drivers/nvme/host/core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * NVM Express device driver
   4  * Copyright (c) 2011-2014, Intel Corporation.
   5  */
   6
   7 #include <linux/blkdev.h>
   8 #include <linux/blk-mq.h>
   9 #include <linux/delay.h>
  10 #include <linux/errno.h>
  11 #include <linux/hdreg.h>
  12 #include <linux/kernel.h>
  13 #include <linux/module.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/list_sort.h>
  16 #include <linux/slab.h>
  17 #include <linux/types.h>
  18 #include <linux/pr.h>
  19 #include <linux/ptrace.h>
  20 #include <linux/nvme_ioctl.h>
  21 #include <linux/t10-pi.h>
  22 #include <linux/pm_qos.h>
  23 #include <asm/unaligned.h>
  24
  25 #include "nvme.h"
  26 #include "fabrics.h"
  27
  28 #define CREATE_TRACE_POINTS
  29 #include "trace.h"
  30
  31 #define NVME_MINORS             (1U << MINORBITS)
  32
  33 unsigned int admin_timeout = 60;
  34 module_param(admin_timeout, uint, 0644);
  35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  36 EXPORT_SYMBOL_GPL(admin_timeout);
  37
  38 unsigned int nvme_io_timeout = 30;
  39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  42
  43 static unsigned char shutdown_timeout = 5;
  44 module_param(shutdown_timeout, byte, 0644);
  45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  46
  47 static u8 nvme_max_retries = 5;
  48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  50
  51 static unsigned long default_ps_max_latency_us = 100000;
  52 module_param(default_ps_max_latency_us, ulong, 0644);
  53 MODULE_PARM_DESC(default_ps_max_latency_us,
  54                  "max power saving latency for new devices; use PM QOS to change per device");
  55
  56 static bool force_apst;
  57 module_param(force_apst, bool, 0644);
  58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  59
  60 static bool streams;
  61 module_param(streams, bool, 0644);
  62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  63
  64 /*
  65  * nvme_wq - hosts nvme related works that are not reset or delete
  66  * nvme_reset_wq - hosts nvme reset works
  67  * nvme_delete_wq - hosts nvme delete works
  68  *
  69  * nvme_wq will host works such are scan, aen handling, fw activation,
  70  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
  71  * runs reset works which also flush works hosted on nvme_wq for
  72  * serialization purposes. nvme_delete_wq host controller deletion
  73  * works which flush reset works for serialization.
  74  */
  75 struct workqueue_struct *nvme_wq;
  76 EXPORT_SYMBOL_GPL(nvme_wq);
  77
  78 struct workqueue_struct *nvme_reset_wq;
  79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  80
  81 struct workqueue_struct *nvme_delete_wq;
  82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  83
  84 static DEFINE_IDA(nvme_subsystems_ida);
  85 static LIST_HEAD(nvme_subsystems);
  86 static DEFINE_MUTEX(nvme_subsystems_lock);
  87
  88 static DEFINE_IDA(nvme_instance_ida);
  89 static dev_t nvme_chr_devt;
  90 static struct class *nvme_class;
  91 static struct class *nvme_subsys_class;
  92
  93 static int nvme_revalidate_disk(struct gendisk *disk);
  94 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
  95 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
  96                                            unsigned nsid);
  97
  98 static void nvme_set_queue_dying(struct nvme_ns *ns)
  99 {
 100         /*
 101          * Revalidating a dead namespace sets capacity to 0. This will end
 102          * buffered writers dirtying pages that can't be synced.
 103          */
 104         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 105                 return;
 106         revalidate_disk(ns->disk);
 107         blk_set_queue_dying(ns->queue);
 108         /* Forcibly unquiesce queues to avoid blocking dispatch */
 109         blk_mq_unquiesce_queue(ns->queue);
 110 }
 111
 112 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 113 {
 114         /*
 115          * Only new queue scan work when admin and IO queues are both alive
 116          */
 117         if (ctrl->state == NVME_CTRL_LIVE)
 118                 queue_work(nvme_wq, &ctrl->scan_work);
 119 }
 120
 121 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 122 {
 123         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 124                 return -EBUSY;
 125         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 126                 return -EBUSY;
 127         return 0;
 128 }
 129 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 130
 131 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 132 {
 133         int ret;
 134
 135         ret = nvme_reset_ctrl(ctrl);
 136         if (!ret) {
 137                 flush_work(&ctrl->reset_work);
 138                 if (ctrl->state != NVME_CTRL_LIVE &&
 139                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
 140                         ret = -ENETRESET;
 141         }
 142
 143         return ret;
 144 }
 145 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 146
 147 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 148 {
 149         dev_info(ctrl->device,
 150                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 151
 152         flush_work(&ctrl->reset_work);
 153         nvme_stop_ctrl(ctrl);
 154         nvme_remove_namespaces(ctrl);
 155         ctrl->ops->delete_ctrl(ctrl);
 156         nvme_uninit_ctrl(ctrl);
 157         nvme_put_ctrl(ctrl);
 158 }
 159
 160 static void nvme_delete_ctrl_work(struct work_struct *work)
 161 {
 162         struct nvme_ctrl *ctrl =
 163                 container_of(work, struct nvme_ctrl, delete_work);
 164
 165         nvme_do_delete_ctrl(ctrl);
 166 }
 167
 168 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 169 {
 170         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 171                 return -EBUSY;
 172         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 173                 return -EBUSY;
 174         return 0;
 175 }
 176 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 177
 178 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 179 {
 180         int ret = 0;
 181
 182         /*
 183          * Keep a reference until nvme_do_delete_ctrl() complete,
 184          * since ->delete_ctrl can free the controller.
 185          */
 186         nvme_get_ctrl(ctrl);
 187         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 188                 ret = -EBUSY;
 189         if (!ret)
 190                 nvme_do_delete_ctrl(ctrl);
 191         nvme_put_ctrl(ctrl);
 192         return ret;
 193 }
 194
 195 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 196 {
 197         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 198 }
 199
 200 static blk_status_t nvme_error_status(u16 status)
 201 {
 202         switch (status & 0x7ff) {
 203         case NVME_SC_SUCCESS:
 204                 return BLK_STS_OK;
 205         case NVME_SC_CAP_EXCEEDED:
 206                 return BLK_STS_NOSPC;
 207         case NVME_SC_LBA_RANGE:
 208                 return BLK_STS_TARGET;
 209         case NVME_SC_BAD_ATTRIBUTES:
 210         case NVME_SC_ONCS_NOT_SUPPORTED:
 211         case NVME_SC_INVALID_OPCODE:
 212         case NVME_SC_INVALID_FIELD:
 213         case NVME_SC_INVALID_NS:
 214                 return BLK_STS_NOTSUPP;
 215         case NVME_SC_WRITE_FAULT:
 216         case NVME_SC_READ_ERROR:
 217         case NVME_SC_UNWRITTEN_BLOCK:
 218         case NVME_SC_ACCESS_DENIED:
 219         case NVME_SC_READ_ONLY:
 220         case NVME_SC_COMPARE_FAILED:
 221                 return BLK_STS_MEDIUM;
 222         case NVME_SC_GUARD_CHECK:
 223         case NVME_SC_APPTAG_CHECK:
 224         case NVME_SC_REFTAG_CHECK:
 225         case NVME_SC_INVALID_PI:
 226                 return BLK_STS_PROTECTION;
 227         case NVME_SC_RESERVATION_CONFLICT:
 228                 return BLK_STS_NEXUS;
 229         case NVME_SC_HOST_PATH_ERROR:
 230                 return BLK_STS_TRANSPORT;
 231         default:
 232                 return BLK_STS_IOERR;
 233         }
 234 }
 235
 236 static inline bool nvme_req_needs_retry(struct request *req)
 237 {
 238         if (blk_noretry_request(req))
 239                 return false;
 240         if (nvme_req(req)->status & NVME_SC_DNR)
 241                 return false;
 242         if (nvme_req(req)->retries >= nvme_max_retries)
 243                 return false;
 244         return true;
 245 }
 246
 247 static void nvme_retry_req(struct request *req)
 248 {
 249         struct nvme_ns *ns = req->q->queuedata;
 250         unsigned long delay = 0;
 251         u16 crd;
 252
 253         /* The mask and shift result must be <= 3 */
 254         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 255         if (ns && crd)
 256                 delay = ns->ctrl->crdt[crd - 1] * 100;
 257
 258         nvme_req(req)->retries++;
 259         blk_mq_requeue_request(req, false);
 260         blk_mq_delay_kick_requeue_list(req->q, delay);
 261 }
 262
 263 void nvme_complete_rq(struct request *req)
 264 {
 265         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 266
 267         trace_nvme_complete_rq(req);
 268
 269         if (nvme_req(req)->ctrl->kas)
 270                 nvme_req(req)->ctrl->comp_seen = true;
 271
 272         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 273                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
 274                     blk_path_error(status)) {
 275                         nvme_failover_req(req);
 276                         return;
 277                 }
 278
 279                 if (!blk_queue_dying(req->q)) {
 280                         nvme_retry_req(req);
 281                         return;
 282                 }
 283         }
 284
 285         nvme_trace_bio_complete(req, status);
 286         blk_mq_end_request(req, status);
 287 }
 288 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 289
 290 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 291 {
 292         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 293                                 "Cancelling I/O %d", req->tag);
 294
 295         /* don't abort one completed request */
 296         if (blk_mq_request_completed(req))
 297                 return true;
 298
 299         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 300         blk_mq_complete_request(req);
 301         return true;
 302 }
 303 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 304
 305 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 306                 enum nvme_ctrl_state new_state)
 307 {
 308         enum nvme_ctrl_state old_state;
 309         unsigned long flags;
 310         bool changed = false;
 311
 312         spin_lock_irqsave(&ctrl->lock, flags);
 313
 314         old_state = ctrl->state;
 315         switch (new_state) {
 316         case NVME_CTRL_ADMIN_ONLY:
 317                 switch (old_state) {
 318                 case NVME_CTRL_CONNECTING:
 319                         changed = true;
 320                         /* FALLTHRU */
 321                 default:
 322                         break;
 323                 }
 324                 break;
 325         case NVME_CTRL_LIVE:
 326                 switch (old_state) {
 327                 case NVME_CTRL_NEW:
 328                 case NVME_CTRL_RESETTING:
 329                 case NVME_CTRL_CONNECTING:
 330                         changed = true;
 331                         /* FALLTHRU */
 332                 default:
 333                         break;
 334                 }
 335                 break;
 336         case NVME_CTRL_RESETTING:
 337                 switch (old_state) {
 338                 case NVME_CTRL_NEW:
 339                 case NVME_CTRL_LIVE:
 340                 case NVME_CTRL_ADMIN_ONLY:
 341                         changed = true;
 342                         /* FALLTHRU */
 343                 default:
 344                         break;
 345                 }
 346                 break;
 347         case NVME_CTRL_CONNECTING:
 348                 switch (old_state) {
 349                 case NVME_CTRL_NEW:
 350                 case NVME_CTRL_RESETTING:
 351                         changed = true;
 352                         /* FALLTHRU */
 353                 default:
 354                         break;
 355                 }
 356                 break;
 357         case NVME_CTRL_DELETING:
 358                 switch (old_state) {
 359                 case NVME_CTRL_LIVE:
 360                 case NVME_CTRL_ADMIN_ONLY:
 361                 case NVME_CTRL_RESETTING:
 362                 case NVME_CTRL_CONNECTING:
 363                         changed = true;
 364                         /* FALLTHRU */
 365                 default:
 366                         break;
 367                 }
 368                 break;
 369         case NVME_CTRL_DEAD:
 370                 switch (old_state) {
 371                 case NVME_CTRL_DELETING:
 372                         changed = true;
 373                         /* FALLTHRU */
 374                 default:
 375                         break;
 376                 }
 377                 break;
 378         default:
 379                 break;
 380         }
 381
 382         if (changed)
 383                 ctrl->state = new_state;
 384
 385         spin_unlock_irqrestore(&ctrl->lock, flags);
 386         if (changed && ctrl->state == NVME_CTRL_LIVE)
 387                 nvme_kick_requeue_lists(ctrl);
 388         return changed;
 389 }
 390 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 391
 392 static void nvme_free_ns_head(struct kref *ref)
 393 {
 394         struct nvme_ns_head *head =
 395                 container_of(ref, struct nvme_ns_head, ref);
 396
 397         nvme_mpath_remove_disk(head);
 398         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 399         list_del_init(&head->entry);
 400         cleanup_srcu_struct(&head->srcu);
 401         nvme_put_subsystem(head->subsys);
 402         kfree(head);
 403 }
 404
 405 static void nvme_put_ns_head(struct nvme_ns_head *head)
 406 {
 407         kref_put(&head->ref, nvme_free_ns_head);
 408 }
 409
 410 static void nvme_free_ns(struct kref *kref)
 411 {
 412         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 413
 414         if (ns->ndev)
 415                 nvme_nvm_unregister(ns);
 416
 417         put_disk(ns->disk);
 418         nvme_put_ns_head(ns->head);
 419         nvme_put_ctrl(ns->ctrl);
 420         kfree(ns);
 421 }
 422
 423 static void nvme_put_ns(struct nvme_ns *ns)
 424 {
 425         kref_put(&ns->kref, nvme_free_ns);
 426 }
 427
 428 static inline void nvme_clear_nvme_request(struct request *req)
 429 {
 430         if (!(req->rq_flags & RQF_DONTPREP)) {
 431                 nvme_req(req)->retries = 0;
 432                 nvme_req(req)->flags = 0;
 433                 req->rq_flags |= RQF_DONTPREP;
 434         }
 435 }
 436
 437 struct request *nvme_alloc_request(struct request_queue *q,
 438                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 439 {
 440         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 441         struct request *req;
 442
 443         if (qid == NVME_QID_ANY) {
 444                 req = blk_mq_alloc_request(q, op, flags);
 445         } else {
 446                 req = blk_mq_alloc_request_hctx(q, op, flags,
 447                                 qid ? qid - 1 : 0);
 448         }
 449         if (IS_ERR(req))
 450                 return req;
 451
 452         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 453         nvme_clear_nvme_request(req);
 454         nvme_req(req)->cmd = cmd;
 455
 456         return req;
 457 }
 458 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 459
 460 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 461 {
 462         struct nvme_command c;
 463
 464         memset(&c, 0, sizeof(c));
 465
 466         c.directive.opcode = nvme_admin_directive_send;
 467         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 468         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 469         c.directive.dtype = NVME_DIR_IDENTIFY;
 470         c.directive.tdtype = NVME_DIR_STREAMS;
 471         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 472
 473         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 474 }
 475
 476 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 477 {
 478         return nvme_toggle_streams(ctrl, false);
 479 }
 480
 481 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 482 {
 483         return nvme_toggle_streams(ctrl, true);
 484 }
 485
 486 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 487                                   struct streams_directive_params *s, u32 nsid)
 488 {
 489         struct nvme_command c;
 490
 491         memset(&c, 0, sizeof(c));
 492         memset(s, 0, sizeof(*s));
 493
 494         c.directive.opcode = nvme_admin_directive_recv;
 495         c.directive.nsid = cpu_to_le32(nsid);
 496         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 497         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 498         c.directive.dtype = NVME_DIR_STREAMS;
 499
 500         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 501 }
 502
 503 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 504 {
 505         struct streams_directive_params s;
 506         int ret;
 507
 508         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 509                 return 0;
 510         if (!streams)
 511                 return 0;
 512
 513         ret = nvme_enable_streams(ctrl);
 514         if (ret)
 515                 return ret;
 516
 517         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 518         if (ret)
 519                 return ret;
 520
 521         ctrl->nssa = le16_to_cpu(s.nssa);
 522         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 523                 dev_info(ctrl->device, "too few streams (%u) available\n",
 524                                         ctrl->nssa);
 525                 nvme_disable_streams(ctrl);
 526                 return 0;
 527         }
 528
 529         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 530         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 531         return 0;
 532 }
 533
 534 /*
 535  * Check if 'req' has a write hint associated with it. If it does, assign
 536  * a valid namespace stream to the write.
 537  */
 538 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 539                                      struct request *req, u16 *control,
 540                                      u32 *dsmgmt)
 541 {
 542         enum rw_hint streamid = req->write_hint;
 543
 544         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 545                 streamid = 0;
 546         else {
 547                 streamid--;
 548                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 549                         return;
 550
 551                 *control |= NVME_RW_DTYPE_STREAMS;
 552                 *dsmgmt |= streamid << 16;
 553         }
 554
 555         if (streamid < ARRAY_SIZE(req->q->write_hints))
 556                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 557 }
 558
 559 static inline void nvme_setup_flush(struct nvme_ns *ns,
 560                 struct nvme_command *cmnd)
 561 {
 562         cmnd->common.opcode = nvme_cmd_flush;
 563         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 564 }
 565
 566 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 567                 struct nvme_command *cmnd)
 568 {
 569         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 570         struct nvme_dsm_range *range;
 571         struct bio *bio;
 572
 573         range = kmalloc_array(segments, sizeof(*range),
 574                                 GFP_ATOMIC | __GFP_NOWARN);
 575         if (!range) {
 576                 /*
 577                  * If we fail allocation our range, fallback to the controller
 578                  * discard page. If that's also busy, it's safe to return
 579                  * busy, as we know we can make progress once that's freed.
 580                  */
 581                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 582                         return BLK_STS_RESOURCE;
 583
 584                 range = page_address(ns->ctrl->discard_page);
 585         }
 586
 587         __rq_for_each_bio(bio, req) {
 588                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 589                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 590
 591                 if (n < segments) {
 592                         range[n].cattr = cpu_to_le32(0);
 593                         range[n].nlb = cpu_to_le32(nlb);
 594                         range[n].slba = cpu_to_le64(slba);
 595                 }
 596                 n++;
 597         }
 598
 599         if (WARN_ON_ONCE(n != segments)) {
 600                 if (virt_to_page(range) == ns->ctrl->discard_page)
 601                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 602                 else
 603                         kfree(range);
 604                 return BLK_STS_IOERR;
 605         }
 606
 607         cmnd->dsm.opcode = nvme_cmd_dsm;
 608         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 609         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 610         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 611
 612         req->special_vec.bv_page = virt_to_page(range);
 613         req->special_vec.bv_offset = offset_in_page(range);
 614         req->special_vec.bv_len = sizeof(*range) * segments;
 615         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 616
 617         return BLK_STS_OK;
 618 }
 619
 620 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 621                 struct request *req, struct nvme_command *cmnd)
 622 {
 623         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 624                 return nvme_setup_discard(ns, req, cmnd);
 625
 626         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 627         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 628         cmnd->write_zeroes.slba =
 629                 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 630         cmnd->write_zeroes.length =
 631                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 632         cmnd->write_zeroes.control = 0;
 633         return BLK_STS_OK;
 634 }
 635
 636 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 637                 struct request *req, struct nvme_command *cmnd)
 638 {
 639         struct nvme_ctrl *ctrl = ns->ctrl;
 640         u16 control = 0;
 641         u32 dsmgmt = 0;
 642
 643         if (req->cmd_flags & REQ_FUA)
 644                 control |= NVME_RW_FUA;
 645         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 646                 control |= NVME_RW_LR;
 647
 648         if (req->cmd_flags & REQ_RAHEAD)
 649                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 650
 651         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 652         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 653         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 654         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 655
 656         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 657                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 658
 659         if (ns->ms) {
 660                 /*
 661                  * If formated with metadata, the block layer always provides a
 662                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 663                  * we enable the PRACT bit for protection information or set the
 664                  * namespace capacity to zero to prevent any I/O.
 665                  */
 666                 if (!blk_integrity_rq(req)) {
 667                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 668                                 return BLK_STS_NOTSUPP;
 669                         control |= NVME_RW_PRINFO_PRACT;
 670                 } else if (req_op(req) == REQ_OP_WRITE) {
 671                         t10_pi_prepare(req, ns->pi_type);
 672                 }
 673
 674                 switch (ns->pi_type) {
 675                 case NVME_NS_DPS_PI_TYPE3:
 676                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 677                         break;
 678                 case NVME_NS_DPS_PI_TYPE1:
 679                 case NVME_NS_DPS_PI_TYPE2:
 680                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 681                                         NVME_RW_PRINFO_PRCHK_REF;
 682                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 683                         break;
 684                 }
 685         }
 686
 687         cmnd->rw.control = cpu_to_le16(control);
 688         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 689         return 0;
 690 }
 691
 692 void nvme_cleanup_cmd(struct request *req)
 693 {
 694         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
 695             nvme_req(req)->status == 0) {
 696                 struct nvme_ns *ns = req->rq_disk->private_data;
 697
 698                 t10_pi_complete(req, ns->pi_type,
 699                                 blk_rq_bytes(req) >> ns->lba_shift);
 700         }
 701         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 702                 struct nvme_ns *ns = req->rq_disk->private_data;
 703                 struct page *page = req->special_vec.bv_page;
 704
 705                 if (page == ns->ctrl->discard_page)
 706                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 707                 else
 708                         kfree(page_address(page) + req->special_vec.bv_offset);
 709         }
 710 }
 711 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 712
 713 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 714                 struct nvme_command *cmd)
 715 {
 716         blk_status_t ret = BLK_STS_OK;
 717
 718         nvme_clear_nvme_request(req);
 719
 720         memset(cmd, 0, sizeof(*cmd));
 721         switch (req_op(req)) {
 722         case REQ_OP_DRV_IN:
 723         case REQ_OP_DRV_OUT:
 724                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 725                 break;
 726         case REQ_OP_FLUSH:
 727                 nvme_setup_flush(ns, cmd);
 728                 break;
 729         case REQ_OP_WRITE_ZEROES:
 730                 ret = nvme_setup_write_zeroes(ns, req, cmd);
 731                 break;
 732         case REQ_OP_DISCARD:
 733                 ret = nvme_setup_discard(ns, req, cmd);
 734                 break;
 735         case REQ_OP_READ:
 736         case REQ_OP_WRITE:
 737                 ret = nvme_setup_rw(ns, req, cmd);
 738                 break;
 739         default:
 740                 WARN_ON_ONCE(1);
 741                 return BLK_STS_IOERR;
 742         }
 743
 744         cmd->common.command_id = req->tag;
 745         trace_nvme_setup_cmd(req, cmd);
 746         return ret;
 747 }
 748 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 749
 750 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 751 {
 752         struct completion *waiting = rq->end_io_data;
 753
 754         rq->end_io_data = NULL;
 755         complete(waiting);
 756 }
 757
 758 static void nvme_execute_rq_polled(struct request_queue *q,
 759                 struct gendisk *bd_disk, struct request *rq, int at_head)
 760 {
 761         DECLARE_COMPLETION_ONSTACK(wait);
 762
 763         WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
 764
 765         rq->cmd_flags |= REQ_HIPRI;
 766         rq->end_io_data = &wait;
 767         blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
 768
 769         while (!completion_done(&wait)) {
 770                 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
 771                 cond_resched();
 772         }
 773 }
 774
 775 /*
 776  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 777  * if the result is positive, it's an NVM Express status code
 778  */
 779 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 780                 union nvme_result *result, void *buffer, unsigned bufflen,
 781                 unsigned timeout, int qid, int at_head,
 782                 blk_mq_req_flags_t flags, bool poll)
 783 {
 784         struct request *req;
 785         int ret;
 786
 787         req = nvme_alloc_request(q, cmd, flags, qid);
 788         if (IS_ERR(req))
 789                 return PTR_ERR(req);
 790
 791         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 792
 793         if (buffer && bufflen) {
 794                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 795                 if (ret)
 796                         goto out;
 797         }
 798
 799         if (poll)
 800                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
 801         else
 802                 blk_execute_rq(req->q, NULL, req, at_head);
 803         if (result)
 804                 *result = nvme_req(req)->result;
 805         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 806                 ret = -EINTR;
 807         else
 808                 ret = nvme_req(req)->status;
 809  out:
 810         blk_mq_free_request(req);
 811         return ret;
 812 }
 813 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 814
 815 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 816                 void *buffer, unsigned bufflen)
 817 {
 818         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 819                         NVME_QID_ANY, 0, 0, false);
 820 }
 821 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 822
 823 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 824                 unsigned len, u32 seed, bool write)
 825 {
 826         struct bio_integrity_payload *bip;
 827         int ret = -ENOMEM;
 828         void *buf;
 829
 830         buf = kmalloc(len, GFP_KERNEL);
 831         if (!buf)
 832                 goto out;
 833
 834         ret = -EFAULT;
 835         if (write && copy_from_user(buf, ubuf, len))
 836                 goto out_free_meta;
 837
 838         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 839         if (IS_ERR(bip)) {
 840                 ret = PTR_ERR(bip);
 841                 goto out_free_meta;
 842         }
 843
 844         bip->bip_iter.bi_size = len;
 845         bip->bip_iter.bi_sector = seed;
 846         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 847                         offset_in_page(buf));
 848         if (ret == len)
 849                 return buf;
 850         ret = -ENOMEM;
 851 out_free_meta:
 852         kfree(buf);
 853 out:
 854         return ERR_PTR(ret);
 855 }
 856
 857 static int nvme_submit_user_cmd(struct request_queue *q,
 858                 struct nvme_command *cmd, void __user *ubuffer,
 859                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 860                 u32 meta_seed, u32 *result, unsigned timeout)
 861 {
 862         bool write = nvme_is_write(cmd);
 863         struct nvme_ns *ns = q->queuedata;
 864         struct gendisk *disk = ns ? ns->disk : NULL;
 865         struct request *req;
 866         struct bio *bio = NULL;
 867         void *meta = NULL;
 868         int ret;
 869
 870         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 871         if (IS_ERR(req))
 872                 return PTR_ERR(req);
 873
 874         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 875         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 876
 877         if (ubuffer && bufflen) {
 878                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 879                                 GFP_KERNEL);
 880                 if (ret)
 881                         goto out;
 882                 bio = req->bio;
 883                 bio->bi_disk = disk;
 884                 if (disk && meta_buffer && meta_len) {
 885                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 886                                         meta_seed, write);
 887                         if (IS_ERR(meta)) {
 888                                 ret = PTR_ERR(meta);
 889                                 goto out_unmap;
 890                         }
 891                         req->cmd_flags |= REQ_INTEGRITY;
 892                 }
 893         }
 894
 895         blk_execute_rq(req->q, disk, req, 0);
 896         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 897                 ret = -EINTR;
 898         else
 899                 ret = nvme_req(req)->status;
 900         if (result)
 901                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 902         if (meta && !ret && !write) {
 903                 if (copy_to_user(meta_buffer, meta, meta_len))
 904                         ret = -EFAULT;
 905         }
 906         kfree(meta);
 907  out_unmap:
 908         if (bio)
 909                 blk_rq_unmap_user(bio);
 910  out:
 911         blk_mq_free_request(req);
 912         return ret;
 913 }
 914
 915 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 916 {
 917         struct nvme_ctrl *ctrl = rq->end_io_data;
 918         unsigned long flags;
 919         bool startka = false;
 920
 921         blk_mq_free_request(rq);
 922
 923         if (status) {
 924                 dev_err(ctrl->device,
 925                         "failed nvme_keep_alive_end_io error=%d\n",
 926                                 status);
 927                 return;
 928         }
 929
 930         ctrl->comp_seen = false;
 931         spin_lock_irqsave(&ctrl->lock, flags);
 932         if (ctrl->state == NVME_CTRL_LIVE ||
 933             ctrl->state == NVME_CTRL_CONNECTING)
 934                 startka = true;
 935         spin_unlock_irqrestore(&ctrl->lock, flags);
 936         if (startka)
 937                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 938 }
 939
 940 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 941 {
 942         struct request *rq;
 943
 944         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 945                         NVME_QID_ANY);
 946         if (IS_ERR(rq))
 947                 return PTR_ERR(rq);
 948
 949         rq->timeout = ctrl->kato * HZ;
 950         rq->end_io_data = ctrl;
 951
 952         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 953
 954         return 0;
 955 }
 956
 957 static void nvme_keep_alive_work(struct work_struct *work)
 958 {
 959         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 960                         struct nvme_ctrl, ka_work);
 961         bool comp_seen = ctrl->comp_seen;
 962
 963         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
 964                 dev_dbg(ctrl->device,
 965                         "reschedule traffic based keep-alive timer\n");
 966                 ctrl->comp_seen = false;
 967                 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 968                 return;
 969         }
 970
 971         if (nvme_keep_alive(ctrl)) {
 972                 /* allocation failure, reset the controller */
 973                 dev_err(ctrl->device, "keep-alive failed\n");
 974                 nvme_reset_ctrl(ctrl);
 975                 return;
 976         }
 977 }
 978
 979 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 980 {
 981         if (unlikely(ctrl->kato == 0))
 982                 return;
 983
 984         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 985 }
 986
 987 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 988 {
 989         if (unlikely(ctrl->kato == 0))
 990                 return;
 991
 992         cancel_delayed_work_sync(&ctrl->ka_work);
 993 }
 994 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 995
 996 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 997 {
 998         struct nvme_command c = { };
 999         int error;
1000
1001         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1002         c.identify.opcode = nvme_admin_identify;
1003         c.identify.cns = NVME_ID_CNS_CTRL;
1004
1005         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1006         if (!*id)
1007                 return -ENOMEM;
1008
1009         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1010                         sizeof(struct nvme_id_ctrl));
1011         if (error)
1012                 kfree(*id);
1013         return error;
1014 }
1015
1016 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1017                 struct nvme_ns_ids *ids)
1018 {
1019         struct nvme_command c = { };
1020         int status;
1021         void *data;
1022         int pos;
1023         int len;
1024
1025         c.identify.opcode = nvme_admin_identify;
1026         c.identify.nsid = cpu_to_le32(nsid);
1027         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1028
1029         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1030         if (!data)
1031                 return -ENOMEM;
1032
1033         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1034                                       NVME_IDENTIFY_DATA_SIZE);
1035         if (status)
1036                 goto free_data;
1037
1038         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1039                 struct nvme_ns_id_desc *cur = data + pos;
1040
1041                 if (cur->nidl == 0)
1042                         break;
1043
1044                 switch (cur->nidt) {
1045                 case NVME_NIDT_EUI64:
1046                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1047                                 dev_warn(ctrl->device,
1048                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1049                                          cur->nidl);
1050                                 goto free_data;
1051                         }
1052                         len = NVME_NIDT_EUI64_LEN;
1053                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1054                         break;
1055                 case NVME_NIDT_NGUID:
1056                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1057                                 dev_warn(ctrl->device,
1058                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1059                                          cur->nidl);
1060                                 goto free_data;
1061                         }
1062                         len = NVME_NIDT_NGUID_LEN;
1063                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1064                         break;
1065                 case NVME_NIDT_UUID:
1066                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
1067                                 dev_warn(ctrl->device,
1068                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1069                                          cur->nidl);
1070                                 goto free_data;
1071                         }
1072                         len = NVME_NIDT_UUID_LEN;
1073                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1074                         break;
1075                 default:
1076                         /* Skip unknown types */
1077                         len = cur->nidl;
1078                         break;
1079                 }
1080
1081                 len += sizeof(*cur);
1082         }
1083 free_data:
1084         kfree(data);
1085         return status;
1086 }
1087
1088 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1089 {
1090         struct nvme_command c = { };
1091
1092         c.identify.opcode = nvme_admin_identify;
1093         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1094         c.identify.nsid = cpu_to_le32(nsid);
1095         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1096                                     NVME_IDENTIFY_DATA_SIZE);
1097 }
1098
1099 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1100                 unsigned nsid, struct nvme_id_ns **id)
1101 {
1102         struct nvme_command c = { };
1103         int error;
1104
1105         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1106         c.identify.opcode = nvme_admin_identify;
1107         c.identify.nsid = cpu_to_le32(nsid);
1108         c.identify.cns = NVME_ID_CNS_NS;
1109
1110         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1111         if (!*id)
1112                 return -ENOMEM;
1113
1114         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1115         if (error) {
1116                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1117                 kfree(*id);
1118         }
1119
1120         return error;
1121 }
1122
1123 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1124                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1125 {
1126         struct nvme_command c;
1127         union nvme_result res;
1128         int ret;
1129
1130         memset(&c, 0, sizeof(c));
1131         c.features.opcode = op;
1132         c.features.fid = cpu_to_le32(fid);
1133         c.features.dword11 = cpu_to_le32(dword11);
1134
1135         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1136                         buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1137         if (ret >= 0 && result)
1138                 *result = le32_to_cpu(res.u32);
1139         return ret;
1140 }
1141
1142 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1143                       unsigned int dword11, void *buffer, size_t buflen,
1144                       u32 *result)
1145 {
1146         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1147                              buflen, result);
1148 }
1149 EXPORT_SYMBOL_GPL(nvme_set_features);
1150
1151 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1152                       unsigned int dword11, void *buffer, size_t buflen,
1153                       u32 *result)
1154 {
1155         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1156                              buflen, result);
1157 }
1158 EXPORT_SYMBOL_GPL(nvme_get_features);
1159
1160 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1161 {
1162         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1163         u32 result;
1164         int status, nr_io_queues;
1165
1166         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1167                         &result);
1168         if (status < 0)
1169                 return status;
1170
1171         /*
1172          * Degraded controllers might return an error when setting the queue
1173          * count.  We still want to be able to bring them online and offer
1174          * access to the admin queue, as that might be only way to fix them up.
1175          */
1176         if (status > 0) {
1177                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1178                 *count = 0;
1179         } else {
1180                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1181                 *count = min(*count, nr_io_queues);
1182         }
1183
1184         return 0;
1185 }
1186 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1187
1188 #define NVME_AEN_SUPPORTED \
1189         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1190
1191 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1192 {
1193         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1194         int status;
1195
1196         if (!supported_aens)
1197                 return;
1198
1199         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1200                         NULL, 0, &result);
1201         if (status)
1202                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1203                          supported_aens);
1204 }
1205
1206 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1207 {
1208         struct nvme_user_io io;
1209         struct nvme_command c;
1210         unsigned length, meta_len;
1211         void __user *metadata;
1212
1213         if (copy_from_user(&io, uio, sizeof(io)))
1214                 return -EFAULT;
1215         if (io.flags)
1216                 return -EINVAL;
1217
1218         switch (io.opcode) {
1219         case nvme_cmd_write:
1220         case nvme_cmd_read:
1221         case nvme_cmd_compare:
1222                 break;
1223         default:
1224                 return -EINVAL;
1225         }
1226
1227         length = (io.nblocks + 1) << ns->lba_shift;
1228         meta_len = (io.nblocks + 1) * ns->ms;
1229         metadata = (void __user *)(uintptr_t)io.metadata;
1230
1231         if (ns->ext) {
1232                 length += meta_len;
1233                 meta_len = 0;
1234         } else if (meta_len) {
1235                 if ((io.metadata & 3) || !io.metadata)
1236                         return -EINVAL;
1237         }
1238
1239         memset(&c, 0, sizeof(c));
1240         c.rw.opcode = io.opcode;
1241         c.rw.flags = io.flags;
1242         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1243         c.rw.slba = cpu_to_le64(io.slba);
1244         c.rw.length = cpu_to_le16(io.nblocks);
1245         c.rw.control = cpu_to_le16(io.control);
1246         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1247         c.rw.reftag = cpu_to_le32(io.reftag);
1248         c.rw.apptag = cpu_to_le16(io.apptag);
1249         c.rw.appmask = cpu_to_le16(io.appmask);
1250
1251         return nvme_submit_user_cmd(ns->queue, &c,
1252                         (void __user *)(uintptr_t)io.addr, length,
1253                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1254 }
1255
1256 static u32 nvme_known_admin_effects(u8 opcode)
1257 {
1258         switch (opcode) {
1259         case nvme_admin_format_nvm:
1260                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1261                                         NVME_CMD_EFFECTS_CSE_MASK;
1262         case nvme_admin_sanitize_nvm:
1263                 return NVME_CMD_EFFECTS_CSE_MASK;
1264         default:
1265                 break;
1266         }
1267         return 0;
1268 }
1269
1270 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1271                                                                 u8 opcode)
1272 {
1273         u32 effects = 0;
1274
1275         if (ns) {
1276                 if (ctrl->effects)
1277                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1278                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1279                         dev_warn(ctrl->device,
1280                                  "IO command:%02x has unhandled effects:%08x\n",
1281                                  opcode, effects);
1282                 return 0;
1283         }
1284
1285         if (ctrl->effects)
1286                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1287         effects |= nvme_known_admin_effects(opcode);
1288
1289         /*
1290          * For simplicity, IO to all namespaces is quiesced even if the command
1291          * effects say only one namespace is affected.
1292          */
1293         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1294                 mutex_lock(&ctrl->scan_lock);
1295                 nvme_start_freeze(ctrl);
1296                 nvme_wait_freeze(ctrl);
1297         }
1298         return effects;
1299 }
1300
1301 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1302 {
1303         struct nvme_ns *ns;
1304
1305         down_read(&ctrl->namespaces_rwsem);
1306         list_for_each_entry(ns, &ctrl->namespaces, list)
1307                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1308                         nvme_set_queue_dying(ns);
1309         up_read(&ctrl->namespaces_rwsem);
1310
1311         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1312 }
1313
1314 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1315 {
1316         /*
1317          * Revalidate LBA changes prior to unfreezing. This is necessary to
1318          * prevent memory corruption if a logical block size was changed by
1319          * this command.
1320          */
1321         if (effects & NVME_CMD_EFFECTS_LBCC)
1322                 nvme_update_formats(ctrl);
1323         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1324                 nvme_unfreeze(ctrl);
1325                 mutex_unlock(&ctrl->scan_lock);
1326         }
1327         if (effects & NVME_CMD_EFFECTS_CCC)
1328                 nvme_init_identify(ctrl);
1329         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1330                 nvme_queue_scan(ctrl);
1331 }
1332
1333 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1334                         struct nvme_passthru_cmd __user *ucmd)
1335 {
1336         struct nvme_passthru_cmd cmd;
1337         struct nvme_command c;
1338         unsigned timeout = 0;
1339         u32 effects;
1340         int status;
1341
1342         if (!capable(CAP_SYS_ADMIN))
1343                 return -EACCES;
1344         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1345                 return -EFAULT;
1346         if (cmd.flags)
1347                 return -EINVAL;
1348
1349         memset(&c, 0, sizeof(c));
1350         c.common.opcode = cmd.opcode;
1351         c.common.flags = cmd.flags;
1352         c.common.nsid = cpu_to_le32(cmd.nsid);
1353         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1354         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1355         c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1356         c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1357         c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1358         c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1359         c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1360         c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1361
1362         if (cmd.timeout_ms)
1363                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1364
1365         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1366         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1367                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1368                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1369                         0, &cmd.result, timeout);
1370         nvme_passthru_end(ctrl, effects);
1371
1372         if (status >= 0) {
1373                 if (put_user(cmd.result, &ucmd->result))
1374                         return -EFAULT;
1375         }
1376
1377         return status;
1378 }
1379
1380 /*
1381  * Issue ioctl requests on the first available path.  Note that unlike normal
1382  * block layer requests we will not retry failed request on another controller.
1383  */
1384 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1385                 struct nvme_ns_head **head, int *srcu_idx)
1386 {
1387 #ifdef CONFIG_NVME_MULTIPATH
1388         if (disk->fops == &nvme_ns_head_ops) {
1389                 struct nvme_ns *ns;
1390
1391                 *head = disk->private_data;
1392                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1393                 ns = nvme_find_path(*head);
1394                 if (!ns)
1395                         srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1396                 return ns;
1397         }
1398 #endif
1399         *head = NULL;
1400         *srcu_idx = -1;
1401         return disk->private_data;
1402 }
1403
1404 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1405 {
1406         if (head)
1407                 srcu_read_unlock(&head->srcu, idx);
1408 }
1409
1410 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1411                 unsigned int cmd, unsigned long arg)
1412 {
1413         struct nvme_ns_head *head = NULL;
1414         void __user *argp = (void __user *)arg;
1415         struct nvme_ns *ns;
1416         int srcu_idx, ret;
1417
1418         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1419         if (unlikely(!ns))
1420                 return -EWOULDBLOCK;
1421
1422         /*
1423          * Handle ioctls that apply to the controller instead of the namespace
1424          * seperately and drop the ns SRCU reference early.  This avoids a
1425          * deadlock when deleting namespaces using the passthrough interface.
1426          */
1427         if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1428                 struct nvme_ctrl *ctrl = ns->ctrl;
1429
1430                 nvme_get_ctrl(ns->ctrl);
1431                 nvme_put_ns_from_disk(head, srcu_idx);
1432
1433                 if (cmd == NVME_IOCTL_ADMIN_CMD)
1434                         ret = nvme_user_cmd(ctrl, NULL, argp);
1435                 else
1436                         ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1437
1438                 nvme_put_ctrl(ctrl);
1439                 return ret;
1440         }
1441
1442         switch (cmd) {
1443         case NVME_IOCTL_ID:
1444                 force_successful_syscall_return();
1445                 ret = ns->head->ns_id;
1446                 break;
1447         case NVME_IOCTL_IO_CMD:
1448                 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1449                 break;
1450         case NVME_IOCTL_SUBMIT_IO:
1451                 ret = nvme_submit_io(ns, argp);
1452                 break;
1453         default:
1454                 if (ns->ndev)
1455                         ret = nvme_nvm_ioctl(ns, cmd, arg);
1456                 else
1457                         ret = -ENOTTY;
1458         }
1459
1460         nvme_put_ns_from_disk(head, srcu_idx);
1461         return ret;
1462 }
1463
1464 static int nvme_open(struct block_device *bdev, fmode_t mode)
1465 {
1466         struct nvme_ns *ns = bdev->bd_disk->private_data;
1467
1468 #ifdef CONFIG_NVME_MULTIPATH
1469         /* should never be called due to GENHD_FL_HIDDEN */
1470         if (WARN_ON_ONCE(ns->head->disk))
1471                 goto fail;
1472 #endif
1473         if (!kref_get_unless_zero(&ns->kref))
1474                 goto fail;
1475         if (!try_module_get(ns->ctrl->ops->module))
1476                 goto fail_put_ns;
1477
1478         return 0;
1479
1480 fail_put_ns:
1481         nvme_put_ns(ns);
1482 fail:
1483         return -ENXIO;
1484 }
1485
1486 static void nvme_release(struct gendisk *disk, fmode_t mode)
1487 {
1488         struct nvme_ns *ns = disk->private_data;
1489
1490         module_put(ns->ctrl->ops->module);
1491         nvme_put_ns(ns);
1492 }
1493
1494 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1495 {
1496         /* some standard values */
1497         geo->heads = 1 << 6;
1498         geo->sectors = 1 << 5;
1499         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1500         return 0;
1501 }
1502
1503 #ifdef CONFIG_BLK_DEV_INTEGRITY
1504 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1505 {
1506         struct blk_integrity integrity;
1507
1508         memset(&integrity, 0, sizeof(integrity));
1509         switch (pi_type) {
1510         case NVME_NS_DPS_PI_TYPE3:
1511                 integrity.profile = &t10_pi_type3_crc;
1512                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1513                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1514                 break;
1515         case NVME_NS_DPS_PI_TYPE1:
1516         case NVME_NS_DPS_PI_TYPE2:
1517                 integrity.profile = &t10_pi_type1_crc;
1518                 integrity.tag_size = sizeof(u16);
1519                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1520                 break;
1521         default:
1522                 integrity.profile = NULL;
1523                 break;
1524         }
1525         integrity.tuple_size = ms;
1526         blk_integrity_register(disk, &integrity);
1527         blk_queue_max_integrity_segments(disk->queue, 1);
1528 }
1529 #else
1530 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1531 {
1532 }
1533 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1534
1535 static void nvme_set_chunk_size(struct nvme_ns *ns)
1536 {
1537         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1538         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1539 }
1540
1541 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1542 {
1543         struct nvme_ctrl *ctrl = ns->ctrl;
1544         struct request_queue *queue = disk->queue;
1545         u32 size = queue_logical_block_size(queue);
1546
1547         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1548                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1549                 return;
1550         }
1551
1552         if (ctrl->nr_streams && ns->sws && ns->sgs)
1553                 size *= ns->sws * ns->sgs;
1554
1555         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1556                         NVME_DSM_MAX_RANGES);
1557
1558         queue->limits.discard_alignment = 0;
1559         queue->limits.discard_granularity = size;
1560
1561         /* If discard is already enabled, don't reset queue limits */
1562         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1563                 return;
1564
1565         blk_queue_max_discard_sectors(queue, UINT_MAX);
1566         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1567
1568         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1569                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1570 }
1571
1572 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1573 {
1574         u32 max_sectors;
1575         unsigned short bs = 1 << ns->lba_shift;
1576
1577         if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1578             (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1579                 return;
1580         /*
1581          * Even though NVMe spec explicitly states that MDTS is not
1582          * applicable to the write-zeroes:- "The restriction does not apply to
1583          * commands that do not transfer data between the host and the
1584          * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1585          * In order to be more cautious use controller's max_hw_sectors value
1586          * to configure the maximum sectors for the write-zeroes which is
1587          * configured based on the controller's MDTS field in the
1588          * nvme_init_identify() if available.
1589          */
1590         if (ns->ctrl->max_hw_sectors == UINT_MAX)
1591                 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1592         else
1593                 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1594
1595         blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1596 }
1597
1598 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1599                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1600 {
1601         int ret = 0;
1602
1603         memset(ids, 0, sizeof(*ids));
1604
1605         if (ctrl->vs >= NVME_VS(1, 1, 0))
1606                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1607         if (ctrl->vs >= NVME_VS(1, 2, 0))
1608                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1609         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1610                  /* Don't treat error as fatal we potentially
1611                   * already have a NGUID or EUI-64
1612                   */
1613                 ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1614                 if (ret)
1615                         dev_warn(ctrl->device,
1616                                  "Identify Descriptors failed (%d)\n", ret);
1617         }
1618         return ret;
1619 }
1620
1621 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1622 {
1623         return !uuid_is_null(&ids->uuid) ||
1624                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1625                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1626 }
1627
1628 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1629 {
1630         return uuid_equal(&a->uuid, &b->uuid) &&
1631                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1632                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1633 }
1634
1635 static void nvme_update_disk_info(struct gendisk *disk,
1636                 struct nvme_ns *ns, struct nvme_id_ns *id)
1637 {
1638         sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1639         unsigned short bs = 1 << ns->lba_shift;
1640         u32 atomic_bs, phys_bs, io_opt;
1641
1642         if (ns->lba_shift > PAGE_SHIFT) {
1643                 /* unsupported block size, set capacity to 0 later */
1644                 bs = (1 << 9);
1645         }
1646         blk_mq_freeze_queue(disk->queue);
1647         blk_integrity_unregister(disk);
1648
1649         if (id->nabo == 0) {
1650                 /*
1651                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1652                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1653                  * 0 then AWUPF must be used instead.
1654                  */
1655                 if (id->nsfeat & (1 << 1) && id->nawupf)
1656                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1657                 else
1658                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1659         } else {
1660                 atomic_bs = bs;
1661         }
1662         phys_bs = bs;
1663         io_opt = bs;
1664         if (id->nsfeat & (1 << 4)) {
1665                 /* NPWG = Namespace Preferred Write Granularity */
1666                 phys_bs *= 1 + le16_to_cpu(id->npwg);
1667                 /* NOWS = Namespace Optimal Write Size */
1668                 io_opt *= 1 + le16_to_cpu(id->nows);
1669         }
1670
1671         blk_queue_logical_block_size(disk->queue, bs);
1672         /*
1673          * Linux filesystems assume writing a single physical block is
1674          * an atomic operation. Hence limit the physical block size to the
1675          * value of the Atomic Write Unit Power Fail parameter.
1676          */
1677         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1678         blk_queue_io_min(disk->queue, phys_bs);
1679         blk_queue_io_opt(disk->queue, io_opt);
1680
1681         if (ns->ms && !ns->ext &&
1682             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1683                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1684         if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1685             ns->lba_shift > PAGE_SHIFT)
1686                 capacity = 0;
1687
1688         set_capacity(disk, capacity);
1689
1690         nvme_config_discard(disk, ns);
1691         nvme_config_write_zeroes(disk, ns);
1692
1693         if (id->nsattr & (1 << 0))
1694                 set_disk_ro(disk, true);
1695         else
1696                 set_disk_ro(disk, false);
1697
1698         blk_mq_unfreeze_queue(disk->queue);
1699 }
1700
1701 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1702 {
1703         struct nvme_ns *ns = disk->private_data;
1704
1705         /*
1706          * If identify namespace failed, use default 512 byte block size so
1707          * block layer can use before failing read/write for 0 capacity.
1708          */
1709         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1710         if (ns->lba_shift == 0)
1711                 ns->lba_shift = 9;
1712         ns->noiob = le16_to_cpu(id->noiob);
1713         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1714         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1715         /* the PI implementation requires metadata equal t10 pi tuple size */
1716         if (ns->ms == sizeof(struct t10_pi_tuple))
1717                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1718         else
1719                 ns->pi_type = 0;
1720
1721         if (ns->noiob)
1722                 nvme_set_chunk_size(ns);
1723         nvme_update_disk_info(disk, ns, id);
1724 #ifdef CONFIG_NVME_MULTIPATH
1725         if (ns->head->disk) {
1726                 nvme_update_disk_info(ns->head->disk, ns, id);
1727                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1728         }
1729 #endif
1730 }
1731
1732 static int nvme_revalidate_disk(struct gendisk *disk)
1733 {
1734         struct nvme_ns *ns = disk->private_data;
1735         struct nvme_ctrl *ctrl = ns->ctrl;
1736         struct nvme_id_ns *id;
1737         struct nvme_ns_ids ids;
1738         int ret = 0;
1739
1740         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1741                 set_capacity(disk, 0);
1742                 return -ENODEV;
1743         }
1744
1745         ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1746         if (ret)
1747                 goto out;
1748
1749         if (id->ncap == 0) {
1750                 ret = -ENODEV;
1751                 goto free_id;
1752         }
1753
1754         __nvme_revalidate_disk(disk, id);
1755         ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1756         if (ret)
1757                 goto free_id;
1758
1759         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1760                 dev_err(ctrl->device,
1761                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1762                 ret = -ENODEV;
1763         }
1764
1765 free_id:
1766         kfree(id);
1767 out:
1768         if (ret > 0)
1769                 ret = blk_status_to_errno(nvme_error_status(ret));
1770         return ret;
1771 }
1772
1773 static char nvme_pr_type(enum pr_type type)
1774 {
1775         switch (type) {
1776         case PR_WRITE_EXCLUSIVE:
1777                 return 1;
1778         case PR_EXCLUSIVE_ACCESS:
1779                 return 2;
1780         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1781                 return 3;
1782         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1783                 return 4;
1784         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1785                 return 5;
1786         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1787                 return 6;
1788         default:
1789                 return 0;
1790         }
1791 };
1792
1793 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1794                                 u64 key, u64 sa_key, u8 op)
1795 {
1796         struct nvme_ns_head *head = NULL;
1797         struct nvme_ns *ns;
1798         struct nvme_command c;
1799         int srcu_idx, ret;
1800         u8 data[16] = { 0, };
1801
1802         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1803         if (unlikely(!ns))
1804                 return -EWOULDBLOCK;
1805
1806         put_unaligned_le64(key, &data[0]);
1807         put_unaligned_le64(sa_key, &data[8]);
1808
1809         memset(&c, 0, sizeof(c));
1810         c.common.opcode = op;
1811         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1812         c.common.cdw10 = cpu_to_le32(cdw10);
1813
1814         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1815         nvme_put_ns_from_disk(head, srcu_idx);
1816         return ret;
1817 }
1818
1819 static int nvme_pr_register(struct block_device *bdev, u64 old,
1820                 u64 new, unsigned flags)
1821 {
1822         u32 cdw10;
1823
1824         if (flags & ~PR_FL_IGNORE_KEY)
1825                 return -EOPNOTSUPP;
1826
1827         cdw10 = old ? 2 : 0;
1828         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1829         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1830         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1831 }
1832
1833 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1834                 enum pr_type type, unsigned flags)
1835 {
1836         u32 cdw10;
1837
1838         if (flags & ~PR_FL_IGNORE_KEY)
1839                 return -EOPNOTSUPP;
1840
1841         cdw10 = nvme_pr_type(type) << 8;
1842         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1843         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1844 }
1845
1846 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1847                 enum pr_type type, bool abort)
1848 {
1849         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1850         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1851 }
1852
1853 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1854 {
1855         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1856         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1857 }
1858
1859 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1860 {
1861         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1862         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1863 }
1864
1865 static const struct pr_ops nvme_pr_ops = {
1866         .pr_register    = nvme_pr_register,
1867         .pr_reserve     = nvme_pr_reserve,
1868         .pr_release     = nvme_pr_release,
1869         .pr_preempt     = nvme_pr_preempt,
1870         .pr_clear       = nvme_pr_clear,
1871 };
1872
1873 #ifdef CONFIG_BLK_SED_OPAL
1874 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1875                 bool send)
1876 {
1877         struct nvme_ctrl *ctrl = data;
1878         struct nvme_command cmd;
1879
1880         memset(&cmd, 0, sizeof(cmd));
1881         if (send)
1882                 cmd.common.opcode = nvme_admin_security_send;
1883         else
1884                 cmd.common.opcode = nvme_admin_security_recv;
1885         cmd.common.nsid = 0;
1886         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1887         cmd.common.cdw11 = cpu_to_le32(len);
1888
1889         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1890                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1891 }
1892 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1893 #endif /* CONFIG_BLK_SED_OPAL */
1894
1895 static const struct block_device_operations nvme_fops = {
1896         .owner          = THIS_MODULE,
1897         .ioctl          = nvme_ioctl,
1898         .compat_ioctl   = nvme_ioctl,
1899         .open           = nvme_open,
1900         .release        = nvme_release,
1901         .getgeo         = nvme_getgeo,
1902         .revalidate_disk= nvme_revalidate_disk,
1903         .pr_ops         = &nvme_pr_ops,
1904 };
1905
1906 #ifdef CONFIG_NVME_MULTIPATH
1907 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1908 {
1909         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1910
1911         if (!kref_get_unless_zero(&head->ref))
1912                 return -ENXIO;
1913         return 0;
1914 }
1915
1916 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1917 {
1918         nvme_put_ns_head(disk->private_data);
1919 }
1920
1921 const struct block_device_operations nvme_ns_head_ops = {
1922         .owner          = THIS_MODULE,
1923         .open           = nvme_ns_head_open,
1924         .release        = nvme_ns_head_release,
1925         .ioctl          = nvme_ioctl,
1926         .compat_ioctl   = nvme_ioctl,
1927         .getgeo         = nvme_getgeo,
1928         .pr_ops         = &nvme_pr_ops,
1929 };
1930 #endif /* CONFIG_NVME_MULTIPATH */
1931
1932 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1933 {
1934         unsigned long timeout =
1935                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1936         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1937         int ret;
1938
1939         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1940                 if (csts == ~0)
1941                         return -ENODEV;
1942                 if ((csts & NVME_CSTS_RDY) == bit)
1943                         break;
1944
1945                 msleep(100);
1946                 if (fatal_signal_pending(current))
1947                         return -EINTR;
1948                 if (time_after(jiffies, timeout)) {
1949                         dev_err(ctrl->device,
1950                                 "Device not ready; aborting %s\n", enabled ?
1951                                                 "initialisation" : "reset");
1952                         return -ENODEV;
1953                 }
1954         }
1955
1956         return ret;
1957 }
1958
1959 /*
1960  * If the device has been passed off to us in an enabled state, just clear
1961  * the enabled bit.  The spec says we should set the 'shutdown notification
1962  * bits', but doing so may cause the device to complete commands to the
1963  * admin queue ... and we don't know what memory that might be pointing at!
1964  */
1965 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
1966 {
1967         int ret;
1968
1969         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1970         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1971
1972         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1973         if (ret)
1974                 return ret;
1975
1976         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1977                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1978
1979         return nvme_wait_ready(ctrl, ctrl->cap, false);
1980 }
1981 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1982
1983 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
1984 {
1985         /*
1986          * Default to a 4K page size, with the intention to update this
1987          * path in the future to accomodate architectures with differing
1988          * kernel and IO page sizes.
1989          */
1990         unsigned dev_page_min, page_shift = 12;
1991         int ret;
1992
1993         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
1994         if (ret) {
1995                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1996                 return ret;
1997         }
1998         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
1999
2000         if (page_shift < dev_page_min) {
2001                 dev_err(ctrl->device,
2002                         "Minimum device page size %u too large for host (%u)\n",
2003                         1 << dev_page_min, 1 << page_shift);
2004                 return -ENODEV;
2005         }
2006
2007         ctrl->page_size = 1 << page_shift;
2008
2009         ctrl->ctrl_config = NVME_CC_CSS_NVM;
2010         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2011         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2012         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2013         ctrl->ctrl_config |= NVME_CC_ENABLE;
2014
2015         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2016         if (ret)
2017                 return ret;
2018         return nvme_wait_ready(ctrl, ctrl->cap, true);
2019 }
2020 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2021
2022 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2023 {
2024         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2025         u32 csts;
2026         int ret;
2027
2028         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2029         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2030
2031         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2032         if (ret)
2033                 return ret;
2034
2035         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2036                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2037                         break;
2038
2039                 msleep(100);
2040                 if (fatal_signal_pending(current))
2041                         return -EINTR;
2042                 if (time_after(jiffies, timeout)) {
2043                         dev_err(ctrl->device,
2044                                 "Device shutdown incomplete; abort shutdown\n");
2045                         return -ENODEV;
2046                 }
2047         }
2048
2049         return ret;
2050 }
2051 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2052
2053 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2054                 struct request_queue *q)
2055 {
2056         bool vwc = false;
2057
2058         if (ctrl->max_hw_sectors) {
2059                 u32 max_segments =
2060                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2061
2062                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2063                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2064                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2065         }
2066         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2067             is_power_of_2(ctrl->max_hw_sectors))
2068                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2069         blk_queue_virt_boundary(q, ctrl->page_size - 1);
2070         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2071                 vwc = true;
2072         blk_queue_write_cache(q, vwc, vwc);
2073 }
2074
2075 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2076 {
2077         __le64 ts;
2078         int ret;
2079
2080         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2081                 return 0;
2082
2083         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2084         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2085                         NULL);
2086         if (ret)
2087                 dev_warn_once(ctrl->device,
2088                         "could not set timestamp (%d)\n", ret);
2089         return ret;
2090 }
2091
2092 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2093 {
2094         struct nvme_feat_host_behavior *host;
2095         int ret;
2096
2097         /* Don't bother enabling the feature if retry delay is not reported */
2098         if (!ctrl->crdt[0])
2099                 return 0;
2100
2101         host = kzalloc(sizeof(*host), GFP_KERNEL);
2102         if (!host)
2103                 return 0;
2104
2105         host->acre = NVME_ENABLE_ACRE;
2106         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2107                                 host, sizeof(*host), NULL);
2108         kfree(host);
2109         return ret;
2110 }
2111
2112 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2113 {
2114         /*
2115          * APST (Autonomous Power State Transition) lets us program a
2116          * table of power state transitions that the controller will
2117          * perform automatically.  We configure it with a simple
2118          * heuristic: we are willing to spend at most 2% of the time
2119          * transitioning between power states.  Therefore, when running
2120          * in any given state, we will enter the next lower-power
2121          * non-operational state after waiting 50 * (enlat + exlat)
2122          * microseconds, as long as that state's exit latency is under
2123          * the requested maximum latency.
2124          *
2125          * We will not autonomously enter any non-operational state for
2126          * which the total latency exceeds ps_max_latency_us.  Users
2127          * can set ps_max_latency_us to zero to turn off APST.
2128          */
2129
2130         unsigned apste;
2131         struct nvme_feat_auto_pst *table;
2132         u64 max_lat_us = 0;
2133         int max_ps = -1;
2134         int ret;
2135
2136         /*
2137          * If APST isn't supported or if we haven't been initialized yet,
2138          * then don't do anything.
2139          */
2140         if (!ctrl->apsta)
2141                 return 0;
2142
2143         if (ctrl->npss > 31) {
2144                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2145                 return 0;
2146         }
2147
2148         table = kzalloc(sizeof(*table), GFP_KERNEL);
2149         if (!table)
2150                 return 0;
2151
2152         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2153                 /* Turn off APST. */
2154                 apste = 0;
2155                 dev_dbg(ctrl->device, "APST disabled\n");
2156         } else {
2157                 __le64 target = cpu_to_le64(0);
2158                 int state;
2159
2160                 /*
2161                  * Walk through all states from lowest- to highest-power.
2162                  * According to the spec, lower-numbered states use more
2163                  * power.  NPSS, despite the name, is the index of the
2164                  * lowest-power state, not the number of states.
2165                  */
2166                 for (state = (int)ctrl->npss; state >= 0; state--) {
2167                         u64 total_latency_us, exit_latency_us, transition_ms;
2168
2169                         if (target)
2170                                 table->entries[state] = target;
2171
2172                         /*
2173                          * Don't allow transitions to the deepest state
2174                          * if it's quirked off.
2175                          */
2176                         if (state == ctrl->npss &&
2177                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2178                                 continue;
2179
2180                         /*
2181                          * Is this state a useful non-operational state for
2182                          * higher-power states to autonomously transition to?
2183                          */
2184                         if (!(ctrl->psd[state].flags &
2185                               NVME_PS_FLAGS_NON_OP_STATE))
2186                                 continue;
2187
2188                         exit_latency_us =
2189                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2190                         if (exit_latency_us > ctrl->ps_max_latency_us)
2191                                 continue;
2192
2193                         total_latency_us =
2194                                 exit_latency_us +
2195                                 le32_to_cpu(ctrl->psd[state].entry_lat);
2196
2197                         /*
2198                          * This state is good.  Use it as the APST idle
2199                          * target for higher power states.
2200                          */
2201                         transition_ms = total_latency_us + 19;
2202                         do_div(transition_ms, 20);
2203                         if (transition_ms > (1 << 24) - 1)
2204                                 transition_ms = (1 << 24) - 1;
2205
2206                         target = cpu_to_le64((state << 3) |
2207                                              (transition_ms << 8));
2208
2209                         if (max_ps == -1)
2210                                 max_ps = state;
2211
2212                         if (total_latency_us > max_lat_us)
2213                                 max_lat_us = total_latency_us;
2214                 }
2215
2216                 apste = 1;
2217
2218                 if (max_ps == -1) {
2219                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2220                 } else {
2221                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2222                                 max_ps, max_lat_us, (int)sizeof(*table), table);
2223                 }
2224         }
2225
2226         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2227                                 table, sizeof(*table), NULL);
2228         if (ret)
2229                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2230
2231         kfree(table);
2232         return ret;
2233 }
2234
2235 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2236 {
2237         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2238         u64 latency;
2239
2240         switch (val) {
2241         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2242         case PM_QOS_LATENCY_ANY:
2243                 latency = U64_MAX;
2244                 break;
2245
2246         default:
2247                 latency = val;
2248         }
2249
2250         if (ctrl->ps_max_latency_us != latency) {
2251                 ctrl->ps_max_latency_us = latency;
2252                 nvme_configure_apst(ctrl);
2253         }
2254 }
2255
2256 struct nvme_core_quirk_entry {
2257         /*
2258          * NVMe model and firmware strings are padded with spaces.  For
2259          * simplicity, strings in the quirk table are padded with NULLs
2260          * instead.
2261          */
2262         u16 vid;
2263         const char *mn;
2264         const char *fr;
2265         unsigned long quirks;
2266 };
2267
2268 static const struct nvme_core_quirk_entry core_quirks[] = {
2269         {
2270                 /*
2271                  * This Toshiba device seems to die using any APST states.  See:
2272                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2273                  */
2274                 .vid = 0x1179,
2275                 .mn = "THNSF5256GPUK TOSHIBA",
2276                 .quirks = NVME_QUIRK_NO_APST,
2277         }
2278 };
2279
2280 /* match is null-terminated but idstr is space-padded. */
2281 static bool string_matches(const char *idstr, const char *match, size_t len)
2282 {
2283         size_t matchlen;
2284
2285         if (!match)
2286                 return true;
2287
2288         matchlen = strlen(match);
2289         WARN_ON_ONCE(matchlen > len);
2290
2291         if (memcmp(idstr, match, matchlen))
2292                 return false;
2293
2294         for (; matchlen < len; matchlen++)
2295                 if (idstr[matchlen] != ' ')
2296                         return false;
2297
2298         return true;
2299 }
2300
2301 static bool quirk_matches(const struct nvme_id_ctrl *id,
2302                           const struct nvme_core_quirk_entry *q)
2303 {
2304         return q->vid == le16_to_cpu(id->vid) &&
2305                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2306                 string_matches(id->fr, q->fr, sizeof(id->fr));
2307 }
2308
2309 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2310                 struct nvme_id_ctrl *id)
2311 {
2312         size_t nqnlen;
2313         int off;
2314
2315         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2316                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2317                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2318                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2319                         return;
2320                 }
2321
2322                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2323                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2324         }
2325
2326         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2327         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2328                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2329                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2330         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2331         off += sizeof(id->sn);
2332         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2333         off += sizeof(id->mn);
2334         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2335 }
2336
2337 static void nvme_release_subsystem(struct device *dev)
2338 {
2339         struct nvme_subsystem *subsys =
2340                 container_of(dev, struct nvme_subsystem, dev);
2341
2342         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2343         kfree(subsys);
2344 }
2345
2346 static void nvme_destroy_subsystem(struct kref *ref)
2347 {
2348         struct nvme_subsystem *subsys =
2349                         container_of(ref, struct nvme_subsystem, ref);
2350
2351         mutex_lock(&nvme_subsystems_lock);
2352         list_del(&subsys->entry);
2353         mutex_unlock(&nvme_subsystems_lock);
2354
2355         ida_destroy(&subsys->ns_ida);
2356         device_del(&subsys->dev);
2357         put_device(&subsys->dev);
2358 }
2359
2360 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2361 {
2362         kref_put(&subsys->ref, nvme_destroy_subsystem);
2363 }
2364
2365 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2366 {
2367         struct nvme_subsystem *subsys;
2368
2369         lockdep_assert_held(&nvme_subsystems_lock);
2370
2371         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2372                 if (strcmp(subsys->subnqn, subsysnqn))
2373                         continue;
2374                 if (!kref_get_unless_zero(&subsys->ref))
2375                         continue;
2376                 return subsys;
2377         }
2378
2379         return NULL;
2380 }
2381
2382 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2383         struct device_attribute subsys_attr_##_name = \
2384                 __ATTR(_name, _mode, _show, NULL)
2385
2386 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2387                                     struct device_attribute *attr,
2388                                     char *buf)
2389 {
2390         struct nvme_subsystem *subsys =
2391                 container_of(dev, struct nvme_subsystem, dev);
2392
2393         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2394 }
2395 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2396
2397 #define nvme_subsys_show_str_function(field)                            \
2398 static ssize_t subsys_##field##_show(struct device *dev,                \
2399                             struct device_attribute *attr, char *buf)   \
2400 {                                                                       \
2401         struct nvme_subsystem *subsys =                                 \
2402                 container_of(dev, struct nvme_subsystem, dev);          \
2403         return sprintf(buf, "%.*s\n",                                   \
2404                        (int)sizeof(subsys->field), subsys->field);      \
2405 }                                                                       \
2406 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2407
2408 nvme_subsys_show_str_function(model);
2409 nvme_subsys_show_str_function(serial);
2410 nvme_subsys_show_str_function(firmware_rev);
2411
2412 static struct attribute *nvme_subsys_attrs[] = {
2413         &subsys_attr_model.attr,
2414         &subsys_attr_serial.attr,
2415         &subsys_attr_firmware_rev.attr,
2416         &subsys_attr_subsysnqn.attr,
2417 #ifdef CONFIG_NVME_MULTIPATH
2418         &subsys_attr_iopolicy.attr,
2419 #endif
2420         NULL,
2421 };
2422
2423 static struct attribute_group nvme_subsys_attrs_group = {
2424         .attrs = nvme_subsys_attrs,
2425 };
2426
2427 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2428         &nvme_subsys_attrs_group,
2429         NULL,
2430 };
2431
2432 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2433                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2434 {
2435         struct nvme_ctrl *tmp;
2436
2437         lockdep_assert_held(&nvme_subsystems_lock);
2438
2439         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2440                 if (tmp->state == NVME_CTRL_DELETING ||
2441                     tmp->state == NVME_CTRL_DEAD)
2442                         continue;
2443
2444                 if (tmp->cntlid == ctrl->cntlid) {
2445                         dev_err(ctrl->device,
2446                                 "Duplicate cntlid %u with %s, rejecting\n",
2447                                 ctrl->cntlid, dev_name(tmp->device));
2448                         return false;
2449                 }
2450
2451                 if ((id->cmic & (1 << 1)) ||
2452                     (ctrl->opts && ctrl->opts->discovery_nqn))
2453                         continue;
2454
2455                 dev_err(ctrl->device,
2456                         "Subsystem does not support multiple controllers\n");
2457                 return false;
2458         }
2459
2460         return true;
2461 }
2462
2463 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2464 {
2465         struct nvme_subsystem *subsys, *found;
2466         int ret;
2467
2468         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2469         if (!subsys)
2470                 return -ENOMEM;
2471         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2472         if (ret < 0) {
2473                 kfree(subsys);
2474                 return ret;
2475         }
2476         subsys->instance = ret;
2477         mutex_init(&subsys->lock);
2478         kref_init(&subsys->ref);
2479         INIT_LIST_HEAD(&subsys->ctrls);
2480         INIT_LIST_HEAD(&subsys->nsheads);
2481         nvme_init_subnqn(subsys, ctrl, id);
2482         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2483         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2484         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2485         subsys->vendor_id = le16_to_cpu(id->vid);
2486         subsys->cmic = id->cmic;
2487         subsys->awupf = le16_to_cpu(id->awupf);
2488 #ifdef CONFIG_NVME_MULTIPATH
2489         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2490 #endif
2491
2492         subsys->dev.class = nvme_subsys_class;
2493         subsys->dev.release = nvme_release_subsystem;
2494         subsys->dev.groups = nvme_subsys_attrs_groups;
2495         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2496         device_initialize(&subsys->dev);
2497
2498         mutex_lock(&nvme_subsystems_lock);
2499         found = __nvme_find_get_subsystem(subsys->subnqn);
2500         if (found) {
2501                 put_device(&subsys->dev);
2502                 subsys = found;
2503
2504                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2505                         ret = -EINVAL;
2506                         goto out_put_subsystem;
2507                 }
2508         } else {
2509                 ret = device_add(&subsys->dev);
2510                 if (ret) {
2511                         dev_err(ctrl->device,
2512                                 "failed to register subsystem device.\n");
2513                         goto out_unlock;
2514                 }
2515                 ida_init(&subsys->ns_ida);
2516                 list_add_tail(&subsys->entry, &nvme_subsystems);
2517         }
2518
2519         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2520                         dev_name(ctrl->device))) {
2521                 dev_err(ctrl->device,
2522                         "failed to create sysfs link from subsystem.\n");
2523                 goto out_put_subsystem;
2524         }
2525
2526         ctrl->subsys = subsys;
2527         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2528         mutex_unlock(&nvme_subsystems_lock);
2529         return 0;
2530
2531 out_put_subsystem:
2532         nvme_put_subsystem(subsys);
2533 out_unlock:
2534         mutex_unlock(&nvme_subsystems_lock);
2535         put_device(&subsys->dev);
2536         return ret;
2537 }
2538
2539 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2540                 void *log, size_t size, u64 offset)
2541 {
2542         struct nvme_command c = { };
2543         unsigned long dwlen = size / 4 - 1;
2544
2545         c.get_log_page.opcode = nvme_admin_get_log_page;
2546         c.get_log_page.nsid = cpu_to_le32(nsid);
2547         c.get_log_page.lid = log_page;
2548         c.get_log_page.lsp = lsp;
2549         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2550         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2551         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2552         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2553
2554         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2555 }
2556
2557 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2558 {
2559         int ret;
2560
2561         if (!ctrl->effects)
2562                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2563
2564         if (!ctrl->effects)
2565                 return 0;
2566
2567         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2568                         ctrl->effects, sizeof(*ctrl->effects), 0);
2569         if (ret) {
2570                 kfree(ctrl->effects);
2571                 ctrl->effects = NULL;
2572         }
2573         return ret;
2574 }
2575
2576 /*
2577  * Initialize the cached copies of the Identify data and various controller
2578  * register in our nvme_ctrl structure.  This should be called as soon as
2579  * the admin queue is fully up and running.
2580  */
2581 int nvme_init_identify(struct nvme_ctrl *ctrl)
2582 {
2583         struct nvme_id_ctrl *id;
2584         int ret, page_shift;
2585         u32 max_hw_sectors;
2586         bool prev_apst_enabled;
2587
2588         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2589         if (ret) {
2590                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2591                 return ret;
2592         }
2593         page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2594         ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2595
2596         if (ctrl->vs >= NVME_VS(1, 1, 0))
2597                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2598
2599         ret = nvme_identify_ctrl(ctrl, &id);
2600         if (ret) {
2601                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2602                 return -EIO;
2603         }
2604
2605         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2606                 ret = nvme_get_effects_log(ctrl);
2607                 if (ret < 0)
2608                         goto out_free;
2609         }
2610
2611         if (!ctrl->identified) {
2612                 int i;
2613
2614                 ret = nvme_init_subsystem(ctrl, id);
2615                 if (ret)
2616                         goto out_free;
2617
2618                 /*
2619                  * Check for quirks.  Quirk can depend on firmware version,
2620                  * so, in principle, the set of quirks present can change
2621                  * across a reset.  As a possible future enhancement, we
2622                  * could re-scan for quirks every time we reinitialize
2623                  * the device, but we'd have to make sure that the driver
2624                  * behaves intelligently if the quirks change.
2625                  */
2626                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2627                         if (quirk_matches(id, &core_quirks[i]))
2628                                 ctrl->quirks |= core_quirks[i].quirks;
2629                 }
2630         }
2631
2632         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2633                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2634                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2635         }
2636
2637         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2638         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2639         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2640
2641         ctrl->oacs = le16_to_cpu(id->oacs);
2642         ctrl->oncs = le16_to_cpu(id->oncs);
2643         ctrl->mtfa = le16_to_cpu(id->mtfa);
2644         ctrl->oaes = le32_to_cpu(id->oaes);
2645         atomic_set(&ctrl->abort_limit, id->acl + 1);
2646         ctrl->vwc = id->vwc;
2647         if (id->mdts)
2648                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2649         else
2650                 max_hw_sectors = UINT_MAX;
2651         ctrl->max_hw_sectors =
2652                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2653
2654         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2655         ctrl->sgls = le32_to_cpu(id->sgls);
2656         ctrl->kas = le16_to_cpu(id->kas);
2657         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2658         ctrl->ctratt = le32_to_cpu(id->ctratt);
2659
2660         if (id->rtd3e) {
2661                 /* us -> s */
2662                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2663
2664                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2665                                                  shutdown_timeout, 60);
2666
2667                 if (ctrl->shutdown_timeout != shutdown_timeout)
2668                         dev_info(ctrl->device,
2669                                  "Shutdown timeout set to %u seconds\n",
2670                                  ctrl->shutdown_timeout);
2671         } else
2672                 ctrl->shutdown_timeout = shutdown_timeout;
2673
2674         ctrl->npss = id->npss;
2675         ctrl->apsta = id->apsta;
2676         prev_apst_enabled = ctrl->apst_enabled;
2677         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2678                 if (force_apst && id->apsta) {
2679                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2680                         ctrl->apst_enabled = true;
2681                 } else {
2682                         ctrl->apst_enabled = false;
2683                 }
2684         } else {
2685                 ctrl->apst_enabled = id->apsta;
2686         }
2687         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2688
2689         if (ctrl->ops->flags & NVME_F_FABRICS) {
2690                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2691                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2692                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2693                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2694
2695                 /*
2696                  * In fabrics we need to verify the cntlid matches the
2697                  * admin connect
2698                  */
2699                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2700                         ret = -EINVAL;
2701                         goto out_free;
2702                 }
2703
2704                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2705                         dev_err(ctrl->device,
2706                                 "keep-alive support is mandatory for fabrics\n");
2707                         ret = -EINVAL;
2708                         goto out_free;
2709                 }
2710         } else {
2711                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2712                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2713                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2714                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2715                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2716         }
2717
2718         ret = nvme_mpath_init(ctrl, id);
2719         kfree(id);
2720
2721         if (ret < 0)
2722                 return ret;
2723
2724         if (ctrl->apst_enabled && !prev_apst_enabled)
2725                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2726         else if (!ctrl->apst_enabled && prev_apst_enabled)
2727                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2728
2729         ret = nvme_configure_apst(ctrl);
2730         if (ret < 0)
2731                 return ret;
2732
2733         ret = nvme_configure_timestamp(ctrl);
2734         if (ret < 0)
2735                 return ret;
2736
2737         ret = nvme_configure_directives(ctrl);
2738         if (ret < 0)
2739                 return ret;
2740
2741         ret = nvme_configure_acre(ctrl);
2742         if (ret < 0)
2743                 return ret;
2744
2745         ctrl->identified = true;
2746
2747         return 0;
2748
2749 out_free:
2750         kfree(id);
2751         return ret;
2752 }
2753 EXPORT_SYMBOL_GPL(nvme_init_identify);
2754
2755 static int nvme_dev_open(struct inode *inode, struct file *file)
2756 {
2757         struct nvme_ctrl *ctrl =
2758                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2759
2760         switch (ctrl->state) {
2761         case NVME_CTRL_LIVE:
2762         case NVME_CTRL_ADMIN_ONLY:
2763                 break;
2764         default:
2765                 return -EWOULDBLOCK;
2766         }
2767
2768         file->private_data = ctrl;
2769         return 0;
2770 }
2771
2772 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2773 {
2774         struct nvme_ns *ns;
2775         int ret;
2776
2777         down_read(&ctrl->namespaces_rwsem);
2778         if (list_empty(&ctrl->namespaces)) {
2779                 ret = -ENOTTY;
2780                 goto out_unlock;
2781         }
2782
2783         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2784         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2785                 dev_warn(ctrl->device,
2786                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2787                 ret = -EINVAL;
2788                 goto out_unlock;
2789         }
2790
2791         dev_warn(ctrl->device,
2792                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2793         kref_get(&ns->kref);
2794         up_read(&ctrl->namespaces_rwsem);
2795
2796         ret = nvme_user_cmd(ctrl, ns, argp);
2797         nvme_put_ns(ns);
2798         return ret;
2799
2800 out_unlock:
2801         up_read(&ctrl->namespaces_rwsem);
2802         return ret;
2803 }
2804
2805 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2806                 unsigned long arg)
2807 {
2808         struct nvme_ctrl *ctrl = file->private_data;
2809         void __user *argp = (void __user *)arg;
2810
2811         switch (cmd) {
2812         case NVME_IOCTL_ADMIN_CMD:
2813                 return nvme_user_cmd(ctrl, NULL, argp);
2814         case NVME_IOCTL_IO_CMD:
2815                 return nvme_dev_user_cmd(ctrl, argp);
2816         case NVME_IOCTL_RESET:
2817                 dev_warn(ctrl->device, "resetting controller\n");
2818                 return nvme_reset_ctrl_sync(ctrl);
2819         case NVME_IOCTL_SUBSYS_RESET:
2820                 return nvme_reset_subsystem(ctrl);
2821         case NVME_IOCTL_RESCAN:
2822                 nvme_queue_scan(ctrl);
2823                 return 0;
2824         default:
2825                 return -ENOTTY;
2826         }
2827 }
2828
2829 static const struct file_operations nvme_dev_fops = {
2830         .owner          = THIS_MODULE,
2831         .open           = nvme_dev_open,
2832         .unlocked_ioctl = nvme_dev_ioctl,
2833         .compat_ioctl   = nvme_dev_ioctl,
2834 };
2835
2836 static ssize_t nvme_sysfs_reset(struct device *dev,
2837                                 struct device_attribute *attr, const char *buf,
2838                                 size_t count)
2839 {
2840         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2841         int ret;
2842
2843         ret = nvme_reset_ctrl_sync(ctrl);
2844         if (ret < 0)
2845                 return ret;
2846         return count;
2847 }
2848 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2849
2850 static ssize_t nvme_sysfs_rescan(struct device *dev,
2851                                 struct device_attribute *attr, const char *buf,
2852                                 size_t count)
2853 {
2854         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2855
2856         nvme_queue_scan(ctrl);
2857         return count;
2858 }
2859 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2860
2861 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2862 {
2863         struct gendisk *disk = dev_to_disk(dev);
2864
2865         if (disk->fops == &nvme_fops)
2866                 return nvme_get_ns_from_dev(dev)->head;
2867         else
2868                 return disk->private_data;
2869 }
2870
2871 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2872                 char *buf)
2873 {
2874         struct nvme_ns_head *head = dev_to_ns_head(dev);
2875         struct nvme_ns_ids *ids = &head->ids;
2876         struct nvme_subsystem *subsys = head->subsys;
2877         int serial_len = sizeof(subsys->serial);
2878         int model_len = sizeof(subsys->model);
2879
2880         if (!uuid_is_null(&ids->uuid))
2881                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2882
2883         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2884                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2885
2886         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2887                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2888
2889         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2890                                   subsys->serial[serial_len - 1] == '\0'))
2891                 serial_len--;
2892         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2893                                  subsys->model[model_len - 1] == '\0'))
2894                 model_len--;
2895
2896         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2897                 serial_len, subsys->serial, model_len, subsys->model,
2898                 head->ns_id);
2899 }
2900 static DEVICE_ATTR_RO(wwid);
2901
2902 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2903                 char *buf)
2904 {
2905         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2906 }
2907 static DEVICE_ATTR_RO(nguid);
2908
2909 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2910                 char *buf)
2911 {
2912         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2913
2914         /* For backward compatibility expose the NGUID to userspace if
2915          * we have no UUID set
2916          */
2917         if (uuid_is_null(&ids->uuid)) {
2918                 printk_ratelimited(KERN_WARNING
2919                                    "No UUID available providing old NGUID\n");
2920                 return sprintf(buf, "%pU\n", ids->nguid);
2921         }
2922         return sprintf(buf, "%pU\n", &ids->uuid);
2923 }
2924 static DEVICE_ATTR_RO(uuid);
2925
2926 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2927                 char *buf)
2928 {
2929         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2930 }
2931 static DEVICE_ATTR_RO(eui);
2932
2933 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2934                 char *buf)
2935 {
2936         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2937 }
2938 static DEVICE_ATTR_RO(nsid);
2939
2940 static struct attribute *nvme_ns_id_attrs[] = {
2941         &dev_attr_wwid.attr,
2942         &dev_attr_uuid.attr,
2943         &dev_attr_nguid.attr,
2944         &dev_attr_eui.attr,
2945         &dev_attr_nsid.attr,
2946 #ifdef CONFIG_NVME_MULTIPATH
2947         &dev_attr_ana_grpid.attr,
2948         &dev_attr_ana_state.attr,
2949 #endif
2950         NULL,
2951 };
2952
2953 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2954                 struct attribute *a, int n)
2955 {
2956         struct device *dev = container_of(kobj, struct device, kobj);
2957         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2958
2959         if (a == &dev_attr_uuid.attr) {
2960                 if (uuid_is_null(&ids->uuid) &&
2961                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2962                         return 0;
2963         }
2964         if (a == &dev_attr_nguid.attr) {
2965                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2966                         return 0;
2967         }
2968         if (a == &dev_attr_eui.attr) {
2969                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2970                         return 0;
2971         }
2972 #ifdef CONFIG_NVME_MULTIPATH
2973         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2974                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2975                         return 0;
2976                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2977                         return 0;
2978         }
2979 #endif
2980         return a->mode;
2981 }
2982
2983 static const struct attribute_group nvme_ns_id_attr_group = {
2984         .attrs          = nvme_ns_id_attrs,
2985         .is_visible     = nvme_ns_id_attrs_are_visible,
2986 };
2987
2988 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2989         &nvme_ns_id_attr_group,
2990 #ifdef CONFIG_NVM
2991         &nvme_nvm_attr_group,
2992 #endif
2993         NULL,
2994 };
2995
2996 #define nvme_show_str_function(field)                                           \
2997 static ssize_t  field##_show(struct device *dev,                                \
2998                             struct device_attribute *attr, char *buf)           \
2999 {                                                                               \
3000         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3001         return sprintf(buf, "%.*s\n",                                           \
3002                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3003 }                                                                               \
3004 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3005
3006 nvme_show_str_function(model);
3007 nvme_show_str_function(serial);
3008 nvme_show_str_function(firmware_rev);
3009
3010 #define nvme_show_int_function(field)                                           \
3011 static ssize_t  field##_show(struct device *dev,                                \
3012                             struct device_attribute *attr, char *buf)           \
3013 {                                                                               \
3014         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3015         return sprintf(buf, "%d\n", ctrl->field);       \
3016 }                                                                               \
3017 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3018
3019 nvme_show_int_function(cntlid);
3020 nvme_show_int_function(numa_node);
3021
3022 static ssize_t nvme_sysfs_delete(struct device *dev,
3023                                 struct device_attribute *attr, const char *buf,
3024                                 size_t count)
3025 {
3026         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3027
3028         if (device_remove_file_self(dev, attr))
3029                 nvme_delete_ctrl_sync(ctrl);
3030         return count;
3031 }
3032 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3033
3034 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3035                                          struct device_attribute *attr,
3036                                          char *buf)
3037 {
3038         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3039
3040         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3041 }
3042 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3043
3044 static ssize_t nvme_sysfs_show_state(struct device *dev,
3045                                      struct device_attribute *attr,
3046                                      char *buf)
3047 {
3048         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3049         static const char *const state_name[] = {
3050                 [NVME_CTRL_NEW]         = "new",
3051                 [NVME_CTRL_LIVE]        = "live",
3052                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
3053                 [NVME_CTRL_RESETTING]   = "resetting",
3054                 [NVME_CTRL_CONNECTING]  = "connecting",
3055                 [NVME_CTRL_DELETING]    = "deleting",
3056                 [NVME_CTRL_DEAD]        = "dead",
3057         };
3058
3059         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3060             state_name[ctrl->state])
3061                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3062
3063         return sprintf(buf, "unknown state\n");
3064 }
3065
3066 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3067
3068 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3069                                          struct device_attribute *attr,
3070                                          char *buf)
3071 {
3072         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3073
3074         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3075 }
3076 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3077
3078 static ssize_t nvme_sysfs_show_address(struct device *dev,
3079                                          struct device_attribute *attr,
3080                                          char *buf)
3081 {
3082         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3083
3084         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3085 }
3086 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3087
3088 static struct attribute *nvme_dev_attrs[] = {
3089         &dev_attr_reset_controller.attr,
3090         &dev_attr_rescan_controller.attr,
3091         &dev_attr_model.attr,
3092         &dev_attr_serial.attr,
3093         &dev_attr_firmware_rev.attr,
3094         &dev_attr_cntlid.attr,
3095         &dev_attr_delete_controller.attr,
3096         &dev_attr_transport.attr,
3097         &dev_attr_subsysnqn.attr,
3098         &dev_attr_address.attr,
3099         &dev_attr_state.attr,
3100         &dev_attr_numa_node.attr,
3101         NULL
3102 };
3103
3104 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3105                 struct attribute *a, int n)
3106 {
3107         struct device *dev = container_of(kobj, struct device, kobj);
3108         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3109
3110         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3111                 return 0;
3112         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3113                 return 0;
3114
3115         return a->mode;
3116 }
3117
3118 static struct attribute_group nvme_dev_attrs_group = {
3119         .attrs          = nvme_dev_attrs,
3120         .is_visible     = nvme_dev_attrs_are_visible,
3121 };
3122
3123 static const struct attribute_group *nvme_dev_attr_groups[] = {
3124         &nvme_dev_attrs_group,
3125         NULL,
3126 };
3127
3128 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3129                 unsigned nsid)
3130 {
3131         struct nvme_ns_head *h;
3132
3133         lockdep_assert_held(&subsys->lock);
3134
3135         list_for_each_entry(h, &subsys->nsheads, entry) {
3136                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3137                         return h;
3138         }
3139
3140         return NULL;
3141 }
3142
3143 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3144                 struct nvme_ns_head *new)
3145 {
3146         struct nvme_ns_head *h;
3147
3148         lockdep_assert_held(&subsys->lock);
3149
3150         list_for_each_entry(h, &subsys->nsheads, entry) {
3151                 if (nvme_ns_ids_valid(&new->ids) &&
3152                     !list_empty(&h->list) &&
3153                     nvme_ns_ids_equal(&new->ids, &h->ids))
3154                         return -EINVAL;
3155         }
3156
3157         return 0;
3158 }
3159
3160 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3161                 unsigned nsid, struct nvme_id_ns *id)
3162 {
3163         struct nvme_ns_head *head;
3164         size_t size = sizeof(*head);
3165         int ret = -ENOMEM;
3166
3167 #ifdef CONFIG_NVME_MULTIPATH
3168         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3169 #endif
3170
3171         head = kzalloc(size, GFP_KERNEL);
3172         if (!head)
3173                 goto out;
3174         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3175         if (ret < 0)
3176                 goto out_free_head;
3177         head->instance = ret;
3178         INIT_LIST_HEAD(&head->list);
3179         ret = init_srcu_struct(&head->srcu);
3180         if (ret)
3181                 goto out_ida_remove;
3182         head->subsys = ctrl->subsys;
3183         head->ns_id = nsid;
3184         kref_init(&head->ref);
3185
3186         ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3187         if (ret)
3188                 goto out_cleanup_srcu;
3189
3190         ret = __nvme_check_ids(ctrl->subsys, head);
3191         if (ret) {
3192                 dev_err(ctrl->device,
3193                         "duplicate IDs for nsid %d\n", nsid);
3194                 goto out_cleanup_srcu;
3195         }
3196
3197         ret = nvme_mpath_alloc_disk(ctrl, head);
3198         if (ret)
3199                 goto out_cleanup_srcu;
3200
3201         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3202
3203         kref_get(&ctrl->subsys->ref);
3204
3205         return head;
3206 out_cleanup_srcu:
3207         cleanup_srcu_struct(&head->srcu);
3208 out_ida_remove:
3209         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3210 out_free_head:
3211         kfree(head);
3212 out:
3213         if (ret > 0)
3214                 ret = blk_status_to_errno(nvme_error_status(ret));
3215         return ERR_PTR(ret);
3216 }
3217
3218 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3219                 struct nvme_id_ns *id)
3220 {
3221         struct nvme_ctrl *ctrl = ns->ctrl;
3222         bool is_shared = id->nmic & (1 << 0);
3223         struct nvme_ns_head *head = NULL;
3224         int ret = 0;
3225
3226         mutex_lock(&ctrl->subsys->lock);
3227         if (is_shared)
3228                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3229         if (!head) {
3230                 head = nvme_alloc_ns_head(ctrl, nsid, id);
3231                 if (IS_ERR(head)) {
3232                         ret = PTR_ERR(head);
3233                         goto out_unlock;
3234                 }
3235         } else {
3236                 struct nvme_ns_ids ids;
3237
3238                 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3239                 if (ret)
3240                         goto out_unlock;
3241
3242                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3243                         dev_err(ctrl->device,
3244                                 "IDs don't match for shared namespace %d\n",
3245                                         nsid);
3246                         ret = -EINVAL;
3247                         goto out_unlock;
3248                 }
3249         }
3250
3251         list_add_tail(&ns->siblings, &head->list);
3252         ns->head = head;
3253
3254 out_unlock:
3255         mutex_unlock(&ctrl->subsys->lock);
3256         if (ret > 0)
3257                 ret = blk_status_to_errno(nvme_error_status(ret));
3258         return ret;
3259 }
3260
3261 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3262 {
3263         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3264         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3265
3266         return nsa->head->ns_id - nsb->head->ns_id;
3267 }
3268
3269 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3270 {
3271         struct nvme_ns *ns, *ret = NULL;
3272
3273         down_read(&ctrl->namespaces_rwsem);
3274         list_for_each_entry(ns, &ctrl->namespaces, list) {
3275                 if (ns->head->ns_id == nsid) {
3276                         if (!kref_get_unless_zero(&ns->kref))
3277                                 continue;
3278                         ret = ns;
3279                         break;
3280                 }
3281                 if (ns->head->ns_id > nsid)
3282                         break;
3283         }
3284         up_read(&ctrl->namespaces_rwsem);
3285         return ret;
3286 }
3287
3288 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3289 {
3290         struct streams_directive_params s;
3291         int ret;
3292
3293         if (!ctrl->nr_streams)
3294                 return 0;
3295
3296         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3297         if (ret)
3298                 return ret;
3299
3300         ns->sws = le32_to_cpu(s.sws);
3301         ns->sgs = le16_to_cpu(s.sgs);
3302
3303         if (ns->sws) {
3304                 unsigned int bs = 1 << ns->lba_shift;
3305
3306                 blk_queue_io_min(ns->queue, bs * ns->sws);
3307                 if (ns->sgs)
3308                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3309         }
3310
3311         return 0;
3312 }
3313
3314 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3315 {
3316         struct nvme_ns *ns;
3317         struct gendisk *disk;
3318         struct nvme_id_ns *id;
3319         char disk_name[DISK_NAME_LEN];
3320         int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3321
3322         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3323         if (!ns)
3324                 return -ENOMEM;
3325
3326         ns->queue = blk_mq_init_queue(ctrl->tagset);
3327         if (IS_ERR(ns->queue)) {
3328                 ret = PTR_ERR(ns->queue);
3329                 goto out_free_ns;
3330         }
3331
3332         if (ctrl->opts && ctrl->opts->data_digest)
3333                 ns->queue->backing_dev_info->capabilities
3334                         |= BDI_CAP_STABLE_WRITES;
3335
3336         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3337         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3338                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3339
3340         ns->queue->queuedata = ns;
3341         ns->ctrl = ctrl;
3342
3343         kref_init(&ns->kref);
3344         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3345
3346         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3347         nvme_set_queue_limits(ctrl, ns->queue);
3348
3349         ret = nvme_identify_ns(ctrl, nsid, &id);
3350         if (ret)
3351                 goto out_free_queue;
3352
3353         if (id->ncap == 0) {
3354                 ret = -EINVAL;
3355                 goto out_free_id;
3356         }
3357
3358         ret = nvme_init_ns_head(ns, nsid, id);
3359         if (ret)
3360                 goto out_free_id;
3361         nvme_setup_streams_ns(ctrl, ns);
3362         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3363
3364         disk = alloc_disk_node(0, node);
3365         if (!disk) {
3366                 ret = -ENOMEM;
3367                 goto out_unlink_ns;
3368         }
3369
3370         disk->fops = &nvme_fops;
3371         disk->private_data = ns;
3372         disk->queue = ns->queue;
3373         disk->flags = flags;
3374         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3375         ns->disk = disk;
3376
3377         __nvme_revalidate_disk(disk, id);
3378
3379         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3380                 ret = nvme_nvm_register(ns, disk_name, node);
3381                 if (ret) {
3382                         dev_warn(ctrl->device, "LightNVM init failure\n");
3383                         goto out_put_disk;
3384                 }
3385         }
3386
3387         down_write(&ctrl->namespaces_rwsem);
3388         list_add_tail(&ns->list, &ctrl->namespaces);
3389         up_write(&ctrl->namespaces_rwsem);
3390
3391         nvme_get_ctrl(ctrl);
3392
3393         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3394
3395         nvme_mpath_add_disk(ns, id);
3396         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3397         kfree(id);
3398
3399         return 0;
3400  out_put_disk:
3401         put_disk(ns->disk);
3402  out_unlink_ns:
3403         mutex_lock(&ctrl->subsys->lock);
3404         list_del_rcu(&ns->siblings);
3405         mutex_unlock(&ctrl->subsys->lock);
3406         nvme_put_ns_head(ns->head);
3407  out_free_id:
3408         kfree(id);
3409  out_free_queue:
3410         blk_cleanup_queue(ns->queue);
3411  out_free_ns:
3412         kfree(ns);
3413         if (ret > 0)
3414                 ret = blk_status_to_errno(nvme_error_status(ret));
3415         return ret;
3416 }
3417
3418 static void nvme_ns_remove(struct nvme_ns *ns)
3419 {
3420         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3421                 return;
3422
3423         nvme_fault_inject_fini(&ns->fault_inject);
3424
3425         mutex_lock(&ns->ctrl->subsys->lock);
3426         list_del_rcu(&ns->siblings);
3427         mutex_unlock(&ns->ctrl->subsys->lock);
3428         synchronize_rcu(); /* guarantee not available in head->list */
3429         nvme_mpath_clear_current_path(ns);
3430         synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3431
3432         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3433                 del_gendisk(ns->disk);
3434                 blk_cleanup_queue(ns->queue);
3435                 if (blk_get_integrity(ns->disk))
3436                         blk_integrity_unregister(ns->disk);
3437         }
3438
3439         down_write(&ns->ctrl->namespaces_rwsem);
3440         list_del_init(&ns->list);
3441         up_write(&ns->ctrl->namespaces_rwsem);
3442
3443         nvme_mpath_check_last_path(ns);
3444         nvme_put_ns(ns);
3445 }
3446
3447 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3448 {
3449         struct nvme_ns *ns;
3450
3451         ns = nvme_find_get_ns(ctrl, nsid);
3452         if (ns) {
3453                 if (ns->disk && revalidate_disk(ns->disk))
3454                         nvme_ns_remove(ns);
3455                 nvme_put_ns(ns);
3456         } else
3457                 nvme_alloc_ns(ctrl, nsid);
3458 }
3459
3460 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3461                                         unsigned nsid)
3462 {
3463         struct nvme_ns *ns, *next;
3464         LIST_HEAD(rm_list);
3465
3466         down_write(&ctrl->namespaces_rwsem);
3467         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3468                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3469                         list_move_tail(&ns->list, &rm_list);
3470         }
3471         up_write(&ctrl->namespaces_rwsem);
3472
3473         list_for_each_entry_safe(ns, next, &rm_list, list)
3474                 nvme_ns_remove(ns);
3475
3476 }
3477
3478 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3479 {
3480         struct nvme_ns *ns;
3481         __le32 *ns_list;
3482         unsigned i, j, nsid, prev = 0;
3483         unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3484         int ret = 0;
3485
3486         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3487         if (!ns_list)
3488                 return -ENOMEM;
3489
3490         for (i = 0; i < num_lists; i++) {
3491                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3492                 if (ret)
3493                         goto free;
3494
3495                 for (j = 0; j < min(nn, 1024U); j++) {
3496                         nsid = le32_to_cpu(ns_list[j]);
3497                         if (!nsid)
3498                                 goto out;
3499
3500                         nvme_validate_ns(ctrl, nsid);
3501
3502                         while (++prev < nsid) {
3503                                 ns = nvme_find_get_ns(ctrl, prev);
3504                                 if (ns) {
3505                                         nvme_ns_remove(ns);
3506                                         nvme_put_ns(ns);
3507                                 }
3508                         }
3509                 }
3510                 nn -= j;
3511         }
3512  out:
3513         nvme_remove_invalid_namespaces(ctrl, prev);
3514  free:
3515         kfree(ns_list);
3516         return ret;
3517 }
3518
3519 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3520 {
3521         unsigned i;
3522
3523         for (i = 1; i <= nn; i++)
3524                 nvme_validate_ns(ctrl, i);
3525
3526         nvme_remove_invalid_namespaces(ctrl, nn);
3527 }
3528
3529 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3530 {
3531         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3532         __le32 *log;
3533         int error;
3534
3535         log = kzalloc(log_size, GFP_KERNEL);
3536         if (!log)
3537                 return;
3538
3539         /*
3540          * We need to read the log to clear the AEN, but we don't want to rely
3541          * on it for the changed namespace information as userspace could have
3542          * raced with us in reading the log page, which could cause us to miss
3543          * updates.
3544          */
3545         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3546                         log_size, 0);
3547         if (error)
3548                 dev_warn(ctrl->device,
3549                         "reading changed ns log failed: %d\n", error);
3550
3551         kfree(log);
3552 }
3553
3554 static void nvme_scan_work(struct work_struct *work)
3555 {
3556         struct nvme_ctrl *ctrl =
3557                 container_of(work, struct nvme_ctrl, scan_work);
3558         struct nvme_id_ctrl *id;
3559         unsigned nn;
3560
3561         if (ctrl->state != NVME_CTRL_LIVE)
3562                 return;
3563
3564         WARN_ON_ONCE(!ctrl->tagset);
3565
3566         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3567                 dev_info(ctrl->device, "rescanning namespaces.\n");
3568                 nvme_clear_changed_ns_log(ctrl);
3569         }
3570
3571         if (nvme_identify_ctrl(ctrl, &id))
3572                 return;
3573
3574         mutex_lock(&ctrl->scan_lock);
3575         nn = le32_to_cpu(id->nn);
3576         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3577             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3578                 if (!nvme_scan_ns_list(ctrl, nn))
3579                         goto out_free_id;
3580         }
3581         nvme_scan_ns_sequential(ctrl, nn);
3582 out_free_id:
3583         mutex_unlock(&ctrl->scan_lock);
3584         kfree(id);
3585         down_write(&ctrl->namespaces_rwsem);
3586         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3587         up_write(&ctrl->namespaces_rwsem);
3588 }
3589
3590 /*
3591  * This function iterates the namespace list unlocked to allow recovery from
3592  * controller failure. It is up to the caller to ensure the namespace list is
3593  * not modified by scan work while this function is executing.
3594  */
3595 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3596 {
3597         struct nvme_ns *ns, *next;
3598         LIST_HEAD(ns_list);
3599
3600         /* prevent racing with ns scanning */
3601         flush_work(&ctrl->scan_work);
3602
3603         /*
3604          * The dead states indicates the controller was not gracefully
3605          * disconnected. In that case, we won't be able to flush any data while
3606          * removing the namespaces' disks; fail all the queues now to avoid
3607          * potentially having to clean up the failed sync later.
3608          */
3609         if (ctrl->state == NVME_CTRL_DEAD)
3610                 nvme_kill_queues(ctrl);
3611
3612         down_write(&ctrl->namespaces_rwsem);
3613         list_splice_init(&ctrl->namespaces, &ns_list);
3614         up_write(&ctrl->namespaces_rwsem);
3615
3616         list_for_each_entry_safe(ns, next, &ns_list, list)
3617                 nvme_ns_remove(ns);
3618 }
3619 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3620
3621 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3622 {
3623         char *envp[2] = { NULL, NULL };
3624         u32 aen_result = ctrl->aen_result;
3625
3626         ctrl->aen_result = 0;
3627         if (!aen_result)
3628                 return;
3629
3630         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3631         if (!envp[0])
3632                 return;
3633         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3634         kfree(envp[0]);
3635 }
3636
3637 static void nvme_async_event_work(struct work_struct *work)
3638 {
3639         struct nvme_ctrl *ctrl =
3640                 container_of(work, struct nvme_ctrl, async_event_work);
3641
3642         nvme_aen_uevent(ctrl);
3643         ctrl->ops->submit_async_event(ctrl);
3644 }
3645
3646 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3647 {
3648
3649         u32 csts;
3650
3651         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3652                 return false;
3653
3654         if (csts == ~0)
3655                 return false;
3656
3657         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3658 }
3659
3660 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3661 {
3662         struct nvme_fw_slot_info_log *log;
3663
3664         log = kmalloc(sizeof(*log), GFP_KERNEL);
3665         if (!log)
3666                 return;
3667
3668         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3669                         sizeof(*log), 0))
3670                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3671         kfree(log);
3672 }
3673
3674 static void nvme_fw_act_work(struct work_struct *work)
3675 {
3676         struct nvme_ctrl *ctrl = container_of(work,
3677                                 struct nvme_ctrl, fw_act_work);
3678         unsigned long fw_act_timeout;
3679
3680         if (ctrl->mtfa)
3681                 fw_act_timeout = jiffies +
3682                                 msecs_to_jiffies(ctrl->mtfa * 100);
3683         else
3684                 fw_act_timeout = jiffies +
3685                                 msecs_to_jiffies(admin_timeout * 1000);
3686
3687         nvme_stop_queues(ctrl);
3688         while (nvme_ctrl_pp_status(ctrl)) {
3689                 if (time_after(jiffies, fw_act_timeout)) {
3690                         dev_warn(ctrl->device,
3691                                 "Fw activation timeout, reset controller\n");
3692                         nvme_reset_ctrl(ctrl);
3693                         break;
3694                 }
3695                 msleep(100);
3696         }
3697
3698         if (ctrl->state != NVME_CTRL_LIVE)
3699                 return;
3700
3701         nvme_start_queues(ctrl);
3702         /* read FW slot information to clear the AER */
3703         nvme_get_fw_slot_info(ctrl);
3704 }
3705
3706 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3707 {
3708         u32 aer_notice_type = (result & 0xff00) >> 8;
3709
3710         trace_nvme_async_event(ctrl, aer_notice_type);
3711
3712         switch (aer_notice_type) {
3713         case NVME_AER_NOTICE_NS_CHANGED:
3714                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3715                 nvme_queue_scan(ctrl);
3716                 break;
3717         case NVME_AER_NOTICE_FW_ACT_STARTING:
3718                 queue_work(nvme_wq, &ctrl->fw_act_work);
3719                 break;
3720 #ifdef CONFIG_NVME_MULTIPATH
3721         case NVME_AER_NOTICE_ANA:
3722                 if (!ctrl->ana_log_buf)
3723                         break;
3724                 queue_work(nvme_wq, &ctrl->ana_work);
3725                 break;
3726 #endif
3727         default:
3728                 dev_warn(ctrl->device, "async event result %08x\n", result);
3729         }
3730 }
3731
3732 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3733                 volatile union nvme_result *res)
3734 {
3735         u32 result = le32_to_cpu(res->u32);
3736         u32 aer_type = result & 0x07;
3737
3738         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3739                 return;
3740
3741         switch (aer_type) {
3742         case NVME_AER_NOTICE:
3743                 nvme_handle_aen_notice(ctrl, result);
3744                 break;
3745         case NVME_AER_ERROR:
3746         case NVME_AER_SMART:
3747         case NVME_AER_CSS:
3748         case NVME_AER_VS:
3749                 trace_nvme_async_event(ctrl, aer_type);
3750                 ctrl->aen_result = result;
3751                 break;
3752         default:
3753                 break;
3754         }
3755         queue_work(nvme_wq, &ctrl->async_event_work);
3756 }
3757 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3758
3759 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3760 {
3761         nvme_mpath_stop(ctrl);
3762         nvme_stop_keep_alive(ctrl);
3763         flush_work(&ctrl->async_event_work);
3764         cancel_work_sync(&ctrl->fw_act_work);
3765 }
3766 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3767
3768 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3769 {
3770         if (ctrl->kato)
3771                 nvme_start_keep_alive(ctrl);
3772
3773         if (ctrl->queue_count > 1) {
3774                 nvme_queue_scan(ctrl);
3775                 nvme_enable_aen(ctrl);
3776                 queue_work(nvme_wq, &ctrl->async_event_work);
3777                 nvme_start_queues(ctrl);
3778         }
3779 }
3780 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3781
3782 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3783 {
3784         nvme_fault_inject_fini(&ctrl->fault_inject);
3785         dev_pm_qos_hide_latency_tolerance(ctrl->device);
3786         cdev_device_del(&ctrl->cdev, ctrl->device);
3787 }
3788 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3789
3790 static void nvme_free_ctrl(struct device *dev)
3791 {
3792         struct nvme_ctrl *ctrl =
3793                 container_of(dev, struct nvme_ctrl, ctrl_device);
3794         struct nvme_subsystem *subsys = ctrl->subsys;
3795
3796         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3797         kfree(ctrl->effects);
3798         nvme_mpath_uninit(ctrl);
3799         __free_page(ctrl->discard_page);
3800
3801         if (subsys) {
3802                 mutex_lock(&nvme_subsystems_lock);
3803                 list_del(&ctrl->subsys_entry);
3804                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3805                 mutex_unlock(&nvme_subsystems_lock);
3806         }
3807
3808         ctrl->ops->free_ctrl(ctrl);
3809
3810         if (subsys)
3811                 nvme_put_subsystem(subsys);
3812 }
3813
3814 /*
3815  * Initialize a NVMe controller structures.  This needs to be called during
3816  * earliest initialization so that we have the initialized structured around
3817  * during probing.
3818  */
3819 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3820                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3821 {
3822         int ret;
3823
3824         ctrl->state = NVME_CTRL_NEW;
3825         spin_lock_init(&ctrl->lock);
3826         mutex_init(&ctrl->scan_lock);
3827         INIT_LIST_HEAD(&ctrl->namespaces);
3828         init_rwsem(&ctrl->namespaces_rwsem);
3829         ctrl->dev = dev;
3830         ctrl->ops = ops;
3831         ctrl->quirks = quirks;
3832         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3833         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3834         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3835         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3836
3837         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3838         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3839         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3840
3841         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3842                         PAGE_SIZE);
3843         ctrl->discard_page = alloc_page(GFP_KERNEL);
3844         if (!ctrl->discard_page) {
3845                 ret = -ENOMEM;
3846                 goto out;
3847         }
3848
3849         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3850         if (ret < 0)
3851                 goto out;
3852         ctrl->instance = ret;
3853
3854         device_initialize(&ctrl->ctrl_device);
3855         ctrl->device = &ctrl->ctrl_device;
3856         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3857         ctrl->device->class = nvme_class;
3858         ctrl->device->parent = ctrl->dev;
3859         ctrl->device->groups = nvme_dev_attr_groups;
3860         ctrl->device->release = nvme_free_ctrl;
3861         dev_set_drvdata(ctrl->device, ctrl);
3862         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3863         if (ret)
3864                 goto out_release_instance;
3865
3866         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3867         ctrl->cdev.owner = ops->module;
3868         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3869         if (ret)
3870                 goto out_free_name;
3871
3872         /*
3873          * Initialize latency tolerance controls.  The sysfs files won't
3874          * be visible to userspace unless the device actually supports APST.
3875          */
3876         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3877         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3878                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3879
3880         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
3881
3882         return 0;
3883 out_free_name:
3884         kfree_const(ctrl->device->kobj.name);
3885 out_release_instance:
3886         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3887 out:
3888         if (ctrl->discard_page)
3889                 __free_page(ctrl->discard_page);
3890         return ret;
3891 }
3892 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3893
3894 /**
3895  * nvme_kill_queues(): Ends all namespace queues
3896  * @ctrl: the dead controller that needs to end
3897  *
3898  * Call this function when the driver determines it is unable to get the
3899  * controller in a state capable of servicing IO.
3900  */
3901 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3902 {
3903         struct nvme_ns *ns;
3904
3905         down_read(&ctrl->namespaces_rwsem);
3906
3907         /* Forcibly unquiesce queues to avoid blocking dispatch */
3908         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3909                 blk_mq_unquiesce_queue(ctrl->admin_q);
3910
3911         list_for_each_entry(ns, &ctrl->namespaces, list)
3912                 nvme_set_queue_dying(ns);
3913
3914         up_read(&ctrl->namespaces_rwsem);
3915 }
3916 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3917
3918 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3919 {
3920         struct nvme_ns *ns;
3921
3922         down_read(&ctrl->namespaces_rwsem);
3923         list_for_each_entry(ns, &ctrl->namespaces, list)
3924                 blk_mq_unfreeze_queue(ns->queue);
3925         up_read(&ctrl->namespaces_rwsem);
3926 }
3927 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3928
3929 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3930 {
3931         struct nvme_ns *ns;
3932
3933         down_read(&ctrl->namespaces_rwsem);
3934         list_for_each_entry(ns, &ctrl->namespaces, list) {
3935                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3936                 if (timeout <= 0)
3937                         break;
3938         }
3939         up_read(&ctrl->namespaces_rwsem);
3940 }
3941 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3942
3943 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3944 {
3945         struct nvme_ns *ns;
3946
3947         down_read(&ctrl->namespaces_rwsem);
3948         list_for_each_entry(ns, &ctrl->namespaces, list)
3949                 blk_mq_freeze_queue_wait(ns->queue);
3950         up_read(&ctrl->namespaces_rwsem);
3951 }
3952 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3953
3954 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3955 {
3956         struct nvme_ns *ns;
3957
3958         down_read(&ctrl->namespaces_rwsem);
3959         list_for_each_entry(ns, &ctrl->namespaces, list)
3960                 blk_freeze_queue_start(ns->queue);
3961         up_read(&ctrl->namespaces_rwsem);
3962 }
3963 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3964
3965 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3966 {
3967         struct nvme_ns *ns;
3968
3969         down_read(&ctrl->namespaces_rwsem);
3970         list_for_each_entry(ns, &ctrl->namespaces, list)
3971                 blk_mq_quiesce_queue(ns->queue);
3972         up_read(&ctrl->namespaces_rwsem);
3973 }
3974 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3975
3976 void nvme_start_queues(struct nvme_ctrl *ctrl)
3977 {
3978         struct nvme_ns *ns;
3979
3980         down_read(&ctrl->namespaces_rwsem);
3981         list_for_each_entry(ns, &ctrl->namespaces, list)
3982                 blk_mq_unquiesce_queue(ns->queue);
3983         up_read(&ctrl->namespaces_rwsem);
3984 }
3985 EXPORT_SYMBOL_GPL(nvme_start_queues);
3986
3987
3988 void nvme_sync_queues(struct nvme_ctrl *ctrl)
3989 {
3990         struct nvme_ns *ns;
3991
3992         down_read(&ctrl->namespaces_rwsem);
3993         list_for_each_entry(ns, &ctrl->namespaces, list)
3994                 blk_sync_queue(ns->queue);
3995         up_read(&ctrl->namespaces_rwsem);
3996 }
3997 EXPORT_SYMBOL_GPL(nvme_sync_queues);
3998
3999 /*
4000  * Check we didn't inadvertently grow the command structure sizes:
4001  */
4002 static inline void _nvme_check_size(void)
4003 {
4004         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4005         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4006         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4007         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4008         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4009         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4010         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4011         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4012         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4013         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4014         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4015         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4016         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4017         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4018         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4019         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4020         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4021 }
4022
4023
4024 static int __init nvme_core_init(void)
4025 {
4026         int result = -ENOMEM;
4027
4028         _nvme_check_size();
4029
4030         nvme_wq = alloc_workqueue("nvme-wq",
4031                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4032         if (!nvme_wq)
4033                 goto out;
4034
4035         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4036                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4037         if (!nvme_reset_wq)
4038                 goto destroy_wq;
4039
4040         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4041                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4042         if (!nvme_delete_wq)
4043                 goto destroy_reset_wq;
4044
4045         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4046         if (result < 0)
4047                 goto destroy_delete_wq;
4048
4049         nvme_class = class_create(THIS_MODULE, "nvme");
4050         if (IS_ERR(nvme_class)) {
4051                 result = PTR_ERR(nvme_class);
4052                 goto unregister_chrdev;
4053         }
4054
4055         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4056         if (IS_ERR(nvme_subsys_class)) {
4057                 result = PTR_ERR(nvme_subsys_class);
4058                 goto destroy_class;
4059         }
4060         return 0;
4061
4062 destroy_class:
4063         class_destroy(nvme_class);
4064 unregister_chrdev:
4065         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4066 destroy_delete_wq:
4067         destroy_workqueue(nvme_delete_wq);
4068 destroy_reset_wq:
4069         destroy_workqueue(nvme_reset_wq);
4070 destroy_wq:
4071         destroy_workqueue(nvme_wq);
4072 out:
4073         return result;
4074 }
4075
4076 static void __exit nvme_core_exit(void)
4077 {
4078         ida_destroy(&nvme_subsystems_ida);
4079         class_destroy(nvme_subsys_class);
4080         class_destroy(nvme_class);
4081         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4082         destroy_workqueue(nvme_delete_wq);
4083         destroy_workqueue(nvme_reset_wq);
4084         destroy_workqueue(nvme_wq);
4085 }
4086
4087 MODULE_LICENSE("GPL");
4088 MODULE_VERSION("1.0");
4089 module_init(nvme_core_init);
4090 module_exit(nvme_core_exit);