drivers/nvme/host/core.c

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