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