drivers/nvme/host/core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * NVM Express device driver
   4  * Copyright (c) 2011-2014, Intel Corporation.
   5  */
   6
   7 #include <linux/blkdev.h>
   8 #include <linux/blk-mq.h>
   9 #include <linux/compat.h>
  10 #include <linux/delay.h>
  11 #include <linux/errno.h>
  12 #include <linux/hdreg.h>
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/backing-dev.h>
  16 #include <linux/slab.h>
  17 #include <linux/types.h>
  18 #include <linux/pr.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/nvme_ioctl.h>
  21 #include <linux/pm_qos.h>
  22 #include <asm/unaligned.h>
  23
  24 #include "nvme.h"
  25 #include "fabrics.h"
  26
  27 #define CREATE_TRACE_POINTS
  28 #include "trace.h"
  29
  30 #define NVME_MINORS             (1U << MINORBITS)
  31
  32 unsigned int admin_timeout = 60;
  33 module_param(admin_timeout, uint, 0644);
  34 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  35 EXPORT_SYMBOL_GPL(admin_timeout);
  36
  37 unsigned int nvme_io_timeout = 30;
  38 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  39 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  40 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  41
  42 static unsigned char shutdown_timeout = 5;
  43 module_param(shutdown_timeout, byte, 0644);
  44 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  45
  46 static u8 nvme_max_retries = 5;
  47 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  48 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  49
  50 static unsigned long default_ps_max_latency_us = 100000;
  51 module_param(default_ps_max_latency_us, ulong, 0644);
  52 MODULE_PARM_DESC(default_ps_max_latency_us,
  53                  "max power saving latency for new devices; use PM QOS to change per device");
  54
  55 static bool force_apst;
  56 module_param(force_apst, bool, 0644);
  57 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  58
  59 static bool streams;
  60 module_param(streams, bool, 0644);
  61 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  62
  63 /*
  64  * nvme_wq - hosts nvme related works that are not reset or delete
  65  * nvme_reset_wq - hosts nvme reset works
  66  * nvme_delete_wq - hosts nvme delete works
  67  *
  68  * nvme_wq will host works such as scan, aen handling, fw activation,
  69  * keep-alive, periodic reconnects etc. nvme_reset_wq
  70  * runs reset works which also flush works hosted on nvme_wq for
  71  * serialization purposes. nvme_delete_wq host controller deletion
  72  * works which flush reset works for serialization.
  73  */
  74 struct workqueue_struct *nvme_wq;
  75 EXPORT_SYMBOL_GPL(nvme_wq);
  76
  77 struct workqueue_struct *nvme_reset_wq;
  78 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  79
  80 struct workqueue_struct *nvme_delete_wq;
  81 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  82
  83 static LIST_HEAD(nvme_subsystems);
  84 static DEFINE_MUTEX(nvme_subsystems_lock);
  85
  86 static DEFINE_IDA(nvme_instance_ida);
  87 static dev_t nvme_chr_devt;
  88 static struct class *nvme_class;
  89 static struct class *nvme_subsys_class;
  90
  91 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  92 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  93                                            unsigned nsid);
  94
  95 static void nvme_update_bdev_size(struct gendisk *disk)
  96 {
  97         struct block_device *bdev = bdget_disk(disk, 0);
  98
  99         if (bdev) {
 100                 bd_set_nr_sectors(bdev, get_capacity(disk));
 101                 bdput(bdev);
 102         }
 103 }
 104
 105 /*
 106  * Prepare a queue for teardown.
 107  *
 108  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
 109  * the capacity to 0 after that to avoid blocking dispatchers that may be
 110  * holding bd_butex.  This will end buffered writers dirtying pages that can't
 111  * be synced.
 112  */
 113 static void nvme_set_queue_dying(struct nvme_ns *ns)
 114 {
 115         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 116                 return;
 117
 118         blk_set_queue_dying(ns->queue);
 119         blk_mq_unquiesce_queue(ns->queue);
 120
 121         set_capacity(ns->disk, 0);
 122         nvme_update_bdev_size(ns->disk);
 123 }
 124
 125 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 126 {
 127         /*
 128          * Only new queue scan work when admin and IO queues are both alive
 129          */
 130         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 131                 queue_work(nvme_wq, &ctrl->scan_work);
 132 }
 133
 134 /*
 135  * Use this function to proceed with scheduling reset_work for a controller
 136  * that had previously been set to the resetting state. This is intended for
 137  * code paths that can't be interrupted by other reset attempts. A hot removal
 138  * may prevent this from succeeding.
 139  */
 140 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 141 {
 142         if (ctrl->state != NVME_CTRL_RESETTING)
 143                 return -EBUSY;
 144         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 145                 return -EBUSY;
 146         return 0;
 147 }
 148 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 149
 150 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 151 {
 152         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 153                 return -EBUSY;
 154         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 155                 return -EBUSY;
 156         return 0;
 157 }
 158 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 159
 160 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 161 {
 162         int ret;
 163
 164         ret = nvme_reset_ctrl(ctrl);
 165         if (!ret) {
 166                 flush_work(&ctrl->reset_work);
 167                 if (ctrl->state != NVME_CTRL_LIVE)
 168                         ret = -ENETRESET;
 169         }
 170
 171         return ret;
 172 }
 173 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 174
 175 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 176 {
 177         dev_info(ctrl->device,
 178                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 179
 180         flush_work(&ctrl->reset_work);
 181         nvme_stop_ctrl(ctrl);
 182         nvme_remove_namespaces(ctrl);
 183         ctrl->ops->delete_ctrl(ctrl);
 184         nvme_uninit_ctrl(ctrl);
 185 }
 186
 187 static void nvme_delete_ctrl_work(struct work_struct *work)
 188 {
 189         struct nvme_ctrl *ctrl =
 190                 container_of(work, struct nvme_ctrl, delete_work);
 191
 192         nvme_do_delete_ctrl(ctrl);
 193 }
 194
 195 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 196 {
 197         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 198                 return -EBUSY;
 199         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 200                 return -EBUSY;
 201         return 0;
 202 }
 203 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 204
 205 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 206 {
 207         /*
 208          * Keep a reference until nvme_do_delete_ctrl() complete,
 209          * since ->delete_ctrl can free the controller.
 210          */
 211         nvme_get_ctrl(ctrl);
 212         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 213                 nvme_do_delete_ctrl(ctrl);
 214         nvme_put_ctrl(ctrl);
 215 }
 216
 217 static blk_status_t nvme_error_status(u16 status)
 218 {
 219         switch (status & 0x7ff) {
 220         case NVME_SC_SUCCESS:
 221                 return BLK_STS_OK;
 222         case NVME_SC_CAP_EXCEEDED:
 223                 return BLK_STS_NOSPC;
 224         case NVME_SC_LBA_RANGE:
 225         case NVME_SC_CMD_INTERRUPTED:
 226         case NVME_SC_NS_NOT_READY:
 227                 return BLK_STS_TARGET;
 228         case NVME_SC_BAD_ATTRIBUTES:
 229         case NVME_SC_ONCS_NOT_SUPPORTED:
 230         case NVME_SC_INVALID_OPCODE:
 231         case NVME_SC_INVALID_FIELD:
 232         case NVME_SC_INVALID_NS:
 233                 return BLK_STS_NOTSUPP;
 234         case NVME_SC_WRITE_FAULT:
 235         case NVME_SC_READ_ERROR:
 236         case NVME_SC_UNWRITTEN_BLOCK:
 237         case NVME_SC_ACCESS_DENIED:
 238         case NVME_SC_READ_ONLY:
 239         case NVME_SC_COMPARE_FAILED:
 240                 return BLK_STS_MEDIUM;
 241         case NVME_SC_GUARD_CHECK:
 242         case NVME_SC_APPTAG_CHECK:
 243         case NVME_SC_REFTAG_CHECK:
 244         case NVME_SC_INVALID_PI:
 245                 return BLK_STS_PROTECTION;
 246         case NVME_SC_RESERVATION_CONFLICT:
 247                 return BLK_STS_NEXUS;
 248         case NVME_SC_HOST_PATH_ERROR:
 249                 return BLK_STS_TRANSPORT;
 250         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 251                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 252         case NVME_SC_ZONE_TOO_MANY_OPEN:
 253                 return BLK_STS_ZONE_OPEN_RESOURCE;
 254         default:
 255                 return BLK_STS_IOERR;
 256         }
 257 }
 258
 259 static void nvme_retry_req(struct request *req)
 260 {
 261         struct nvme_ns *ns = req->q->queuedata;
 262         unsigned long delay = 0;
 263         u16 crd;
 264
 265         /* The mask and shift result must be <= 3 */
 266         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 267         if (ns && crd)
 268                 delay = ns->ctrl->crdt[crd - 1] * 100;
 269
 270         nvme_req(req)->retries++;
 271         blk_mq_requeue_request(req, false);
 272         blk_mq_delay_kick_requeue_list(req->q, delay);
 273 }
 274
 275 enum nvme_disposition {
 276         COMPLETE,
 277         RETRY,
 278         FAILOVER,
 279 };
 280
 281 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 282 {
 283         if (likely(nvme_req(req)->status == 0))
 284                 return COMPLETE;
 285
 286         if (blk_noretry_request(req) ||
 287             (nvme_req(req)->status & NVME_SC_DNR) ||
 288             nvme_req(req)->retries >= nvme_max_retries)
 289                 return COMPLETE;
 290
 291         if (req->cmd_flags & REQ_NVME_MPATH) {
 292                 if (nvme_is_path_error(nvme_req(req)->status) ||
 293                     blk_queue_dying(req->q))
 294                         return FAILOVER;
 295         } else {
 296                 if (blk_queue_dying(req->q))
 297                         return COMPLETE;
 298         }
 299
 300         return RETRY;
 301 }
 302
 303 static inline void nvme_end_req(struct request *req)
 304 {
 305         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 306
 307         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 308             req_op(req) == REQ_OP_ZONE_APPEND)
 309                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 310                         le64_to_cpu(nvme_req(req)->result.u64));
 311
 312         nvme_trace_bio_complete(req, status);
 313         blk_mq_end_request(req, status);
 314 }
 315
 316 void nvme_complete_rq(struct request *req)
 317 {
 318         trace_nvme_complete_rq(req);
 319         nvme_cleanup_cmd(req);
 320
 321         if (nvme_req(req)->ctrl->kas)
 322                 nvme_req(req)->ctrl->comp_seen = true;
 323
 324         switch (nvme_decide_disposition(req)) {
 325         case COMPLETE:
 326                 nvme_end_req(req);
 327                 return;
 328         case RETRY:
 329                 nvme_retry_req(req);
 330                 return;
 331         case FAILOVER:
 332                 nvme_failover_req(req);
 333                 return;
 334         }
 335 }
 336 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 337
 338 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 339 {
 340         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 341                                 "Cancelling I/O %d", req->tag);
 342
 343         /* don't abort one completed request */
 344         if (blk_mq_request_completed(req))
 345                 return true;
 346
 347         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 348         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 349         blk_mq_complete_request(req);
 350         return true;
 351 }
 352 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 353
 354 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 355 {
 356         if (ctrl->tagset) {
 357                 blk_mq_tagset_busy_iter(ctrl->tagset,
 358                                 nvme_cancel_request, ctrl);
 359                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 360         }
 361 }
 362 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 363
 364 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 365 {
 366         if (ctrl->admin_tagset) {
 367                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 368                                 nvme_cancel_request, ctrl);
 369                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 370         }
 371 }
 372 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 373
 374 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 375                 enum nvme_ctrl_state new_state)
 376 {
 377         enum nvme_ctrl_state old_state;
 378         unsigned long flags;
 379         bool changed = false;
 380
 381         spin_lock_irqsave(&ctrl->lock, flags);
 382
 383         old_state = ctrl->state;
 384         switch (new_state) {
 385         case NVME_CTRL_LIVE:
 386                 switch (old_state) {
 387                 case NVME_CTRL_NEW:
 388                 case NVME_CTRL_RESETTING:
 389                 case NVME_CTRL_CONNECTING:
 390                         changed = true;
 391                         fallthrough;
 392                 default:
 393                         break;
 394                 }
 395                 break;
 396         case NVME_CTRL_RESETTING:
 397                 switch (old_state) {
 398                 case NVME_CTRL_NEW:
 399                 case NVME_CTRL_LIVE:
 400                         changed = true;
 401                         fallthrough;
 402                 default:
 403                         break;
 404                 }
 405                 break;
 406         case NVME_CTRL_CONNECTING:
 407                 switch (old_state) {
 408                 case NVME_CTRL_NEW:
 409                 case NVME_CTRL_RESETTING:
 410                         changed = true;
 411                         fallthrough;
 412                 default:
 413                         break;
 414                 }
 415                 break;
 416         case NVME_CTRL_DELETING:
 417                 switch (old_state) {
 418                 case NVME_CTRL_LIVE:
 419                 case NVME_CTRL_RESETTING:
 420                 case NVME_CTRL_CONNECTING:
 421                         changed = true;
 422                         fallthrough;
 423                 default:
 424                         break;
 425                 }
 426                 break;
 427         case NVME_CTRL_DELETING_NOIO:
 428                 switch (old_state) {
 429                 case NVME_CTRL_DELETING:
 430                 case NVME_CTRL_DEAD:
 431                         changed = true;
 432                         fallthrough;
 433                 default:
 434                         break;
 435                 }
 436                 break;
 437         case NVME_CTRL_DEAD:
 438                 switch (old_state) {
 439                 case NVME_CTRL_DELETING:
 440                         changed = true;
 441                         fallthrough;
 442                 default:
 443                         break;
 444                 }
 445                 break;
 446         default:
 447                 break;
 448         }
 449
 450         if (changed) {
 451                 ctrl->state = new_state;
 452                 wake_up_all(&ctrl->state_wq);
 453         }
 454
 455         spin_unlock_irqrestore(&ctrl->lock, flags);
 456         if (changed && ctrl->state == NVME_CTRL_LIVE)
 457                 nvme_kick_requeue_lists(ctrl);
 458         return changed;
 459 }
 460 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 461
 462 /*
 463  * Returns true for sink states that can't ever transition back to live.
 464  */
 465 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 466 {
 467         switch (ctrl->state) {
 468         case NVME_CTRL_NEW:
 469         case NVME_CTRL_LIVE:
 470         case NVME_CTRL_RESETTING:
 471         case NVME_CTRL_CONNECTING:
 472                 return false;
 473         case NVME_CTRL_DELETING:
 474         case NVME_CTRL_DELETING_NOIO:
 475         case NVME_CTRL_DEAD:
 476                 return true;
 477         default:
 478                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 479                 return true;
 480         }
 481 }
 482
 483 /*
 484  * Waits for the controller state to be resetting, or returns false if it is
 485  * not possible to ever transition to that state.
 486  */
 487 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 488 {
 489         wait_event(ctrl->state_wq,
 490                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 491                    nvme_state_terminal(ctrl));
 492         return ctrl->state == NVME_CTRL_RESETTING;
 493 }
 494 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 495
 496 static void nvme_free_ns_head(struct kref *ref)
 497 {
 498         struct nvme_ns_head *head =
 499                 container_of(ref, struct nvme_ns_head, ref);
 500
 501         nvme_mpath_remove_disk(head);
 502         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 503         cleanup_srcu_struct(&head->srcu);
 504         nvme_put_subsystem(head->subsys);
 505         kfree(head);
 506 }
 507
 508 static void nvme_put_ns_head(struct nvme_ns_head *head)
 509 {
 510         kref_put(&head->ref, nvme_free_ns_head);
 511 }
 512
 513 static void nvme_free_ns(struct kref *kref)
 514 {
 515         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 516
 517         if (ns->ndev)
 518                 nvme_nvm_unregister(ns);
 519
 520         put_disk(ns->disk);
 521         nvme_put_ns_head(ns->head);
 522         nvme_put_ctrl(ns->ctrl);
 523         kfree(ns);
 524 }
 525
 526 void nvme_put_ns(struct nvme_ns *ns)
 527 {
 528         kref_put(&ns->kref, nvme_free_ns);
 529 }
 530 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 531
 532 static inline void nvme_clear_nvme_request(struct request *req)
 533 {
 534         nvme_req(req)->retries = 0;
 535         nvme_req(req)->flags = 0;
 536         req->rq_flags |= RQF_DONTPREP;
 537 }
 538
 539 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
 540 {
 541         return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 542 }
 543
 544 static inline void nvme_init_request(struct request *req,
 545                 struct nvme_command *cmd)
 546 {
 547         if (req->q->queuedata)
 548                 req->timeout = NVME_IO_TIMEOUT;
 549         else /* no queuedata implies admin queue */
 550                 req->timeout = ADMIN_TIMEOUT;
 551
 552         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 553         nvme_clear_nvme_request(req);
 554         nvme_req(req)->cmd = cmd;
 555 }
 556
 557 struct request *nvme_alloc_request(struct request_queue *q,
 558                 struct nvme_command *cmd, blk_mq_req_flags_t flags)
 559 {
 560         struct request *req;
 561
 562         req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
 563         if (!IS_ERR(req))
 564                 nvme_init_request(req, cmd);
 565         return req;
 566 }
 567 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 568
 569 struct request *nvme_alloc_request_qid(struct request_queue *q,
 570                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 571 {
 572         struct request *req;
 573
 574         req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
 575                         qid ? qid - 1 : 0);
 576         if (!IS_ERR(req))
 577                 nvme_init_request(req, cmd);
 578         return req;
 579 }
 580 EXPORT_SYMBOL_GPL(nvme_alloc_request_qid);
 581
 582 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 583 {
 584         struct nvme_command c;
 585
 586         memset(&c, 0, sizeof(c));
 587
 588         c.directive.opcode = nvme_admin_directive_send;
 589         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 590         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 591         c.directive.dtype = NVME_DIR_IDENTIFY;
 592         c.directive.tdtype = NVME_DIR_STREAMS;
 593         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 594
 595         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 596 }
 597
 598 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 599 {
 600         return nvme_toggle_streams(ctrl, false);
 601 }
 602
 603 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 604 {
 605         return nvme_toggle_streams(ctrl, true);
 606 }
 607
 608 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 609                                   struct streams_directive_params *s, u32 nsid)
 610 {
 611         struct nvme_command c;
 612
 613         memset(&c, 0, sizeof(c));
 614         memset(s, 0, sizeof(*s));
 615
 616         c.directive.opcode = nvme_admin_directive_recv;
 617         c.directive.nsid = cpu_to_le32(nsid);
 618         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 619         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 620         c.directive.dtype = NVME_DIR_STREAMS;
 621
 622         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 623 }
 624
 625 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 626 {
 627         struct streams_directive_params s;
 628         int ret;
 629
 630         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 631                 return 0;
 632         if (!streams)
 633                 return 0;
 634
 635         ret = nvme_enable_streams(ctrl);
 636         if (ret)
 637                 return ret;
 638
 639         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 640         if (ret)
 641                 goto out_disable_stream;
 642
 643         ctrl->nssa = le16_to_cpu(s.nssa);
 644         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 645                 dev_info(ctrl->device, "too few streams (%u) available\n",
 646                                         ctrl->nssa);
 647                 goto out_disable_stream;
 648         }
 649
 650         ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 651         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 652         return 0;
 653
 654 out_disable_stream:
 655         nvme_disable_streams(ctrl);
 656         return ret;
 657 }
 658
 659 /*
 660  * Check if 'req' has a write hint associated with it. If it does, assign
 661  * a valid namespace stream to the write.
 662  */
 663 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 664                                      struct request *req, u16 *control,
 665                                      u32 *dsmgmt)
 666 {
 667         enum rw_hint streamid = req->write_hint;
 668
 669         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 670                 streamid = 0;
 671         else {
 672                 streamid--;
 673                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 674                         return;
 675
 676                 *control |= NVME_RW_DTYPE_STREAMS;
 677                 *dsmgmt |= streamid << 16;
 678         }
 679
 680         if (streamid < ARRAY_SIZE(req->q->write_hints))
 681                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 682 }
 683
 684 static inline void nvme_setup_passthrough(struct request *req,
 685                 struct nvme_command *cmd)
 686 {
 687         memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 688         /* passthru commands should let the driver set the SGL flags */
 689         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 690 }
 691
 692 static inline void nvme_setup_flush(struct nvme_ns *ns,
 693                 struct nvme_command *cmnd)
 694 {
 695         cmnd->common.opcode = nvme_cmd_flush;
 696         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 697 }
 698
 699 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 700                 struct nvme_command *cmnd)
 701 {
 702         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 703         struct nvme_dsm_range *range;
 704         struct bio *bio;
 705
 706         /*
 707          * Some devices do not consider the DSM 'Number of Ranges' field when
 708          * determining how much data to DMA. Always allocate memory for maximum
 709          * number of segments to prevent device reading beyond end of buffer.
 710          */
 711         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 712
 713         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 714         if (!range) {
 715                 /*
 716                  * If we fail allocation our range, fallback to the controller
 717                  * discard page. If that's also busy, it's safe to return
 718                  * busy, as we know we can make progress once that's freed.
 719                  */
 720                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 721                         return BLK_STS_RESOURCE;
 722
 723                 range = page_address(ns->ctrl->discard_page);
 724         }
 725
 726         __rq_for_each_bio(bio, req) {
 727                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 728                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 729
 730                 if (n < segments) {
 731                         range[n].cattr = cpu_to_le32(0);
 732                         range[n].nlb = cpu_to_le32(nlb);
 733                         range[n].slba = cpu_to_le64(slba);
 734                 }
 735                 n++;
 736         }
 737
 738         if (WARN_ON_ONCE(n != segments)) {
 739                 if (virt_to_page(range) == ns->ctrl->discard_page)
 740                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 741                 else
 742                         kfree(range);
 743                 return BLK_STS_IOERR;
 744         }
 745
 746         cmnd->dsm.opcode = nvme_cmd_dsm;
 747         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 748         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 749         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 750
 751         req->special_vec.bv_page = virt_to_page(range);
 752         req->special_vec.bv_offset = offset_in_page(range);
 753         req->special_vec.bv_len = alloc_size;
 754         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 755
 756         return BLK_STS_OK;
 757 }
 758
 759 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 760                 struct request *req, struct nvme_command *cmnd)
 761 {
 762         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 763                 return nvme_setup_discard(ns, req, cmnd);
 764
 765         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 766         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 767         cmnd->write_zeroes.slba =
 768                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 769         cmnd->write_zeroes.length =
 770                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 771         if (nvme_ns_has_pi(ns))
 772                 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
 773         else
 774                 cmnd->write_zeroes.control = 0;
 775         return BLK_STS_OK;
 776 }
 777
 778 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 779                 struct request *req, struct nvme_command *cmnd,
 780                 enum nvme_opcode op)
 781 {
 782         struct nvme_ctrl *ctrl = ns->ctrl;
 783         u16 control = 0;
 784         u32 dsmgmt = 0;
 785
 786         if (req->cmd_flags & REQ_FUA)
 787                 control |= NVME_RW_FUA;
 788         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 789                 control |= NVME_RW_LR;
 790
 791         if (req->cmd_flags & REQ_RAHEAD)
 792                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 793
 794         cmnd->rw.opcode = op;
 795         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 796         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 797         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 798
 799         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 800                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 801
 802         if (ns->ms) {
 803                 /*
 804                  * If formated with metadata, the block layer always provides a
 805                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 806                  * we enable the PRACT bit for protection information or set the
 807                  * namespace capacity to zero to prevent any I/O.
 808                  */
 809                 if (!blk_integrity_rq(req)) {
 810                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 811                                 return BLK_STS_NOTSUPP;
 812                         control |= NVME_RW_PRINFO_PRACT;
 813                 }
 814
 815                 switch (ns->pi_type) {
 816                 case NVME_NS_DPS_PI_TYPE3:
 817                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 818                         break;
 819                 case NVME_NS_DPS_PI_TYPE1:
 820                 case NVME_NS_DPS_PI_TYPE2:
 821                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 822                                         NVME_RW_PRINFO_PRCHK_REF;
 823                         if (op == nvme_cmd_zone_append)
 824                                 control |= NVME_RW_APPEND_PIREMAP;
 825                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 826                         break;
 827                 }
 828         }
 829
 830         cmnd->rw.control = cpu_to_le16(control);
 831         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 832         return 0;
 833 }
 834
 835 void nvme_cleanup_cmd(struct request *req)
 836 {
 837         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 838                 struct nvme_ns *ns = req->rq_disk->private_data;
 839                 struct page *page = req->special_vec.bv_page;
 840
 841                 if (page == ns->ctrl->discard_page)
 842                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 843                 else
 844                         kfree(page_address(page) + req->special_vec.bv_offset);
 845         }
 846 }
 847 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 848
 849 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 850                 struct nvme_command *cmd)
 851 {
 852         struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 853         blk_status_t ret = BLK_STS_OK;
 854
 855         if (!(req->rq_flags & RQF_DONTPREP))
 856                 nvme_clear_nvme_request(req);
 857
 858         memset(cmd, 0, sizeof(*cmd));
 859         switch (req_op(req)) {
 860         case REQ_OP_DRV_IN:
 861         case REQ_OP_DRV_OUT:
 862                 nvme_setup_passthrough(req, cmd);
 863                 break;
 864         case REQ_OP_FLUSH:
 865                 nvme_setup_flush(ns, cmd);
 866                 break;
 867         case REQ_OP_ZONE_RESET_ALL:
 868         case REQ_OP_ZONE_RESET:
 869                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 870                 break;
 871         case REQ_OP_ZONE_OPEN:
 872                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
 873                 break;
 874         case REQ_OP_ZONE_CLOSE:
 875                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
 876                 break;
 877         case REQ_OP_ZONE_FINISH:
 878                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
 879                 break;
 880         case REQ_OP_WRITE_ZEROES:
 881                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 882                 break;
 883         case REQ_OP_DISCARD:
 884                 ret = nvme_setup_discard(ns, req, cmd);
 885                 break;
 886         case REQ_OP_READ:
 887                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
 888                 break;
 889         case REQ_OP_WRITE:
 890                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
 891                 break;
 892         case REQ_OP_ZONE_APPEND:
 893                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
 894                 break;
 895         default:
 896                 WARN_ON_ONCE(1);
 897                 return BLK_STS_IOERR;
 898         }
 899
 900         if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
 901                 nvme_req(req)->genctr++;
 902         cmd->common.command_id = nvme_cid(req);
 903         trace_nvme_setup_cmd(req, cmd);
 904         return ret;
 905 }
 906 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 907
 908 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 909 {
 910         struct completion *waiting = rq->end_io_data;
 911
 912         rq->end_io_data = NULL;
 913         complete(waiting);
 914 }
 915
 916 static void nvme_execute_rq_polled(struct request_queue *q,
 917                 struct gendisk *bd_disk, struct request *rq, int at_head)
 918 {
 919         DECLARE_COMPLETION_ONSTACK(wait);
 920
 921         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 922
 923         rq->cmd_flags |= REQ_HIPRI;
 924         rq->end_io_data = &wait;
 925         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 926
 927         while (!completion_done(&wait)) {
 928                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 929                 cond_resched();
 930         }
 931 }
 932
 933 /*
 934  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 935  * if the result is positive, it's an NVM Express status code
 936  */
 937 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 938                 union nvme_result *result, void *buffer, unsigned bufflen,
 939                 unsigned timeout, int qid, int at_head,
 940                 blk_mq_req_flags_t flags, bool poll)
 941 {
 942         struct request *req;
 943         int ret;
 944
 945         if (qid == NVME_QID_ANY)
 946                 req = nvme_alloc_request(q, cmd, flags);
 947         else
 948                 req = nvme_alloc_request_qid(q, cmd, flags, qid);
 949         if (IS_ERR(req))
 950                 return PTR_ERR(req);
 951
 952         if (timeout)
 953                 req->timeout = timeout;
 954
 955         if (buffer && bufflen) {
 956                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 957                 if (ret)
 958                         goto out;
 959         }
 960
 961         if (poll)
 962                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 963         else
 964                 blk_execute_rq(req->q, NULL, req, at_head);
 965         if (result)
 966                 *result = nvme_req(req)->result;
 967         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 968                 ret = -EINTR;
 969         else
 970                 ret = nvme_req(req)->status;
 971  out:
 972         blk_mq_free_request(req);
 973         return ret;
 974 }
 975 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 976
 977 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 978                 void *buffer, unsigned bufflen)
 979 {
 980         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 981                         NVME_QID_ANY, 0, 0, false);
 982 }
 983 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 984
 985 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 986                 unsigned len, u32 seed, bool write)
 987 {
 988         struct bio_integrity_payload *bip;
 989         int ret = -ENOMEM;
 990         void *buf;
 991
 992         buf = kmalloc(len, GFP_KERNEL);
 993         if (!buf)
 994                 goto out;
 995
 996         ret = -EFAULT;
 997         if (write && copy_from_user(buf, ubuf, len))
 998                 goto out_free_meta;
 999
1000         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1001         if (IS_ERR(bip)) {
1002                 ret = PTR_ERR(bip);
1003                 goto out_free_meta;
1004         }
1005
1006         bip->bip_iter.bi_size = len;
1007         bip->bip_iter.bi_sector = seed;
1008         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1009                         offset_in_page(buf));
1010         if (ret == len)
1011                 return buf;
1012         ret = -ENOMEM;
1013 out_free_meta:
1014         kfree(buf);
1015 out:
1016         return ERR_PTR(ret);
1017 }
1018
1019 static u32 nvme_known_admin_effects(u8 opcode)
1020 {
1021         switch (opcode) {
1022         case nvme_admin_format_nvm:
1023                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1024                         NVME_CMD_EFFECTS_CSE_MASK;
1025         case nvme_admin_sanitize_nvm:
1026                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1027         default:
1028                 break;
1029         }
1030         return 0;
1031 }
1032
1033 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1034 {
1035         u32 effects = 0;
1036
1037         if (ns) {
1038                 if (ns->head->effects)
1039                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1040                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1041                         dev_warn(ctrl->device,
1042                                  "IO command:%02x has unhandled effects:%08x\n",
1043                                  opcode, effects);
1044                 return 0;
1045         }
1046
1047         if (ctrl->effects)
1048                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1049         effects |= nvme_known_admin_effects(opcode);
1050
1051         return effects;
1052 }
1053 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1054
1055 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1056                                u8 opcode)
1057 {
1058         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1059
1060         /*
1061          * For simplicity, IO to all namespaces is quiesced even if the command
1062          * effects say only one namespace is affected.
1063          */
1064         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1065                 mutex_lock(&ctrl->scan_lock);
1066                 mutex_lock(&ctrl->subsys->lock);
1067                 nvme_mpath_start_freeze(ctrl->subsys);
1068                 nvme_mpath_wait_freeze(ctrl->subsys);
1069                 nvme_start_freeze(ctrl);
1070                 nvme_wait_freeze(ctrl);
1071         }
1072         return effects;
1073 }
1074
1075 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1076 {
1077         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1078                 nvme_unfreeze(ctrl);
1079                 nvme_mpath_unfreeze(ctrl->subsys);
1080                 mutex_unlock(&ctrl->subsys->lock);
1081                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1082                 mutex_unlock(&ctrl->scan_lock);
1083         }
1084         if (effects & NVME_CMD_EFFECTS_CCC)
1085                 nvme_init_identify(ctrl);
1086         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1087                 nvme_queue_scan(ctrl);
1088                 flush_work(&ctrl->scan_work);
1089         }
1090 }
1091
1092 void nvme_execute_passthru_rq(struct request *rq)
1093 {
1094         struct nvme_command *cmd = nvme_req(rq)->cmd;
1095         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1096         struct nvme_ns *ns = rq->q->queuedata;
1097         struct gendisk *disk = ns ? ns->disk : NULL;
1098         u32 effects;
1099
1100         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1101         blk_execute_rq(rq->q, disk, rq, 0);
1102         nvme_passthru_end(ctrl, effects);
1103 }
1104 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1105
1106 static int nvme_submit_user_cmd(struct request_queue *q,
1107                 struct nvme_command *cmd, void __user *ubuffer,
1108                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1109                 u32 meta_seed, u64 *result, unsigned timeout)
1110 {
1111         bool write = nvme_is_write(cmd);
1112         struct nvme_ns *ns = q->queuedata;
1113         struct gendisk *disk = ns ? ns->disk : NULL;
1114         struct request *req;
1115         struct bio *bio = NULL;
1116         void *meta = NULL;
1117         int ret;
1118
1119         req = nvme_alloc_request(q, cmd, 0);
1120         if (IS_ERR(req))
1121                 return PTR_ERR(req);
1122
1123         if (timeout)
1124                 req->timeout = timeout;
1125         nvme_req(req)->flags |= NVME_REQ_USERCMD;
1126
1127         if (ubuffer && bufflen) {
1128                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1129                                 GFP_KERNEL);
1130                 if (ret)
1131                         goto out;
1132                 bio = req->bio;
1133                 bio->bi_disk = disk;
1134                 if (disk && meta_buffer && meta_len) {
1135                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1136                                         meta_seed, write);
1137                         if (IS_ERR(meta)) {
1138                                 ret = PTR_ERR(meta);
1139                                 goto out_unmap;
1140                         }
1141                         req->cmd_flags |= REQ_INTEGRITY;
1142                 }
1143         }
1144
1145         nvme_execute_passthru_rq(req);
1146         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1147                 ret = -EINTR;
1148         else
1149                 ret = nvme_req(req)->status;
1150         if (result)
1151                 *result = le64_to_cpu(nvme_req(req)->result.u64);
1152         if (meta && !ret && !write) {
1153                 if (copy_to_user(meta_buffer, meta, meta_len))
1154                         ret = -EFAULT;
1155         }
1156         kfree(meta);
1157  out_unmap:
1158         if (bio)
1159                 blk_rq_unmap_user(bio);
1160  out:
1161         blk_mq_free_request(req);
1162         return ret;
1163 }
1164
1165 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1166 {
1167         struct nvme_ctrl *ctrl = rq->end_io_data;
1168         unsigned long flags;
1169         bool startka = false;
1170
1171         blk_mq_free_request(rq);
1172
1173         if (status) {
1174                 dev_err(ctrl->device,
1175                         "failed nvme_keep_alive_end_io error=%d\n",
1176                                 status);
1177                 return;
1178         }
1179
1180         ctrl->comp_seen = false;
1181         spin_lock_irqsave(&ctrl->lock, flags);
1182         if (ctrl->state == NVME_CTRL_LIVE ||
1183             ctrl->state == NVME_CTRL_CONNECTING)
1184                 startka = true;
1185         spin_unlock_irqrestore(&ctrl->lock, flags);
1186         if (startka)
1187                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1188 }
1189
1190 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1191 {
1192         struct request *rq;
1193
1194         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1195                         BLK_MQ_REQ_RESERVED);
1196         if (IS_ERR(rq))
1197                 return PTR_ERR(rq);
1198
1199         rq->timeout = ctrl->kato * HZ;
1200         rq->end_io_data = ctrl;
1201
1202         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
1203
1204         return 0;
1205 }
1206
1207 static void nvme_keep_alive_work(struct work_struct *work)
1208 {
1209         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1210                         struct nvme_ctrl, ka_work);
1211         bool comp_seen = ctrl->comp_seen;
1212
1213         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1214                 dev_dbg(ctrl->device,
1215                         "reschedule traffic based keep-alive timer\n");
1216                 ctrl->comp_seen = false;
1217                 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1218                 return;
1219         }
1220
1221         if (nvme_keep_alive(ctrl)) {
1222                 /* allocation failure, reset the controller */
1223                 dev_err(ctrl->device, "keep-alive failed\n");
1224                 nvme_reset_ctrl(ctrl);
1225                 return;
1226         }
1227 }
1228
1229 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1230 {
1231         if (unlikely(ctrl->kato == 0))
1232                 return;
1233
1234         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1235 }
1236
1237 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1238 {
1239         if (unlikely(ctrl->kato == 0))
1240                 return;
1241
1242         cancel_delayed_work_sync(&ctrl->ka_work);
1243 }
1244 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1245
1246 /*
1247  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1248  * flag, thus sending any new CNS opcodes has a big chance of not working.
1249  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1250  * (but not for any later version).
1251  */
1252 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1253 {
1254         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1255                 return ctrl->vs < NVME_VS(1, 2, 0);
1256         return ctrl->vs < NVME_VS(1, 1, 0);
1257 }
1258
1259 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1260 {
1261         struct nvme_command c = { };
1262         int error;
1263
1264         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1265         c.identify.opcode = nvme_admin_identify;
1266         c.identify.cns = NVME_ID_CNS_CTRL;
1267
1268         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1269         if (!*id)
1270                 return -ENOMEM;
1271
1272         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1273                         sizeof(struct nvme_id_ctrl));
1274         if (error)
1275                 kfree(*id);
1276         return error;
1277 }
1278
1279 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1280 {
1281         return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1282 }
1283
1284 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1285                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1286 {
1287         const char *warn_str = "ctrl returned bogus length:";
1288         void *data = cur;
1289
1290         switch (cur->nidt) {
1291         case NVME_NIDT_EUI64:
1292                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1293                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1294                                  warn_str, cur->nidl);
1295                         return -1;
1296                 }
1297                 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1298                         return NVME_NIDT_EUI64_LEN;
1299                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1300                 return NVME_NIDT_EUI64_LEN;
1301         case NVME_NIDT_NGUID:
1302                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1303                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1304                                  warn_str, cur->nidl);
1305                         return -1;
1306                 }
1307                 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1308                         return NVME_NIDT_NGUID_LEN;
1309                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1310                 return NVME_NIDT_NGUID_LEN;
1311         case NVME_NIDT_UUID:
1312                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1313                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1314                                  warn_str, cur->nidl);
1315                         return -1;
1316                 }
1317                 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1318                         return NVME_NIDT_UUID_LEN;
1319                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1320                 return NVME_NIDT_UUID_LEN;
1321         case NVME_NIDT_CSI:
1322                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1323                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1324                                  warn_str, cur->nidl);
1325                         return -1;
1326                 }
1327                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1328                 *csi_seen = true;
1329                 return NVME_NIDT_CSI_LEN;
1330         default:
1331                 /* Skip unknown types */
1332                 return cur->nidl;
1333         }
1334 }
1335
1336 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1337                 struct nvme_ns_ids *ids)
1338 {
1339         struct nvme_command c = { };
1340         bool csi_seen = false;
1341         int status, pos, len;
1342         void *data;
1343
1344         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1345                 return 0;
1346         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1347                 return 0;
1348
1349         c.identify.opcode = nvme_admin_identify;
1350         c.identify.nsid = cpu_to_le32(nsid);
1351         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1352
1353         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1354         if (!data)
1355                 return -ENOMEM;
1356
1357         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1358                                       NVME_IDENTIFY_DATA_SIZE);
1359         if (status) {
1360                 dev_warn(ctrl->device,
1361                         "Identify Descriptors failed (%d)\n", status);
1362                 goto free_data;
1363         }
1364
1365         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1366                 struct nvme_ns_id_desc *cur = data + pos;
1367
1368                 if (cur->nidl == 0)
1369                         break;
1370
1371                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1372                 if (len < 0)
1373                         break;
1374
1375                 len += sizeof(*cur);
1376         }
1377
1378         if (nvme_multi_css(ctrl) && !csi_seen) {
1379                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1380                          nsid);
1381                 status = -EINVAL;
1382         }
1383
1384 free_data:
1385         kfree(data);
1386         return status;
1387 }
1388
1389 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1390                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1391 {
1392         struct nvme_command c = { };
1393         int error;
1394
1395         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1396         c.identify.opcode = nvme_admin_identify;
1397         c.identify.nsid = cpu_to_le32(nsid);
1398         c.identify.cns = NVME_ID_CNS_NS;
1399
1400         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1401         if (!*id)
1402                 return -ENOMEM;
1403
1404         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1405         if (error) {
1406                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1407                 goto out_free_id;
1408         }
1409
1410         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1411         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1412                 goto out_free_id;
1413
1414
1415         if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1416                 dev_info(ctrl->device,
1417                          "Ignoring bogus Namespace Identifiers\n");
1418         } else {
1419                 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1420                     !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1421                         memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1422                 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1423                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1424                         memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1425         }
1426
1427         return 0;
1428
1429 out_free_id:
1430         kfree(*id);
1431         return error;
1432 }
1433
1434 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1435                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1436 {
1437         union nvme_result res = { 0 };
1438         struct nvme_command c;
1439         int ret;
1440
1441         memset(&c, 0, sizeof(c));
1442         c.features.opcode = op;
1443         c.features.fid = cpu_to_le32(fid);
1444         c.features.dword11 = cpu_to_le32(dword11);
1445
1446         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1447                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1448         if (ret >= 0 && result)
1449                 *result = le32_to_cpu(res.u32);
1450         return ret;
1451 }
1452
1453 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1454                       unsigned int dword11, void *buffer, size_t buflen,
1455                       u32 *result)
1456 {
1457         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1458                              buflen, result);
1459 }
1460 EXPORT_SYMBOL_GPL(nvme_set_features);
1461
1462 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1463                       unsigned int dword11, void *buffer, size_t buflen,
1464                       u32 *result)
1465 {
1466         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1467                              buflen, result);
1468 }
1469 EXPORT_SYMBOL_GPL(nvme_get_features);
1470
1471 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1472 {
1473         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1474         u32 result;
1475         int status, nr_io_queues;
1476
1477         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1478                         &result);
1479         if (status < 0)
1480                 return status;
1481
1482         /*
1483          * Degraded controllers might return an error when setting the queue
1484          * count.  We still want to be able to bring them online and offer
1485          * access to the admin queue, as that might be only way to fix them up.
1486          */
1487         if (status > 0) {
1488                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1489                 *count = 0;
1490         } else {
1491                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1492                 *count = min(*count, nr_io_queues);
1493         }
1494
1495         return 0;
1496 }
1497 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1498
1499 #define NVME_AEN_SUPPORTED \
1500         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1501          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1502
1503 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1504 {
1505         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1506         int status;
1507
1508         if (!supported_aens)
1509                 return;
1510
1511         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1512                         NULL, 0, &result);
1513         if (status)
1514                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1515                          supported_aens);
1516
1517         queue_work(nvme_wq, &ctrl->async_event_work);
1518 }
1519
1520 /*
1521  * Convert integer values from ioctl structures to user pointers, silently
1522  * ignoring the upper bits in the compat case to match behaviour of 32-bit
1523  * kernels.
1524  */
1525 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1526 {
1527         if (in_compat_syscall())
1528                 ptrval = (compat_uptr_t)ptrval;
1529         return (void __user *)ptrval;
1530 }
1531
1532 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1533 {
1534         struct nvme_user_io io;
1535         struct nvme_command c;
1536         unsigned length, meta_len;
1537         void __user *metadata;
1538
1539         if (copy_from_user(&io, uio, sizeof(io)))
1540                 return -EFAULT;
1541         if (io.flags)
1542                 return -EINVAL;
1543
1544         switch (io.opcode) {
1545         case nvme_cmd_write:
1546         case nvme_cmd_read:
1547         case nvme_cmd_compare:
1548                 break;
1549         default:
1550                 return -EINVAL;
1551         }
1552
1553         length = (io.nblocks + 1) << ns->lba_shift;
1554
1555         if ((io.control & NVME_RW_PRINFO_PRACT) &&
1556             ns->ms == sizeof(struct t10_pi_tuple)) {
1557                 /*
1558                  * Protection information is stripped/inserted by the
1559                  * controller.
1560                  */
1561                 if (nvme_to_user_ptr(io.metadata))
1562                         return -EINVAL;
1563                 meta_len = 0;
1564                 metadata = NULL;
1565         } else {
1566                 meta_len = (io.nblocks + 1) * ns->ms;
1567                 metadata = nvme_to_user_ptr(io.metadata);
1568         }
1569
1570         if (ns->features & NVME_NS_EXT_LBAS) {
1571                 length += meta_len;
1572                 meta_len = 0;
1573         } else if (meta_len) {
1574                 if ((io.metadata & 3) || !io.metadata)
1575                         return -EINVAL;
1576         }
1577
1578         memset(&c, 0, sizeof(c));
1579         c.rw.opcode = io.opcode;
1580         c.rw.flags = io.flags;
1581         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1582         c.rw.slba = cpu_to_le64(io.slba);
1583         c.rw.length = cpu_to_le16(io.nblocks);
1584         c.rw.control = cpu_to_le16(io.control);
1585         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1586         c.rw.reftag = cpu_to_le32(io.reftag);
1587         c.rw.apptag = cpu_to_le16(io.apptag);
1588         c.rw.appmask = cpu_to_le16(io.appmask);
1589
1590         return nvme_submit_user_cmd(ns->queue, &c,
1591                         nvme_to_user_ptr(io.addr), length,
1592                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1593 }
1594
1595 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1596                         struct nvme_passthru_cmd __user *ucmd)
1597 {
1598         struct nvme_passthru_cmd cmd;
1599         struct nvme_command c;
1600         unsigned timeout = 0;
1601         u64 result;
1602         int status;
1603
1604         if (!capable(CAP_SYS_ADMIN))
1605                 return -EACCES;
1606         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1607                 return -EFAULT;
1608         if (cmd.flags)
1609                 return -EINVAL;
1610
1611         memset(&c, 0, sizeof(c));
1612         c.common.opcode = cmd.opcode;
1613         c.common.flags = cmd.flags;
1614         c.common.nsid = cpu_to_le32(cmd.nsid);
1615         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1616         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1617         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1618         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1619         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1620         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1621         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1622         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1623
1624         if (cmd.timeout_ms)
1625                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1626
1627         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1628                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1629                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1630                         0, &result, timeout);
1631
1632         if (status >= 0) {
1633                 if (put_user(result, &ucmd->result))
1634                         return -EFAULT;
1635         }
1636
1637         return status;
1638 }
1639
1640 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1641                         struct nvme_passthru_cmd64 __user *ucmd)
1642 {
1643         struct nvme_passthru_cmd64 cmd;
1644         struct nvme_command c;
1645         unsigned timeout = 0;
1646         int status;
1647
1648         if (!capable(CAP_SYS_ADMIN))
1649                 return -EACCES;
1650         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1651                 return -EFAULT;
1652         if (cmd.flags)
1653                 return -EINVAL;
1654
1655         memset(&c, 0, sizeof(c));
1656         c.common.opcode = cmd.opcode;
1657         c.common.flags = cmd.flags;
1658         c.common.nsid = cpu_to_le32(cmd.nsid);
1659         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1660         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1661         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1662         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1663         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1664         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1665         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1666         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1667
1668         if (cmd.timeout_ms)
1669                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1670
1671         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1672                         nvme_to_user_ptr(cmd.addr), cmd.data_len,
1673                         nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1674                         0, &cmd.result, timeout);
1675
1676         if (status >= 0) {
1677                 if (put_user(cmd.result, &ucmd->result))
1678                         return -EFAULT;
1679         }
1680
1681         return status;
1682 }
1683
1684 /*
1685  * Issue ioctl requests on the first available path.  Note that unlike normal
1686  * block layer requests we will not retry failed request on another controller.
1687  */
1688 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1689                 struct nvme_ns_head **head, int *srcu_idx)
1690 {
1691 #ifdef CONFIG_NVME_MULTIPATH
1692         if (disk->fops == &nvme_ns_head_ops) {
1693                 struct nvme_ns *ns;
1694
1695                 *head = disk->private_data;
1696                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1697                 ns = nvme_find_path(*head);
1698                 if (!ns)
1699                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1700                 return ns;
1701         }
1702 #endif
1703         *head = NULL;
1704         *srcu_idx = -1;
1705         return disk->private_data;
1706 }
1707
1708 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1709 {
1710         if (head)
1711                 srcu_read_unlock(&head->srcu, idx);
1712 }
1713
1714 static bool is_ctrl_ioctl(unsigned int cmd)
1715 {
1716         if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1717                 return true;
1718         if (is_sed_ioctl(cmd))
1719                 return true;
1720         return false;
1721 }
1722
1723 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1724                                   void __user *argp,
1725                                   struct nvme_ns_head *head,
1726                                   int srcu_idx)
1727 {
1728         struct nvme_ctrl *ctrl = ns->ctrl;
1729         int ret;
1730
1731         nvme_get_ctrl(ns->ctrl);
1732         nvme_put_ns_from_disk(head, srcu_idx);
1733
1734         switch (cmd) {
1735         case NVME_IOCTL_ADMIN_CMD:
1736                 ret = nvme_user_cmd(ctrl, NULL, argp);
1737                 break;
1738         case NVME_IOCTL_ADMIN64_CMD:
1739                 ret = nvme_user_cmd64(ctrl, NULL, argp);
1740                 break;
1741         default:
1742                 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1743                 break;
1744         }
1745         nvme_put_ctrl(ctrl);
1746         return ret;
1747 }
1748
1749 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1750                 unsigned int cmd, unsigned long arg)
1751 {
1752         struct nvme_ns_head *head = NULL;
1753         void __user *argp = (void __user *)arg;
1754         struct nvme_ns *ns;
1755         int srcu_idx, ret;
1756
1757         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1758         if (unlikely(!ns))
1759                 return -EWOULDBLOCK;
1760
1761         /*
1762          * Handle ioctls that apply to the controller instead of the namespace
1763          * seperately and drop the ns SRCU reference early.  This avoids a
1764          * deadlock when deleting namespaces using the passthrough interface.
1765          */
1766         if (is_ctrl_ioctl(cmd))
1767                 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1768
1769         switch (cmd) {
1770         case NVME_IOCTL_ID:
1771                 force_successful_syscall_return();
1772                 ret = ns->head->ns_id;
1773                 break;
1774         case NVME_IOCTL_IO_CMD:
1775                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1776                 break;
1777         case NVME_IOCTL_SUBMIT_IO:
1778                 ret = nvme_submit_io(ns, argp);
1779                 break;
1780         case NVME_IOCTL_IO64_CMD:
1781                 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1782                 break;
1783         default:
1784                 if (ns->ndev)
1785                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1786                 else
1787                         ret = -ENOTTY;
1788         }
1789
1790         nvme_put_ns_from_disk(head, srcu_idx);
1791         return ret;
1792 }
1793
1794 #ifdef CONFIG_COMPAT
1795 struct nvme_user_io32 {
1796         __u8    opcode;
1797         __u8    flags;
1798         __u16   control;
1799         __u16   nblocks;
1800         __u16   rsvd;
1801         __u64   metadata;
1802         __u64   addr;
1803         __u64   slba;
1804         __u32   dsmgmt;
1805         __u32   reftag;
1806         __u16   apptag;
1807         __u16   appmask;
1808 } __attribute__((__packed__));
1809
1810 #define NVME_IOCTL_SUBMIT_IO32  _IOW('N', 0x42, struct nvme_user_io32)
1811
1812 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1813                 unsigned int cmd, unsigned long arg)
1814 {
1815         /*
1816          * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1817          * between 32 bit programs and 64 bit kernel.
1818          * The cause is that the results of sizeof(struct nvme_user_io),
1819          * which is used to define NVME_IOCTL_SUBMIT_IO,
1820          * are not same between 32 bit compiler and 64 bit compiler.
1821          * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1822          * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1823          * Other IOCTL numbers are same between 32 bit and 64 bit.
1824          * So there is nothing to do regarding to other IOCTL numbers.
1825          */
1826         if (cmd == NVME_IOCTL_SUBMIT_IO32)
1827                 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1828
1829         return nvme_ioctl(bdev, mode, cmd, arg);
1830 }
1831 #else
1832 #define nvme_compat_ioctl       NULL
1833 #endif /* CONFIG_COMPAT */
1834
1835 static int nvme_open(struct block_device *bdev, fmode_t mode)
1836 {
1837         struct nvme_ns *ns = bdev->bd_disk->private_data;
1838
1839 #ifdef CONFIG_NVME_MULTIPATH
1840         /* should never be called due to GENHD_FL_HIDDEN */
1841         if (WARN_ON_ONCE(ns->head->disk))
1842                 goto fail;
1843 #endif
1844         if (!kref_get_unless_zero(&ns->kref))
1845                 goto fail;
1846         if (!try_module_get(ns->ctrl->ops->module))
1847                 goto fail_put_ns;
1848
1849         return 0;
1850
1851 fail_put_ns:
1852         nvme_put_ns(ns);
1853 fail:
1854         return -ENXIO;
1855 }
1856
1857 static void nvme_release(struct gendisk *disk, fmode_t mode)
1858 {
1859         struct nvme_ns *ns = disk->private_data;
1860
1861         module_put(ns->ctrl->ops->module);
1862         nvme_put_ns(ns);
1863 }
1864
1865 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1866 {
1867         /* some standard values */
1868         geo->heads = 1 << 6;
1869         geo->sectors = 1 << 5;
1870         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1871         return 0;
1872 }
1873
1874 #ifdef CONFIG_BLK_DEV_INTEGRITY
1875 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1876                                 u32 max_integrity_segments)
1877 {
1878         struct blk_integrity integrity;
1879
1880         memset(&integrity, 0, sizeof(integrity));
1881         switch (pi_type) {
1882         case NVME_NS_DPS_PI_TYPE3:
1883                 integrity.profile = &t10_pi_type3_crc;
1884                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1885                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1886                 break;
1887         case NVME_NS_DPS_PI_TYPE1:
1888         case NVME_NS_DPS_PI_TYPE2:
1889                 integrity.profile = &t10_pi_type1_crc;
1890                 integrity.tag_size = sizeof(u16);
1891                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1892                 break;
1893         default:
1894                 integrity.profile = NULL;
1895                 break;
1896         }
1897         integrity.tuple_size = ms;
1898         blk_integrity_register(disk, &integrity);
1899         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1900 }
1901 #else
1902 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1903                                 u32 max_integrity_segments)
1904 {
1905 }
1906 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1907
1908 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1909 {
1910         struct nvme_ctrl *ctrl = ns->ctrl;
1911         struct request_queue *queue = disk->queue;
1912         u32 size = queue_logical_block_size(queue);
1913
1914         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1915                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1916                 return;
1917         }
1918
1919         if (ctrl->nr_streams && ns->sws && ns->sgs)
1920                 size *= ns->sws * ns->sgs;
1921
1922         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1923                         NVME_DSM_MAX_RANGES);
1924
1925         queue->limits.discard_alignment = 0;
1926         queue->limits.discard_granularity = size;
1927
1928         /* If discard is already enabled, don't reset queue limits */
1929         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1930                 return;
1931
1932         blk_queue_max_discard_sectors(queue, UINT_MAX);
1933         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1934
1935         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1936                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1937 }
1938
1939 /*
1940  * Even though NVMe spec explicitly states that MDTS is not applicable to the
1941  * write-zeroes, we are cautious and limit the size to the controllers
1942  * max_hw_sectors value, which is based on the MDTS field and possibly other
1943  * limiting factors.
1944  */
1945 static void nvme_config_write_zeroes(struct request_queue *q,
1946                 struct nvme_ctrl *ctrl)
1947 {
1948         if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
1949             !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1950                 blk_queue_max_write_zeroes_sectors(q, ctrl->max_hw_sectors);
1951 }
1952
1953 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1954 {
1955         return !uuid_is_null(&ids->uuid) ||
1956                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1957                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1958 }
1959
1960 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1961 {
1962         return uuid_equal(&a->uuid, &b->uuid) &&
1963                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1964                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1965                 a->csi == b->csi;
1966 }
1967
1968 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1969                                  u32 *phys_bs, u32 *io_opt)
1970 {
1971         struct streams_directive_params s;
1972         int ret;
1973
1974         if (!ctrl->nr_streams)
1975                 return 0;
1976
1977         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1978         if (ret)
1979                 return ret;
1980
1981         ns->sws = le32_to_cpu(s.sws);
1982         ns->sgs = le16_to_cpu(s.sgs);
1983
1984         if (ns->sws) {
1985                 *phys_bs = ns->sws * (1 << ns->lba_shift);
1986                 if (ns->sgs)
1987                         *io_opt = *phys_bs * ns->sgs;
1988         }
1989
1990         return 0;
1991 }
1992
1993 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1994 {
1995         struct nvme_ctrl *ctrl = ns->ctrl;
1996
1997         /*
1998          * The PI implementation requires the metadata size to be equal to the
1999          * t10 pi tuple size.
2000          */
2001         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2002         if (ns->ms == sizeof(struct t10_pi_tuple))
2003                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2004         else
2005                 ns->pi_type = 0;
2006
2007         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2008         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2009                 return 0;
2010         if (ctrl->ops->flags & NVME_F_FABRICS) {
2011                 /*
2012                  * The NVMe over Fabrics specification only supports metadata as
2013                  * part of the extended data LBA.  We rely on HCA/HBA support to
2014                  * remap the separate metadata buffer from the block layer.
2015                  */
2016                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2017                         return -EINVAL;
2018                 if (ctrl->max_integrity_segments)
2019                         ns->features |=
2020                                 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2021         } else {
2022                 /*
2023                  * For PCIe controllers, we can't easily remap the separate
2024                  * metadata buffer from the block layer and thus require a
2025                  * separate metadata buffer for block layer metadata/PI support.
2026                  * We allow extended LBAs for the passthrough interface, though.
2027                  */
2028                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2029                         ns->features |= NVME_NS_EXT_LBAS;
2030                 else
2031                         ns->features |= NVME_NS_METADATA_SUPPORTED;
2032         }
2033
2034         return 0;
2035 }
2036
2037 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2038                 struct request_queue *q)
2039 {
2040         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2041
2042         if (ctrl->max_hw_sectors) {
2043                 u32 max_segments =
2044                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2045
2046                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2047                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2048                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2049         }
2050         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2051         blk_queue_dma_alignment(q, 3);
2052         blk_queue_write_cache(q, vwc, vwc);
2053 }
2054
2055 static void nvme_update_disk_info(struct gendisk *disk,
2056                 struct nvme_ns *ns, struct nvme_id_ns *id)
2057 {
2058         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2059         unsigned short bs = 1 << ns->lba_shift;
2060         u32 atomic_bs, phys_bs, io_opt = 0;
2061
2062         /*
2063          * The block layer can't support LBA sizes larger than the page size
2064          * yet, so catch this early and don't allow block I/O.
2065          */
2066         if (ns->lba_shift > PAGE_SHIFT) {
2067                 capacity = 0;
2068                 bs = (1 << 9);
2069         }
2070
2071         blk_integrity_unregister(disk);
2072
2073         atomic_bs = phys_bs = bs;
2074         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2075         if (id->nabo == 0) {
2076                 /*
2077                  * Bit 1 indicates whether NAWUPF is defined for this namespace
2078                  * and whether it should be used instead of AWUPF. If NAWUPF ==
2079                  * 0 then AWUPF must be used instead.
2080                  */
2081                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2082                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2083                 else
2084                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2085         }
2086
2087         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2088                 /* NPWG = Namespace Preferred Write Granularity */
2089                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2090                 /* NOWS = Namespace Optimal Write Size */
2091                 io_opt = bs * (1 + le16_to_cpu(id->nows));
2092         }
2093
2094         blk_queue_logical_block_size(disk->queue, bs);
2095         /*
2096          * Linux filesystems assume writing a single physical block is
2097          * an atomic operation. Hence limit the physical block size to the
2098          * value of the Atomic Write Unit Power Fail parameter.
2099          */
2100         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2101         blk_queue_io_min(disk->queue, phys_bs);
2102         blk_queue_io_opt(disk->queue, io_opt);
2103
2104         /*
2105          * Register a metadata profile for PI, or the plain non-integrity NVMe
2106          * metadata masquerading as Type 0 if supported, otherwise reject block
2107          * I/O to namespaces with metadata except when the namespace supports
2108          * PI, as it can strip/insert in that case.
2109          */
2110         if (ns->ms) {
2111                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2112                     (ns->features & NVME_NS_METADATA_SUPPORTED))
2113                         nvme_init_integrity(disk, ns->ms, ns->pi_type,
2114                                             ns->ctrl->max_integrity_segments);
2115                 else if (!nvme_ns_has_pi(ns))
2116                         capacity = 0;
2117         }
2118
2119         set_capacity_revalidate_and_notify(disk, capacity, false);
2120
2121         nvme_config_discard(disk, ns);
2122         nvme_config_write_zeroes(disk->queue, ns->ctrl);
2123
2124         if (id->nsattr & NVME_NS_ATTR_RO)
2125                 set_disk_ro(disk, true);
2126 }
2127
2128 static inline bool nvme_first_scan(struct gendisk *disk)
2129 {
2130         /* nvme_alloc_ns() scans the disk prior to adding it */
2131         return !(disk->flags & GENHD_FL_UP);
2132 }
2133
2134 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2135 {
2136         struct nvme_ctrl *ctrl = ns->ctrl;
2137         u32 iob;
2138
2139         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2140             is_power_of_2(ctrl->max_hw_sectors))
2141                 iob = ctrl->max_hw_sectors;
2142         else
2143                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2144
2145         if (!iob)
2146                 return;
2147
2148         if (!is_power_of_2(iob)) {
2149                 if (nvme_first_scan(ns->disk))
2150                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2151                                 ns->disk->disk_name, iob);
2152                 return;
2153         }
2154
2155         if (blk_queue_is_zoned(ns->disk->queue)) {
2156                 if (nvme_first_scan(ns->disk))
2157                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
2158                                 ns->disk->disk_name);
2159                 return;
2160         }
2161
2162         blk_queue_chunk_sectors(ns->queue, iob);
2163 }
2164
2165 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2166 {
2167         unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2168         int ret;
2169
2170         blk_mq_freeze_queue(ns->disk->queue);
2171         ns->lba_shift = id->lbaf[lbaf].ds;
2172         nvme_set_queue_limits(ns->ctrl, ns->queue);
2173
2174         if (ns->head->ids.csi == NVME_CSI_ZNS) {
2175                 ret = nvme_update_zone_info(ns, lbaf);
2176                 if (ret)
2177                         goto out_unfreeze;
2178         }
2179
2180         ret = nvme_configure_metadata(ns, id);
2181         if (ret)
2182                 goto out_unfreeze;
2183         nvme_set_chunk_sectors(ns, id);
2184         nvme_update_disk_info(ns->disk, ns, id);
2185         blk_mq_unfreeze_queue(ns->disk->queue);
2186
2187         if (blk_queue_is_zoned(ns->queue)) {
2188                 ret = nvme_revalidate_zones(ns);
2189                 if (ret && !nvme_first_scan(ns->disk))
2190                         return ret;
2191         }
2192
2193 #ifdef CONFIG_NVME_MULTIPATH
2194         if (ns->head->disk) {
2195                 blk_mq_freeze_queue(ns->head->disk->queue);
2196                 nvme_update_disk_info(ns->head->disk, ns, id);
2197                 blk_stack_limits(&ns->head->disk->queue->limits,
2198                                  &ns->queue->limits, 0);
2199                 blk_queue_update_readahead(ns->head->disk->queue);
2200                 nvme_update_bdev_size(ns->head->disk);
2201                 blk_mq_unfreeze_queue(ns->head->disk->queue);
2202         }
2203 #endif
2204         return 0;
2205
2206 out_unfreeze:
2207         blk_mq_unfreeze_queue(ns->disk->queue);
2208         return ret;
2209 }
2210
2211 static char nvme_pr_type(enum pr_type type)
2212 {
2213         switch (type) {
2214         case PR_WRITE_EXCLUSIVE:
2215                 return 1;
2216         case PR_EXCLUSIVE_ACCESS:
2217                 return 2;
2218         case PR_WRITE_EXCLUSIVE_REG_ONLY:
2219                 return 3;
2220         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2221                 return 4;
2222         case PR_WRITE_EXCLUSIVE_ALL_REGS:
2223                 return 5;
2224         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2225                 return 6;
2226         default:
2227                 return 0;
2228         }
2229 };
2230
2231 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2232                                 u64 key, u64 sa_key, u8 op)
2233 {
2234         struct nvme_ns_head *head = NULL;
2235         struct nvme_ns *ns;
2236         struct nvme_command c;
2237         int srcu_idx, ret;
2238         u8 data[16] = { 0, };
2239
2240         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2241         if (unlikely(!ns))
2242                 return -EWOULDBLOCK;
2243
2244         put_unaligned_le64(key, &data[0]);
2245         put_unaligned_le64(sa_key, &data[8]);
2246
2247         memset(&c, 0, sizeof(c));
2248         c.common.opcode = op;
2249         c.common.nsid = cpu_to_le32(ns->head->ns_id);
2250         c.common.cdw10 = cpu_to_le32(cdw10);
2251
2252         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2253         nvme_put_ns_from_disk(head, srcu_idx);
2254         return ret;
2255 }
2256
2257 static int nvme_pr_register(struct block_device *bdev, u64 old,
2258                 u64 new, unsigned flags)
2259 {
2260         u32 cdw10;
2261
2262         if (flags & ~PR_FL_IGNORE_KEY)
2263                 return -EOPNOTSUPP;
2264
2265         cdw10 = old ? 2 : 0;
2266         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2267         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2268         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2269 }
2270
2271 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2272                 enum pr_type type, unsigned flags)
2273 {
2274         u32 cdw10;
2275
2276         if (flags & ~PR_FL_IGNORE_KEY)
2277                 return -EOPNOTSUPP;
2278
2279         cdw10 = nvme_pr_type(type) << 8;
2280         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2281         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2282 }
2283
2284 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2285                 enum pr_type type, bool abort)
2286 {
2287         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2288
2289         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2290 }
2291
2292 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2293 {
2294         u32 cdw10 = 1 | (key ? 0 : 1 << 3);
2295
2296         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2297 }
2298
2299 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2300 {
2301         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
2302
2303         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2304 }
2305
2306 static const struct pr_ops nvme_pr_ops = {
2307         .pr_register    = nvme_pr_register,
2308         .pr_reserve     = nvme_pr_reserve,
2309         .pr_release     = nvme_pr_release,
2310         .pr_preempt     = nvme_pr_preempt,
2311         .pr_clear       = nvme_pr_clear,
2312 };
2313
2314 #ifdef CONFIG_BLK_SED_OPAL
2315 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2316                 bool send)
2317 {
2318         struct nvme_ctrl *ctrl = data;
2319         struct nvme_command cmd;
2320
2321         memset(&cmd, 0, sizeof(cmd));
2322         if (send)
2323                 cmd.common.opcode = nvme_admin_security_send;
2324         else
2325                 cmd.common.opcode = nvme_admin_security_recv;
2326         cmd.common.nsid = 0;
2327         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2328         cmd.common.cdw11 = cpu_to_le32(len);
2329
2330         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2331                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2332 }
2333 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2334 #endif /* CONFIG_BLK_SED_OPAL */
2335
2336 static const struct block_device_operations nvme_fops = {
2337         .owner          = THIS_MODULE,
2338         .ioctl          = nvme_ioctl,
2339         .compat_ioctl   = nvme_compat_ioctl,
2340         .open           = nvme_open,
2341         .release        = nvme_release,
2342         .getgeo         = nvme_getgeo,
2343         .report_zones   = nvme_report_zones,
2344         .pr_ops         = &nvme_pr_ops,
2345 };
2346
2347 #ifdef CONFIG_NVME_MULTIPATH
2348 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2349 {
2350         struct nvme_ns_head *head = bdev->bd_disk->private_data;
2351
2352         if (!kref_get_unless_zero(&head->ref))
2353                 return -ENXIO;
2354         return 0;
2355 }
2356
2357 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2358 {
2359         nvme_put_ns_head(disk->private_data);
2360 }
2361
2362 const struct block_device_operations nvme_ns_head_ops = {
2363         .owner          = THIS_MODULE,
2364         .submit_bio     = nvme_ns_head_submit_bio,
2365         .open           = nvme_ns_head_open,
2366         .release        = nvme_ns_head_release,
2367         .ioctl          = nvme_ioctl,
2368         .compat_ioctl   = nvme_compat_ioctl,
2369         .getgeo         = nvme_getgeo,
2370         .report_zones   = nvme_report_zones,
2371         .pr_ops         = &nvme_pr_ops,
2372 };
2373 #endif /* CONFIG_NVME_MULTIPATH */
2374
2375 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2376 {
2377         unsigned long timeout =
2378                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2379         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2380         int ret;
2381
2382         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2383                 if (csts == ~0)
2384                         return -ENODEV;
2385                 if ((csts & NVME_CSTS_RDY) == bit)
2386                         break;
2387
2388                 usleep_range(1000, 2000);
2389                 if (fatal_signal_pending(current))
2390                         return -EINTR;
2391                 if (time_after(jiffies, timeout)) {
2392                         dev_err(ctrl->device,
2393                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2394                                 enabled ? "initialisation" : "reset", csts);
2395                         return -ENODEV;
2396                 }
2397         }
2398
2399         return ret;
2400 }
2401
2402 /*
2403  * If the device has been passed off to us in an enabled state, just clear
2404  * the enabled bit.  The spec says we should set the 'shutdown notification
2405  * bits', but doing so may cause the device to complete commands to the
2406  * admin queue ... and we don't know what memory that might be pointing at!
2407  */
2408 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2409 {
2410         int ret;
2411
2412         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2413         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2414
2415         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2416         if (ret)
2417                 return ret;
2418
2419         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2420                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2421
2422         return nvme_wait_ready(ctrl, ctrl->cap, false);
2423 }
2424 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2425
2426 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2427 {
2428         unsigned dev_page_min;
2429         int ret;
2430
2431         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2432         if (ret) {
2433                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2434                 return ret;
2435         }
2436         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2437
2438         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2439                 dev_err(ctrl->device,
2440                         "Minimum device page size %u too large for host (%u)\n",
2441                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2442                 return -ENODEV;
2443         }
2444
2445         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2446                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2447         else
2448                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2449         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2450         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2451         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2452         ctrl->ctrl_config |= NVME_CC_ENABLE;
2453
2454         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2455         if (ret)
2456                 return ret;
2457         return nvme_wait_ready(ctrl, ctrl->cap, true);
2458 }
2459 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2460
2461 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2462 {
2463         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2464         u32 csts;
2465         int ret;
2466
2467         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2468         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2469
2470         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2471         if (ret)
2472                 return ret;
2473
2474         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2475                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2476                         break;
2477
2478                 msleep(100);
2479                 if (fatal_signal_pending(current))
2480                         return -EINTR;
2481                 if (time_after(jiffies, timeout)) {
2482                         dev_err(ctrl->device,
2483                                 "Device shutdown incomplete; abort shutdown\n");
2484                         return -ENODEV;
2485                 }
2486         }
2487
2488         return ret;
2489 }
2490 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2491
2492 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2493 {
2494         __le64 ts;
2495         int ret;
2496
2497         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2498                 return 0;
2499
2500         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2501         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2502                         NULL);
2503         if (ret)
2504                 dev_warn_once(ctrl->device,
2505                         "could not set timestamp (%d)\n", ret);
2506         return ret;
2507 }
2508
2509 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2510 {
2511         struct nvme_feat_host_behavior *host;
2512         int ret;
2513
2514         /* Don't bother enabling the feature if retry delay is not reported */
2515         if (!ctrl->crdt[0])
2516                 return 0;
2517
2518         host = kzalloc(sizeof(*host), GFP_KERNEL);
2519         if (!host)
2520                 return 0;
2521
2522         host->acre = NVME_ENABLE_ACRE;
2523         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2524                                 host, sizeof(*host), NULL);
2525         kfree(host);
2526         return ret;
2527 }
2528
2529 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2530 {
2531         /*
2532          * APST (Autonomous Power State Transition) lets us program a
2533          * table of power state transitions that the controller will
2534          * perform automatically.  We configure it with a simple
2535          * heuristic: we are willing to spend at most 2% of the time
2536          * transitioning between power states.  Therefore, when running
2537          * in any given state, we will enter the next lower-power
2538          * non-operational state after waiting 50 * (enlat + exlat)
2539          * microseconds, as long as that state's exit latency is under
2540          * the requested maximum latency.
2541          *
2542          * We will not autonomously enter any non-operational state for
2543          * which the total latency exceeds ps_max_latency_us.  Users
2544          * can set ps_max_latency_us to zero to turn off APST.
2545          */
2546
2547         unsigned apste;
2548         struct nvme_feat_auto_pst *table;
2549         u64 max_lat_us = 0;
2550         int max_ps = -1;
2551         int ret;
2552
2553         /*
2554          * If APST isn't supported or if we haven't been initialized yet,
2555          * then don't do anything.
2556          */
2557         if (!ctrl->apsta)
2558                 return 0;
2559
2560         if (ctrl->npss > 31) {
2561                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2562                 return 0;
2563         }
2564
2565         table = kzalloc(sizeof(*table), GFP_KERNEL);
2566         if (!table)
2567                 return 0;
2568
2569         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2570                 /* Turn off APST. */
2571                 apste = 0;
2572                 dev_dbg(ctrl->device, "APST disabled\n");
2573         } else {
2574                 __le64 target = cpu_to_le64(0);
2575                 int state;
2576
2577                 /*
2578                  * Walk through all states from lowest- to highest-power.
2579                  * According to the spec, lower-numbered states use more
2580                  * power.  NPSS, despite the name, is the index of the
2581                  * lowest-power state, not the number of states.
2582                  */
2583                 for (state = (int)ctrl->npss; state >= 0; state--) {
2584                         u64 total_latency_us, exit_latency_us, transition_ms;
2585
2586                         if (target)
2587                                 table->entries[state] = target;
2588
2589                         /*
2590                          * Don't allow transitions to the deepest state
2591                          * if it's quirked off.
2592                          */
2593                         if (state == ctrl->npss &&
2594                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2595                                 continue;
2596
2597                         /*
2598                          * Is this state a useful non-operational state for
2599                          * higher-power states to autonomously transition to?
2600                          */
2601                         if (!(ctrl->psd[state].flags &
2602                               NVME_PS_FLAGS_NON_OP_STATE))
2603                                 continue;
2604
2605                         exit_latency_us =
2606                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2607                         if (exit_latency_us > ctrl->ps_max_latency_us)
2608                                 continue;
2609
2610                         total_latency_us =
2611                                 exit_latency_us +
2612                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2613
2614                         /*
2615                          * This state is good.  Use it as the APST idle
2616                          * target for higher power states.
2617                          */
2618                         transition_ms = total_latency_us + 19;
2619                         do_div(transition_ms, 20);
2620                         if (transition_ms > (1 << 24) - 1)
2621                                 transition_ms = (1 << 24) - 1;
2622
2623                         target = cpu_to_le64((state << 3) |
2624                                              (transition_ms << 8));
2625
2626                         if (max_ps == -1)
2627                                 max_ps = state;
2628
2629                         if (total_latency_us > max_lat_us)
2630                                 max_lat_us = total_latency_us;
2631                 }
2632
2633                 apste = 1;
2634
2635                 if (max_ps == -1) {
2636                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2637                 } else {
2638                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2639                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2640                 }
2641         }
2642
2643         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2644                                 table, sizeof(*table), NULL);
2645         if (ret)
2646                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2647
2648         kfree(table);
2649         return ret;
2650 }
2651
2652 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2653 {
2654         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2655         u64 latency;
2656
2657         switch (val) {
2658         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2659         case PM_QOS_LATENCY_ANY:
2660                 latency = U64_MAX;
2661                 break;
2662
2663         default:
2664                 latency = val;
2665         }
2666
2667         if (ctrl->ps_max_latency_us != latency) {
2668                 ctrl->ps_max_latency_us = latency;
2669                 if (ctrl->state == NVME_CTRL_LIVE)
2670                         nvme_configure_apst(ctrl);
2671         }
2672 }
2673
2674 struct nvme_core_quirk_entry {
2675         /*
2676          * NVMe model and firmware strings are padded with spaces.  For
2677          * simplicity, strings in the quirk table are padded with NULLs
2678          * instead.
2679          */
2680         u16 vid;
2681         const char *mn;
2682         const char *fr;
2683         unsigned long quirks;
2684 };
2685
2686 static const struct nvme_core_quirk_entry core_quirks[] = {
2687         {
2688                 /*
2689                  * This Toshiba device seems to die using any APST states.  See:
2690                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2691                  */
2692                 .vid = 0x1179,
2693                 .mn = "THNSF5256GPUK TOSHIBA",
2694                 .quirks = NVME_QUIRK_NO_APST,
2695         },
2696         {
2697                 /*
2698                  * This LiteON CL1-3D*-Q11 firmware version has a race
2699                  * condition associated with actions related to suspend to idle
2700                  * LiteON has resolved the problem in future firmware
2701                  */
2702                 .vid = 0x14a4,
2703                 .fr = "22301111",
2704                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2705         },
2706         {
2707                 /*
2708                  * This Kioxia CD6-V Series / HPE PE8030 device times out and
2709                  * aborts I/O during any load, but more easily reproducible
2710                  * with discards (fstrim).
2711                  *
2712                  * The device is left in a state where it is also not possible
2713                  * to use "nvme set-feature" to disable APST, but booting with
2714                  * nvme_core.default_ps_max_latency=0 works.
2715                  */
2716                 .vid = 0x1e0f,
2717                 .mn = "KCD6XVUL6T40",
2718                 .quirks = NVME_QUIRK_NO_APST,
2719         },
2720         {
2721                 /*
2722                  * The external Samsung X5 SSD fails initialization without a
2723                  * delay before checking if it is ready and has a whole set of
2724                  * other problems.  To make this even more interesting, it
2725                  * shares the PCI ID with internal Samsung 970 Evo Plus that
2726                  * does not need or want these quirks.
2727                  */
2728                 .vid = 0x144d,
2729                 .mn = "Samsung Portable SSD X5",
2730                 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2731                           NVME_QUIRK_NO_DEEPEST_PS |
2732                           NVME_QUIRK_IGNORE_DEV_SUBNQN,
2733         }
2734 };
2735
2736 /* match is null-terminated but idstr is space-padded. */
2737 static bool string_matches(const char *idstr, const char *match, size_t len)
2738 {
2739         size_t matchlen;
2740
2741         if (!match)
2742                 return true;
2743
2744         matchlen = strlen(match);
2745         WARN_ON_ONCE(matchlen > len);
2746
2747         if (memcmp(idstr, match, matchlen))
2748                 return false;
2749
2750         for (; matchlen < len; matchlen++)
2751                 if (idstr[matchlen] != ' ')
2752                         return false;
2753
2754         return true;
2755 }
2756
2757 static bool quirk_matches(const struct nvme_id_ctrl *id,
2758                           const struct nvme_core_quirk_entry *q)
2759 {
2760         return q->vid == le16_to_cpu(id->vid) &&
2761                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2762                 string_matches(id->fr, q->fr, sizeof(id->fr));
2763 }
2764
2765 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2766                 struct nvme_id_ctrl *id)
2767 {
2768         size_t nqnlen;
2769         int off;
2770
2771         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2772                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2773                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2774                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2775                         return;
2776                 }
2777
2778                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2779                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2780         }
2781
2782         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2783         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2784                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2785                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2786         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2787         off += sizeof(id->sn);
2788         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2789         off += sizeof(id->mn);
2790         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2791 }
2792
2793 static void nvme_release_subsystem(struct device *dev)
2794 {
2795         struct nvme_subsystem *subsys =
2796                 container_of(dev, struct nvme_subsystem, dev);
2797
2798         if (subsys->instance >= 0)
2799                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2800         kfree(subsys);
2801 }
2802
2803 static void nvme_destroy_subsystem(struct kref *ref)
2804 {
2805         struct nvme_subsystem *subsys =
2806                         container_of(ref, struct nvme_subsystem, ref);
2807
2808         mutex_lock(&nvme_subsystems_lock);
2809         list_del(&subsys->entry);
2810         mutex_unlock(&nvme_subsystems_lock);
2811
2812         ida_destroy(&subsys->ns_ida);
2813         device_del(&subsys->dev);
2814         put_device(&subsys->dev);
2815 }
2816
2817 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2818 {
2819         kref_put(&subsys->ref, nvme_destroy_subsystem);
2820 }
2821
2822 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2823 {
2824         struct nvme_subsystem *subsys;
2825
2826         lockdep_assert_held(&nvme_subsystems_lock);
2827
2828         /*
2829          * Fail matches for discovery subsystems. This results
2830          * in each discovery controller bound to a unique subsystem.
2831          * This avoids issues with validating controller values
2832          * that can only be true when there is a single unique subsystem.
2833          * There may be multiple and completely independent entities
2834          * that provide discovery controllers.
2835          */
2836         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2837                 return NULL;
2838
2839         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2840                 if (strcmp(subsys->subnqn, subsysnqn))
2841                         continue;
2842                 if (!kref_get_unless_zero(&subsys->ref))
2843                         continue;
2844                 return subsys;
2845         }
2846
2847         return NULL;
2848 }
2849
2850 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2851         struct device_attribute subsys_attr_##_name = \
2852                 __ATTR(_name, _mode, _show, NULL)
2853
2854 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2855                                     struct device_attribute *attr,
2856                                     char *buf)
2857 {
2858         struct nvme_subsystem *subsys =
2859                 container_of(dev, struct nvme_subsystem, dev);
2860
2861         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2862 }
2863 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2864
2865 #define nvme_subsys_show_str_function(field)                            \
2866 static ssize_t subsys_##field##_show(struct device *dev,                \
2867                             struct device_attribute *attr, char *buf)   \
2868 {                                                                       \
2869         struct nvme_subsystem *subsys =                                 \
2870                 container_of(dev, struct nvme_subsystem, dev);          \
2871         return sysfs_emit(buf, "%.*s\n",                                \
2872                            (int)sizeof(subsys->field), subsys->field);  \
2873 }                                                                       \
2874 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2875
2876 nvme_subsys_show_str_function(model);
2877 nvme_subsys_show_str_function(serial);
2878 nvme_subsys_show_str_function(firmware_rev);
2879
2880 static struct attribute *nvme_subsys_attrs[] = {
2881         &subsys_attr_model.attr,
2882         &subsys_attr_serial.attr,
2883         &subsys_attr_firmware_rev.attr,
2884         &subsys_attr_subsysnqn.attr,
2885 #ifdef CONFIG_NVME_MULTIPATH
2886         &subsys_attr_iopolicy.attr,
2887 #endif
2888         NULL,
2889 };
2890
2891 static struct attribute_group nvme_subsys_attrs_group = {
2892         .attrs = nvme_subsys_attrs,
2893 };
2894
2895 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2896         &nvme_subsys_attrs_group,
2897         NULL,
2898 };
2899
2900 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2901 {
2902         return ctrl->opts && ctrl->opts->discovery_nqn;
2903 }
2904
2905 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2906                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2907 {
2908         struct nvme_ctrl *tmp;
2909
2910         lockdep_assert_held(&nvme_subsystems_lock);
2911
2912         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2913                 if (nvme_state_terminal(tmp))
2914                         continue;
2915
2916                 if (tmp->cntlid == ctrl->cntlid) {
2917                         dev_err(ctrl->device,
2918                                 "Duplicate cntlid %u with %s, rejecting\n",
2919                                 ctrl->cntlid, dev_name(tmp->device));
2920                         return false;
2921                 }
2922
2923                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2924                     nvme_discovery_ctrl(ctrl))
2925                         continue;
2926
2927                 dev_err(ctrl->device,
2928                         "Subsystem does not support multiple controllers\n");
2929                 return false;
2930         }
2931
2932         return true;
2933 }
2934
2935 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2936 {
2937         struct nvme_subsystem *subsys, *found;
2938         int ret;
2939
2940         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2941         if (!subsys)
2942                 return -ENOMEM;
2943
2944         subsys->instance = -1;
2945         mutex_init(&subsys->lock);
2946         kref_init(&subsys->ref);
2947         INIT_LIST_HEAD(&subsys->ctrls);
2948         INIT_LIST_HEAD(&subsys->nsheads);
2949         nvme_init_subnqn(subsys, ctrl, id);
2950         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2951         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2952         subsys->vendor_id = le16_to_cpu(id->vid);
2953         subsys->cmic = id->cmic;
2954         subsys->awupf = le16_to_cpu(id->awupf);
2955 #ifdef CONFIG_NVME_MULTIPATH
2956         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2957 #endif
2958
2959         subsys->dev.class = nvme_subsys_class;
2960         subsys->dev.release = nvme_release_subsystem;
2961         subsys->dev.groups = nvme_subsys_attrs_groups;
2962         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2963         device_initialize(&subsys->dev);
2964
2965         mutex_lock(&nvme_subsystems_lock);
2966         found = __nvme_find_get_subsystem(subsys->subnqn);
2967         if (found) {
2968                 put_device(&subsys->dev);
2969                 subsys = found;
2970
2971                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2972                         ret = -EINVAL;
2973                         goto out_put_subsystem;
2974                 }
2975         } else {
2976                 ret = device_add(&subsys->dev);
2977                 if (ret) {
2978                         dev_err(ctrl->device,
2979                                 "failed to register subsystem device.\n");
2980                         put_device(&subsys->dev);
2981                         goto out_unlock;
2982                 }
2983                 ida_init(&subsys->ns_ida);
2984                 list_add_tail(&subsys->entry, &nvme_subsystems);
2985         }
2986
2987         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2988                                 dev_name(ctrl->device));
2989         if (ret) {
2990                 dev_err(ctrl->device,
2991                         "failed to create sysfs link from subsystem.\n");
2992                 goto out_put_subsystem;
2993         }
2994
2995         if (!found)
2996                 subsys->instance = ctrl->instance;
2997         ctrl->subsys = subsys;
2998         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2999         mutex_unlock(&nvme_subsystems_lock);
3000         return 0;
3001
3002 out_put_subsystem:
3003         nvme_put_subsystem(subsys);
3004 out_unlock:
3005         mutex_unlock(&nvme_subsystems_lock);
3006         return ret;
3007 }
3008
3009 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3010                 void *log, size_t size, u64 offset)
3011 {
3012         struct nvme_command c = { };
3013         u32 dwlen = nvme_bytes_to_numd(size);
3014
3015         c.get_log_page.opcode = nvme_admin_get_log_page;
3016         c.get_log_page.nsid = cpu_to_le32(nsid);
3017         c.get_log_page.lid = log_page;
3018         c.get_log_page.lsp = lsp;
3019         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3020         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3021         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3022         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3023         c.get_log_page.csi = csi;
3024
3025         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3026 }
3027
3028 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3029                                 struct nvme_effects_log **log)
3030 {
3031         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3032         int ret;
3033
3034         if (cel)
3035                 goto out;
3036
3037         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3038         if (!cel)
3039                 return -ENOMEM;
3040
3041         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3042                         cel, sizeof(*cel), 0);
3043         if (ret) {
3044                 kfree(cel);
3045                 return ret;
3046         }
3047
3048         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3049 out:
3050         *log = cel;
3051         return 0;
3052 }
3053
3054 /*
3055  * Initialize the cached copies of the Identify data and various controller
3056  * register in our nvme_ctrl structure.  This should be called as soon as
3057  * the admin queue is fully up and running.
3058  */
3059 int nvme_init_identify(struct nvme_ctrl *ctrl)
3060 {
3061         struct nvme_id_ctrl *id;
3062         int ret, page_shift;
3063         u32 max_hw_sectors;
3064         bool prev_apst_enabled;
3065
3066         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3067         if (ret) {
3068                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3069                 return ret;
3070         }
3071         page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3072         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3073
3074         if (ctrl->vs >= NVME_VS(1, 1, 0))
3075                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3076
3077         ret = nvme_identify_ctrl(ctrl, &id);
3078         if (ret) {
3079                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3080                 return -EIO;
3081         }
3082
3083         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3084                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3085                 if (ret < 0)
3086                         goto out_free;
3087         }
3088
3089         if (!(ctrl->ops->flags & NVME_F_FABRICS))
3090                 ctrl->cntlid = le16_to_cpu(id->cntlid);
3091
3092         if (!ctrl->identified) {
3093                 int i;
3094
3095                 ret = nvme_init_subsystem(ctrl, id);
3096                 if (ret)
3097                         goto out_free;
3098
3099                 /*
3100                  * Check for quirks.  Quirk can depend on firmware version,
3101                  * so, in principle, the set of quirks present can change
3102                  * across a reset.  As a possible future enhancement, we
3103                  * could re-scan for quirks every time we reinitialize
3104                  * the device, but we'd have to make sure that the driver
3105                  * behaves intelligently if the quirks change.
3106                  */
3107                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3108                         if (quirk_matches(id, &core_quirks[i]))
3109                                 ctrl->quirks |= core_quirks[i].quirks;
3110                 }
3111         }
3112         memcpy(ctrl->subsys->firmware_rev, id->fr,
3113                sizeof(ctrl->subsys->firmware_rev));
3114
3115         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3116                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3117                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3118         }
3119
3120         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3121         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3122         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3123
3124         ctrl->oacs = le16_to_cpu(id->oacs);
3125         ctrl->oncs = le16_to_cpu(id->oncs);
3126         ctrl->mtfa = le16_to_cpu(id->mtfa);
3127         ctrl->oaes = le32_to_cpu(id->oaes);
3128         ctrl->wctemp = le16_to_cpu(id->wctemp);
3129         ctrl->cctemp = le16_to_cpu(id->cctemp);
3130
3131         atomic_set(&ctrl->abort_limit, id->acl + 1);
3132         ctrl->vwc = id->vwc;
3133         if (id->mdts)
3134                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3135         else
3136                 max_hw_sectors = UINT_MAX;
3137         ctrl->max_hw_sectors =
3138                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3139
3140         nvme_set_queue_limits(ctrl, ctrl->admin_q);
3141         ctrl->sgls = le32_to_cpu(id->sgls);
3142         ctrl->kas = le16_to_cpu(id->kas);
3143         ctrl->max_namespaces = le32_to_cpu(id->mnan);
3144         ctrl->ctratt = le32_to_cpu(id->ctratt);
3145
3146         if (id->rtd3e) {
3147                 /* us -> s */
3148                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3149
3150                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3151                                                  shutdown_timeout, 60);
3152
3153                 if (ctrl->shutdown_timeout != shutdown_timeout)
3154                         dev_info(ctrl->device,
3155                                  "Shutdown timeout set to %u seconds\n",
3156                                  ctrl->shutdown_timeout);
3157         } else
3158                 ctrl->shutdown_timeout = shutdown_timeout;
3159
3160         ctrl->npss = id->npss;
3161         ctrl->apsta = id->apsta;
3162         prev_apst_enabled = ctrl->apst_enabled;
3163         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3164                 if (force_apst && id->apsta) {
3165                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3166                         ctrl->apst_enabled = true;
3167                 } else {
3168                         ctrl->apst_enabled = false;
3169                 }
3170         } else {
3171                 ctrl->apst_enabled = id->apsta;
3172         }
3173         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3174
3175         if (ctrl->ops->flags & NVME_F_FABRICS) {
3176                 ctrl->icdoff = le16_to_cpu(id->icdoff);
3177                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3178                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3179                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3180
3181                 /*
3182                  * In fabrics we need to verify the cntlid matches the
3183                  * admin connect
3184                  */
3185                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3186                         dev_err(ctrl->device,
3187                                 "Mismatching cntlid: Connect %u vs Identify "
3188                                 "%u, rejecting\n",
3189                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
3190                         ret = -EINVAL;
3191                         goto out_free;
3192                 }
3193
3194                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3195                         dev_err(ctrl->device,
3196                                 "keep-alive support is mandatory for fabrics\n");
3197                         ret = -EINVAL;
3198                         goto out_free;
3199                 }
3200         } else {
3201                 ctrl->hmpre = le32_to_cpu(id->hmpre);
3202                 ctrl->hmmin = le32_to_cpu(id->hmmin);
3203                 ctrl->hmminds = le32_to_cpu(id->hmminds);
3204                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3205         }
3206
3207         ret = nvme_mpath_init_identify(ctrl, id);
3208         kfree(id);
3209
3210         if (ret < 0)
3211                 return ret;
3212
3213         if (ctrl->apst_enabled && !prev_apst_enabled)
3214                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3215         else if (!ctrl->apst_enabled && prev_apst_enabled)
3216                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3217
3218         ret = nvme_configure_apst(ctrl);
3219         if (ret < 0)
3220                 return ret;
3221
3222         ret = nvme_configure_timestamp(ctrl);
3223         if (ret < 0)
3224                 return ret;
3225
3226         ret = nvme_configure_directives(ctrl);
3227         if (ret < 0)
3228                 return ret;
3229
3230         ret = nvme_configure_acre(ctrl);
3231         if (ret < 0)
3232                 return ret;
3233
3234         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3235                 /*
3236                  * Do not return errors unless we are in a controller reset,
3237                  * the controller works perfectly fine without hwmon.
3238                  */
3239                 ret = nvme_hwmon_init(ctrl);
3240                 if (ret == -EINTR)
3241                         return ret;
3242         }
3243
3244         ctrl->identified = true;
3245
3246         return 0;
3247
3248 out_free:
3249         kfree(id);
3250         return ret;
3251 }
3252 EXPORT_SYMBOL_GPL(nvme_init_identify);
3253
3254 static int nvme_dev_open(struct inode *inode, struct file *file)
3255 {
3256         struct nvme_ctrl *ctrl =
3257                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3258
3259         switch (ctrl->state) {
3260         case NVME_CTRL_LIVE:
3261                 break;
3262         default:
3263                 return -EWOULDBLOCK;
3264         }
3265
3266         nvme_get_ctrl(ctrl);
3267         if (!try_module_get(ctrl->ops->module)) {
3268                 nvme_put_ctrl(ctrl);
3269                 return -EINVAL;
3270         }
3271
3272         file->private_data = ctrl;
3273         return 0;
3274 }
3275
3276 static int nvme_dev_release(struct inode *inode, struct file *file)
3277 {
3278         struct nvme_ctrl *ctrl =
3279                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3280
3281         module_put(ctrl->ops->module);
3282         nvme_put_ctrl(ctrl);
3283         return 0;
3284 }
3285
3286 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3287 {
3288         struct nvme_ns *ns;
3289         int ret;
3290
3291         down_read(&ctrl->namespaces_rwsem);
3292         if (list_empty(&ctrl->namespaces)) {
3293                 ret = -ENOTTY;
3294                 goto out_unlock;
3295         }
3296
3297         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3298         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3299                 dev_warn(ctrl->device,
3300                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3301                 ret = -EINVAL;
3302                 goto out_unlock;
3303         }
3304
3305         dev_warn(ctrl->device,
3306                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3307         kref_get(&ns->kref);
3308         up_read(&ctrl->namespaces_rwsem);
3309
3310         ret = nvme_user_cmd(ctrl, ns, argp);
3311         nvme_put_ns(ns);
3312         return ret;
3313
3314 out_unlock:
3315         up_read(&ctrl->namespaces_rwsem);
3316         return ret;
3317 }
3318
3319 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3320                 unsigned long arg)
3321 {
3322         struct nvme_ctrl *ctrl = file->private_data;
3323         void __user *argp = (void __user *)arg;
3324
3325         switch (cmd) {
3326         case NVME_IOCTL_ADMIN_CMD:
3327                 return nvme_user_cmd(ctrl, NULL, argp);
3328         case NVME_IOCTL_ADMIN64_CMD:
3329                 return nvme_user_cmd64(ctrl, NULL, argp);
3330         case NVME_IOCTL_IO_CMD:
3331                 return nvme_dev_user_cmd(ctrl, argp);
3332         case NVME_IOCTL_RESET:
3333                 dev_warn(ctrl->device, "resetting controller\n");
3334                 return nvme_reset_ctrl_sync(ctrl);
3335         case NVME_IOCTL_SUBSYS_RESET:
3336                 return nvme_reset_subsystem(ctrl);
3337         case NVME_IOCTL_RESCAN:
3338                 nvme_queue_scan(ctrl);
3339                 return 0;
3340         default:
3341                 return -ENOTTY;
3342         }
3343 }
3344
3345 static const struct file_operations nvme_dev_fops = {
3346         .owner          = THIS_MODULE,
3347         .open           = nvme_dev_open,
3348         .release        = nvme_dev_release,
3349         .unlocked_ioctl = nvme_dev_ioctl,
3350         .compat_ioctl   = compat_ptr_ioctl,
3351 };
3352
3353 static ssize_t nvme_sysfs_reset(struct device *dev,
3354                                 struct device_attribute *attr, const char *buf,
3355                                 size_t count)
3356 {
3357         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3358         int ret;
3359
3360         ret = nvme_reset_ctrl_sync(ctrl);
3361         if (ret < 0)
3362                 return ret;
3363         return count;
3364 }
3365 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3366
3367 static ssize_t nvme_sysfs_rescan(struct device *dev,
3368                                 struct device_attribute *attr, const char *buf,
3369                                 size_t count)
3370 {
3371         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3372
3373         nvme_queue_scan(ctrl);
3374         return count;
3375 }
3376 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3377
3378 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3379 {
3380         struct gendisk *disk = dev_to_disk(dev);
3381
3382         if (disk->fops == &nvme_fops)
3383                 return nvme_get_ns_from_dev(dev)->head;
3384         else
3385                 return disk->private_data;
3386 }
3387
3388 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3389                 char *buf)
3390 {
3391         struct nvme_ns_head *head = dev_to_ns_head(dev);
3392         struct nvme_ns_ids *ids = &head->ids;
3393         struct nvme_subsystem *subsys = head->subsys;
3394         int serial_len = sizeof(subsys->serial);
3395         int model_len = sizeof(subsys->model);
3396
3397         if (!uuid_is_null(&ids->uuid))
3398                 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3399
3400         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3401                 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3402
3403         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3404                 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3405
3406         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3407                                   subsys->serial[serial_len - 1] == '\0'))
3408                 serial_len--;
3409         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3410                                  subsys->model[model_len - 1] == '\0'))
3411                 model_len--;
3412
3413         return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3414                 serial_len, subsys->serial, model_len, subsys->model,
3415                 head->ns_id);
3416 }
3417 static DEVICE_ATTR_RO(wwid);
3418
3419 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3420                 char *buf)
3421 {
3422         return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3423 }
3424 static DEVICE_ATTR_RO(nguid);
3425
3426 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3427                 char *buf)
3428 {
3429         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3430
3431         /* For backward compatibility expose the NGUID to userspace if
3432          * we have no UUID set
3433          */
3434         if (uuid_is_null(&ids->uuid)) {
3435                 dev_warn_ratelimited(dev,
3436                         "No UUID available providing old NGUID\n");
3437                 return sysfs_emit(buf, "%pU\n", ids->nguid);
3438         }
3439         return sysfs_emit(buf, "%pU\n", &ids->uuid);
3440 }
3441 static DEVICE_ATTR_RO(uuid);
3442
3443 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3444                 char *buf)
3445 {
3446         return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3447 }
3448 static DEVICE_ATTR_RO(eui);
3449
3450 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3451                 char *buf)
3452 {
3453         return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3454 }
3455 static DEVICE_ATTR_RO(nsid);
3456
3457 static struct attribute *nvme_ns_id_attrs[] = {
3458         &dev_attr_wwid.attr,
3459         &dev_attr_uuid.attr,
3460         &dev_attr_nguid.attr,
3461         &dev_attr_eui.attr,
3462         &dev_attr_nsid.attr,
3463 #ifdef CONFIG_NVME_MULTIPATH
3464         &dev_attr_ana_grpid.attr,
3465         &dev_attr_ana_state.attr,
3466 #endif
3467         NULL,
3468 };
3469
3470 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3471                 struct attribute *a, int n)
3472 {
3473         struct device *dev = container_of(kobj, struct device, kobj);
3474         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3475
3476         if (a == &dev_attr_uuid.attr) {
3477                 if (uuid_is_null(&ids->uuid) &&
3478                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3479                         return 0;
3480         }
3481         if (a == &dev_attr_nguid.attr) {
3482                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3483                         return 0;
3484         }
3485         if (a == &dev_attr_eui.attr) {
3486                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3487                         return 0;
3488         }
3489 #ifdef CONFIG_NVME_MULTIPATH
3490         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3491                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3492                         return 0;
3493                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3494                         return 0;
3495         }
3496 #endif
3497         return a->mode;
3498 }
3499
3500 static const struct attribute_group nvme_ns_id_attr_group = {
3501         .attrs          = nvme_ns_id_attrs,
3502         .is_visible     = nvme_ns_id_attrs_are_visible,
3503 };
3504
3505 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3506         &nvme_ns_id_attr_group,
3507 #ifdef CONFIG_NVM
3508         &nvme_nvm_attr_group,
3509 #endif
3510         NULL,
3511 };
3512
3513 #define nvme_show_str_function(field)                                           \
3514 static ssize_t  field##_show(struct device *dev,                                \
3515                             struct device_attribute *attr, char *buf)           \
3516 {                                                                               \
3517         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3518         return sysfs_emit(buf, "%.*s\n",                                        \
3519                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3520 }                                                                               \
3521 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3522
3523 nvme_show_str_function(model);
3524 nvme_show_str_function(serial);
3525 nvme_show_str_function(firmware_rev);
3526
3527 #define nvme_show_int_function(field)                                           \
3528 static ssize_t  field##_show(struct device *dev,                                \
3529                             struct device_attribute *attr, char *buf)           \
3530 {                                                                               \
3531         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3532         return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3533 }                                                                               \
3534 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3535
3536 nvme_show_int_function(cntlid);
3537 nvme_show_int_function(numa_node);
3538 nvme_show_int_function(queue_count);
3539 nvme_show_int_function(sqsize);
3540
3541 static ssize_t nvme_sysfs_delete(struct device *dev,
3542                                 struct device_attribute *attr, const char *buf,
3543                                 size_t count)
3544 {
3545         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3546
3547         if (device_remove_file_self(dev, attr))
3548                 nvme_delete_ctrl_sync(ctrl);
3549         return count;
3550 }
3551 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3552
3553 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3554                                          struct device_attribute *attr,
3555                                          char *buf)
3556 {
3557         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3558
3559         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3560 }
3561 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3562
3563 static ssize_t nvme_sysfs_show_state(struct device *dev,
3564                                      struct device_attribute *attr,
3565                                      char *buf)
3566 {
3567         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3568         static const char *const state_name[] = {
3569                 [NVME_CTRL_NEW]         = "new",
3570                 [NVME_CTRL_LIVE]        = "live",
3571                 [NVME_CTRL_RESETTING]   = "resetting",
3572                 [NVME_CTRL_CONNECTING]  = "connecting",
3573                 [NVME_CTRL_DELETING]    = "deleting",
3574                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3575                 [NVME_CTRL_DEAD]        = "dead",
3576         };
3577
3578         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3579             state_name[ctrl->state])
3580                 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3581
3582         return sysfs_emit(buf, "unknown state\n");
3583 }
3584
3585 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3586
3587 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3588                                          struct device_attribute *attr,
3589                                          char *buf)
3590 {
3591         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3592
3593         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3594 }
3595 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3596
3597 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3598                                         struct device_attribute *attr,
3599                                         char *buf)
3600 {
3601         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3602
3603         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3604 }
3605 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3606
3607 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3608                                         struct device_attribute *attr,
3609                                         char *buf)
3610 {
3611         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3612
3613         return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3614 }
3615 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3616
3617 static ssize_t nvme_sysfs_show_address(struct device *dev,
3618                                          struct device_attribute *attr,
3619                                          char *buf)
3620 {
3621         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3622
3623         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3624 }
3625 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3626
3627 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3628                 struct device_attribute *attr, char *buf)
3629 {
3630         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3631         struct nvmf_ctrl_options *opts = ctrl->opts;
3632
3633         if (ctrl->opts->max_reconnects == -1)
3634                 return sysfs_emit(buf, "off\n");
3635         return sysfs_emit(buf, "%d\n",
3636                           opts->max_reconnects * opts->reconnect_delay);
3637 }
3638
3639 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3640                 struct device_attribute *attr, const char *buf, size_t count)
3641 {
3642         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3643         struct nvmf_ctrl_options *opts = ctrl->opts;
3644         int ctrl_loss_tmo, err;
3645
3646         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3647         if (err)
3648                 return -EINVAL;
3649
3650         else if (ctrl_loss_tmo < 0)
3651                 opts->max_reconnects = -1;
3652         else
3653                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3654                                                 opts->reconnect_delay);
3655         return count;
3656 }
3657 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3658         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3659
3660 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3661                 struct device_attribute *attr, char *buf)
3662 {
3663         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3664
3665         if (ctrl->opts->reconnect_delay == -1)
3666                 return sysfs_emit(buf, "off\n");
3667         return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3668 }
3669
3670 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3671                 struct device_attribute *attr, const char *buf, size_t count)
3672 {
3673         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3674         unsigned int v;
3675         int err;
3676
3677         err = kstrtou32(buf, 10, &v);
3678         if (err)
3679                 return err;
3680
3681         ctrl->opts->reconnect_delay = v;
3682         return count;
3683 }
3684 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3685         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3686
3687 static struct attribute *nvme_dev_attrs[] = {
3688         &dev_attr_reset_controller.attr,
3689         &dev_attr_rescan_controller.attr,
3690         &dev_attr_model.attr,
3691         &dev_attr_serial.attr,
3692         &dev_attr_firmware_rev.attr,
3693         &dev_attr_cntlid.attr,
3694         &dev_attr_delete_controller.attr,
3695         &dev_attr_transport.attr,
3696         &dev_attr_subsysnqn.attr,
3697         &dev_attr_address.attr,
3698         &dev_attr_state.attr,
3699         &dev_attr_numa_node.attr,
3700         &dev_attr_queue_count.attr,
3701         &dev_attr_sqsize.attr,
3702         &dev_attr_hostnqn.attr,
3703         &dev_attr_hostid.attr,
3704         &dev_attr_ctrl_loss_tmo.attr,
3705         &dev_attr_reconnect_delay.attr,
3706         NULL
3707 };
3708
3709 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3710                 struct attribute *a, int n)
3711 {
3712         struct device *dev = container_of(kobj, struct device, kobj);
3713         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3714
3715         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3716                 return 0;
3717         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3718                 return 0;
3719         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3720                 return 0;
3721         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3722                 return 0;
3723         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3724                 return 0;
3725         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3726                 return 0;
3727
3728         return a->mode;
3729 }
3730
3731 static struct attribute_group nvme_dev_attrs_group = {
3732         .attrs          = nvme_dev_attrs,
3733         .is_visible     = nvme_dev_attrs_are_visible,
3734 };
3735
3736 static const struct attribute_group *nvme_dev_attr_groups[] = {
3737         &nvme_dev_attrs_group,
3738         NULL,
3739 };
3740
3741 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3742                 unsigned nsid)
3743 {
3744         struct nvme_ns_head *h;
3745
3746         lockdep_assert_held(&subsys->lock);
3747
3748         list_for_each_entry(h, &subsys->nsheads, entry) {
3749                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3750                         return h;
3751         }
3752
3753         return NULL;
3754 }
3755
3756 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3757                 struct nvme_ns_ids *ids)
3758 {
3759         struct nvme_ns_head *h;
3760
3761         lockdep_assert_held(&subsys->lock);
3762
3763         list_for_each_entry(h, &subsys->nsheads, entry) {
3764                 if (nvme_ns_ids_valid(ids) && nvme_ns_ids_equal(ids, &h->ids))
3765                         return -EINVAL;
3766         }
3767
3768         return 0;
3769 }
3770
3771 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3772                 unsigned nsid, struct nvme_ns_ids *ids)
3773 {
3774         struct nvme_ns_head *head;
3775         size_t size = sizeof(*head);
3776         int ret = -ENOMEM;
3777
3778 #ifdef CONFIG_NVME_MULTIPATH
3779         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3780 #endif
3781
3782         head = kzalloc(size, GFP_KERNEL);
3783         if (!head)
3784                 goto out;
3785         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3786         if (ret < 0)
3787                 goto out_free_head;
3788         head->instance = ret;
3789         INIT_LIST_HEAD(&head->list);
3790         ret = init_srcu_struct(&head->srcu);
3791         if (ret)
3792                 goto out_ida_remove;
3793         head->subsys = ctrl->subsys;
3794         head->ns_id = nsid;
3795         head->ids = *ids;
3796         kref_init(&head->ref);
3797
3798         ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &head->ids);
3799         if (ret) {
3800                 dev_err(ctrl->device,
3801                         "duplicate IDs for nsid %d\n", nsid);
3802                 goto out_cleanup_srcu;
3803         }
3804
3805         if (head->ids.csi) {
3806                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3807                 if (ret)
3808                         goto out_cleanup_srcu;
3809         } else
3810                 head->effects = ctrl->effects;
3811
3812         ret = nvme_mpath_alloc_disk(ctrl, head);
3813         if (ret)
3814                 goto out_cleanup_srcu;
3815
3816         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3817
3818         kref_get(&ctrl->subsys->ref);
3819
3820         return head;
3821 out_cleanup_srcu:
3822         cleanup_srcu_struct(&head->srcu);
3823 out_ida_remove:
3824         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3825 out_free_head:
3826         kfree(head);
3827 out:
3828         if (ret > 0)
3829                 ret = blk_status_to_errno(nvme_error_status(ret));
3830         return ERR_PTR(ret);
3831 }
3832
3833 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3834                 struct nvme_ns_ids *ids, bool is_shared)
3835 {
3836         struct nvme_ctrl *ctrl = ns->ctrl;
3837         struct nvme_ns_head *head = NULL;
3838         int ret = 0;
3839
3840         mutex_lock(&ctrl->subsys->lock);
3841         head = nvme_find_ns_head(ctrl->subsys, nsid);
3842         if (!head) {
3843                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3844                 if (IS_ERR(head)) {
3845                         ret = PTR_ERR(head);
3846                         goto out_unlock;
3847                 }
3848                 head->shared = is_shared;
3849         } else {
3850                 ret = -EINVAL;
3851                 if (!is_shared || !head->shared) {
3852                         dev_err(ctrl->device,
3853                                 "Duplicate unshared namespace %d\n", nsid);
3854                         goto out_put_ns_head;
3855                 }
3856                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3857                         dev_err(ctrl->device,
3858                                 "IDs don't match for shared namespace %d\n",
3859                                         nsid);
3860                         goto out_put_ns_head;
3861                 }
3862         }
3863
3864         list_add_tail(&ns->siblings, &head->list);
3865         ns->head = head;
3866         mutex_unlock(&ctrl->subsys->lock);
3867         return 0;
3868
3869 out_put_ns_head:
3870         nvme_put_ns_head(head);
3871 out_unlock:
3872         mutex_unlock(&ctrl->subsys->lock);
3873         return ret;
3874 }
3875
3876 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3877 {
3878         struct nvme_ns *ns, *ret = NULL;
3879
3880         down_read(&ctrl->namespaces_rwsem);
3881         list_for_each_entry(ns, &ctrl->namespaces, list) {
3882                 if (ns->head->ns_id == nsid) {
3883                         if (!kref_get_unless_zero(&ns->kref))
3884                                 continue;
3885                         ret = ns;
3886                         break;
3887                 }
3888                 if (ns->head->ns_id > nsid)
3889                         break;
3890         }
3891         up_read(&ctrl->namespaces_rwsem);
3892         return ret;
3893 }
3894 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3895
3896 /*
3897  * Add the namespace to the controller list while keeping the list ordered.
3898  */
3899 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3900 {
3901         struct nvme_ns *tmp;
3902
3903         list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3904                 if (tmp->head->ns_id < ns->head->ns_id) {
3905                         list_add(&ns->list, &tmp->list);
3906                         return;
3907                 }
3908         }
3909         list_add(&ns->list, &ns->ctrl->namespaces);
3910 }
3911
3912 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3913                 struct nvme_ns_ids *ids)
3914 {
3915         struct nvme_ns *ns;
3916         struct gendisk *disk;
3917         struct nvme_id_ns *id;
3918         char disk_name[DISK_NAME_LEN];
3919         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3920
3921         if (nvme_identify_ns(ctrl, nsid, ids, &id))
3922                 return;
3923
3924         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3925         if (!ns)
3926                 goto out_free_id;
3927
3928         ns->queue = blk_mq_init_queue(ctrl->tagset);
3929         if (IS_ERR(ns->queue))
3930                 goto out_free_ns;
3931
3932         if (ctrl->opts && ctrl->opts->data_digest)
3933                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3934
3935         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3936         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3937                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3938
3939         ns->queue->queuedata = ns;
3940         ns->ctrl = ctrl;
3941         kref_init(&ns->kref);
3942
3943         ret = nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED);
3944         if (ret)
3945                 goto out_free_queue;
3946         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3947
3948         disk = alloc_disk_node(0, node);
3949         if (!disk)
3950                 goto out_unlink_ns;
3951
3952         disk->fops = &nvme_fops;
3953         disk->private_data = ns;
3954         disk->queue = ns->queue;
3955         disk->flags = flags;
3956         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3957         ns->disk = disk;
3958
3959         if (nvme_update_ns_info(ns, id))
3960                 goto out_put_disk;
3961
3962         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3963                 ret = nvme_nvm_register(ns, disk_name, node);
3964                 if (ret) {
3965                         dev_warn(ctrl->device, "LightNVM init failure\n");
3966                         goto out_put_disk;
3967                 }
3968         }
3969
3970         down_write(&ctrl->namespaces_rwsem);
3971         nvme_ns_add_to_ctrl_list(ns);
3972         up_write(&ctrl->namespaces_rwsem);
3973         nvme_get_ctrl(ctrl);
3974
3975         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3976
3977         nvme_mpath_add_disk(ns, id);
3978         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3979         kfree(id);
3980
3981         return;
3982  out_put_disk:
3983         /* prevent double queue cleanup */
3984         ns->disk->queue = NULL;
3985         put_disk(ns->disk);
3986  out_unlink_ns:
3987         mutex_lock(&ctrl->subsys->lock);
3988         list_del_rcu(&ns->siblings);
3989         if (list_empty(&ns->head->list))
3990                 list_del_init(&ns->head->entry);
3991         mutex_unlock(&ctrl->subsys->lock);
3992         nvme_put_ns_head(ns->head);
3993  out_free_queue:
3994         blk_cleanup_queue(ns->queue);
3995  out_free_ns:
3996         kfree(ns);
3997  out_free_id:
3998         kfree(id);
3999 }
4000
4001 static void nvme_ns_remove(struct nvme_ns *ns)
4002 {
4003         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4004                 return;
4005
4006         set_capacity(ns->disk, 0);
4007         nvme_fault_inject_fini(&ns->fault_inject);
4008
4009         mutex_lock(&ns->ctrl->subsys->lock);
4010         list_del_rcu(&ns->siblings);
4011         if (list_empty(&ns->head->list))
4012                 list_del_init(&ns->head->entry);
4013         mutex_unlock(&ns->ctrl->subsys->lock);
4014
4015         synchronize_rcu(); /* guarantee not available in head->list */
4016         nvme_mpath_clear_current_path(ns);
4017         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
4018
4019         if (ns->disk->flags & GENHD_FL_UP) {
4020                 del_gendisk(ns->disk);
4021                 blk_cleanup_queue(ns->queue);
4022                 if (blk_get_integrity(ns->disk))
4023                         blk_integrity_unregister(ns->disk);
4024         }
4025
4026         down_write(&ns->ctrl->namespaces_rwsem);
4027         list_del_init(&ns->list);
4028         up_write(&ns->ctrl->namespaces_rwsem);
4029
4030         nvme_mpath_check_last_path(ns);
4031         nvme_put_ns(ns);
4032 }
4033
4034 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4035 {
4036         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4037
4038         if (ns) {
4039                 nvme_ns_remove(ns);
4040                 nvme_put_ns(ns);
4041         }
4042 }
4043
4044 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4045 {
4046         struct nvme_id_ns *id;
4047         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4048
4049         if (test_bit(NVME_NS_DEAD, &ns->flags))
4050                 goto out;
4051
4052         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4053         if (ret)
4054                 goto out;
4055
4056         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4057         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4058                 dev_err(ns->ctrl->device,
4059                         "identifiers changed for nsid %d\n", ns->head->ns_id);
4060                 goto out_free_id;
4061         }
4062
4063         ret = nvme_update_ns_info(ns, id);
4064
4065 out_free_id:
4066         kfree(id);
4067 out:
4068         /*
4069          * Only remove the namespace if we got a fatal error back from the
4070          * device, otherwise ignore the error and just move on.
4071          *
4072          * TODO: we should probably schedule a delayed retry here.
4073          */
4074         if (ret > 0 && (ret & NVME_SC_DNR))
4075                 nvme_ns_remove(ns);
4076         else
4077                 revalidate_disk_size(ns->disk, true);
4078 }
4079
4080 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4081 {
4082         struct nvme_ns_ids ids = { };
4083         struct nvme_ns *ns;
4084
4085         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4086                 return;
4087
4088         ns = nvme_find_get_ns(ctrl, nsid);
4089         if (ns) {
4090                 nvme_validate_ns(ns, &ids);
4091                 nvme_put_ns(ns);
4092                 return;
4093         }
4094
4095         switch (ids.csi) {
4096         case NVME_CSI_NVM:
4097                 nvme_alloc_ns(ctrl, nsid, &ids);
4098                 break;
4099         case NVME_CSI_ZNS:
4100                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4101                         dev_warn(ctrl->device,
4102                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4103                                 nsid);
4104                         break;
4105                 }
4106                 if (!nvme_multi_css(ctrl)) {
4107                         dev_warn(ctrl->device,
4108                                 "command set not reported for nsid: %d\n",
4109                                 nsid);
4110                         break;
4111                 }
4112                 nvme_alloc_ns(ctrl, nsid, &ids);
4113                 break;
4114         default:
4115                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4116                         ids.csi, nsid);
4117                 break;
4118         }
4119 }
4120
4121 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4122                                         unsigned nsid)
4123 {
4124         struct nvme_ns *ns, *next;
4125         LIST_HEAD(rm_list);
4126
4127         down_write(&ctrl->namespaces_rwsem);
4128         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4129                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4130                         list_move_tail(&ns->list, &rm_list);
4131         }
4132         up_write(&ctrl->namespaces_rwsem);
4133
4134         list_for_each_entry_safe(ns, next, &rm_list, list)
4135                 nvme_ns_remove(ns);
4136
4137 }
4138
4139 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4140 {
4141         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4142         __le32 *ns_list;
4143         u32 prev = 0;
4144         int ret = 0, i;
4145
4146         if (nvme_ctrl_limited_cns(ctrl))
4147                 return -EOPNOTSUPP;
4148
4149         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4150         if (!ns_list)
4151                 return -ENOMEM;
4152
4153         for (;;) {
4154                 struct nvme_command cmd = {
4155                         .identify.opcode        = nvme_admin_identify,
4156                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
4157                         .identify.nsid          = cpu_to_le32(prev),
4158                 };
4159
4160                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4161                                             NVME_IDENTIFY_DATA_SIZE);
4162                 if (ret)
4163                         goto free;
4164
4165                 for (i = 0; i < nr_entries; i++) {
4166                         u32 nsid = le32_to_cpu(ns_list[i]);
4167
4168                         if (!nsid)      /* end of the list? */
4169                                 goto out;
4170                         nvme_validate_or_alloc_ns(ctrl, nsid);
4171                         while (++prev < nsid)
4172                                 nvme_ns_remove_by_nsid(ctrl, prev);
4173                 }
4174         }
4175  out:
4176         nvme_remove_invalid_namespaces(ctrl, prev);
4177  free:
4178         kfree(ns_list);
4179         return ret;
4180 }
4181
4182 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4183 {
4184         struct nvme_id_ctrl *id;
4185         u32 nn, i;
4186
4187         if (nvme_identify_ctrl(ctrl, &id))
4188                 return;
4189         nn = le32_to_cpu(id->nn);
4190         kfree(id);
4191
4192         for (i = 1; i <= nn; i++)
4193                 nvme_validate_or_alloc_ns(ctrl, i);
4194
4195         nvme_remove_invalid_namespaces(ctrl, nn);
4196 }
4197
4198 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4199 {
4200         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4201         __le32 *log;
4202         int error;
4203
4204         log = kzalloc(log_size, GFP_KERNEL);
4205         if (!log)
4206                 return;
4207
4208         /*
4209          * We need to read the log to clear the AEN, but we don't want to rely
4210          * on it for the changed namespace information as userspace could have
4211          * raced with us in reading the log page, which could cause us to miss
4212          * updates.
4213          */
4214         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4215                         NVME_CSI_NVM, log, log_size, 0);
4216         if (error)
4217                 dev_warn(ctrl->device,
4218                         "reading changed ns log failed: %d\n", error);
4219
4220         kfree(log);
4221 }
4222
4223 static void nvme_scan_work(struct work_struct *work)
4224 {
4225         struct nvme_ctrl *ctrl =
4226                 container_of(work, struct nvme_ctrl, scan_work);
4227
4228         /* No tagset on a live ctrl means IO queues could not created */
4229         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4230                 return;
4231
4232         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4233                 dev_info(ctrl->device, "rescanning namespaces.\n");
4234                 nvme_clear_changed_ns_log(ctrl);
4235         }
4236
4237         mutex_lock(&ctrl->scan_lock);
4238         if (nvme_scan_ns_list(ctrl) != 0)
4239                 nvme_scan_ns_sequential(ctrl);
4240         mutex_unlock(&ctrl->scan_lock);
4241 }
4242
4243 /*
4244  * This function iterates the namespace list unlocked to allow recovery from
4245  * controller failure. It is up to the caller to ensure the namespace list is
4246  * not modified by scan work while this function is executing.
4247  */
4248 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4249 {
4250         struct nvme_ns *ns, *next;
4251         LIST_HEAD(ns_list);
4252
4253         /*
4254          * make sure to requeue I/O to all namespaces as these
4255          * might result from the scan itself and must complete
4256          * for the scan_work to make progress
4257          */
4258         nvme_mpath_clear_ctrl_paths(ctrl);
4259
4260         /* prevent racing with ns scanning */
4261         flush_work(&ctrl->scan_work);
4262
4263         /*
4264          * The dead states indicates the controller was not gracefully
4265          * disconnected. In that case, we won't be able to flush any data while
4266          * removing the namespaces' disks; fail all the queues now to avoid
4267          * potentially having to clean up the failed sync later.
4268          */
4269         if (ctrl->state == NVME_CTRL_DEAD)
4270                 nvme_kill_queues(ctrl);
4271
4272         /* this is a no-op when called from the controller reset handler */
4273         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4274
4275         down_write(&ctrl->namespaces_rwsem);
4276         list_splice_init(&ctrl->namespaces, &ns_list);
4277         up_write(&ctrl->namespaces_rwsem);
4278
4279         list_for_each_entry_safe(ns, next, &ns_list, list)
4280                 nvme_ns_remove(ns);
4281 }
4282 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4283
4284 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4285 {
4286         struct nvme_ctrl *ctrl =
4287                 container_of(dev, struct nvme_ctrl, ctrl_device);
4288         struct nvmf_ctrl_options *opts = ctrl->opts;
4289         int ret;
4290
4291         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4292         if (ret)
4293                 return ret;
4294
4295         if (opts) {
4296                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4297                 if (ret)
4298                         return ret;
4299
4300                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4301                                 opts->trsvcid ?: "none");
4302                 if (ret)
4303                         return ret;
4304
4305                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4306                                 opts->host_traddr ?: "none");
4307         }
4308         return ret;
4309 }
4310
4311 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4312 {
4313         char *envp[2] = { NULL, NULL };
4314         u32 aen_result = ctrl->aen_result;
4315
4316         ctrl->aen_result = 0;
4317         if (!aen_result)
4318                 return;
4319
4320         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4321         if (!envp[0])
4322                 return;
4323         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4324         kfree(envp[0]);
4325 }
4326
4327 static void nvme_async_event_work(struct work_struct *work)
4328 {
4329         struct nvme_ctrl *ctrl =
4330                 container_of(work, struct nvme_ctrl, async_event_work);
4331
4332         nvme_aen_uevent(ctrl);
4333
4334         /*
4335          * The transport drivers must guarantee AER submission here is safe by
4336          * flushing ctrl async_event_work after changing the controller state
4337          * from LIVE and before freeing the admin queue.
4338         */
4339         if (ctrl->state == NVME_CTRL_LIVE)
4340                 ctrl->ops->submit_async_event(ctrl);
4341 }
4342
4343 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4344 {
4345
4346         u32 csts;
4347
4348         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4349                 return false;
4350
4351         if (csts == ~0)
4352                 return false;
4353
4354         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4355 }
4356
4357 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4358 {
4359         struct nvme_fw_slot_info_log *log;
4360
4361         log = kmalloc(sizeof(*log), GFP_KERNEL);
4362         if (!log)
4363                 return;
4364
4365         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4366                         log, sizeof(*log), 0))
4367                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4368         kfree(log);
4369 }
4370
4371 static void nvme_fw_act_work(struct work_struct *work)
4372 {
4373         struct nvme_ctrl *ctrl = container_of(work,
4374                                 struct nvme_ctrl, fw_act_work);
4375         unsigned long fw_act_timeout;
4376
4377         if (ctrl->mtfa)
4378                 fw_act_timeout = jiffies +
4379                                 msecs_to_jiffies(ctrl->mtfa * 100);
4380         else
4381                 fw_act_timeout = jiffies +
4382                                 msecs_to_jiffies(admin_timeout * 1000);
4383
4384         nvme_stop_queues(ctrl);
4385         while (nvme_ctrl_pp_status(ctrl)) {
4386                 if (time_after(jiffies, fw_act_timeout)) {
4387                         dev_warn(ctrl->device,
4388                                 "Fw activation timeout, reset controller\n");
4389                         nvme_try_sched_reset(ctrl);
4390                         return;
4391                 }
4392                 msleep(100);
4393         }
4394
4395         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4396                 return;
4397
4398         nvme_start_queues(ctrl);
4399         /* read FW slot information to clear the AER */
4400         nvme_get_fw_slot_info(ctrl);
4401 }
4402
4403 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4404 {
4405         u32 aer_notice_type = (result & 0xff00) >> 8;
4406
4407         trace_nvme_async_event(ctrl, aer_notice_type);
4408
4409         switch (aer_notice_type) {
4410         case NVME_AER_NOTICE_NS_CHANGED:
4411                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4412                 nvme_queue_scan(ctrl);
4413                 break;
4414         case NVME_AER_NOTICE_FW_ACT_STARTING:
4415                 /*
4416                  * We are (ab)using the RESETTING state to prevent subsequent
4417                  * recovery actions from interfering with the controller's
4418                  * firmware activation.
4419                  */
4420                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4421                         queue_work(nvme_wq, &ctrl->fw_act_work);
4422                 break;
4423 #ifdef CONFIG_NVME_MULTIPATH
4424         case NVME_AER_NOTICE_ANA:
4425                 if (!ctrl->ana_log_buf)
4426                         break;
4427                 queue_work(nvme_wq, &ctrl->ana_work);
4428                 break;
4429 #endif
4430         case NVME_AER_NOTICE_DISC_CHANGED:
4431                 ctrl->aen_result = result;
4432                 break;
4433         default:
4434                 dev_warn(ctrl->device, "async event result %08x\n", result);
4435         }
4436 }
4437
4438 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4439                 volatile union nvme_result *res)
4440 {
4441         u32 result = le32_to_cpu(res->u32);
4442         u32 aer_type = result & 0x07;
4443
4444         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4445                 return;
4446
4447         switch (aer_type) {
4448         case NVME_AER_NOTICE:
4449                 nvme_handle_aen_notice(ctrl, result);
4450                 break;
4451         case NVME_AER_ERROR:
4452         case NVME_AER_SMART:
4453         case NVME_AER_CSS:
4454         case NVME_AER_VS:
4455                 trace_nvme_async_event(ctrl, aer_type);
4456                 ctrl->aen_result = result;
4457                 break;
4458         default:
4459                 break;
4460         }
4461         queue_work(nvme_wq, &ctrl->async_event_work);
4462 }
4463 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4464
4465 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4466 {
4467         nvme_mpath_stop(ctrl);
4468         nvme_stop_keep_alive(ctrl);
4469         flush_work(&ctrl->async_event_work);
4470         cancel_work_sync(&ctrl->fw_act_work);
4471         if (ctrl->ops->stop_ctrl)
4472                 ctrl->ops->stop_ctrl(ctrl);
4473 }
4474 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4475
4476 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4477 {
4478         nvme_start_keep_alive(ctrl);
4479
4480         nvme_enable_aen(ctrl);
4481
4482         if (ctrl->queue_count > 1) {
4483                 nvme_queue_scan(ctrl);
4484                 nvme_start_queues(ctrl);
4485                 nvme_mpath_update(ctrl);
4486         }
4487 }
4488 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4489
4490 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4491 {
4492         nvme_hwmon_exit(ctrl);
4493         nvme_fault_inject_fini(&ctrl->fault_inject);
4494         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4495         cdev_device_del(&ctrl->cdev, ctrl->device);
4496         nvme_put_ctrl(ctrl);
4497 }
4498 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4499
4500 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4501 {
4502         struct nvme_effects_log *cel;
4503         unsigned long i;
4504
4505         xa_for_each (&ctrl->cels, i, cel) {
4506                 xa_erase(&ctrl->cels, i);
4507                 kfree(cel);
4508         }
4509
4510         xa_destroy(&ctrl->cels);
4511 }
4512
4513 static void nvme_free_ctrl(struct device *dev)
4514 {
4515         struct nvme_ctrl *ctrl =
4516                 container_of(dev, struct nvme_ctrl, ctrl_device);
4517         struct nvme_subsystem *subsys = ctrl->subsys;
4518
4519         if (!subsys || ctrl->instance != subsys->instance)
4520                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4521
4522         nvme_free_cels(ctrl);
4523         nvme_mpath_uninit(ctrl);
4524         __free_page(ctrl->discard_page);
4525
4526         if (subsys) {
4527                 mutex_lock(&nvme_subsystems_lock);
4528                 list_del(&ctrl->subsys_entry);
4529                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4530                 mutex_unlock(&nvme_subsystems_lock);
4531         }
4532
4533         ctrl->ops->free_ctrl(ctrl);
4534
4535         if (subsys)
4536                 nvme_put_subsystem(subsys);
4537 }
4538
4539 /*
4540  * Initialize a NVMe controller structures.  This needs to be called during
4541  * earliest initialization so that we have the initialized structured around
4542  * during probing.
4543  */
4544 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4545                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4546 {
4547         int ret;
4548
4549         ctrl->state = NVME_CTRL_NEW;
4550         spin_lock_init(&ctrl->lock);
4551         mutex_init(&ctrl->scan_lock);
4552         INIT_LIST_HEAD(&ctrl->namespaces);
4553         xa_init(&ctrl->cels);
4554         init_rwsem(&ctrl->namespaces_rwsem);
4555         ctrl->dev = dev;
4556         ctrl->ops = ops;
4557         ctrl->quirks = quirks;
4558         ctrl->numa_node = NUMA_NO_NODE;
4559         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4560         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4561         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4562         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4563         init_waitqueue_head(&ctrl->state_wq);
4564
4565         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4566         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4567         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4568
4569         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4570                         PAGE_SIZE);
4571         ctrl->discard_page = alloc_page(GFP_KERNEL);
4572         if (!ctrl->discard_page) {
4573                 ret = -ENOMEM;
4574                 goto out;
4575         }
4576
4577         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4578         if (ret < 0)
4579                 goto out;
4580         ctrl->instance = ret;
4581
4582         device_initialize(&ctrl->ctrl_device);
4583         ctrl->device = &ctrl->ctrl_device;
4584         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4585         ctrl->device->class = nvme_class;
4586         ctrl->device->parent = ctrl->dev;
4587         ctrl->device->groups = nvme_dev_attr_groups;
4588         ctrl->device->release = nvme_free_ctrl;
4589         dev_set_drvdata(ctrl->device, ctrl);
4590         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4591         if (ret)
4592                 goto out_release_instance;
4593
4594         nvme_get_ctrl(ctrl);
4595         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4596         ctrl->cdev.owner = ops->module;
4597         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4598         if (ret)
4599                 goto out_free_name;
4600
4601         /*
4602          * Initialize latency tolerance controls.  The sysfs files won't
4603          * be visible to userspace unless the device actually supports APST.
4604          */
4605         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4606         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4607                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4608
4609         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4610         nvme_mpath_init_ctrl(ctrl);
4611
4612         return 0;
4613 out_free_name:
4614         nvme_put_ctrl(ctrl);
4615         kfree_const(ctrl->device->kobj.name);
4616 out_release_instance:
4617         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4618 out:
4619         if (ctrl->discard_page)
4620                 __free_page(ctrl->discard_page);
4621         return ret;
4622 }
4623 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4624
4625 /**
4626  * nvme_kill_queues(): Ends all namespace queues
4627  * @ctrl: the dead controller that needs to end
4628  *
4629  * Call this function when the driver determines it is unable to get the
4630  * controller in a state capable of servicing IO.
4631  */
4632 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4633 {
4634         struct nvme_ns *ns;
4635
4636         down_read(&ctrl->namespaces_rwsem);
4637
4638         /* Forcibly unquiesce queues to avoid blocking dispatch */
4639         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4640                 blk_mq_unquiesce_queue(ctrl->admin_q);
4641
4642         list_for_each_entry(ns, &ctrl->namespaces, list)
4643                 nvme_set_queue_dying(ns);
4644
4645         up_read(&ctrl->namespaces_rwsem);
4646 }
4647 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4648
4649 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4650 {
4651         struct nvme_ns *ns;
4652
4653         down_read(&ctrl->namespaces_rwsem);
4654         list_for_each_entry(ns, &ctrl->namespaces, list)
4655                 blk_mq_unfreeze_queue(ns->queue);
4656         up_read(&ctrl->namespaces_rwsem);
4657 }
4658 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4659
4660 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4661 {
4662         struct nvme_ns *ns;
4663
4664         down_read(&ctrl->namespaces_rwsem);
4665         list_for_each_entry(ns, &ctrl->namespaces, list) {
4666                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4667                 if (timeout <= 0)
4668                         break;
4669         }
4670         up_read(&ctrl->namespaces_rwsem);
4671         return timeout;
4672 }
4673 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4674
4675 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4676 {
4677         struct nvme_ns *ns;
4678
4679         down_read(&ctrl->namespaces_rwsem);
4680         list_for_each_entry(ns, &ctrl->namespaces, list)
4681                 blk_mq_freeze_queue_wait(ns->queue);
4682         up_read(&ctrl->namespaces_rwsem);
4683 }
4684 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4685
4686 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4687 {
4688         struct nvme_ns *ns;
4689
4690         down_read(&ctrl->namespaces_rwsem);
4691         list_for_each_entry(ns, &ctrl->namespaces, list)
4692                 blk_freeze_queue_start(ns->queue);
4693         up_read(&ctrl->namespaces_rwsem);
4694 }
4695 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4696
4697 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4698 {
4699         struct nvme_ns *ns;
4700
4701         down_read(&ctrl->namespaces_rwsem);
4702         list_for_each_entry(ns, &ctrl->namespaces, list)
4703                 blk_mq_quiesce_queue(ns->queue);
4704         up_read(&ctrl->namespaces_rwsem);
4705 }
4706 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4707
4708 void nvme_start_queues(struct nvme_ctrl *ctrl)
4709 {
4710         struct nvme_ns *ns;
4711
4712         down_read(&ctrl->namespaces_rwsem);
4713         list_for_each_entry(ns, &ctrl->namespaces, list)
4714                 blk_mq_unquiesce_queue(ns->queue);
4715         up_read(&ctrl->namespaces_rwsem);
4716 }
4717 EXPORT_SYMBOL_GPL(nvme_start_queues);
4718
4719 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4720 {
4721         struct nvme_ns *ns;
4722
4723         down_read(&ctrl->namespaces_rwsem);
4724         list_for_each_entry(ns, &ctrl->namespaces, list)
4725                 blk_sync_queue(ns->queue);
4726         up_read(&ctrl->namespaces_rwsem);
4727 }
4728 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4729
4730 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4731 {
4732         nvme_sync_io_queues(ctrl);
4733         if (ctrl->admin_q)
4734                 blk_sync_queue(ctrl->admin_q);
4735 }
4736 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4737
4738 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4739 {
4740         if (file->f_op != &nvme_dev_fops)
4741                 return NULL;
4742         return file->private_data;
4743 }
4744 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4745
4746 /*
4747  * Check we didn't inadvertently grow the command structure sizes:
4748  */
4749 static inline void _nvme_check_size(void)
4750 {
4751         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4752         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4753         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4754         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4755         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4756         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4757         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4758         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4759         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4760         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4761         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4762         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4763         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4764         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4765         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4766         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4767         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4768         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4769         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4770 }
4771
4772
4773 static int __init nvme_core_init(void)
4774 {
4775         int result = -ENOMEM;
4776
4777         _nvme_check_size();
4778
4779         nvme_wq = alloc_workqueue("nvme-wq",
4780                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4781         if (!nvme_wq)
4782                 goto out;
4783
4784         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4785                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4786         if (!nvme_reset_wq)
4787                 goto destroy_wq;
4788
4789         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4790                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4791         if (!nvme_delete_wq)
4792                 goto destroy_reset_wq;
4793
4794         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4795         if (result < 0)
4796                 goto destroy_delete_wq;
4797
4798         nvme_class = class_create(THIS_MODULE, "nvme");
4799         if (IS_ERR(nvme_class)) {
4800                 result = PTR_ERR(nvme_class);
4801                 goto unregister_chrdev;
4802         }
4803         nvme_class->dev_uevent = nvme_class_uevent;
4804
4805         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4806         if (IS_ERR(nvme_subsys_class)) {
4807                 result = PTR_ERR(nvme_subsys_class);
4808                 goto destroy_class;
4809         }
4810         return 0;
4811
4812 destroy_class:
4813         class_destroy(nvme_class);
4814 unregister_chrdev:
4815         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4816 destroy_delete_wq:
4817         destroy_workqueue(nvme_delete_wq);
4818 destroy_reset_wq:
4819         destroy_workqueue(nvme_reset_wq);
4820 destroy_wq:
4821         destroy_workqueue(nvme_wq);
4822 out:
4823         return result;
4824 }
4825
4826 static void __exit nvme_core_exit(void)
4827 {
4828         class_destroy(nvme_subsys_class);
4829         class_destroy(nvme_class);
4830         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4831         destroy_workqueue(nvme_delete_wq);
4832         destroy_workqueue(nvme_reset_wq);
4833         destroy_workqueue(nvme_wq);
4834         ida_destroy(&nvme_instance_ida);
4835 }
4836
4837 MODULE_LICENSE("GPL");
4838 MODULE_VERSION("1.0");
4839 module_init(nvme_core_init);
4840 module_exit(nvme_core_exit);