linux/drivers/nvme/host/core.c
<<
>>
Prefs
   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
  33unsigned int admin_timeout = 60;
  34module_param(admin_timeout, uint, 0644);
  35MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  36EXPORT_SYMBOL_GPL(admin_timeout);
  37
  38unsigned int nvme_io_timeout = 30;
  39module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  40MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  41EXPORT_SYMBOL_GPL(nvme_io_timeout);
  42
  43static unsigned char shutdown_timeout = 5;
  44module_param(shutdown_timeout, byte, 0644);
  45MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  46
  47static u8 nvme_max_retries = 5;
  48module_param_named(max_retries, nvme_max_retries, byte, 0644);
  49MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  50
  51static unsigned long default_ps_max_latency_us = 100000;
  52module_param(default_ps_max_latency_us, ulong, 0644);
  53MODULE_PARM_DESC(default_ps_max_latency_us,
  54                 "max power saving latency for new devices; use PM QOS to change per device");
  55
  56static bool force_apst;
  57module_param(force_apst, bool, 0644);
  58MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  59
  60static unsigned long apst_primary_timeout_ms = 100;
  61module_param(apst_primary_timeout_ms, ulong, 0644);
  62MODULE_PARM_DESC(apst_primary_timeout_ms,
  63        "primary APST timeout in ms");
  64
  65static unsigned long apst_secondary_timeout_ms = 2000;
  66module_param(apst_secondary_timeout_ms, ulong, 0644);
  67MODULE_PARM_DESC(apst_secondary_timeout_ms,
  68        "secondary APST timeout in ms");
  69
  70static unsigned long apst_primary_latency_tol_us = 15000;
  71module_param(apst_primary_latency_tol_us, ulong, 0644);
  72MODULE_PARM_DESC(apst_primary_latency_tol_us,
  73        "primary APST latency tolerance in us");
  74
  75static unsigned long apst_secondary_latency_tol_us = 100000;
  76module_param(apst_secondary_latency_tol_us, ulong, 0644);
  77MODULE_PARM_DESC(apst_secondary_latency_tol_us,
  78        "secondary APST latency tolerance in us");
  79
  80static bool streams;
  81module_param(streams, bool, 0644);
  82MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  83
  84/*
  85 * nvme_wq - hosts nvme related works that are not reset or delete
  86 * nvme_reset_wq - hosts nvme reset works
  87 * nvme_delete_wq - hosts nvme delete works
  88 *
  89 * nvme_wq will host works such as scan, aen handling, fw activation,
  90 * keep-alive, periodic reconnects etc. nvme_reset_wq
  91 * runs reset works which also flush works hosted on nvme_wq for
  92 * serialization purposes. nvme_delete_wq host controller deletion
  93 * works which flush reset works for serialization.
  94 */
  95struct workqueue_struct *nvme_wq;
  96EXPORT_SYMBOL_GPL(nvme_wq);
  97
  98struct workqueue_struct *nvme_reset_wq;
  99EXPORT_SYMBOL_GPL(nvme_reset_wq);
 100
 101struct workqueue_struct *nvme_delete_wq;
 102EXPORT_SYMBOL_GPL(nvme_delete_wq);
 103
 104static LIST_HEAD(nvme_subsystems);
 105static DEFINE_MUTEX(nvme_subsystems_lock);
 106
 107static DEFINE_IDA(nvme_instance_ida);
 108static dev_t nvme_ctrl_base_chr_devt;
 109static struct class *nvme_class;
 110static struct class *nvme_subsys_class;
 111
 112static DEFINE_IDA(nvme_ns_chr_minor_ida);
 113static dev_t nvme_ns_chr_devt;
 114static struct class *nvme_ns_chr_class;
 115
 116static void nvme_put_subsystem(struct nvme_subsystem *subsys);
 117static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
 118                                           unsigned nsid);
 119
 120/*
 121 * Prepare a queue for teardown.
 122 *
 123 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
 124 * the capacity to 0 after that to avoid blocking dispatchers that may be
 125 * holding bd_butex.  This will end buffered writers dirtying pages that can't
 126 * be synced.
 127 */
 128static void nvme_set_queue_dying(struct nvme_ns *ns)
 129{
 130        if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 131                return;
 132
 133        blk_set_queue_dying(ns->queue);
 134        blk_mq_unquiesce_queue(ns->queue);
 135
 136        set_capacity_and_notify(ns->disk, 0);
 137}
 138
 139void nvme_queue_scan(struct nvme_ctrl *ctrl)
 140{
 141        /*
 142         * Only new queue scan work when admin and IO queues are both alive
 143         */
 144        if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 145                queue_work(nvme_wq, &ctrl->scan_work);
 146}
 147
 148/*
 149 * Use this function to proceed with scheduling reset_work for a controller
 150 * that had previously been set to the resetting state. This is intended for
 151 * code paths that can't be interrupted by other reset attempts. A hot removal
 152 * may prevent this from succeeding.
 153 */
 154int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 155{
 156        if (ctrl->state != NVME_CTRL_RESETTING)
 157                return -EBUSY;
 158        if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 159                return -EBUSY;
 160        return 0;
 161}
 162EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 163
 164static void nvme_failfast_work(struct work_struct *work)
 165{
 166        struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 167                        struct nvme_ctrl, failfast_work);
 168
 169        if (ctrl->state != NVME_CTRL_CONNECTING)
 170                return;
 171
 172        set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 173        dev_info(ctrl->device, "failfast expired\n");
 174        nvme_kick_requeue_lists(ctrl);
 175}
 176
 177static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 178{
 179        if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 180                return;
 181
 182        schedule_delayed_work(&ctrl->failfast_work,
 183                              ctrl->opts->fast_io_fail_tmo * HZ);
 184}
 185
 186static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 187{
 188        if (!ctrl->opts)
 189                return;
 190
 191        cancel_delayed_work_sync(&ctrl->failfast_work);
 192        clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 193}
 194
 195
 196int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 197{
 198        if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 199                return -EBUSY;
 200        if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 201                return -EBUSY;
 202        return 0;
 203}
 204EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 205
 206int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 207{
 208        int ret;
 209
 210        ret = nvme_reset_ctrl(ctrl);
 211        if (!ret) {
 212                flush_work(&ctrl->reset_work);
 213                if (ctrl->state != NVME_CTRL_LIVE)
 214                        ret = -ENETRESET;
 215        }
 216
 217        return ret;
 218}
 219
 220static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 221{
 222        dev_info(ctrl->device,
 223                 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 224
 225        flush_work(&ctrl->reset_work);
 226        nvme_stop_ctrl(ctrl);
 227        nvme_remove_namespaces(ctrl);
 228        ctrl->ops->delete_ctrl(ctrl);
 229        nvme_uninit_ctrl(ctrl);
 230}
 231
 232static void nvme_delete_ctrl_work(struct work_struct *work)
 233{
 234        struct nvme_ctrl *ctrl =
 235                container_of(work, struct nvme_ctrl, delete_work);
 236
 237        nvme_do_delete_ctrl(ctrl);
 238}
 239
 240int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 241{
 242        if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 243                return -EBUSY;
 244        if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 245                return -EBUSY;
 246        return 0;
 247}
 248EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 249
 250static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 251{
 252        /*
 253         * Keep a reference until nvme_do_delete_ctrl() complete,
 254         * since ->delete_ctrl can free the controller.
 255         */
 256        nvme_get_ctrl(ctrl);
 257        if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 258                nvme_do_delete_ctrl(ctrl);
 259        nvme_put_ctrl(ctrl);
 260}
 261
 262static blk_status_t nvme_error_status(u16 status)
 263{
 264        switch (status & 0x7ff) {
 265        case NVME_SC_SUCCESS:
 266                return BLK_STS_OK;
 267        case NVME_SC_CAP_EXCEEDED:
 268                return BLK_STS_NOSPC;
 269        case NVME_SC_LBA_RANGE:
 270        case NVME_SC_CMD_INTERRUPTED:
 271        case NVME_SC_NS_NOT_READY:
 272                return BLK_STS_TARGET;
 273        case NVME_SC_BAD_ATTRIBUTES:
 274        case NVME_SC_ONCS_NOT_SUPPORTED:
 275        case NVME_SC_INVALID_OPCODE:
 276        case NVME_SC_INVALID_FIELD:
 277        case NVME_SC_INVALID_NS:
 278                return BLK_STS_NOTSUPP;
 279        case NVME_SC_WRITE_FAULT:
 280        case NVME_SC_READ_ERROR:
 281        case NVME_SC_UNWRITTEN_BLOCK:
 282        case NVME_SC_ACCESS_DENIED:
 283        case NVME_SC_READ_ONLY:
 284        case NVME_SC_COMPARE_FAILED:
 285                return BLK_STS_MEDIUM;
 286        case NVME_SC_GUARD_CHECK:
 287        case NVME_SC_APPTAG_CHECK:
 288        case NVME_SC_REFTAG_CHECK:
 289        case NVME_SC_INVALID_PI:
 290                return BLK_STS_PROTECTION;
 291        case NVME_SC_RESERVATION_CONFLICT:
 292                return BLK_STS_NEXUS;
 293        case NVME_SC_HOST_PATH_ERROR:
 294                return BLK_STS_TRANSPORT;
 295        case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 296                return BLK_STS_ZONE_ACTIVE_RESOURCE;
 297        case NVME_SC_ZONE_TOO_MANY_OPEN:
 298                return BLK_STS_ZONE_OPEN_RESOURCE;
 299        default:
 300                return BLK_STS_IOERR;
 301        }
 302}
 303
 304static void nvme_retry_req(struct request *req)
 305{
 306        unsigned long delay = 0;
 307        u16 crd;
 308
 309        /* The mask and shift result must be <= 3 */
 310        crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 311        if (crd)
 312                delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 313
 314        nvme_req(req)->retries++;
 315        blk_mq_requeue_request(req, false);
 316        blk_mq_delay_kick_requeue_list(req->q, delay);
 317}
 318
 319enum nvme_disposition {
 320        COMPLETE,
 321        RETRY,
 322        FAILOVER,
 323};
 324
 325static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 326{
 327        if (likely(nvme_req(req)->status == 0))
 328                return COMPLETE;
 329
 330        if (blk_noretry_request(req) ||
 331            (nvme_req(req)->status & NVME_SC_DNR) ||
 332            nvme_req(req)->retries >= nvme_max_retries)
 333                return COMPLETE;
 334
 335        if (req->cmd_flags & REQ_NVME_MPATH) {
 336                if (nvme_is_path_error(nvme_req(req)->status) ||
 337                    blk_queue_dying(req->q))
 338                        return FAILOVER;
 339        } else {
 340                if (blk_queue_dying(req->q))
 341                        return COMPLETE;
 342        }
 343
 344        return RETRY;
 345}
 346
 347static inline void nvme_end_req(struct request *req)
 348{
 349        blk_status_t status = nvme_error_status(nvme_req(req)->status);
 350
 351        if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 352            req_op(req) == REQ_OP_ZONE_APPEND)
 353                req->__sector = nvme_lba_to_sect(req->q->queuedata,
 354                        le64_to_cpu(nvme_req(req)->result.u64));
 355
 356        nvme_trace_bio_complete(req);
 357        blk_mq_end_request(req, status);
 358}
 359
 360void nvme_complete_rq(struct request *req)
 361{
 362        trace_nvme_complete_rq(req);
 363        nvme_cleanup_cmd(req);
 364
 365        if (nvme_req(req)->ctrl->kas)
 366                nvme_req(req)->ctrl->comp_seen = true;
 367
 368        switch (nvme_decide_disposition(req)) {
 369        case COMPLETE:
 370                nvme_end_req(req);
 371                return;
 372        case RETRY:
 373                nvme_retry_req(req);
 374                return;
 375        case FAILOVER:
 376                nvme_failover_req(req);
 377                return;
 378        }
 379}
 380EXPORT_SYMBOL_GPL(nvme_complete_rq);
 381
 382/*
 383 * Called to unwind from ->queue_rq on a failed command submission so that the
 384 * multipathing code gets called to potentially failover to another path.
 385 * The caller needs to unwind all transport specific resource allocations and
 386 * must return propagate the return value.
 387 */
 388blk_status_t nvme_host_path_error(struct request *req)
 389{
 390        nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 391        blk_mq_set_request_complete(req);
 392        nvme_complete_rq(req);
 393        return BLK_STS_OK;
 394}
 395EXPORT_SYMBOL_GPL(nvme_host_path_error);
 396
 397bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 398{
 399        dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 400                                "Cancelling I/O %d", req->tag);
 401
 402        /* don't abort one completed request */
 403        if (blk_mq_request_completed(req))
 404                return true;
 405
 406        nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 407        nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 408        blk_mq_complete_request(req);
 409        return true;
 410}
 411EXPORT_SYMBOL_GPL(nvme_cancel_request);
 412
 413void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 414{
 415        if (ctrl->tagset) {
 416                blk_mq_tagset_busy_iter(ctrl->tagset,
 417                                nvme_cancel_request, ctrl);
 418                blk_mq_tagset_wait_completed_request(ctrl->tagset);
 419        }
 420}
 421EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 422
 423void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 424{
 425        if (ctrl->admin_tagset) {
 426                blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 427                                nvme_cancel_request, ctrl);
 428                blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 429        }
 430}
 431EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 432
 433bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 434                enum nvme_ctrl_state new_state)
 435{
 436        enum nvme_ctrl_state old_state;
 437        unsigned long flags;
 438        bool changed = false;
 439
 440        spin_lock_irqsave(&ctrl->lock, flags);
 441
 442        old_state = ctrl->state;
 443        switch (new_state) {
 444        case NVME_CTRL_LIVE:
 445                switch (old_state) {
 446                case NVME_CTRL_NEW:
 447                case NVME_CTRL_RESETTING:
 448                case NVME_CTRL_CONNECTING:
 449                        changed = true;
 450                        fallthrough;
 451                default:
 452                        break;
 453                }
 454                break;
 455        case NVME_CTRL_RESETTING:
 456                switch (old_state) {
 457                case NVME_CTRL_NEW:
 458                case NVME_CTRL_LIVE:
 459                        changed = true;
 460                        fallthrough;
 461                default:
 462                        break;
 463                }
 464                break;
 465        case NVME_CTRL_CONNECTING:
 466                switch (old_state) {
 467                case NVME_CTRL_NEW:
 468                case NVME_CTRL_RESETTING:
 469                        changed = true;
 470                        fallthrough;
 471                default:
 472                        break;
 473                }
 474                break;
 475        case NVME_CTRL_DELETING:
 476                switch (old_state) {
 477                case NVME_CTRL_LIVE:
 478                case NVME_CTRL_RESETTING:
 479                case NVME_CTRL_CONNECTING:
 480                        changed = true;
 481                        fallthrough;
 482                default:
 483                        break;
 484                }
 485                break;
 486        case NVME_CTRL_DELETING_NOIO:
 487                switch (old_state) {
 488                case NVME_CTRL_DELETING:
 489                case NVME_CTRL_DEAD:
 490                        changed = true;
 491                        fallthrough;
 492                default:
 493                        break;
 494                }
 495                break;
 496        case NVME_CTRL_DEAD:
 497                switch (old_state) {
 498                case NVME_CTRL_DELETING:
 499                        changed = true;
 500                        fallthrough;
 501                default:
 502                        break;
 503                }
 504                break;
 505        default:
 506                break;
 507        }
 508
 509        if (changed) {
 510                ctrl->state = new_state;
 511                wake_up_all(&ctrl->state_wq);
 512        }
 513
 514        spin_unlock_irqrestore(&ctrl->lock, flags);
 515        if (!changed)
 516                return false;
 517
 518        if (ctrl->state == NVME_CTRL_LIVE) {
 519                if (old_state == NVME_CTRL_CONNECTING)
 520                        nvme_stop_failfast_work(ctrl);
 521                nvme_kick_requeue_lists(ctrl);
 522        } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 523                old_state == NVME_CTRL_RESETTING) {
 524                nvme_start_failfast_work(ctrl);
 525        }
 526        return changed;
 527}
 528EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 529
 530/*
 531 * Returns true for sink states that can't ever transition back to live.
 532 */
 533static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 534{
 535        switch (ctrl->state) {
 536        case NVME_CTRL_NEW:
 537        case NVME_CTRL_LIVE:
 538        case NVME_CTRL_RESETTING:
 539        case NVME_CTRL_CONNECTING:
 540                return false;
 541        case NVME_CTRL_DELETING:
 542        case NVME_CTRL_DELETING_NOIO:
 543        case NVME_CTRL_DEAD:
 544                return true;
 545        default:
 546                WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 547                return true;
 548        }
 549}
 550
 551/*
 552 * Waits for the controller state to be resetting, or returns false if it is
 553 * not possible to ever transition to that state.
 554 */
 555bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 556{
 557        wait_event(ctrl->state_wq,
 558                   nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 559                   nvme_state_terminal(ctrl));
 560        return ctrl->state == NVME_CTRL_RESETTING;
 561}
 562EXPORT_SYMBOL_GPL(nvme_wait_reset);
 563
 564static void nvme_free_ns_head(struct kref *ref)
 565{
 566        struct nvme_ns_head *head =
 567                container_of(ref, struct nvme_ns_head, ref);
 568
 569        nvme_mpath_remove_disk(head);
 570        ida_simple_remove(&head->subsys->ns_ida, head->instance);
 571        cleanup_srcu_struct(&head->srcu);
 572        nvme_put_subsystem(head->subsys);
 573        kfree(head);
 574}
 575
 576bool nvme_tryget_ns_head(struct nvme_ns_head *head)
 577{
 578        return kref_get_unless_zero(&head->ref);
 579}
 580
 581void nvme_put_ns_head(struct nvme_ns_head *head)
 582{
 583        kref_put(&head->ref, nvme_free_ns_head);
 584}
 585
 586static void nvme_free_ns(struct kref *kref)
 587{
 588        struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 589
 590        if (ns->ndev)
 591                nvme_nvm_unregister(ns);
 592
 593        put_disk(ns->disk);
 594        nvme_put_ns_head(ns->head);
 595        nvme_put_ctrl(ns->ctrl);
 596        kfree(ns);
 597}
 598
 599static inline bool nvme_get_ns(struct nvme_ns *ns)
 600{
 601        return kref_get_unless_zero(&ns->kref);
 602}
 603
 604void nvme_put_ns(struct nvme_ns *ns)
 605{
 606        kref_put(&ns->kref, nvme_free_ns);
 607}
 608EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 609
 610static inline void nvme_clear_nvme_request(struct request *req)
 611{
 612        nvme_req(req)->status = 0;
 613        nvme_req(req)->retries = 0;
 614        nvme_req(req)->flags = 0;
 615        req->rq_flags |= RQF_DONTPREP;
 616}
 617
 618static inline unsigned int nvme_req_op(struct nvme_command *cmd)
 619{
 620        return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 621}
 622
 623static inline void nvme_init_request(struct request *req,
 624                struct nvme_command *cmd)
 625{
 626        if (req->q->queuedata)
 627                req->timeout = NVME_IO_TIMEOUT;
 628        else /* no queuedata implies admin queue */
 629                req->timeout = NVME_ADMIN_TIMEOUT;
 630
 631        /* passthru commands should let the driver set the SGL flags */
 632        cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 633
 634        req->cmd_flags |= REQ_FAILFAST_DRIVER;
 635        if (req->mq_hctx->type == HCTX_TYPE_POLL)
 636                req->cmd_flags |= REQ_HIPRI;
 637        nvme_clear_nvme_request(req);
 638        memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
 639}
 640
 641struct request *nvme_alloc_request(struct request_queue *q,
 642                struct nvme_command *cmd, blk_mq_req_flags_t flags)
 643{
 644        struct request *req;
 645
 646        req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
 647        if (!IS_ERR(req))
 648                nvme_init_request(req, cmd);
 649        return req;
 650}
 651EXPORT_SYMBOL_GPL(nvme_alloc_request);
 652
 653static struct request *nvme_alloc_request_qid(struct request_queue *q,
 654                struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 655{
 656        struct request *req;
 657
 658        req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
 659                        qid ? qid - 1 : 0);
 660        if (!IS_ERR(req))
 661                nvme_init_request(req, cmd);
 662        return req;
 663}
 664
 665/*
 666 * For something we're not in a state to send to the device the default action
 667 * is to busy it and retry it after the controller state is recovered.  However,
 668 * if the controller is deleting or if anything is marked for failfast or
 669 * nvme multipath it is immediately failed.
 670 *
 671 * Note: commands used to initialize the controller will be marked for failfast.
 672 * Note: nvme cli/ioctl commands are marked for failfast.
 673 */
 674blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
 675                struct request *rq)
 676{
 677        if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
 678            ctrl->state != NVME_CTRL_DEAD &&
 679            !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
 680            !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 681                return BLK_STS_RESOURCE;
 682        return nvme_host_path_error(rq);
 683}
 684EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
 685
 686bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 687                bool queue_live)
 688{
 689        struct nvme_request *req = nvme_req(rq);
 690
 691        /*
 692         * currently we have a problem sending passthru commands
 693         * on the admin_q if the controller is not LIVE because we can't
 694         * make sure that they are going out after the admin connect,
 695         * controller enable and/or other commands in the initialization
 696         * sequence. until the controller will be LIVE, fail with
 697         * BLK_STS_RESOURCE so that they will be rescheduled.
 698         */
 699        if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
 700                return false;
 701
 702        if (ctrl->ops->flags & NVME_F_FABRICS) {
 703                /*
 704                 * Only allow commands on a live queue, except for the connect
 705                 * command, which is require to set the queue live in the
 706                 * appropinquate states.
 707                 */
 708                switch (ctrl->state) {
 709                case NVME_CTRL_CONNECTING:
 710                        if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
 711                            req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
 712                                return true;
 713                        break;
 714                default:
 715                        break;
 716                case NVME_CTRL_DEAD:
 717                        return false;
 718                }
 719        }
 720
 721        return queue_live;
 722}
 723EXPORT_SYMBOL_GPL(__nvme_check_ready);
 724
 725static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 726{
 727        struct nvme_command c = { };
 728
 729        c.directive.opcode = nvme_admin_directive_send;
 730        c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 731        c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 732        c.directive.dtype = NVME_DIR_IDENTIFY;
 733        c.directive.tdtype = NVME_DIR_STREAMS;
 734        c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 735
 736        return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 737}
 738
 739static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 740{
 741        return nvme_toggle_streams(ctrl, false);
 742}
 743
 744static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 745{
 746        return nvme_toggle_streams(ctrl, true);
 747}
 748
 749static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 750                                  struct streams_directive_params *s, u32 nsid)
 751{
 752        struct nvme_command c = { };
 753
 754        memset(s, 0, sizeof(*s));
 755
 756        c.directive.opcode = nvme_admin_directive_recv;
 757        c.directive.nsid = cpu_to_le32(nsid);
 758        c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 759        c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 760        c.directive.dtype = NVME_DIR_STREAMS;
 761
 762        return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 763}
 764
 765static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 766{
 767        struct streams_directive_params s;
 768        int ret;
 769
 770        if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 771                return 0;
 772        if (!streams)
 773                return 0;
 774
 775        ret = nvme_enable_streams(ctrl);
 776        if (ret)
 777                return ret;
 778
 779        ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 780        if (ret)
 781                goto out_disable_stream;
 782
 783        ctrl->nssa = le16_to_cpu(s.nssa);
 784        if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 785                dev_info(ctrl->device, "too few streams (%u) available\n",
 786                                        ctrl->nssa);
 787                goto out_disable_stream;
 788        }
 789
 790        ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 791        dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 792        return 0;
 793
 794out_disable_stream:
 795        nvme_disable_streams(ctrl);
 796        return ret;
 797}
 798
 799/*
 800 * Check if 'req' has a write hint associated with it. If it does, assign
 801 * a valid namespace stream to the write.
 802 */
 803static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 804                                     struct request *req, u16 *control,
 805                                     u32 *dsmgmt)
 806{
 807        enum rw_hint streamid = req->write_hint;
 808
 809        if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 810                streamid = 0;
 811        else {
 812                streamid--;
 813                if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 814                        return;
 815
 816                *control |= NVME_RW_DTYPE_STREAMS;
 817                *dsmgmt |= streamid << 16;
 818        }
 819
 820        if (streamid < ARRAY_SIZE(req->q->write_hints))
 821                req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 822}
 823
 824static inline void nvme_setup_flush(struct nvme_ns *ns,
 825                struct nvme_command *cmnd)
 826{
 827        cmnd->common.opcode = nvme_cmd_flush;
 828        cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 829}
 830
 831static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 832                struct nvme_command *cmnd)
 833{
 834        unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 835        struct nvme_dsm_range *range;
 836        struct bio *bio;
 837
 838        /*
 839         * Some devices do not consider the DSM 'Number of Ranges' field when
 840         * determining how much data to DMA. Always allocate memory for maximum
 841         * number of segments to prevent device reading beyond end of buffer.
 842         */
 843        static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 844
 845        range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 846        if (!range) {
 847                /*
 848                 * If we fail allocation our range, fallback to the controller
 849                 * discard page. If that's also busy, it's safe to return
 850                 * busy, as we know we can make progress once that's freed.
 851                 */
 852                if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 853                        return BLK_STS_RESOURCE;
 854
 855                range = page_address(ns->ctrl->discard_page);
 856        }
 857
 858        __rq_for_each_bio(bio, req) {
 859                u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 860                u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 861
 862                if (n < segments) {
 863                        range[n].cattr = cpu_to_le32(0);
 864                        range[n].nlb = cpu_to_le32(nlb);
 865                        range[n].slba = cpu_to_le64(slba);
 866                }
 867                n++;
 868        }
 869
 870        if (WARN_ON_ONCE(n != segments)) {
 871                if (virt_to_page(range) == ns->ctrl->discard_page)
 872                        clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 873                else
 874                        kfree(range);
 875                return BLK_STS_IOERR;
 876        }
 877
 878        cmnd->dsm.opcode = nvme_cmd_dsm;
 879        cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 880        cmnd->dsm.nr = cpu_to_le32(segments - 1);
 881        cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 882
 883        req->special_vec.bv_page = virt_to_page(range);
 884        req->special_vec.bv_offset = offset_in_page(range);
 885        req->special_vec.bv_len = alloc_size;
 886        req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 887
 888        return BLK_STS_OK;
 889}
 890
 891static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 892                struct request *req, struct nvme_command *cmnd)
 893{
 894        if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 895                return nvme_setup_discard(ns, req, cmnd);
 896
 897        cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 898        cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 899        cmnd->write_zeroes.slba =
 900                cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 901        cmnd->write_zeroes.length =
 902                cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 903        if (nvme_ns_has_pi(ns))
 904                cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
 905        else
 906                cmnd->write_zeroes.control = 0;
 907        return BLK_STS_OK;
 908}
 909
 910static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 911                struct request *req, struct nvme_command *cmnd,
 912                enum nvme_opcode op)
 913{
 914        struct nvme_ctrl *ctrl = ns->ctrl;
 915        u16 control = 0;
 916        u32 dsmgmt = 0;
 917
 918        if (req->cmd_flags & REQ_FUA)
 919                control |= NVME_RW_FUA;
 920        if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 921                control |= NVME_RW_LR;
 922
 923        if (req->cmd_flags & REQ_RAHEAD)
 924                dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 925
 926        cmnd->rw.opcode = op;
 927        cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 928        cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 929        cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 930
 931        if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 932                nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 933
 934        if (ns->ms) {
 935                /*
 936                 * If formated with metadata, the block layer always provides a
 937                 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 938                 * we enable the PRACT bit for protection information or set the
 939                 * namespace capacity to zero to prevent any I/O.
 940                 */
 941                if (!blk_integrity_rq(req)) {
 942                        if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 943                                return BLK_STS_NOTSUPP;
 944                        control |= NVME_RW_PRINFO_PRACT;
 945                }
 946
 947                switch (ns->pi_type) {
 948                case NVME_NS_DPS_PI_TYPE3:
 949                        control |= NVME_RW_PRINFO_PRCHK_GUARD;
 950                        break;
 951                case NVME_NS_DPS_PI_TYPE1:
 952                case NVME_NS_DPS_PI_TYPE2:
 953                        control |= NVME_RW_PRINFO_PRCHK_GUARD |
 954                                        NVME_RW_PRINFO_PRCHK_REF;
 955                        if (op == nvme_cmd_zone_append)
 956                                control |= NVME_RW_APPEND_PIREMAP;
 957                        cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 958                        break;
 959                }
 960        }
 961
 962        cmnd->rw.control = cpu_to_le16(control);
 963        cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 964        return 0;
 965}
 966
 967void nvme_cleanup_cmd(struct request *req)
 968{
 969        if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 970                struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 971                struct page *page = req->special_vec.bv_page;
 972
 973                if (page == ctrl->discard_page)
 974                        clear_bit_unlock(0, &ctrl->discard_page_busy);
 975                else
 976                        kfree(page_address(page) + req->special_vec.bv_offset);
 977        }
 978}
 979EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 980
 981blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
 982{
 983        struct nvme_command *cmd = nvme_req(req)->cmd;
 984        blk_status_t ret = BLK_STS_OK;
 985
 986        if (!(req->rq_flags & RQF_DONTPREP)) {
 987                nvme_clear_nvme_request(req);
 988                memset(cmd, 0, sizeof(*cmd));
 989        }
 990
 991        switch (req_op(req)) {
 992        case REQ_OP_DRV_IN:
 993        case REQ_OP_DRV_OUT:
 994                /* these are setup prior to execution in nvme_init_request() */
 995                break;
 996        case REQ_OP_FLUSH:
 997                nvme_setup_flush(ns, cmd);
 998                break;
 999        case REQ_OP_ZONE_RESET_ALL:
1000        case REQ_OP_ZONE_RESET:
1001                ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1002                break;
1003        case REQ_OP_ZONE_OPEN:
1004                ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1005                break;
1006        case REQ_OP_ZONE_CLOSE:
1007                ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1008                break;
1009        case REQ_OP_ZONE_FINISH:
1010                ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1011                break;
1012        case REQ_OP_WRITE_ZEROES:
1013                ret = nvme_setup_write_zeroes(ns, req, cmd);
1014                break;
1015        case REQ_OP_DISCARD:
1016                ret = nvme_setup_discard(ns, req, cmd);
1017                break;
1018        case REQ_OP_READ:
1019                ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1020                break;
1021        case REQ_OP_WRITE:
1022                ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1023                break;
1024        case REQ_OP_ZONE_APPEND:
1025                ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1026                break;
1027        default:
1028                WARN_ON_ONCE(1);
1029                return BLK_STS_IOERR;
1030        }
1031
1032        cmd->common.command_id = req->tag;
1033        trace_nvme_setup_cmd(req, cmd);
1034        return ret;
1035}
1036EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1037
1038/*
1039 * Return values:
1040 * 0:  success
1041 * >0: nvme controller's cqe status response
1042 * <0: kernel error in lieu of controller response
1043 */
1044static int nvme_execute_rq(struct gendisk *disk, struct request *rq,
1045                bool at_head)
1046{
1047        blk_status_t status;
1048
1049        status = blk_execute_rq(disk, rq, at_head);
1050        if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1051                return -EINTR;
1052        if (nvme_req(rq)->status)
1053                return nvme_req(rq)->status;
1054        return blk_status_to_errno(status);
1055}
1056
1057/*
1058 * Returns 0 on success.  If the result is negative, it's a Linux error code;
1059 * if the result is positive, it's an NVM Express status code
1060 */
1061int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1062                union nvme_result *result, void *buffer, unsigned bufflen,
1063                unsigned timeout, int qid, int at_head,
1064                blk_mq_req_flags_t flags)
1065{
1066        struct request *req;
1067        int ret;
1068
1069        if (qid == NVME_QID_ANY)
1070                req = nvme_alloc_request(q, cmd, flags);
1071        else
1072                req = nvme_alloc_request_qid(q, cmd, flags, qid);
1073        if (IS_ERR(req))
1074                return PTR_ERR(req);
1075
1076        if (timeout)
1077                req->timeout = timeout;
1078
1079        if (buffer && bufflen) {
1080                ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1081                if (ret)
1082                        goto out;
1083        }
1084
1085        ret = nvme_execute_rq(NULL, req, at_head);
1086        if (result && ret >= 0)
1087                *result = nvme_req(req)->result;
1088 out:
1089        blk_mq_free_request(req);
1090        return ret;
1091}
1092EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1093
1094int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1095                void *buffer, unsigned bufflen)
1096{
1097        return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1098                        NVME_QID_ANY, 0, 0);
1099}
1100EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1101
1102static u32 nvme_known_admin_effects(u8 opcode)
1103{
1104        switch (opcode) {
1105        case nvme_admin_format_nvm:
1106                return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1107                        NVME_CMD_EFFECTS_CSE_MASK;
1108        case nvme_admin_sanitize_nvm:
1109                return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1110        default:
1111                break;
1112        }
1113        return 0;
1114}
1115
1116u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1117{
1118        u32 effects = 0;
1119
1120        if (ns) {
1121                if (ns->head->effects)
1122                        effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1123                if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1124                        dev_warn_once(ctrl->device,
1125                                "IO command:%02x has unhandled effects:%08x\n",
1126                                opcode, effects);
1127                return 0;
1128        }
1129
1130        if (ctrl->effects)
1131                effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1132        effects |= nvme_known_admin_effects(opcode);
1133
1134        return effects;
1135}
1136EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1137
1138static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1139                               u8 opcode)
1140{
1141        u32 effects = nvme_command_effects(ctrl, ns, opcode);
1142
1143        /*
1144         * For simplicity, IO to all namespaces is quiesced even if the command
1145         * effects say only one namespace is affected.
1146         */
1147        if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1148                mutex_lock(&ctrl->scan_lock);
1149                mutex_lock(&ctrl->subsys->lock);
1150                nvme_mpath_start_freeze(ctrl->subsys);
1151                nvme_mpath_wait_freeze(ctrl->subsys);
1152                nvme_start_freeze(ctrl);
1153                nvme_wait_freeze(ctrl);
1154        }
1155        return effects;
1156}
1157
1158static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1159{
1160        if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1161                nvme_unfreeze(ctrl);
1162                nvme_mpath_unfreeze(ctrl->subsys);
1163                mutex_unlock(&ctrl->subsys->lock);
1164                nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1165                mutex_unlock(&ctrl->scan_lock);
1166        }
1167        if (effects & NVME_CMD_EFFECTS_CCC)
1168                nvme_init_ctrl_finish(ctrl);
1169        if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1170                nvme_queue_scan(ctrl);
1171                flush_work(&ctrl->scan_work);
1172        }
1173}
1174
1175int nvme_execute_passthru_rq(struct request *rq)
1176{
1177        struct nvme_command *cmd = nvme_req(rq)->cmd;
1178        struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1179        struct nvme_ns *ns = rq->q->queuedata;
1180        struct gendisk *disk = ns ? ns->disk : NULL;
1181        u32 effects;
1182        int  ret;
1183
1184        effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1185        ret = nvme_execute_rq(disk, rq, false);
1186        if (effects) /* nothing to be done for zero cmd effects */
1187                nvme_passthru_end(ctrl, effects);
1188
1189        return ret;
1190}
1191EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1192
1193/*
1194 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1195 * 
1196 *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1197 *   accounting for transport roundtrip times [..].
1198 */
1199static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1200{
1201        queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1202}
1203
1204static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1205{
1206        struct nvme_ctrl *ctrl = rq->end_io_data;
1207        unsigned long flags;
1208        bool startka = false;
1209
1210        blk_mq_free_request(rq);
1211
1212        if (status) {
1213                dev_err(ctrl->device,
1214                        "failed nvme_keep_alive_end_io error=%d\n",
1215                                status);
1216                return;
1217        }
1218
1219        ctrl->comp_seen = false;
1220        spin_lock_irqsave(&ctrl->lock, flags);
1221        if (ctrl->state == NVME_CTRL_LIVE ||
1222            ctrl->state == NVME_CTRL_CONNECTING)
1223                startka = true;
1224        spin_unlock_irqrestore(&ctrl->lock, flags);
1225        if (startka)
1226                nvme_queue_keep_alive_work(ctrl);
1227}
1228
1229static void nvme_keep_alive_work(struct work_struct *work)
1230{
1231        struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1232                        struct nvme_ctrl, ka_work);
1233        bool comp_seen = ctrl->comp_seen;
1234        struct request *rq;
1235
1236        if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1237                dev_dbg(ctrl->device,
1238                        "reschedule traffic based keep-alive timer\n");
1239                ctrl->comp_seen = false;
1240                nvme_queue_keep_alive_work(ctrl);
1241                return;
1242        }
1243
1244        rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1245                                BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1246        if (IS_ERR(rq)) {
1247                /* allocation failure, reset the controller */
1248                dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1249                nvme_reset_ctrl(ctrl);
1250                return;
1251        }
1252
1253        rq->timeout = ctrl->kato * HZ;
1254        rq->end_io_data = ctrl;
1255        blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1256}
1257
1258static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1259{
1260        if (unlikely(ctrl->kato == 0))
1261                return;
1262
1263        nvme_queue_keep_alive_work(ctrl);
1264}
1265
1266void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1267{
1268        if (unlikely(ctrl->kato == 0))
1269                return;
1270
1271        cancel_delayed_work_sync(&ctrl->ka_work);
1272}
1273EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1274
1275/*
1276 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1277 * flag, thus sending any new CNS opcodes has a big chance of not working.
1278 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1279 * (but not for any later version).
1280 */
1281static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1282{
1283        if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1284                return ctrl->vs < NVME_VS(1, 2, 0);
1285        return ctrl->vs < NVME_VS(1, 1, 0);
1286}
1287
1288static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1289{
1290        struct nvme_command c = { };
1291        int error;
1292
1293        /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1294        c.identify.opcode = nvme_admin_identify;
1295        c.identify.cns = NVME_ID_CNS_CTRL;
1296
1297        *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1298        if (!*id)
1299                return -ENOMEM;
1300
1301        error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1302                        sizeof(struct nvme_id_ctrl));
1303        if (error)
1304                kfree(*id);
1305        return error;
1306}
1307
1308static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1309{
1310        return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1311}
1312
1313static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1314                struct nvme_ns_id_desc *cur, bool *csi_seen)
1315{
1316        const char *warn_str = "ctrl returned bogus length:";
1317        void *data = cur;
1318
1319        switch (cur->nidt) {
1320        case NVME_NIDT_EUI64:
1321                if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1322                        dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1323                                 warn_str, cur->nidl);
1324                        return -1;
1325                }
1326                memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1327                return NVME_NIDT_EUI64_LEN;
1328        case NVME_NIDT_NGUID:
1329                if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1330                        dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1331                                 warn_str, cur->nidl);
1332                        return -1;
1333                }
1334                memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1335                return NVME_NIDT_NGUID_LEN;
1336        case NVME_NIDT_UUID:
1337                if (cur->nidl != NVME_NIDT_UUID_LEN) {
1338                        dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1339                                 warn_str, cur->nidl);
1340                        return -1;
1341                }
1342                uuid_copy(&ids->uuid, data + sizeof(*cur));
1343                return NVME_NIDT_UUID_LEN;
1344        case NVME_NIDT_CSI:
1345                if (cur->nidl != NVME_NIDT_CSI_LEN) {
1346                        dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1347                                 warn_str, cur->nidl);
1348                        return -1;
1349                }
1350                memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1351                *csi_seen = true;
1352                return NVME_NIDT_CSI_LEN;
1353        default:
1354                /* Skip unknown types */
1355                return cur->nidl;
1356        }
1357}
1358
1359static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1360                struct nvme_ns_ids *ids)
1361{
1362        struct nvme_command c = { };
1363        bool csi_seen = false;
1364        int status, pos, len;
1365        void *data;
1366
1367        if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1368                return 0;
1369        if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1370                return 0;
1371
1372        c.identify.opcode = nvme_admin_identify;
1373        c.identify.nsid = cpu_to_le32(nsid);
1374        c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1375
1376        data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1377        if (!data)
1378                return -ENOMEM;
1379
1380        status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1381                                      NVME_IDENTIFY_DATA_SIZE);
1382        if (status) {
1383                dev_warn(ctrl->device,
1384                        "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1385                        nsid, status);
1386                goto free_data;
1387        }
1388
1389        for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1390                struct nvme_ns_id_desc *cur = data + pos;
1391
1392                if (cur->nidl == 0)
1393                        break;
1394
1395                len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1396                if (len < 0)
1397                        break;
1398
1399                len += sizeof(*cur);
1400        }
1401
1402        if (nvme_multi_css(ctrl) && !csi_seen) {
1403                dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1404                         nsid);
1405                status = -EINVAL;
1406        }
1407
1408free_data:
1409        kfree(data);
1410        return status;
1411}
1412
1413static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1414                        struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1415{
1416        struct nvme_command c = { };
1417        int error;
1418
1419        /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1420        c.identify.opcode = nvme_admin_identify;
1421        c.identify.nsid = cpu_to_le32(nsid);
1422        c.identify.cns = NVME_ID_CNS_NS;
1423
1424        *id = kmalloc(sizeof(**id), GFP_KERNEL);
1425        if (!*id)
1426                return -ENOMEM;
1427
1428        error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1429        if (error) {
1430                dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1431                goto out_free_id;
1432        }
1433
1434        error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1435        if ((*id)->ncap == 0) /* namespace not allocated or attached */
1436                goto out_free_id;
1437
1438        if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1439            !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1440                memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1441        if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1442            !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1443                memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1444
1445        return 0;
1446
1447out_free_id:
1448        kfree(*id);
1449        return error;
1450}
1451
1452static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1453                unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1454{
1455        union nvme_result res = { 0 };
1456        struct nvme_command c = { };
1457        int ret;
1458
1459        c.features.opcode = op;
1460        c.features.fid = cpu_to_le32(fid);
1461        c.features.dword11 = cpu_to_le32(dword11);
1462
1463        ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1464                        buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1465        if (ret >= 0 && result)
1466                *result = le32_to_cpu(res.u32);
1467        return ret;
1468}
1469
1470int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1471                      unsigned int dword11, void *buffer, size_t buflen,
1472                      u32 *result)
1473{
1474        return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1475                             buflen, result);
1476}
1477EXPORT_SYMBOL_GPL(nvme_set_features);
1478
1479int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1480                      unsigned int dword11, void *buffer, size_t buflen,
1481                      u32 *result)
1482{
1483        return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1484                             buflen, result);
1485}
1486EXPORT_SYMBOL_GPL(nvme_get_features);
1487
1488int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1489{
1490        u32 q_count = (*count - 1) | ((*count - 1) << 16);
1491        u32 result;
1492        int status, nr_io_queues;
1493
1494        status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1495                        &result);
1496        if (status < 0)
1497                return status;
1498
1499        /*
1500         * Degraded controllers might return an error when setting the queue
1501         * count.  We still want to be able to bring them online and offer
1502         * access to the admin queue, as that might be only way to fix them up.
1503         */
1504        if (status > 0) {
1505                dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1506                *count = 0;
1507        } else {
1508                nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1509                *count = min(*count, nr_io_queues);
1510        }
1511
1512        return 0;
1513}
1514EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1515
1516#define NVME_AEN_SUPPORTED \
1517        (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1518         NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1519
1520static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1521{
1522        u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1523        int status;
1524
1525        if (!supported_aens)
1526                return;
1527
1528        status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1529                        NULL, 0, &result);
1530        if (status)
1531                dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1532                         supported_aens);
1533
1534        queue_work(nvme_wq, &ctrl->async_event_work);
1535}
1536
1537static int nvme_ns_open(struct nvme_ns *ns)
1538{
1539
1540        /* should never be called due to GENHD_FL_HIDDEN */
1541        if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1542                goto fail;
1543        if (!nvme_get_ns(ns))
1544                goto fail;
1545        if (!try_module_get(ns->ctrl->ops->module))
1546                goto fail_put_ns;
1547
1548        return 0;
1549
1550fail_put_ns:
1551        nvme_put_ns(ns);
1552fail:
1553        return -ENXIO;
1554}
1555
1556static void nvme_ns_release(struct nvme_ns *ns)
1557{
1558
1559        module_put(ns->ctrl->ops->module);
1560        nvme_put_ns(ns);
1561}
1562
1563static int nvme_open(struct block_device *bdev, fmode_t mode)
1564{
1565        return nvme_ns_open(bdev->bd_disk->private_data);
1566}
1567
1568static void nvme_release(struct gendisk *disk, fmode_t mode)
1569{
1570        nvme_ns_release(disk->private_data);
1571}
1572
1573int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1574{
1575        /* some standard values */
1576        geo->heads = 1 << 6;
1577        geo->sectors = 1 << 5;
1578        geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1579        return 0;
1580}
1581
1582#ifdef CONFIG_BLK_DEV_INTEGRITY
1583static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1584                                u32 max_integrity_segments)
1585{
1586        struct blk_integrity integrity = { };
1587
1588        switch (pi_type) {
1589        case NVME_NS_DPS_PI_TYPE3:
1590                integrity.profile = &t10_pi_type3_crc;
1591                integrity.tag_size = sizeof(u16) + sizeof(u32);
1592                integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1593                break;
1594        case NVME_NS_DPS_PI_TYPE1:
1595        case NVME_NS_DPS_PI_TYPE2:
1596                integrity.profile = &t10_pi_type1_crc;
1597                integrity.tag_size = sizeof(u16);
1598                integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1599                break;
1600        default:
1601                integrity.profile = NULL;
1602                break;
1603        }
1604        integrity.tuple_size = ms;
1605        blk_integrity_register(disk, &integrity);
1606        blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1607}
1608#else
1609static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1610                                u32 max_integrity_segments)
1611{
1612}
1613#endif /* CONFIG_BLK_DEV_INTEGRITY */
1614
1615static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1616{
1617        struct nvme_ctrl *ctrl = ns->ctrl;
1618        struct request_queue *queue = disk->queue;
1619        u32 size = queue_logical_block_size(queue);
1620
1621        if (ctrl->max_discard_sectors == 0) {
1622                blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1623                return;
1624        }
1625
1626        if (ctrl->nr_streams && ns->sws && ns->sgs)
1627                size *= ns->sws * ns->sgs;
1628
1629        BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1630                        NVME_DSM_MAX_RANGES);
1631
1632        queue->limits.discard_alignment = 0;
1633        queue->limits.discard_granularity = size;
1634
1635        /* If discard is already enabled, don't reset queue limits */
1636        if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1637                return;
1638
1639        blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1640        blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1641
1642        if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1643                blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1644}
1645
1646static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1647{
1648        return !uuid_is_null(&ids->uuid) ||
1649                memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1650                memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1651}
1652
1653static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1654{
1655        return uuid_equal(&a->uuid, &b->uuid) &&
1656                memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1657                memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1658                a->csi == b->csi;
1659}
1660
1661static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1662                                 u32 *phys_bs, u32 *io_opt)
1663{
1664        struct streams_directive_params s;
1665        int ret;
1666
1667        if (!ctrl->nr_streams)
1668                return 0;
1669
1670        ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1671        if (ret)
1672                return ret;
1673
1674        ns->sws = le32_to_cpu(s.sws);
1675        ns->sgs = le16_to_cpu(s.sgs);
1676
1677        if (ns->sws) {
1678                *phys_bs = ns->sws * (1 << ns->lba_shift);
1679                if (ns->sgs)
1680                        *io_opt = *phys_bs * ns->sgs;
1681        }
1682
1683        return 0;
1684}
1685
1686static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1687{
1688        struct nvme_ctrl *ctrl = ns->ctrl;
1689
1690        /*
1691         * The PI implementation requires the metadata size to be equal to the
1692         * t10 pi tuple size.
1693         */
1694        ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1695        if (ns->ms == sizeof(struct t10_pi_tuple))
1696                ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1697        else
1698                ns->pi_type = 0;
1699
1700        ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1701        if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1702                return 0;
1703        if (ctrl->ops->flags & NVME_F_FABRICS) {
1704                /*
1705                 * The NVMe over Fabrics specification only supports metadata as
1706                 * part of the extended data LBA.  We rely on HCA/HBA support to
1707                 * remap the separate metadata buffer from the block layer.
1708                 */
1709                if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1710                        return -EINVAL;
1711                if (ctrl->max_integrity_segments)
1712                        ns->features |=
1713                                (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1714        } else {
1715                /*
1716                 * For PCIe controllers, we can't easily remap the separate
1717                 * metadata buffer from the block layer and thus require a
1718                 * separate metadata buffer for block layer metadata/PI support.
1719                 * We allow extended LBAs for the passthrough interface, though.
1720                 */
1721                if (id->flbas & NVME_NS_FLBAS_META_EXT)
1722                        ns->features |= NVME_NS_EXT_LBAS;
1723                else
1724                        ns->features |= NVME_NS_METADATA_SUPPORTED;
1725        }
1726
1727        return 0;
1728}
1729
1730static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1731                struct request_queue *q)
1732{
1733        bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1734
1735        if (ctrl->max_hw_sectors) {
1736                u32 max_segments =
1737                        (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1738
1739                max_segments = min_not_zero(max_segments, ctrl->max_segments);
1740                blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1741                blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1742        }
1743        blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1744        blk_queue_dma_alignment(q, 7);
1745        blk_queue_write_cache(q, vwc, vwc);
1746}
1747
1748static void nvme_update_disk_info(struct gendisk *disk,
1749                struct nvme_ns *ns, struct nvme_id_ns *id)
1750{
1751        sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1752        unsigned short bs = 1 << ns->lba_shift;
1753        u32 atomic_bs, phys_bs, io_opt = 0;
1754
1755        /*
1756         * The block layer can't support LBA sizes larger than the page size
1757         * yet, so catch this early and don't allow block I/O.
1758         */
1759        if (ns->lba_shift > PAGE_SHIFT) {
1760                capacity = 0;
1761                bs = (1 << 9);
1762        }
1763
1764        blk_integrity_unregister(disk);
1765
1766        atomic_bs = phys_bs = bs;
1767        nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1768        if (id->nabo == 0) {
1769                /*
1770                 * Bit 1 indicates whether NAWUPF is defined for this namespace
1771                 * and whether it should be used instead of AWUPF. If NAWUPF ==
1772                 * 0 then AWUPF must be used instead.
1773                 */
1774                if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1775                        atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1776                else
1777                        atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1778        }
1779
1780        if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1781                /* NPWG = Namespace Preferred Write Granularity */
1782                phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1783                /* NOWS = Namespace Optimal Write Size */
1784                io_opt = bs * (1 + le16_to_cpu(id->nows));
1785        }
1786
1787        blk_queue_logical_block_size(disk->queue, bs);
1788        /*
1789         * Linux filesystems assume writing a single physical block is
1790         * an atomic operation. Hence limit the physical block size to the
1791         * value of the Atomic Write Unit Power Fail parameter.
1792         */
1793        blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1794        blk_queue_io_min(disk->queue, phys_bs);
1795        blk_queue_io_opt(disk->queue, io_opt);
1796
1797        /*
1798         * Register a metadata profile for PI, or the plain non-integrity NVMe
1799         * metadata masquerading as Type 0 if supported, otherwise reject block
1800         * I/O to namespaces with metadata except when the namespace supports
1801         * PI, as it can strip/insert in that case.
1802         */
1803        if (ns->ms) {
1804                if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1805                    (ns->features & NVME_NS_METADATA_SUPPORTED))
1806                        nvme_init_integrity(disk, ns->ms, ns->pi_type,
1807                                            ns->ctrl->max_integrity_segments);
1808                else if (!nvme_ns_has_pi(ns))
1809                        capacity = 0;
1810        }
1811
1812        set_capacity_and_notify(disk, capacity);
1813
1814        nvme_config_discard(disk, ns);
1815        blk_queue_max_write_zeroes_sectors(disk->queue,
1816                                           ns->ctrl->max_zeroes_sectors);
1817
1818        set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1819                test_bit(NVME_NS_FORCE_RO, &ns->flags));
1820}
1821
1822static inline bool nvme_first_scan(struct gendisk *disk)
1823{
1824        /* nvme_alloc_ns() scans the disk prior to adding it */
1825        return !(disk->flags & GENHD_FL_UP);
1826}
1827
1828static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1829{
1830        struct nvme_ctrl *ctrl = ns->ctrl;
1831        u32 iob;
1832
1833        if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1834            is_power_of_2(ctrl->max_hw_sectors))
1835                iob = ctrl->max_hw_sectors;
1836        else
1837                iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1838
1839        if (!iob)
1840                return;
1841
1842        if (!is_power_of_2(iob)) {
1843                if (nvme_first_scan(ns->disk))
1844                        pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1845                                ns->disk->disk_name, iob);
1846                return;
1847        }
1848
1849        if (blk_queue_is_zoned(ns->disk->queue)) {
1850                if (nvme_first_scan(ns->disk))
1851                        pr_warn("%s: ignoring zoned namespace IO boundary\n",
1852                                ns->disk->disk_name);
1853                return;
1854        }
1855
1856        blk_queue_chunk_sectors(ns->queue, iob);
1857}
1858
1859static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1860{
1861        unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1862        int ret;
1863
1864        blk_mq_freeze_queue(ns->disk->queue);
1865        ns->lba_shift = id->lbaf[lbaf].ds;
1866        nvme_set_queue_limits(ns->ctrl, ns->queue);
1867
1868        ret = nvme_configure_metadata(ns, id);
1869        if (ret)
1870                goto out_unfreeze;
1871        nvme_set_chunk_sectors(ns, id);
1872        nvme_update_disk_info(ns->disk, ns, id);
1873
1874        if (ns->head->ids.csi == NVME_CSI_ZNS) {
1875                ret = nvme_update_zone_info(ns, lbaf);
1876                if (ret)
1877                        goto out_unfreeze;
1878        }
1879
1880        blk_mq_unfreeze_queue(ns->disk->queue);
1881
1882        if (blk_queue_is_zoned(ns->queue)) {
1883                ret = nvme_revalidate_zones(ns);
1884                if (ret && !nvme_first_scan(ns->disk))
1885                        goto out;
1886        }
1887
1888        if (nvme_ns_head_multipath(ns->head)) {
1889                blk_mq_freeze_queue(ns->head->disk->queue);
1890                nvme_update_disk_info(ns->head->disk, ns, id);
1891                blk_stack_limits(&ns->head->disk->queue->limits,
1892                                 &ns->queue->limits, 0);
1893                blk_queue_update_readahead(ns->head->disk->queue);
1894                blk_mq_unfreeze_queue(ns->head->disk->queue);
1895        }
1896        return 0;
1897
1898out_unfreeze:
1899        blk_mq_unfreeze_queue(ns->disk->queue);
1900out:
1901        /*
1902         * If probing fails due an unsupported feature, hide the block device,
1903         * but still allow other access.
1904         */
1905        if (ret == -ENODEV) {
1906                ns->disk->flags |= GENHD_FL_HIDDEN;
1907                ret = 0;
1908        }
1909        return ret;
1910}
1911
1912static char nvme_pr_type(enum pr_type type)
1913{
1914        switch (type) {
1915        case PR_WRITE_EXCLUSIVE:
1916                return 1;
1917        case PR_EXCLUSIVE_ACCESS:
1918                return 2;
1919        case PR_WRITE_EXCLUSIVE_REG_ONLY:
1920                return 3;
1921        case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1922                return 4;
1923        case PR_WRITE_EXCLUSIVE_ALL_REGS:
1924                return 5;
1925        case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1926                return 6;
1927        default:
1928                return 0;
1929        }
1930};
1931
1932static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1933                struct nvme_command *c, u8 data[16])
1934{
1935        struct nvme_ns_head *head = bdev->bd_disk->private_data;
1936        int srcu_idx = srcu_read_lock(&head->srcu);
1937        struct nvme_ns *ns = nvme_find_path(head);
1938        int ret = -EWOULDBLOCK;
1939
1940        if (ns) {
1941                c->common.nsid = cpu_to_le32(ns->head->ns_id);
1942                ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1943        }
1944        srcu_read_unlock(&head->srcu, srcu_idx);
1945        return ret;
1946}
1947        
1948static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
1949                u8 data[16])
1950{
1951        c->common.nsid = cpu_to_le32(ns->head->ns_id);
1952        return nvme_submit_sync_cmd(ns->queue, c, data, 16);
1953}
1954
1955static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1956                                u64 key, u64 sa_key, u8 op)
1957{
1958        struct nvme_command c = { };
1959        u8 data[16] = { 0, };
1960
1961        put_unaligned_le64(key, &data[0]);
1962        put_unaligned_le64(sa_key, &data[8]);
1963
1964        c.common.opcode = op;
1965        c.common.cdw10 = cpu_to_le32(cdw10);
1966
1967        if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
1968            bdev->bd_disk->fops == &nvme_ns_head_ops)
1969                return nvme_send_ns_head_pr_command(bdev, &c, data);
1970        return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
1971}
1972
1973static int nvme_pr_register(struct block_device *bdev, u64 old,
1974                u64 new, unsigned flags)
1975{
1976        u32 cdw10;
1977
1978        if (flags & ~PR_FL_IGNORE_KEY)
1979                return -EOPNOTSUPP;
1980
1981        cdw10 = old ? 2 : 0;
1982        cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1983        cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1984        return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1985}
1986
1987static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1988                enum pr_type type, unsigned flags)
1989{
1990        u32 cdw10;
1991
1992        if (flags & ~PR_FL_IGNORE_KEY)
1993                return -EOPNOTSUPP;
1994
1995        cdw10 = nvme_pr_type(type) << 8;
1996        cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1997        return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1998}
1999
2000static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2001                enum pr_type type, bool abort)
2002{
2003        u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2004
2005        return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2006}
2007
2008static int nvme_pr_clear(struct block_device *bdev, u64 key)
2009{
2010        u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2011
2012        return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2013}
2014
2015static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2016{
2017        u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2018
2019        return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2020}
2021
2022const struct pr_ops nvme_pr_ops = {
2023        .pr_register    = nvme_pr_register,
2024        .pr_reserve     = nvme_pr_reserve,
2025        .pr_release     = nvme_pr_release,
2026        .pr_preempt     = nvme_pr_preempt,
2027        .pr_clear       = nvme_pr_clear,
2028};
2029
2030#ifdef CONFIG_BLK_SED_OPAL
2031int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2032                bool send)
2033{
2034        struct nvme_ctrl *ctrl = data;
2035        struct nvme_command cmd = { };
2036
2037        if (send)
2038                cmd.common.opcode = nvme_admin_security_send;
2039        else
2040                cmd.common.opcode = nvme_admin_security_recv;
2041        cmd.common.nsid = 0;
2042        cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2043        cmd.common.cdw11 = cpu_to_le32(len);
2044
2045        return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2046                        NVME_QID_ANY, 1, 0);
2047}
2048EXPORT_SYMBOL_GPL(nvme_sec_submit);
2049#endif /* CONFIG_BLK_SED_OPAL */
2050
2051#ifdef CONFIG_BLK_DEV_ZONED
2052static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2053                unsigned int nr_zones, report_zones_cb cb, void *data)
2054{
2055        return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2056                        data);
2057}
2058#else
2059#define nvme_report_zones       NULL
2060#endif /* CONFIG_BLK_DEV_ZONED */
2061
2062static const struct block_device_operations nvme_bdev_ops = {
2063        .owner          = THIS_MODULE,
2064        .ioctl          = nvme_ioctl,
2065        .open           = nvme_open,
2066        .release        = nvme_release,
2067        .getgeo         = nvme_getgeo,
2068        .report_zones   = nvme_report_zones,
2069        .pr_ops         = &nvme_pr_ops,
2070};
2071
2072static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2073{
2074        unsigned long timeout =
2075                ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2076        u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2077        int ret;
2078
2079        while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2080                if (csts == ~0)
2081                        return -ENODEV;
2082                if ((csts & NVME_CSTS_RDY) == bit)
2083                        break;
2084
2085                usleep_range(1000, 2000);
2086                if (fatal_signal_pending(current))
2087                        return -EINTR;
2088                if (time_after(jiffies, timeout)) {
2089                        dev_err(ctrl->device,
2090                                "Device not ready; aborting %s, CSTS=0x%x\n",
2091                                enabled ? "initialisation" : "reset", csts);
2092                        return -ENODEV;
2093                }
2094        }
2095
2096        return ret;
2097}
2098
2099/*
2100 * If the device has been passed off to us in an enabled state, just clear
2101 * the enabled bit.  The spec says we should set the 'shutdown notification
2102 * bits', but doing so may cause the device to complete commands to the
2103 * admin queue ... and we don't know what memory that might be pointing at!
2104 */
2105int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2106{
2107        int ret;
2108
2109        ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2110        ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2111
2112        ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2113        if (ret)
2114                return ret;
2115
2116        if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2117                msleep(NVME_QUIRK_DELAY_AMOUNT);
2118
2119        return nvme_wait_ready(ctrl, ctrl->cap, false);
2120}
2121EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2122
2123int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2124{
2125        unsigned dev_page_min;
2126        int ret;
2127
2128        ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2129        if (ret) {
2130                dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2131                return ret;
2132        }
2133        dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2134
2135        if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2136                dev_err(ctrl->device,
2137                        "Minimum device page size %u too large for host (%u)\n",
2138                        1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2139                return -ENODEV;
2140        }
2141
2142        if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2143                ctrl->ctrl_config = NVME_CC_CSS_CSI;
2144        else
2145                ctrl->ctrl_config = NVME_CC_CSS_NVM;
2146        ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2147        ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2148        ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2149        ctrl->ctrl_config |= NVME_CC_ENABLE;
2150
2151        ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2152        if (ret)
2153                return ret;
2154        return nvme_wait_ready(ctrl, ctrl->cap, true);
2155}
2156EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2157
2158int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2159{
2160        unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2161        u32 csts;
2162        int ret;
2163
2164        ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2165        ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2166
2167        ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2168        if (ret)
2169                return ret;
2170
2171        while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2172                if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2173                        break;
2174
2175                msleep(100);
2176                if (fatal_signal_pending(current))
2177                        return -EINTR;
2178                if (time_after(jiffies, timeout)) {
2179                        dev_err(ctrl->device,
2180                                "Device shutdown incomplete; abort shutdown\n");
2181                        return -ENODEV;
2182                }
2183        }
2184
2185        return ret;
2186}
2187EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2188
2189static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2190{
2191        __le64 ts;
2192        int ret;
2193
2194        if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2195                return 0;
2196
2197        ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2198        ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2199                        NULL);
2200        if (ret)
2201                dev_warn_once(ctrl->device,
2202                        "could not set timestamp (%d)\n", ret);
2203        return ret;
2204}
2205
2206static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2207{
2208        struct nvme_feat_host_behavior *host;
2209        int ret;
2210
2211        /* Don't bother enabling the feature if retry delay is not reported */
2212        if (!ctrl->crdt[0])
2213                return 0;
2214
2215        host = kzalloc(sizeof(*host), GFP_KERNEL);
2216        if (!host)
2217                return 0;
2218
2219        host->acre = NVME_ENABLE_ACRE;
2220        ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2221                                host, sizeof(*host), NULL);
2222        kfree(host);
2223        return ret;
2224}
2225
2226/*
2227 * The function checks whether the given total (exlat + enlat) latency of
2228 * a power state allows the latter to be used as an APST transition target.
2229 * It does so by comparing the latency to the primary and secondary latency
2230 * tolerances defined by module params. If there's a match, the corresponding
2231 * timeout value is returned and the matching tolerance index (1 or 2) is
2232 * reported.
2233 */
2234static bool nvme_apst_get_transition_time(u64 total_latency,
2235                u64 *transition_time, unsigned *last_index)
2236{
2237        if (total_latency <= apst_primary_latency_tol_us) {
2238                if (*last_index == 1)
2239                        return false;
2240                *last_index = 1;
2241                *transition_time = apst_primary_timeout_ms;
2242                return true;
2243        }
2244        if (apst_secondary_timeout_ms &&
2245                total_latency <= apst_secondary_latency_tol_us) {
2246                if (*last_index <= 2)
2247                        return false;
2248                *last_index = 2;
2249                *transition_time = apst_secondary_timeout_ms;
2250                return true;
2251        }
2252        return false;
2253}
2254
2255/*
2256 * APST (Autonomous Power State Transition) lets us program a table of power
2257 * state transitions that the controller will perform automatically.
2258 *
2259 * Depending on module params, one of the two supported techniques will be used:
2260 *
2261 * - If the parameters provide explicit timeouts and tolerances, they will be
2262 *   used to build a table with up to 2 non-operational states to transition to.
2263 *   The default parameter values were selected based on the values used by
2264 *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2265 *   regeneration of the APST table in the event of switching between external
2266 *   and battery power, the timeouts and tolerances reflect a compromise
2267 *   between values used by Microsoft for AC and battery scenarios.
2268 * - If not, we'll configure the table with a simple heuristic: we are willing
2269 *   to spend at most 2% of the time transitioning between power states.
2270 *   Therefore, when running in any given state, we will enter the next
2271 *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2272 *   microseconds, as long as that state's exit latency is under the requested
2273 *   maximum latency.
2274 *
2275 * We will not autonomously enter any non-operational state for which the total
2276 * latency exceeds ps_max_latency_us.
2277 *
2278 * Users can set ps_max_latency_us to zero to turn off APST.
2279 */
2280static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2281{
2282        struct nvme_feat_auto_pst *table;
2283        unsigned apste = 0;
2284        u64 max_lat_us = 0;
2285        __le64 target = 0;
2286        int max_ps = -1;
2287        int state;
2288        int ret;
2289        unsigned last_lt_index = UINT_MAX;
2290
2291        /*
2292         * If APST isn't supported or if we haven't been initialized yet,
2293         * then don't do anything.
2294         */
2295        if (!ctrl->apsta)
2296                return 0;
2297
2298        if (ctrl->npss > 31) {
2299                dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2300                return 0;
2301        }
2302
2303        table = kzalloc(sizeof(*table), GFP_KERNEL);
2304        if (!table)
2305                return 0;
2306
2307        if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2308                /* Turn off APST. */
2309                dev_dbg(ctrl->device, "APST disabled\n");
2310                goto done;
2311        }
2312
2313        /*
2314         * Walk through all states from lowest- to highest-power.
2315         * According to the spec, lower-numbered states use more power.  NPSS,
2316         * despite the name, is the index of the lowest-power state, not the
2317         * number of states.
2318         */
2319        for (state = (int)ctrl->npss; state >= 0; state--) {
2320                u64 total_latency_us, exit_latency_us, transition_ms;
2321
2322                if (target)
2323                        table->entries[state] = target;
2324
2325                /*
2326                 * Don't allow transitions to the deepest state if it's quirked
2327                 * off.
2328                 */
2329                if (state == ctrl->npss &&
2330                    (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2331                        continue;
2332
2333                /*
2334                 * Is this state a useful non-operational state for higher-power
2335                 * states to autonomously transition to?
2336                 */
2337                if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2338                        continue;
2339
2340                exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2341                if (exit_latency_us > ctrl->ps_max_latency_us)
2342                        continue;
2343
2344                total_latency_us = exit_latency_us +
2345                        le32_to_cpu(ctrl->psd[state].entry_lat);
2346
2347                /*
2348                 * This state is good. It can be used as the APST idle target
2349                 * for higher power states.
2350                 */
2351                if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2352                        if (!nvme_apst_get_transition_time(total_latency_us,
2353                                        &transition_ms, &last_lt_index))
2354                                continue;
2355                } else {
2356                        transition_ms = total_latency_us + 19;
2357                        do_div(transition_ms, 20);
2358                        if (transition_ms > (1 << 24) - 1)
2359                                transition_ms = (1 << 24) - 1;
2360                }
2361
2362                target = cpu_to_le64((state << 3) | (transition_ms << 8));
2363                if (max_ps == -1)
2364                        max_ps = state;
2365                if (total_latency_us > max_lat_us)
2366                        max_lat_us = total_latency_us;
2367        }
2368
2369        if (max_ps == -1)
2370                dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2371        else
2372                dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2373                        max_ps, max_lat_us, (int)sizeof(*table), table);
2374        apste = 1;
2375
2376done:
2377        ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2378                                table, sizeof(*table), NULL);
2379        if (ret)
2380                dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2381        kfree(table);
2382        return ret;
2383}
2384
2385static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2386{
2387        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2388        u64 latency;
2389
2390        switch (val) {
2391        case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2392        case PM_QOS_LATENCY_ANY:
2393                latency = U64_MAX;
2394                break;
2395
2396        default:
2397                latency = val;
2398        }
2399
2400        if (ctrl->ps_max_latency_us != latency) {
2401                ctrl->ps_max_latency_us = latency;
2402                if (ctrl->state == NVME_CTRL_LIVE)
2403                        nvme_configure_apst(ctrl);
2404        }
2405}
2406
2407struct nvme_core_quirk_entry {
2408        /*
2409         * NVMe model and firmware strings are padded with spaces.  For
2410         * simplicity, strings in the quirk table are padded with NULLs
2411         * instead.
2412         */
2413        u16 vid;
2414        const char *mn;
2415        const char *fr;
2416        unsigned long quirks;
2417};
2418
2419static const struct nvme_core_quirk_entry core_quirks[] = {
2420        {
2421                /*
2422                 * This Toshiba device seems to die using any APST states.  See:
2423                 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2424                 */
2425                .vid = 0x1179,
2426                .mn = "THNSF5256GPUK TOSHIBA",
2427                .quirks = NVME_QUIRK_NO_APST,
2428        },
2429        {
2430                /*
2431                 * This LiteON CL1-3D*-Q11 firmware version has a race
2432                 * condition associated with actions related to suspend to idle
2433                 * LiteON has resolved the problem in future firmware
2434                 */
2435                .vid = 0x14a4,
2436                .fr = "22301111",
2437                .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2438        }
2439};
2440
2441/* match is null-terminated but idstr is space-padded. */
2442static bool string_matches(const char *idstr, const char *match, size_t len)
2443{
2444        size_t matchlen;
2445
2446        if (!match)
2447                return true;
2448
2449        matchlen = strlen(match);
2450        WARN_ON_ONCE(matchlen > len);
2451
2452        if (memcmp(idstr, match, matchlen))
2453                return false;
2454
2455        for (; matchlen < len; matchlen++)
2456                if (idstr[matchlen] != ' ')
2457                        return false;
2458
2459        return true;
2460}
2461
2462static bool quirk_matches(const struct nvme_id_ctrl *id,
2463                          const struct nvme_core_quirk_entry *q)
2464{
2465        return q->vid == le16_to_cpu(id->vid) &&
2466                string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2467                string_matches(id->fr, q->fr, sizeof(id->fr));
2468}
2469
2470static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2471                struct nvme_id_ctrl *id)
2472{
2473        size_t nqnlen;
2474        int off;
2475
2476        if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2477                nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2478                if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2479                        strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2480                        return;
2481                }
2482
2483                if (ctrl->vs >= NVME_VS(1, 2, 1))
2484                        dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2485        }
2486
2487        /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2488        off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2489                        "nqn.2014.08.org.nvmexpress:%04x%04x",
2490                        le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2491        memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2492        off += sizeof(id->sn);
2493        memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2494        off += sizeof(id->mn);
2495        memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2496}
2497
2498static void nvme_release_subsystem(struct device *dev)
2499{
2500        struct nvme_subsystem *subsys =
2501                container_of(dev, struct nvme_subsystem, dev);
2502
2503        if (subsys->instance >= 0)
2504                ida_simple_remove(&nvme_instance_ida, subsys->instance);
2505        kfree(subsys);
2506}
2507
2508static void nvme_destroy_subsystem(struct kref *ref)
2509{
2510        struct nvme_subsystem *subsys =
2511                        container_of(ref, struct nvme_subsystem, ref);
2512
2513        mutex_lock(&nvme_subsystems_lock);
2514        list_del(&subsys->entry);
2515        mutex_unlock(&nvme_subsystems_lock);
2516
2517        ida_destroy(&subsys->ns_ida);
2518        device_del(&subsys->dev);
2519        put_device(&subsys->dev);
2520}
2521
2522static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2523{
2524        kref_put(&subsys->ref, nvme_destroy_subsystem);
2525}
2526
2527static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2528{
2529        struct nvme_subsystem *subsys;
2530
2531        lockdep_assert_held(&nvme_subsystems_lock);
2532
2533        /*
2534         * Fail matches for discovery subsystems. This results
2535         * in each discovery controller bound to a unique subsystem.
2536         * This avoids issues with validating controller values
2537         * that can only be true when there is a single unique subsystem.
2538         * There may be multiple and completely independent entities
2539         * that provide discovery controllers.
2540         */
2541        if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2542                return NULL;
2543
2544        list_for_each_entry(subsys, &nvme_subsystems, entry) {
2545                if (strcmp(subsys->subnqn, subsysnqn))
2546                        continue;
2547                if (!kref_get_unless_zero(&subsys->ref))
2548                        continue;
2549                return subsys;
2550        }
2551
2552        return NULL;
2553}
2554
2555#define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2556        struct device_attribute subsys_attr_##_name = \
2557                __ATTR(_name, _mode, _show, NULL)
2558
2559static ssize_t nvme_subsys_show_nqn(struct device *dev,
2560                                    struct device_attribute *attr,
2561                                    char *buf)
2562{
2563        struct nvme_subsystem *subsys =
2564                container_of(dev, struct nvme_subsystem, dev);
2565
2566        return sysfs_emit(buf, "%s\n", subsys->subnqn);
2567}
2568static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2569
2570#define nvme_subsys_show_str_function(field)                            \
2571static ssize_t subsys_##field##_show(struct device *dev,                \
2572                            struct device_attribute *attr, char *buf)   \
2573{                                                                       \
2574        struct nvme_subsystem *subsys =                                 \
2575                container_of(dev, struct nvme_subsystem, dev);          \
2576        return sysfs_emit(buf, "%.*s\n",                                \
2577                           (int)sizeof(subsys->field), subsys->field);  \
2578}                                                                       \
2579static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2580
2581nvme_subsys_show_str_function(model);
2582nvme_subsys_show_str_function(serial);
2583nvme_subsys_show_str_function(firmware_rev);
2584
2585static struct attribute *nvme_subsys_attrs[] = {
2586        &subsys_attr_model.attr,
2587        &subsys_attr_serial.attr,
2588        &subsys_attr_firmware_rev.attr,
2589        &subsys_attr_subsysnqn.attr,
2590#ifdef CONFIG_NVME_MULTIPATH
2591        &subsys_attr_iopolicy.attr,
2592#endif
2593        NULL,
2594};
2595
2596static const struct attribute_group nvme_subsys_attrs_group = {
2597        .attrs = nvme_subsys_attrs,
2598};
2599
2600static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2601        &nvme_subsys_attrs_group,
2602        NULL,
2603};
2604
2605static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2606{
2607        return ctrl->opts && ctrl->opts->discovery_nqn;
2608}
2609
2610static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2611                struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2612{
2613        struct nvme_ctrl *tmp;
2614
2615        lockdep_assert_held(&nvme_subsystems_lock);
2616
2617        list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2618                if (nvme_state_terminal(tmp))
2619                        continue;
2620
2621                if (tmp->cntlid == ctrl->cntlid) {
2622                        dev_err(ctrl->device,
2623                                "Duplicate cntlid %u with %s, rejecting\n",
2624                                ctrl->cntlid, dev_name(tmp->device));
2625                        return false;
2626                }
2627
2628                if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2629                    nvme_discovery_ctrl(ctrl))
2630                        continue;
2631
2632                dev_err(ctrl->device,
2633                        "Subsystem does not support multiple controllers\n");
2634                return false;
2635        }
2636
2637        return true;
2638}
2639
2640static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2641{
2642        struct nvme_subsystem *subsys, *found;
2643        int ret;
2644
2645        subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2646        if (!subsys)
2647                return -ENOMEM;
2648
2649        subsys->instance = -1;
2650        mutex_init(&subsys->lock);
2651        kref_init(&subsys->ref);
2652        INIT_LIST_HEAD(&subsys->ctrls);
2653        INIT_LIST_HEAD(&subsys->nsheads);
2654        nvme_init_subnqn(subsys, ctrl, id);
2655        memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2656        memcpy(subsys->model, id->mn, sizeof(subsys->model));
2657        memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2658        subsys->vendor_id = le16_to_cpu(id->vid);
2659        subsys->cmic = id->cmic;
2660        subsys->awupf = le16_to_cpu(id->awupf);
2661#ifdef CONFIG_NVME_MULTIPATH
2662        subsys->iopolicy = NVME_IOPOLICY_NUMA;
2663#endif
2664
2665        subsys->dev.class = nvme_subsys_class;
2666        subsys->dev.release = nvme_release_subsystem;
2667        subsys->dev.groups = nvme_subsys_attrs_groups;
2668        dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2669        device_initialize(&subsys->dev);
2670
2671        mutex_lock(&nvme_subsystems_lock);
2672        found = __nvme_find_get_subsystem(subsys->subnqn);
2673        if (found) {
2674                put_device(&subsys->dev);
2675                subsys = found;
2676
2677                if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2678                        ret = -EINVAL;
2679                        goto out_put_subsystem;
2680                }
2681        } else {
2682                ret = device_add(&subsys->dev);
2683                if (ret) {
2684                        dev_err(ctrl->device,
2685                                "failed to register subsystem device.\n");
2686                        put_device(&subsys->dev);
2687                        goto out_unlock;
2688                }
2689                ida_init(&subsys->ns_ida);
2690                list_add_tail(&subsys->entry, &nvme_subsystems);
2691        }
2692
2693        ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2694                                dev_name(ctrl->device));
2695        if (ret) {
2696                dev_err(ctrl->device,
2697                        "failed to create sysfs link from subsystem.\n");
2698                goto out_put_subsystem;
2699        }
2700
2701        if (!found)
2702                subsys->instance = ctrl->instance;
2703        ctrl->subsys = subsys;
2704        list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2705        mutex_unlock(&nvme_subsystems_lock);
2706        return 0;
2707
2708out_put_subsystem:
2709        nvme_put_subsystem(subsys);
2710out_unlock:
2711        mutex_unlock(&nvme_subsystems_lock);
2712        return ret;
2713}
2714
2715int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2716                void *log, size_t size, u64 offset)
2717{
2718        struct nvme_command c = { };
2719        u32 dwlen = nvme_bytes_to_numd(size);
2720
2721        c.get_log_page.opcode = nvme_admin_get_log_page;
2722        c.get_log_page.nsid = cpu_to_le32(nsid);
2723        c.get_log_page.lid = log_page;
2724        c.get_log_page.lsp = lsp;
2725        c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2726        c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2727        c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2728        c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2729        c.get_log_page.csi = csi;
2730
2731        return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2732}
2733
2734static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2735                                struct nvme_effects_log **log)
2736{
2737        struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2738        int ret;
2739
2740        if (cel)
2741                goto out;
2742
2743        cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2744        if (!cel)
2745                return -ENOMEM;
2746
2747        ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2748                        cel, sizeof(*cel), 0);
2749        if (ret) {
2750                kfree(cel);
2751                return ret;
2752        }
2753
2754        xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2755out:
2756        *log = cel;
2757        return 0;
2758}
2759
2760static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2761{
2762        u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2763
2764        if (check_shl_overflow(1U, units + page_shift - 9, &val))
2765                return UINT_MAX;
2766        return val;
2767}
2768
2769static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2770{
2771        struct nvme_command c = { };
2772        struct nvme_id_ctrl_nvm *id;
2773        int ret;
2774
2775        if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2776                ctrl->max_discard_sectors = UINT_MAX;
2777                ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2778        } else {
2779                ctrl->max_discard_sectors = 0;
2780                ctrl->max_discard_segments = 0;
2781        }
2782
2783        /*
2784         * Even though NVMe spec explicitly states that MDTS is not applicable
2785         * to the write-zeroes, we are cautious and limit the size to the
2786         * controllers max_hw_sectors value, which is based on the MDTS field
2787         * and possibly other limiting factors.
2788         */
2789        if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2790            !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2791                ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2792        else
2793                ctrl->max_zeroes_sectors = 0;
2794
2795        if (nvme_ctrl_limited_cns(ctrl))
2796                return 0;
2797
2798        id = kzalloc(sizeof(*id), GFP_KERNEL);
2799        if (!id)
2800                return 0;
2801
2802        c.identify.opcode = nvme_admin_identify;
2803        c.identify.cns = NVME_ID_CNS_CS_CTRL;
2804        c.identify.csi = NVME_CSI_NVM;
2805
2806        ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2807        if (ret)
2808                goto free_data;
2809
2810        if (id->dmrl)
2811                ctrl->max_discard_segments = id->dmrl;
2812        if (id->dmrsl)
2813                ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2814        if (id->wzsl)
2815                ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2816
2817free_data:
2818        kfree(id);
2819        return ret;
2820}
2821
2822static int nvme_init_identify(struct nvme_ctrl *ctrl)
2823{
2824        struct nvme_id_ctrl *id;
2825        u32 max_hw_sectors;
2826        bool prev_apst_enabled;
2827        int ret;
2828
2829        ret = nvme_identify_ctrl(ctrl, &id);
2830        if (ret) {
2831                dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2832                return -EIO;
2833        }
2834
2835        if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2836                ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2837                if (ret < 0)
2838                        goto out_free;
2839        }
2840
2841        if (!(ctrl->ops->flags & NVME_F_FABRICS))
2842                ctrl->cntlid = le16_to_cpu(id->cntlid);
2843
2844        if (!ctrl->identified) {
2845                unsigned int i;
2846
2847                ret = nvme_init_subsystem(ctrl, id);
2848                if (ret)
2849                        goto out_free;
2850
2851                /*
2852                 * Check for quirks.  Quirk can depend on firmware version,
2853                 * so, in principle, the set of quirks present can change
2854                 * across a reset.  As a possible future enhancement, we
2855                 * could re-scan for quirks every time we reinitialize
2856                 * the device, but we'd have to make sure that the driver
2857                 * behaves intelligently if the quirks change.
2858                 */
2859                for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2860                        if (quirk_matches(id, &core_quirks[i]))
2861                                ctrl->quirks |= core_quirks[i].quirks;
2862                }
2863        }
2864
2865        if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2866                dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2867                ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2868        }
2869
2870        ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2871        ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2872        ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2873
2874        ctrl->oacs = le16_to_cpu(id->oacs);
2875        ctrl->oncs = le16_to_cpu(id->oncs);
2876        ctrl->mtfa = le16_to_cpu(id->mtfa);
2877        ctrl->oaes = le32_to_cpu(id->oaes);
2878        ctrl->wctemp = le16_to_cpu(id->wctemp);
2879        ctrl->cctemp = le16_to_cpu(id->cctemp);
2880
2881        atomic_set(&ctrl->abort_limit, id->acl + 1);
2882        ctrl->vwc = id->vwc;
2883        if (id->mdts)
2884                max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2885        else
2886                max_hw_sectors = UINT_MAX;
2887        ctrl->max_hw_sectors =
2888                min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2889
2890        nvme_set_queue_limits(ctrl, ctrl->admin_q);
2891        ctrl->sgls = le32_to_cpu(id->sgls);
2892        ctrl->kas = le16_to_cpu(id->kas);
2893        ctrl->max_namespaces = le32_to_cpu(id->mnan);
2894        ctrl->ctratt = le32_to_cpu(id->ctratt);
2895
2896        if (id->rtd3e) {
2897                /* us -> s */
2898                u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2899
2900                ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2901                                                 shutdown_timeout, 60);
2902
2903                if (ctrl->shutdown_timeout != shutdown_timeout)
2904                        dev_info(ctrl->device,
2905                                 "Shutdown timeout set to %u seconds\n",
2906                                 ctrl->shutdown_timeout);
2907        } else
2908                ctrl->shutdown_timeout = shutdown_timeout;
2909
2910        ctrl->npss = id->npss;
2911        ctrl->apsta = id->apsta;
2912        prev_apst_enabled = ctrl->apst_enabled;
2913        if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2914                if (force_apst && id->apsta) {
2915                        dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2916                        ctrl->apst_enabled = true;
2917                } else {
2918                        ctrl->apst_enabled = false;
2919                }
2920        } else {
2921                ctrl->apst_enabled = id->apsta;
2922        }
2923        memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2924
2925        if (ctrl->ops->flags & NVME_F_FABRICS) {
2926                ctrl->icdoff = le16_to_cpu(id->icdoff);
2927                ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2928                ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2929                ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2930
2931                /*
2932                 * In fabrics we need to verify the cntlid matches the
2933                 * admin connect
2934                 */
2935                if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2936                        dev_err(ctrl->device,
2937                                "Mismatching cntlid: Connect %u vs Identify "
2938                                "%u, rejecting\n",
2939                                ctrl->cntlid, le16_to_cpu(id->cntlid));
2940                        ret = -EINVAL;
2941                        goto out_free;
2942                }
2943
2944                if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
2945                        dev_err(ctrl->device,
2946                                "keep-alive support is mandatory for fabrics\n");
2947                        ret = -EINVAL;
2948                        goto out_free;
2949                }
2950        } else {
2951                ctrl->hmpre = le32_to_cpu(id->hmpre);
2952                ctrl->hmmin = le32_to_cpu(id->hmmin);
2953                ctrl->hmminds = le32_to_cpu(id->hmminds);
2954                ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2955        }
2956
2957        ret = nvme_mpath_init_identify(ctrl, id);
2958        if (ret < 0)
2959                goto out_free;
2960
2961        if (ctrl->apst_enabled && !prev_apst_enabled)
2962                dev_pm_qos_expose_latency_tolerance(ctrl->device);
2963        else if (!ctrl->apst_enabled && prev_apst_enabled)
2964                dev_pm_qos_hide_latency_tolerance(ctrl->device);
2965
2966out_free:
2967        kfree(id);
2968        return ret;
2969}
2970
2971/*
2972 * Initialize the cached copies of the Identify data and various controller
2973 * register in our nvme_ctrl structure.  This should be called as soon as
2974 * the admin queue is fully up and running.
2975 */
2976int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
2977{
2978        int ret;
2979
2980        ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2981        if (ret) {
2982                dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2983                return ret;
2984        }
2985
2986        ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2987
2988        if (ctrl->vs >= NVME_VS(1, 1, 0))
2989                ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2990
2991        ret = nvme_init_identify(ctrl);
2992        if (ret)
2993                return ret;
2994
2995        ret = nvme_init_non_mdts_limits(ctrl);
2996        if (ret < 0)
2997                return ret;
2998
2999        ret = nvme_configure_apst(ctrl);
3000        if (ret < 0)
3001                return ret;
3002
3003        ret = nvme_configure_timestamp(ctrl);
3004        if (ret < 0)
3005                return ret;
3006
3007        ret = nvme_configure_directives(ctrl);
3008        if (ret < 0)
3009                return ret;
3010
3011        ret = nvme_configure_acre(ctrl);
3012        if (ret < 0)
3013                return ret;
3014
3015        if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3016                ret = nvme_hwmon_init(ctrl);
3017                if (ret < 0)
3018                        return ret;
3019        }
3020
3021        ctrl->identified = true;
3022
3023        return 0;
3024}
3025EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3026
3027static int nvme_dev_open(struct inode *inode, struct file *file)
3028{
3029        struct nvme_ctrl *ctrl =
3030                container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3031
3032        switch (ctrl->state) {
3033        case NVME_CTRL_LIVE:
3034                break;
3035        default:
3036                return -EWOULDBLOCK;
3037        }
3038
3039        nvme_get_ctrl(ctrl);
3040        if (!try_module_get(ctrl->ops->module)) {
3041                nvme_put_ctrl(ctrl);
3042                return -EINVAL;
3043        }
3044
3045        file->private_data = ctrl;
3046        return 0;
3047}
3048
3049static int nvme_dev_release(struct inode *inode, struct file *file)
3050{
3051        struct nvme_ctrl *ctrl =
3052                container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3053
3054        module_put(ctrl->ops->module);
3055        nvme_put_ctrl(ctrl);
3056        return 0;
3057}
3058
3059static const struct file_operations nvme_dev_fops = {
3060        .owner          = THIS_MODULE,
3061        .open           = nvme_dev_open,
3062        .release        = nvme_dev_release,
3063        .unlocked_ioctl = nvme_dev_ioctl,
3064        .compat_ioctl   = compat_ptr_ioctl,
3065};
3066
3067static ssize_t nvme_sysfs_reset(struct device *dev,
3068                                struct device_attribute *attr, const char *buf,
3069                                size_t count)
3070{
3071        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3072        int ret;
3073
3074        ret = nvme_reset_ctrl_sync(ctrl);
3075        if (ret < 0)
3076                return ret;
3077        return count;
3078}
3079static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3080
3081static ssize_t nvme_sysfs_rescan(struct device *dev,
3082                                struct device_attribute *attr, const char *buf,
3083                                size_t count)
3084{
3085        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3086
3087        nvme_queue_scan(ctrl);
3088        return count;
3089}
3090static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3091
3092static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3093{
3094        struct gendisk *disk = dev_to_disk(dev);
3095
3096        if (disk->fops == &nvme_bdev_ops)
3097                return nvme_get_ns_from_dev(dev)->head;
3098        else
3099                return disk->private_data;
3100}
3101
3102static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3103                char *buf)
3104{
3105        struct nvme_ns_head *head = dev_to_ns_head(dev);
3106        struct nvme_ns_ids *ids = &head->ids;
3107        struct nvme_subsystem *subsys = head->subsys;
3108        int serial_len = sizeof(subsys->serial);
3109        int model_len = sizeof(subsys->model);
3110
3111        if (!uuid_is_null(&ids->uuid))
3112                return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3113
3114        if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3115                return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3116
3117        if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3118                return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3119
3120        while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3121                                  subsys->serial[serial_len - 1] == '\0'))
3122                serial_len--;
3123        while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3124                                 subsys->model[model_len - 1] == '\0'))
3125                model_len--;
3126
3127        return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3128                serial_len, subsys->serial, model_len, subsys->model,
3129                head->ns_id);
3130}
3131static DEVICE_ATTR_RO(wwid);
3132
3133static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3134                char *buf)
3135{
3136        return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3137}
3138static DEVICE_ATTR_RO(nguid);
3139
3140static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3141                char *buf)
3142{
3143        struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3144
3145        /* For backward compatibility expose the NGUID to userspace if
3146         * we have no UUID set
3147         */
3148        if (uuid_is_null(&ids->uuid)) {
3149                printk_ratelimited(KERN_WARNING
3150                                   "No UUID available providing old NGUID\n");
3151                return sysfs_emit(buf, "%pU\n", ids->nguid);
3152        }
3153        return sysfs_emit(buf, "%pU\n", &ids->uuid);
3154}
3155static DEVICE_ATTR_RO(uuid);
3156
3157static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3158                char *buf)
3159{
3160        return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3161}
3162static DEVICE_ATTR_RO(eui);
3163
3164static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3165                char *buf)
3166{
3167        return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3168}
3169static DEVICE_ATTR_RO(nsid);
3170
3171static struct attribute *nvme_ns_id_attrs[] = {
3172        &dev_attr_wwid.attr,
3173        &dev_attr_uuid.attr,
3174        &dev_attr_nguid.attr,
3175        &dev_attr_eui.attr,
3176        &dev_attr_nsid.attr,
3177#ifdef CONFIG_NVME_MULTIPATH
3178        &dev_attr_ana_grpid.attr,
3179        &dev_attr_ana_state.attr,
3180#endif
3181        NULL,
3182};
3183
3184static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3185                struct attribute *a, int n)
3186{
3187        struct device *dev = container_of(kobj, struct device, kobj);
3188        struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3189
3190        if (a == &dev_attr_uuid.attr) {
3191                if (uuid_is_null(&ids->uuid) &&
3192                    !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3193                        return 0;
3194        }
3195        if (a == &dev_attr_nguid.attr) {
3196                if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3197                        return 0;
3198        }
3199        if (a == &dev_attr_eui.attr) {
3200                if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3201                        return 0;
3202        }
3203#ifdef CONFIG_NVME_MULTIPATH
3204        if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3205                if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3206                        return 0;
3207                if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3208                        return 0;
3209        }
3210#endif
3211        return a->mode;
3212}
3213
3214static const struct attribute_group nvme_ns_id_attr_group = {
3215        .attrs          = nvme_ns_id_attrs,
3216        .is_visible     = nvme_ns_id_attrs_are_visible,
3217};
3218
3219const struct attribute_group *nvme_ns_id_attr_groups[] = {
3220        &nvme_ns_id_attr_group,
3221#ifdef CONFIG_NVM
3222        &nvme_nvm_attr_group,
3223#endif
3224        NULL,
3225};
3226
3227#define nvme_show_str_function(field)                                           \
3228static ssize_t  field##_show(struct device *dev,                                \
3229                            struct device_attribute *attr, char *buf)           \
3230{                                                                               \
3231        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3232        return sysfs_emit(buf, "%.*s\n",                                        \
3233                (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3234}                                                                               \
3235static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3236
3237nvme_show_str_function(model);
3238nvme_show_str_function(serial);
3239nvme_show_str_function(firmware_rev);
3240
3241#define nvme_show_int_function(field)                                           \
3242static ssize_t  field##_show(struct device *dev,                                \
3243                            struct device_attribute *attr, char *buf)           \
3244{                                                                               \
3245        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3246        return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3247}                                                                               \
3248static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3249
3250nvme_show_int_function(cntlid);
3251nvme_show_int_function(numa_node);
3252nvme_show_int_function(queue_count);
3253nvme_show_int_function(sqsize);
3254nvme_show_int_function(kato);
3255
3256static ssize_t nvme_sysfs_delete(struct device *dev,
3257                                struct device_attribute *attr, const char *buf,
3258                                size_t count)
3259{
3260        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3261
3262        if (device_remove_file_self(dev, attr))
3263                nvme_delete_ctrl_sync(ctrl);
3264        return count;
3265}
3266static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3267
3268static ssize_t nvme_sysfs_show_transport(struct device *dev,
3269                                         struct device_attribute *attr,
3270                                         char *buf)
3271{
3272        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3273
3274        return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3275}
3276static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3277
3278static ssize_t nvme_sysfs_show_state(struct device *dev,
3279                                     struct device_attribute *attr,
3280                                     char *buf)
3281{
3282        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3283        static const char *const state_name[] = {
3284                [NVME_CTRL_NEW]         = "new",
3285                [NVME_CTRL_LIVE]        = "live",
3286                [NVME_CTRL_RESETTING]   = "resetting",
3287                [NVME_CTRL_CONNECTING]  = "connecting",
3288                [NVME_CTRL_DELETING]    = "deleting",
3289                [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3290                [NVME_CTRL_DEAD]        = "dead",
3291        };
3292
3293        if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3294            state_name[ctrl->state])
3295                return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3296
3297        return sysfs_emit(buf, "unknown state\n");
3298}
3299
3300static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3301
3302static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3303                                         struct device_attribute *attr,
3304                                         char *buf)
3305{
3306        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3307
3308        return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3309}
3310static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3311
3312static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3313                                        struct device_attribute *attr,
3314                                        char *buf)
3315{
3316        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3317
3318        return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3319}
3320static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3321
3322static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3323                                        struct device_attribute *attr,
3324                                        char *buf)
3325{
3326        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3327
3328        return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3329}
3330static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3331
3332static ssize_t nvme_sysfs_show_address(struct device *dev,
3333                                         struct device_attribute *attr,
3334                                         char *buf)
3335{
3336        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3337
3338        return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3339}
3340static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3341
3342static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3343                struct device_attribute *attr, char *buf)
3344{
3345        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3346        struct nvmf_ctrl_options *opts = ctrl->opts;
3347
3348        if (ctrl->opts->max_reconnects == -1)
3349                return sysfs_emit(buf, "off\n");
3350        return sysfs_emit(buf, "%d\n",
3351                          opts->max_reconnects * opts->reconnect_delay);
3352}
3353
3354static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3355                struct device_attribute *attr, const char *buf, size_t count)
3356{
3357        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3358        struct nvmf_ctrl_options *opts = ctrl->opts;
3359        int ctrl_loss_tmo, err;
3360
3361        err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3362        if (err)
3363                return -EINVAL;
3364
3365        if (ctrl_loss_tmo < 0)
3366                opts->max_reconnects = -1;
3367        else
3368                opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3369                                                opts->reconnect_delay);
3370        return count;
3371}
3372static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3373        nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3374
3375static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3376                struct device_attribute *attr, char *buf)
3377{
3378        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3379
3380        if (ctrl->opts->reconnect_delay == -1)
3381                return sysfs_emit(buf, "off\n");
3382        return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3383}
3384
3385static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3386                struct device_attribute *attr, const char *buf, size_t count)
3387{
3388        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3389        unsigned int v;
3390        int err;
3391
3392        err = kstrtou32(buf, 10, &v);
3393        if (err)
3394                return err;
3395
3396        ctrl->opts->reconnect_delay = v;
3397        return count;
3398}
3399static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3400        nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3401
3402static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3403                struct device_attribute *attr, char *buf)
3404{
3405        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3406
3407        if (ctrl->opts->fast_io_fail_tmo == -1)
3408                return sysfs_emit(buf, "off\n");
3409        return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3410}
3411
3412static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3413                struct device_attribute *attr, const char *buf, size_t count)
3414{
3415        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3416        struct nvmf_ctrl_options *opts = ctrl->opts;
3417        int fast_io_fail_tmo, err;
3418
3419        err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3420        if (err)
3421                return -EINVAL;
3422
3423        if (fast_io_fail_tmo < 0)
3424                opts->fast_io_fail_tmo = -1;
3425        else
3426                opts->fast_io_fail_tmo = fast_io_fail_tmo;
3427        return count;
3428}
3429static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3430        nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3431
3432static struct attribute *nvme_dev_attrs[] = {
3433        &dev_attr_reset_controller.attr,
3434        &dev_attr_rescan_controller.attr,
3435        &dev_attr_model.attr,
3436        &dev_attr_serial.attr,
3437        &dev_attr_firmware_rev.attr,
3438        &dev_attr_cntlid.attr,
3439        &dev_attr_delete_controller.attr,
3440        &dev_attr_transport.attr,
3441        &dev_attr_subsysnqn.attr,
3442        &dev_attr_address.attr,
3443        &dev_attr_state.attr,
3444        &dev_attr_numa_node.attr,
3445        &dev_attr_queue_count.attr,
3446        &dev_attr_sqsize.attr,
3447        &dev_attr_hostnqn.attr,
3448        &dev_attr_hostid.attr,
3449        &dev_attr_ctrl_loss_tmo.attr,
3450        &dev_attr_reconnect_delay.attr,
3451        &dev_attr_fast_io_fail_tmo.attr,
3452        &dev_attr_kato.attr,
3453        NULL
3454};
3455
3456static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3457                struct attribute *a, int n)
3458{
3459        struct device *dev = container_of(kobj, struct device, kobj);
3460        struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3461
3462        if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3463                return 0;
3464        if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3465                return 0;
3466        if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3467                return 0;
3468        if (a == &dev_attr_hostid.attr && !ctrl->opts)
3469                return 0;
3470        if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3471                return 0;
3472        if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3473                return 0;
3474        if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3475                return 0;
3476
3477        return a->mode;
3478}
3479
3480static const struct attribute_group nvme_dev_attrs_group = {
3481        .attrs          = nvme_dev_attrs,
3482        .is_visible     = nvme_dev_attrs_are_visible,
3483};
3484
3485static const struct attribute_group *nvme_dev_attr_groups[] = {
3486        &nvme_dev_attrs_group,
3487        NULL,
3488};
3489
3490static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3491                unsigned nsid)
3492{
3493        struct nvme_ns_head *h;
3494
3495        lockdep_assert_held(&subsys->lock);
3496
3497        list_for_each_entry(h, &subsys->nsheads, entry) {
3498                if (h->ns_id == nsid && nvme_tryget_ns_head(h))
3499                        return h;
3500        }
3501
3502        return NULL;
3503}
3504
3505static int __nvme_check_ids(struct nvme_subsystem *subsys,
3506                struct nvme_ns_head *new)
3507{
3508        struct nvme_ns_head *h;
3509
3510        lockdep_assert_held(&subsys->lock);
3511
3512        list_for_each_entry(h, &subsys->nsheads, entry) {
3513                if (nvme_ns_ids_valid(&new->ids) &&
3514                    nvme_ns_ids_equal(&new->ids, &h->ids))
3515                        return -EINVAL;
3516        }
3517
3518        return 0;
3519}
3520
3521void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3522{
3523        cdev_device_del(cdev, cdev_device);
3524        ida_simple_remove(&nvme_ns_chr_minor_ida, MINOR(cdev_device->devt));
3525}
3526
3527int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3528                const struct file_operations *fops, struct module *owner)
3529{
3530        int minor, ret;
3531
3532        minor = ida_simple_get(&nvme_ns_chr_minor_ida, 0, 0, GFP_KERNEL);
3533        if (minor < 0)
3534                return minor;
3535        cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3536        cdev_device->class = nvme_ns_chr_class;
3537        device_initialize(cdev_device);
3538        cdev_init(cdev, fops);
3539        cdev->owner = owner;
3540        ret = cdev_device_add(cdev, cdev_device);
3541        if (ret) {
3542                put_device(cdev_device);
3543                ida_simple_remove(&nvme_ns_chr_minor_ida, minor);
3544        }
3545        return ret;
3546}
3547
3548static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3549{
3550        return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3551}
3552
3553static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3554{
3555        nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3556        return 0;
3557}
3558
3559static const struct file_operations nvme_ns_chr_fops = {
3560        .owner          = THIS_MODULE,
3561        .open           = nvme_ns_chr_open,
3562        .release        = nvme_ns_chr_release,
3563        .unlocked_ioctl = nvme_ns_chr_ioctl,
3564        .compat_ioctl   = compat_ptr_ioctl,
3565};
3566
3567static int nvme_add_ns_cdev(struct nvme_ns *ns)
3568{
3569        int ret;
3570
3571        ns->cdev_device.parent = ns->ctrl->device;
3572        ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3573                           ns->ctrl->instance, ns->head->instance);
3574        if (ret)
3575                return ret;
3576        ret = nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3577                            ns->ctrl->ops->module);
3578        if (ret)
3579                kfree_const(ns->cdev_device.kobj.name);
3580        return ret;
3581}
3582
3583static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3584                unsigned nsid, struct nvme_ns_ids *ids)
3585{
3586        struct nvme_ns_head *head;
3587        size_t size = sizeof(*head);
3588        int ret = -ENOMEM;
3589
3590#ifdef CONFIG_NVME_MULTIPATH
3591        size += num_possible_nodes() * sizeof(struct nvme_ns *);
3592#endif
3593
3594        head = kzalloc(size, GFP_KERNEL);
3595        if (!head)
3596                goto out;
3597        ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3598        if (ret < 0)
3599                goto out_free_head;
3600        head->instance = ret;
3601        INIT_LIST_HEAD(&head->list);
3602        ret = init_srcu_struct(&head->srcu);
3603        if (ret)
3604                goto out_ida_remove;
3605        head->subsys = ctrl->subsys;
3606        head->ns_id = nsid;
3607        head->ids = *ids;
3608        kref_init(&head->ref);
3609
3610        ret = __nvme_check_ids(ctrl->subsys, head);
3611        if (ret) {
3612                dev_err(ctrl->device,
3613                        "duplicate IDs for nsid %d\n", nsid);
3614                goto out_cleanup_srcu;
3615        }
3616
3617        if (head->ids.csi) {
3618                ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3619                if (ret)
3620                        goto out_cleanup_srcu;
3621        } else
3622                head->effects = ctrl->effects;
3623
3624        ret = nvme_mpath_alloc_disk(ctrl, head);
3625        if (ret)
3626                goto out_cleanup_srcu;
3627
3628        list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3629
3630        kref_get(&ctrl->subsys->ref);
3631
3632        return head;
3633out_cleanup_srcu:
3634        cleanup_srcu_struct(&head->srcu);
3635out_ida_remove:
3636        ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3637out_free_head:
3638        kfree(head);
3639out:
3640        if (ret > 0)
3641                ret = blk_status_to_errno(nvme_error_status(ret));
3642        return ERR_PTR(ret);
3643}
3644
3645static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3646                struct nvme_ns_ids *ids, bool is_shared)
3647{
3648        struct nvme_ctrl *ctrl = ns->ctrl;
3649        struct nvme_ns_head *head = NULL;
3650        int ret = 0;
3651
3652        mutex_lock(&ctrl->subsys->lock);
3653        head = nvme_find_ns_head(ctrl->subsys, nsid);
3654        if (!head) {
3655                head = nvme_alloc_ns_head(ctrl, nsid, ids);
3656                if (IS_ERR(head)) {
3657                        ret = PTR_ERR(head);
3658                        goto out_unlock;
3659                }
3660                head->shared = is_shared;
3661        } else {
3662                ret = -EINVAL;
3663                if (!is_shared || !head->shared) {
3664                        dev_err(ctrl->device,
3665                                "Duplicate unshared namespace %d\n", nsid);
3666                        goto out_put_ns_head;
3667                }
3668                if (!nvme_ns_ids_equal(&head->ids, ids)) {
3669                        dev_err(ctrl->device,
3670                                "IDs don't match for shared namespace %d\n",
3671                                        nsid);
3672                        goto out_put_ns_head;
3673                }
3674        }
3675
3676        list_add_tail_rcu(&ns->siblings, &head->list);
3677        ns->head = head;
3678        mutex_unlock(&ctrl->subsys->lock);
3679        return 0;
3680
3681out_put_ns_head:
3682        nvme_put_ns_head(head);
3683out_unlock:
3684        mutex_unlock(&ctrl->subsys->lock);
3685        return ret;
3686}
3687
3688static int ns_cmp(void *priv, const struct list_head *a,
3689                const struct list_head *b)
3690{
3691        struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3692        struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3693
3694        return nsa->head->ns_id - nsb->head->ns_id;
3695}
3696
3697struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3698{
3699        struct nvme_ns *ns, *ret = NULL;
3700
3701        down_read(&ctrl->namespaces_rwsem);
3702        list_for_each_entry(ns, &ctrl->namespaces, list) {
3703                if (ns->head->ns_id == nsid) {
3704                        if (!nvme_get_ns(ns))
3705                                continue;
3706                        ret = ns;
3707                        break;
3708                }
3709                if (ns->head->ns_id > nsid)
3710                        break;
3711        }
3712        up_read(&ctrl->namespaces_rwsem);
3713        return ret;
3714}
3715EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3716
3717static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3718                struct nvme_ns_ids *ids)
3719{
3720        struct nvme_ns *ns;
3721        struct gendisk *disk;
3722        struct nvme_id_ns *id;
3723        int node = ctrl->numa_node;
3724
3725        if (nvme_identify_ns(ctrl, nsid, ids, &id))
3726                return;
3727
3728        ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3729        if (!ns)
3730                goto out_free_id;
3731
3732        ns->queue = blk_mq_init_queue(ctrl->tagset);
3733        if (IS_ERR(ns->queue))
3734                goto out_free_ns;
3735
3736        if (ctrl->opts && ctrl->opts->data_digest)
3737                blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3738
3739        blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3740        if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3741                blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3742
3743        ns->queue->queuedata = ns;
3744        ns->ctrl = ctrl;
3745        kref_init(&ns->kref);
3746
3747        if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3748                goto out_free_queue;
3749
3750        disk = alloc_disk_node(0, node);
3751        if (!disk)
3752                goto out_unlink_ns;
3753
3754        disk->fops = &nvme_bdev_ops;
3755        disk->private_data = ns;
3756        disk->queue = ns->queue;
3757        /*
3758         * Without the multipath code enabled, multiple controller per
3759         * subsystems are visible as devices and thus we cannot use the
3760         * subsystem instance.
3761         */
3762        if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
3763                sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3764                        ns->head->instance);
3765        ns->disk = disk;
3766
3767        if (nvme_update_ns_info(ns, id))
3768                goto out_put_disk;
3769
3770        if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3771                if (nvme_nvm_register(ns, disk->disk_name, node)) {
3772                        dev_warn(ctrl->device, "LightNVM init failure\n");
3773                        goto out_put_disk;
3774                }
3775        }
3776
3777        down_write(&ctrl->namespaces_rwsem);
3778        list_add_tail(&ns->list, &ctrl->namespaces);
3779        up_write(&ctrl->namespaces_rwsem);
3780
3781        nvme_get_ctrl(ctrl);
3782
3783        device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3784        if (!nvme_ns_head_multipath(ns->head))
3785                nvme_add_ns_cdev(ns);
3786
3787        nvme_mpath_add_disk(ns, id);
3788        nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3789        kfree(id);
3790
3791        return;
3792 out_put_disk:
3793        /* prevent double queue cleanup */
3794        ns->disk->queue = NULL;
3795        put_disk(ns->disk);
3796 out_unlink_ns:
3797        mutex_lock(&ctrl->subsys->lock);
3798        list_del_rcu(&ns->siblings);
3799        if (list_empty(&ns->head->list))
3800                list_del_init(&ns->head->entry);
3801        mutex_unlock(&ctrl->subsys->lock);
3802        nvme_put_ns_head(ns->head);
3803 out_free_queue:
3804        blk_cleanup_queue(ns->queue);
3805 out_free_ns:
3806        kfree(ns);
3807 out_free_id:
3808        kfree(id);
3809}
3810
3811static void nvme_ns_remove(struct nvme_ns *ns)
3812{
3813        bool last_path = false;
3814
3815        if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3816                return;
3817
3818        set_capacity(ns->disk, 0);
3819        nvme_fault_inject_fini(&ns->fault_inject);
3820
3821        mutex_lock(&ns->ctrl->subsys->lock);
3822        list_del_rcu(&ns->siblings);
3823        mutex_unlock(&ns->ctrl->subsys->lock);
3824
3825        synchronize_rcu(); /* guarantee not available in head->list */
3826        nvme_mpath_clear_current_path(ns);
3827        synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3828
3829        if (ns->disk->flags & GENHD_FL_UP) {
3830                if (!nvme_ns_head_multipath(ns->head))
3831                        nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3832                del_gendisk(ns->disk);
3833                blk_cleanup_queue(ns->queue);
3834                if (blk_get_integrity(ns->disk))
3835                        blk_integrity_unregister(ns->disk);
3836        }
3837
3838        down_write(&ns->ctrl->namespaces_rwsem);
3839        list_del_init(&ns->list);
3840        up_write(&ns->ctrl->namespaces_rwsem);
3841
3842        /* Synchronize with nvme_init_ns_head() */
3843        mutex_lock(&ns->head->subsys->lock);
3844        if (list_empty(&ns->head->list)) {
3845                list_del_init(&ns->head->entry);
3846                last_path = true;
3847        }
3848        mutex_unlock(&ns->head->subsys->lock);
3849        if (last_path)
3850                nvme_mpath_shutdown_disk(ns->head);
3851        nvme_put_ns(ns);
3852}
3853
3854static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3855{
3856        struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3857
3858        if (ns) {
3859                nvme_ns_remove(ns);
3860                nvme_put_ns(ns);
3861        }
3862}
3863
3864static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3865{
3866        struct nvme_id_ns *id;
3867        int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3868
3869        if (test_bit(NVME_NS_DEAD, &ns->flags))
3870                goto out;
3871
3872        ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3873        if (ret)
3874                goto out;
3875
3876        ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3877        if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3878                dev_err(ns->ctrl->device,
3879                        "identifiers changed for nsid %d\n", ns->head->ns_id);
3880                goto out_free_id;
3881        }
3882
3883        ret = nvme_update_ns_info(ns, id);
3884
3885out_free_id:
3886        kfree(id);
3887out:
3888        /*
3889         * Only remove the namespace if we got a fatal error back from the
3890         * device, otherwise ignore the error and just move on.
3891         *
3892         * TODO: we should probably schedule a delayed retry here.
3893         */
3894        if (ret > 0 && (ret & NVME_SC_DNR))
3895                nvme_ns_remove(ns);
3896}
3897
3898static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3899{
3900        struct nvme_ns_ids ids = { };
3901        struct nvme_ns *ns;
3902
3903        if (nvme_identify_ns_descs(ctrl, nsid, &ids))
3904                return;
3905
3906        ns = nvme_find_get_ns(ctrl, nsid);
3907        if (ns) {
3908                nvme_validate_ns(ns, &ids);
3909                nvme_put_ns(ns);
3910                return;
3911        }
3912
3913        switch (ids.csi) {
3914        case NVME_CSI_NVM:
3915                nvme_alloc_ns(ctrl, nsid, &ids);
3916                break;
3917        case NVME_CSI_ZNS:
3918                if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
3919                        dev_warn(ctrl->device,
3920                                "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
3921                                nsid);
3922                        break;
3923                }
3924                if (!nvme_multi_css(ctrl)) {
3925                        dev_warn(ctrl->device,
3926                                "command set not reported for nsid: %d\n",
3927                                nsid);
3928                        break;
3929                }
3930                nvme_alloc_ns(ctrl, nsid, &ids);
3931                break;
3932        default:
3933                dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
3934                        ids.csi, nsid);
3935                break;
3936        }
3937}
3938
3939static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3940                                        unsigned nsid)
3941{
3942        struct nvme_ns *ns, *next;
3943        LIST_HEAD(rm_list);
3944
3945        down_write(&ctrl->namespaces_rwsem);
3946        list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3947                if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3948                        list_move_tail(&ns->list, &rm_list);
3949        }
3950        up_write(&ctrl->namespaces_rwsem);
3951
3952        list_for_each_entry_safe(ns, next, &rm_list, list)
3953                nvme_ns_remove(ns);
3954
3955}
3956
3957static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3958{
3959        const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3960        __le32 *ns_list;
3961        u32 prev = 0;
3962        int ret = 0, i;
3963
3964        if (nvme_ctrl_limited_cns(ctrl))
3965                return -EOPNOTSUPP;
3966
3967        ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3968        if (!ns_list)
3969                return -ENOMEM;
3970
3971        for (;;) {
3972                struct nvme_command cmd = {
3973                        .identify.opcode        = nvme_admin_identify,
3974                        .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
3975                        .identify.nsid          = cpu_to_le32(prev),
3976                };
3977
3978                ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3979                                            NVME_IDENTIFY_DATA_SIZE);
3980                if (ret) {
3981                        dev_warn(ctrl->device,
3982                                "Identify NS List failed (status=0x%x)\n", ret);
3983                        goto free;
3984                }
3985
3986                for (i = 0; i < nr_entries; i++) {
3987                        u32 nsid = le32_to_cpu(ns_list[i]);
3988
3989                        if (!nsid)      /* end of the list? */
3990                                goto out;
3991                        nvme_validate_or_alloc_ns(ctrl, nsid);
3992                        while (++prev < nsid)
3993                                nvme_ns_remove_by_nsid(ctrl, prev);
3994                }
3995        }
3996 out:
3997        nvme_remove_invalid_namespaces(ctrl, prev);
3998 free:
3999        kfree(ns_list);
4000        return ret;
4001}
4002
4003static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4004{
4005        struct nvme_id_ctrl *id;
4006        u32 nn, i;
4007
4008        if (nvme_identify_ctrl(ctrl, &id))
4009                return;
4010        nn = le32_to_cpu(id->nn);
4011        kfree(id);
4012
4013        for (i = 1; i <= nn; i++)
4014                nvme_validate_or_alloc_ns(ctrl, i);
4015
4016        nvme_remove_invalid_namespaces(ctrl, nn);
4017}
4018
4019static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4020{
4021        size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4022        __le32 *log;
4023        int error;
4024
4025        log = kzalloc(log_size, GFP_KERNEL);
4026        if (!log)
4027                return;
4028
4029        /*
4030         * We need to read the log to clear the AEN, but we don't want to rely
4031         * on it for the changed namespace information as userspace could have
4032         * raced with us in reading the log page, which could cause us to miss
4033         * updates.
4034         */
4035        error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4036                        NVME_CSI_NVM, log, log_size, 0);
4037        if (error)
4038                dev_warn(ctrl->device,
4039                        "reading changed ns log failed: %d\n", error);
4040
4041        kfree(log);
4042}
4043
4044static void nvme_scan_work(struct work_struct *work)
4045{
4046        struct nvme_ctrl *ctrl =
4047                container_of(work, struct nvme_ctrl, scan_work);
4048
4049        /* No tagset on a live ctrl means IO queues could not created */
4050        if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4051                return;
4052
4053        if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4054                dev_info(ctrl->device, "rescanning namespaces.\n");
4055                nvme_clear_changed_ns_log(ctrl);
4056        }
4057
4058        mutex_lock(&ctrl->scan_lock);
4059        if (nvme_scan_ns_list(ctrl) != 0)
4060                nvme_scan_ns_sequential(ctrl);
4061        mutex_unlock(&ctrl->scan_lock);
4062
4063        down_write(&ctrl->namespaces_rwsem);
4064        list_sort(NULL, &ctrl->namespaces, ns_cmp);
4065        up_write(&ctrl->namespaces_rwsem);
4066}
4067
4068/*
4069 * This function iterates the namespace list unlocked to allow recovery from
4070 * controller failure. It is up to the caller to ensure the namespace list is
4071 * not modified by scan work while this function is executing.
4072 */
4073void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4074{
4075        struct nvme_ns *ns, *next;
4076        LIST_HEAD(ns_list);
4077
4078        /*
4079         * make sure to requeue I/O to all namespaces as these
4080         * might result from the scan itself and must complete
4081         * for the scan_work to make progress
4082         */
4083        nvme_mpath_clear_ctrl_paths(ctrl);
4084
4085        /* prevent racing with ns scanning */
4086        flush_work(&ctrl->scan_work);
4087
4088        /*
4089         * The dead states indicates the controller was not gracefully
4090         * disconnected. In that case, we won't be able to flush any data while
4091         * removing the namespaces' disks; fail all the queues now to avoid
4092         * potentially having to clean up the failed sync later.
4093         */
4094        if (ctrl->state == NVME_CTRL_DEAD)
4095                nvme_kill_queues(ctrl);
4096
4097        /* this is a no-op when called from the controller reset handler */
4098        nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4099
4100        down_write(&ctrl->namespaces_rwsem);
4101        list_splice_init(&ctrl->namespaces, &ns_list);
4102        up_write(&ctrl->namespaces_rwsem);
4103
4104        list_for_each_entry_safe(ns, next, &ns_list, list)
4105                nvme_ns_remove(ns);
4106}
4107EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4108
4109static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4110{
4111        struct nvme_ctrl *ctrl =
4112                container_of(dev, struct nvme_ctrl, ctrl_device);
4113        struct nvmf_ctrl_options *opts = ctrl->opts;
4114        int ret;
4115
4116        ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4117        if (ret)
4118                return ret;
4119
4120        if (opts) {
4121                ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4122                if (ret)
4123                        return ret;
4124
4125                ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4126                                opts->trsvcid ?: "none");
4127                if (ret)
4128                        return ret;
4129
4130                ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4131                                opts->host_traddr ?: "none");
4132                if (ret)
4133                        return ret;
4134
4135                ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4136                                opts->host_iface ?: "none");
4137        }
4138        return ret;
4139}
4140
4141static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4142{
4143        char *envp[2] = { NULL, NULL };
4144        u32 aen_result = ctrl->aen_result;
4145
4146        ctrl->aen_result = 0;
4147        if (!aen_result)
4148                return;
4149
4150        envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4151        if (!envp[0])
4152                return;
4153        kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4154        kfree(envp[0]);
4155}
4156
4157static void nvme_async_event_work(struct work_struct *work)
4158{
4159        struct nvme_ctrl *ctrl =
4160                container_of(work, struct nvme_ctrl,