_sync_queues(struct nvme_ctrl *ctrl)
5123 {
5124         nvme_sync_io_queues(ctrl);
5125         if (ctrl->admin_q)
5126                 blk_sync_queue(ctrl->admin_q);
5127 }
5128 EXPORT_SYMBOL_GPL(nvme_sync_queues);
5129
5130 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
5131 {
5132         if (file->f_op != &nvme_dev_fops)
5133                 return NULL;
5134         return file->private_data;
5135 }
5136 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
5137
5138 /*
5139  * Check we didn't inadvertently grow the command structure sizes:
5140  */
5141 static inline void _nvme_check_size(void)
5142 {
5143         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
5144         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
5145         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
5146         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
5147         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
5148         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
5149         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
5150         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
5151         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
5152         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
5153         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
5154         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
5155         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
5156         BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
5157                         NVME_IDENTIFY_DATA_SIZE);
5158         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
5159         BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
5160         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
5161         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
5162         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
5163         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
5164         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
5165         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
5166         BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
5167 }
5168
5169
5170 static int __init nvme_core_init(void)
5171 {
5172         int result = -ENOMEM;
5173
5174         _nvme_check_size();
5175
5176         nvme_wq = alloc_workqueue("nvme-wq",
5177                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5178         if (!nvme_wq)
5179                 goto out;
5180
5181         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
5182                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5183         if (!nvme_reset_wq)
5184                 goto destroy_wq;
5185
5186         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
5187                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5188         if (!nvme_delete_wq)
5189                 goto destroy_reset_wq;
5190
5191         result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
5192                         NVME_MINORS, "nvme");
5193         if (result < 0)
5194                 goto destroy_delete_wq;
5195
5196         nvme_class = class_create(THIS_MODULE, "nvme");
5197         if (IS_ERR(nvme_class)) {
5198                 result = PTR_ERR(nvme_class);
5199                 goto unregister_chrdev;
5200         }
5201         nvme_class->dev_uevent = nvme_class_uevent;
5202
5203         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
5204         if (IS_ERR(nvme_subsys_class)) {
5205                 result = PTR_ERR(nvme_subsys_class);
5206                 goto destroy_class;
5207         }
5208
5209         result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
5210                                      "nvme-generic");
5211         if (result < 0)
5212                 goto destroy_subsys_class;
5213
5214         nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
5215         if (IS_ERR(nvme_ns_chr_class)) {
5216                 result = PTR_ERR(nvme_ns_chr_class);
5217                 goto unregister_generic_ns;
5218         }
5219
5220         return 0;
5221
5222 unregister_generic_ns:
5223         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5224 destroy_subsys_class:
5225         class_destroy(nvme_subsys_class);
5226 destroy_class:
5227         class_destroy(nvme_class);
5228 unregister_chrdev:
5229         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5230 destroy_delete_wq:
5231         destroy_workqueue(nvme_delete_wq);
5232 destroy_reset_wq:
5233         destroy_workqueue(nvme_reset_wq);
5234 destroy_wq:
5235         destroy_workqueue(nvme_wq);
5236 out:
5237         return result;
5238 }
5239
5240 static void __exit nvme_core_exit(void)
5241 {
5242         class_destroy(nvme_ns_chr_class);
5243         class_destroy(nvme_subsys_class);
5244         class_destroy(nvme_class);
5245         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5246         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5247         destroy_workqueue(nvme_delete_wq);
5248         destroy_workqueue(nvme_reset_wq);
5249         destroy_workqueue(nvme_wq);
5250         ida_destroy(&nvme_ns_chr_minor_ida);
5251         ida_destroy(&nvme_instance_ida);
5252 }
5253
5254 MODULE_LICENSE("GPL");
5255 MODULE_VERSION("1.0");
5256 module_init(nvme_core_init);
5257 module_exit(nvme_core_exit);