drivers/nvme/host/core.c

   1 /*
   2  * NVM Express device driver
   3  * Copyright (c) 2011-2014, Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14
  15 #include <linux/blkdev.h>
  16 #include <linux/blk-mq.h>
  17 #include <linux/delay.h>
  18 #include <linux/errno.h>
  19 #include <linux/hdreg.h>
  20 #include <linux/kernel.h>
  21 #include <linux/module.h>
  22 #include <linux/list_sort.h>
  23 #include <linux/slab.h>
  24 #include <linux/types.h>
  25 #include <linux/pr.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/nvme_ioctl.h>
  28 #include <linux/t10-pi.h>
  29 #include <linux/pm_qos.h>
  30 #include <asm/unaligned.h>
  31
  32 #define CREATE_TRACE_POINTS
  33 #include "trace.h"
  34
  35 #include "nvme.h"
  36 #include "fabrics.h"
  37
  38 #define NVME_MINORS             (1U << MINORBITS)
  39
  40 unsigned int admin_timeout = 60;
  41 module_param(admin_timeout, uint, 0644);
  42 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  43 EXPORT_SYMBOL_GPL(admin_timeout);
  44
  45 unsigned int nvme_io_timeout = 30;
  46 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  47 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  48 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  49
  50 static unsigned char shutdown_timeout = 5;
  51 module_param(shutdown_timeout, byte, 0644);
  52 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  53
  54 static u8 nvme_max_retries = 5;
  55 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  56 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  57
  58 static unsigned long default_ps_max_latency_us = 100000;
  59 module_param(default_ps_max_latency_us, ulong, 0644);
  60 MODULE_PARM_DESC(default_ps_max_latency_us,
  61                  "max power saving latency for new devices; use PM QOS to change per device");
  62
  63 static bool force_apst;
  64 module_param(force_apst, bool, 0644);
  65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  66
  67 static bool streams;
  68 module_param(streams, bool, 0644);
  69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  70
  71 /*
  72  * nvme_wq - hosts nvme related works that are not reset or delete
  73  * nvme_reset_wq - hosts nvme reset works
  74  * nvme_delete_wq - hosts nvme delete works
  75  *
  76  * nvme_wq will host works such are scan, aen handling, fw activation,
  77  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
  78  * runs reset works which also flush works hosted on nvme_wq for
  79  * serialization purposes. nvme_delete_wq host controller deletion
  80  * works which flush reset works for serialization.
  81  */
  82 struct workqueue_struct *nvme_wq;
  83 EXPORT_SYMBOL_GPL(nvme_wq);
  84
  85 struct workqueue_struct *nvme_reset_wq;
  86 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  87
  88 struct workqueue_struct *nvme_delete_wq;
  89 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  90
  91 static DEFINE_IDA(nvme_subsystems_ida);
  92 static LIST_HEAD(nvme_subsystems);
  93 static DEFINE_MUTEX(nvme_subsystems_lock);
  94
  95 static DEFINE_IDA(nvme_instance_ida);
  96 static dev_t nvme_chr_devt;
  97 static struct class *nvme_class;
  98 static struct class *nvme_subsys_class;
  99
 100 static void nvme_ns_remove(struct nvme_ns *ns);
 101 static int nvme_revalidate_disk(struct gendisk *disk);
 102
 103 static __le32 nvme_get_log_dw10(u8 lid, size_t size)
 104 {
 105         return cpu_to_le32((((size / 4) - 1) << 16) | lid);
 106 }
 107
 108 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 109 {
 110         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 111                 return -EBUSY;
 112         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 113                 return -EBUSY;
 114         return 0;
 115 }
 116 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 117
 118 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 119 {
 120         int ret;
 121
 122         ret = nvme_reset_ctrl(ctrl);
 123         if (!ret)
 124                 flush_work(&ctrl->reset_work);
 125         return ret;
 126 }
 127 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 128
 129 static void nvme_delete_ctrl_work(struct work_struct *work)
 130 {
 131         struct nvme_ctrl *ctrl =
 132                 container_of(work, struct nvme_ctrl, delete_work);
 133
 134         flush_work(&ctrl->reset_work);
 135         nvme_stop_ctrl(ctrl);
 136         nvme_remove_namespaces(ctrl);
 137         ctrl->ops->delete_ctrl(ctrl);
 138         nvme_uninit_ctrl(ctrl);
 139         nvme_put_ctrl(ctrl);
 140 }
 141
 142 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 143 {
 144         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 145                 return -EBUSY;
 146         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 147                 return -EBUSY;
 148         return 0;
 149 }
 150 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 151
 152 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 153 {
 154         int ret = 0;
 155
 156         /*
 157          * Keep a reference until the work is flushed since ->delete_ctrl
 158          * can free the controller.
 159          */
 160         nvme_get_ctrl(ctrl);
 161         ret = nvme_delete_ctrl(ctrl);
 162         if (!ret)
 163                 flush_work(&ctrl->delete_work);
 164         nvme_put_ctrl(ctrl);
 165         return ret;
 166 }
 167 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
 168
 169 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 170 {
 171         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 172 }
 173
 174 static blk_status_t nvme_error_status(struct request *req)
 175 {
 176         switch (nvme_req(req)->status & 0x7ff) {
 177         case NVME_SC_SUCCESS:
 178                 return BLK_STS_OK;
 179         case NVME_SC_CAP_EXCEEDED:
 180                 return BLK_STS_NOSPC;
 181         case NVME_SC_LBA_RANGE:
 182                 return BLK_STS_TARGET;
 183         case NVME_SC_BAD_ATTRIBUTES:
 184         case NVME_SC_ONCS_NOT_SUPPORTED:
 185         case NVME_SC_INVALID_OPCODE:
 186         case NVME_SC_INVALID_FIELD:
 187         case NVME_SC_INVALID_NS:
 188                 return BLK_STS_NOTSUPP;
 189         case NVME_SC_WRITE_FAULT:
 190         case NVME_SC_READ_ERROR:
 191         case NVME_SC_UNWRITTEN_BLOCK:
 192         case NVME_SC_ACCESS_DENIED:
 193         case NVME_SC_READ_ONLY:
 194         case NVME_SC_COMPARE_FAILED:
 195                 return BLK_STS_MEDIUM;
 196         case NVME_SC_GUARD_CHECK:
 197         case NVME_SC_APPTAG_CHECK:
 198         case NVME_SC_REFTAG_CHECK:
 199         case NVME_SC_INVALID_PI:
 200                 return BLK_STS_PROTECTION;
 201         case NVME_SC_RESERVATION_CONFLICT:
 202                 return BLK_STS_NEXUS;
 203         default:
 204                 return BLK_STS_IOERR;
 205         }
 206 }
 207
 208 static inline bool nvme_req_needs_retry(struct request *req)
 209 {
 210         if (blk_noretry_request(req))
 211                 return false;
 212         if (nvme_req(req)->status & NVME_SC_DNR)
 213                 return false;
 214         if (nvme_req(req)->retries >= nvme_max_retries)
 215                 return false;
 216         return true;
 217 }
 218
 219 void nvme_complete_rq(struct request *req)
 220 {
 221         blk_status_t status = nvme_error_status(req);
 222
 223         trace_nvme_complete_rq(req);
 224
 225         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 226                 if (nvme_req_needs_failover(req, status)) {
 227                         nvme_failover_req(req);
 228                         return;
 229                 }
 230
 231                 if (!blk_queue_dying(req->q)) {
 232                         nvme_req(req)->retries++;
 233                         blk_mq_requeue_request(req, true);
 234                         return;
 235                 }
 236         }
 237         blk_mq_end_request(req, status);
 238 }
 239 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 240
 241 void nvme_cancel_request(struct request *req, void *data, bool reserved)
 242 {
 243         if (!blk_mq_request_started(req))
 244                 return;
 245
 246         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 247                                 "Cancelling I/O %d", req->tag);
 248
 249         nvme_req(req)->status = NVME_SC_ABORT_REQ;
 250         blk_mq_complete_request(req);
 251
 252 }
 253 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 254
 255 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 256                 enum nvme_ctrl_state new_state)
 257 {
 258         enum nvme_ctrl_state old_state;
 259         unsigned long flags;
 260         bool changed = false;
 261
 262         spin_lock_irqsave(&ctrl->lock, flags);
 263
 264         old_state = ctrl->state;
 265         switch (new_state) {
 266         case NVME_CTRL_ADMIN_ONLY:
 267                 switch (old_state) {
 268                 case NVME_CTRL_RECONNECTING:
 269                         changed = true;
 270                         /* FALLTHRU */
 271                 default:
 272                         break;
 273                 }
 274                 break;
 275         case NVME_CTRL_LIVE:
 276                 switch (old_state) {
 277                 case NVME_CTRL_NEW:
 278                 case NVME_CTRL_RESETTING:
 279                 case NVME_CTRL_RECONNECTING:
 280                         changed = true;
 281                         /* FALLTHRU */
 282                 default:
 283                         break;
 284                 }
 285                 break;
 286         case NVME_CTRL_RESETTING:
 287                 switch (old_state) {
 288                 case NVME_CTRL_NEW:
 289                 case NVME_CTRL_LIVE:
 290                 case NVME_CTRL_ADMIN_ONLY:
 291                         changed = true;
 292                         /* FALLTHRU */
 293                 default:
 294                         break;
 295                 }
 296                 break;
 297         case NVME_CTRL_RECONNECTING:
 298                 switch (old_state) {
 299                 case NVME_CTRL_LIVE:
 300                 case NVME_CTRL_RESETTING:
 301                         changed = true;
 302                         /* FALLTHRU */
 303                 default:
 304                         break;
 305                 }
 306                 break;
 307         case NVME_CTRL_DELETING:
 308                 switch (old_state) {
 309                 case NVME_CTRL_LIVE:
 310                 case NVME_CTRL_ADMIN_ONLY:
 311                 case NVME_CTRL_RESETTING:
 312                 case NVME_CTRL_RECONNECTING:
 313                         changed = true;
 314                         /* FALLTHRU */
 315                 default:
 316                         break;
 317                 }
 318                 break;
 319         case NVME_CTRL_DEAD:
 320                 switch (old_state) {
 321                 case NVME_CTRL_DELETING:
 322                         changed = true;
 323                         /* FALLTHRU */
 324                 default:
 325                         break;
 326                 }
 327                 break;
 328         default:
 329                 break;
 330         }
 331
 332         if (changed)
 333                 ctrl->state = new_state;
 334
 335         spin_unlock_irqrestore(&ctrl->lock, flags);
 336         if (changed && ctrl->state == NVME_CTRL_LIVE)
 337                 nvme_kick_requeue_lists(ctrl);
 338         return changed;
 339 }
 340 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 341
 342 static void nvme_free_ns_head(struct kref *ref)
 343 {
 344         struct nvme_ns_head *head =
 345                 container_of(ref, struct nvme_ns_head, ref);
 346
 347         nvme_mpath_remove_disk(head);
 348         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 349         list_del_init(&head->entry);
 350         cleanup_srcu_struct(&head->srcu);
 351         kfree(head);
 352 }
 353
 354 static void nvme_put_ns_head(struct nvme_ns_head *head)
 355 {
 356         kref_put(&head->ref, nvme_free_ns_head);
 357 }
 358
 359 static void nvme_free_ns(struct kref *kref)
 360 {
 361         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 362
 363         if (ns->ndev)
 364                 nvme_nvm_unregister(ns);
 365
 366         put_disk(ns->disk);
 367         nvme_put_ns_head(ns->head);
 368         nvme_put_ctrl(ns->ctrl);
 369         kfree(ns);
 370 }
 371
 372 static void nvme_put_ns(struct nvme_ns *ns)
 373 {
 374         kref_put(&ns->kref, nvme_free_ns);
 375 }
 376
 377 struct request *nvme_alloc_request(struct request_queue *q,
 378                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 379 {
 380         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 381         struct request *req;
 382
 383         if (qid == NVME_QID_ANY) {
 384                 req = blk_mq_alloc_request(q, op, flags);
 385         } else {
 386                 req = blk_mq_alloc_request_hctx(q, op, flags,
 387                                 qid ? qid - 1 : 0);
 388         }
 389         if (IS_ERR(req))
 390                 return req;
 391
 392         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 393         nvme_req(req)->cmd = cmd;
 394
 395         return req;
 396 }
 397 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 398
 399 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 400 {
 401         struct nvme_command c;
 402
 403         memset(&c, 0, sizeof(c));
 404
 405         c.directive.opcode = nvme_admin_directive_send;
 406         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 407         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 408         c.directive.dtype = NVME_DIR_IDENTIFY;
 409         c.directive.tdtype = NVME_DIR_STREAMS;
 410         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 411
 412         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 413 }
 414
 415 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 416 {
 417         return nvme_toggle_streams(ctrl, false);
 418 }
 419
 420 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 421 {
 422         return nvme_toggle_streams(ctrl, true);
 423 }
 424
 425 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 426                                   struct streams_directive_params *s, u32 nsid)
 427 {
 428         struct nvme_command c;
 429
 430         memset(&c, 0, sizeof(c));
 431         memset(s, 0, sizeof(*s));
 432
 433         c.directive.opcode = nvme_admin_directive_recv;
 434         c.directive.nsid = cpu_to_le32(nsid);
 435         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 436         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 437         c.directive.dtype = NVME_DIR_STREAMS;
 438
 439         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 440 }
 441
 442 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 443 {
 444         struct streams_directive_params s;
 445         int ret;
 446
 447         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 448                 return 0;
 449         if (!streams)
 450                 return 0;
 451
 452         ret = nvme_enable_streams(ctrl);
 453         if (ret)
 454                 return ret;
 455
 456         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 457         if (ret)
 458                 return ret;
 459
 460         ctrl->nssa = le16_to_cpu(s.nssa);
 461         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 462                 dev_info(ctrl->device, "too few streams (%u) available\n",
 463                                         ctrl->nssa);
 464                 nvme_disable_streams(ctrl);
 465                 return 0;
 466         }
 467
 468         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 469         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 470         return 0;
 471 }
 472
 473 /*
 474  * Check if 'req' has a write hint associated with it. If it does, assign
 475  * a valid namespace stream to the write.
 476  */
 477 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 478                                      struct request *req, u16 *control,
 479                                      u32 *dsmgmt)
 480 {
 481         enum rw_hint streamid = req->write_hint;
 482
 483         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 484                 streamid = 0;
 485         else {
 486                 streamid--;
 487                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 488                         return;
 489
 490                 *control |= NVME_RW_DTYPE_STREAMS;
 491                 *dsmgmt |= streamid << 16;
 492         }
 493
 494         if (streamid < ARRAY_SIZE(req->q->write_hints))
 495                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 496 }
 497
 498 static inline void nvme_setup_flush(struct nvme_ns *ns,
 499                 struct nvme_command *cmnd)
 500 {
 501         memset(cmnd, 0, sizeof(*cmnd));
 502         cmnd->common.opcode = nvme_cmd_flush;
 503         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 504 }
 505
 506 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 507                 struct nvme_command *cmnd)
 508 {
 509         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 510         struct nvme_dsm_range *range;
 511         struct bio *bio;
 512
 513         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 514         if (!range)
 515                 return BLK_STS_RESOURCE;
 516
 517         __rq_for_each_bio(bio, req) {
 518                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 519                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 520
 521                 range[n].cattr = cpu_to_le32(0);
 522                 range[n].nlb = cpu_to_le32(nlb);
 523                 range[n].slba = cpu_to_le64(slba);
 524                 n++;
 525         }
 526
 527         if (WARN_ON_ONCE(n != segments)) {
 528                 kfree(range);
 529                 return BLK_STS_IOERR;
 530         }
 531
 532         memset(cmnd, 0, sizeof(*cmnd));
 533         cmnd->dsm.opcode = nvme_cmd_dsm;
 534         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 535         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 536         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 537
 538         req->special_vec.bv_page = virt_to_page(range);
 539         req->special_vec.bv_offset = offset_in_page(range);
 540         req->special_vec.bv_len = sizeof(*range) * segments;
 541         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 542
 543         return BLK_STS_OK;
 544 }
 545
 546 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 547                 struct request *req, struct nvme_command *cmnd)
 548 {
 549         struct nvme_ctrl *ctrl = ns->ctrl;
 550         u16 control = 0;
 551         u32 dsmgmt = 0;
 552
 553         if (req->cmd_flags & REQ_FUA)
 554                 control |= NVME_RW_FUA;
 555         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 556                 control |= NVME_RW_LR;
 557
 558         if (req->cmd_flags & REQ_RAHEAD)
 559                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 560
 561         memset(cmnd, 0, sizeof(*cmnd));
 562         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 563         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 564         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 565         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 566
 567         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 568                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 569
 570         if (ns->ms) {
 571                 /*
 572                  * If formated with metadata, the block layer always provides a
 573                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 574                  * we enable the PRACT bit for protection information or set the
 575                  * namespace capacity to zero to prevent any I/O.
 576                  */
 577                 if (!blk_integrity_rq(req)) {
 578                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 579                                 return BLK_STS_NOTSUPP;
 580                         control |= NVME_RW_PRINFO_PRACT;
 581                 }
 582
 583                 switch (ns->pi_type) {
 584                 case NVME_NS_DPS_PI_TYPE3:
 585                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 586                         break;
 587                 case NVME_NS_DPS_PI_TYPE1:
 588                 case NVME_NS_DPS_PI_TYPE2:
 589                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 590                                         NVME_RW_PRINFO_PRCHK_REF;
 591                         cmnd->rw.reftag = cpu_to_le32(
 592                                         nvme_block_nr(ns, blk_rq_pos(req)));
 593                         break;
 594                 }
 595         }
 596
 597         cmnd->rw.control = cpu_to_le16(control);
 598         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 599         return 0;
 600 }
 601
 602 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 603                 struct nvme_command *cmd)
 604 {
 605         blk_status_t ret = BLK_STS_OK;
 606
 607         if (!(req->rq_flags & RQF_DONTPREP)) {
 608                 nvme_req(req)->retries = 0;
 609                 nvme_req(req)->flags = 0;
 610                 req->rq_flags |= RQF_DONTPREP;
 611         }
 612
 613         switch (req_op(req)) {
 614         case REQ_OP_DRV_IN:
 615         case REQ_OP_DRV_OUT:
 616                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 617                 break;
 618         case REQ_OP_FLUSH:
 619                 nvme_setup_flush(ns, cmd);
 620                 break;
 621         case REQ_OP_WRITE_ZEROES:
 622                 /* currently only aliased to deallocate for a few ctrls: */
 623         case REQ_OP_DISCARD:
 624                 ret = nvme_setup_discard(ns, req, cmd);
 625                 break;
 626         case REQ_OP_READ:
 627         case REQ_OP_WRITE:
 628                 ret = nvme_setup_rw(ns, req, cmd);
 629                 break;
 630         default:
 631                 WARN_ON_ONCE(1);
 632                 return BLK_STS_IOERR;
 633         }
 634
 635         cmd->common.command_id = req->tag;
 636         if (ns)
 637                 trace_nvme_setup_nvm_cmd(req->q->id, cmd);
 638         else
 639                 trace_nvme_setup_admin_cmd(cmd);
 640         return ret;
 641 }
 642 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 643
 644 /*
 645  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 646  * if the result is positive, it's an NVM Express status code
 647  */
 648 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 649                 union nvme_result *result, void *buffer, unsigned bufflen,
 650                 unsigned timeout, int qid, int at_head,
 651                 blk_mq_req_flags_t flags)
 652 {
 653         struct request *req;
 654         int ret;
 655
 656         req = nvme_alloc_request(q, cmd, flags, qid);
 657         if (IS_ERR(req))
 658                 return PTR_ERR(req);
 659
 660         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 661
 662         if (buffer && bufflen) {
 663                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 664                 if (ret)
 665                         goto out;
 666         }
 667
 668         blk_execute_rq(req->q, NULL, req, at_head);
 669         if (result)
 670                 *result = nvme_req(req)->result;
 671         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 672                 ret = -EINTR;
 673         else
 674                 ret = nvme_req(req)->status;
 675  out:
 676         blk_mq_free_request(req);
 677         return ret;
 678 }
 679 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 680
 681 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 682                 void *buffer, unsigned bufflen)
 683 {
 684         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 685                         NVME_QID_ANY, 0, 0);
 686 }
 687 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 688
 689 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 690                 unsigned len, u32 seed, bool write)
 691 {
 692         struct bio_integrity_payload *bip;
 693         int ret = -ENOMEM;
 694         void *buf;
 695
 696         buf = kmalloc(len, GFP_KERNEL);
 697         if (!buf)
 698                 goto out;
 699
 700         ret = -EFAULT;
 701         if (write && copy_from_user(buf, ubuf, len))
 702                 goto out_free_meta;
 703
 704         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 705         if (IS_ERR(bip)) {
 706                 ret = PTR_ERR(bip);
 707                 goto out_free_meta;
 708         }
 709
 710         bip->bip_iter.bi_size = len;
 711         bip->bip_iter.bi_sector = seed;
 712         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 713                         offset_in_page(buf));
 714         if (ret == len)
 715                 return buf;
 716         ret = -ENOMEM;
 717 out_free_meta:
 718         kfree(buf);
 719 out:
 720         return ERR_PTR(ret);
 721 }
 722
 723 static int nvme_submit_user_cmd(struct request_queue *q,
 724                 struct nvme_command *cmd, void __user *ubuffer,
 725                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 726                 u32 meta_seed, u32 *result, unsigned timeout)
 727 {
 728         bool write = nvme_is_write(cmd);
 729         struct nvme_ns *ns = q->queuedata;
 730         struct gendisk *disk = ns ? ns->disk : NULL;
 731         struct request *req;
 732         struct bio *bio = NULL;
 733         void *meta = NULL;
 734         int ret;
 735
 736         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 737         if (IS_ERR(req))
 738                 return PTR_ERR(req);
 739
 740         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 741
 742         if (ubuffer && bufflen) {
 743                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 744                                 GFP_KERNEL);
 745                 if (ret)
 746                         goto out;
 747                 bio = req->bio;
 748                 bio->bi_disk = disk;
 749                 if (disk && meta_buffer && meta_len) {
 750                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 751                                         meta_seed, write);
 752                         if (IS_ERR(meta)) {
 753                                 ret = PTR_ERR(meta);
 754                                 goto out_unmap;
 755                         }
 756                 }
 757         }
 758
 759         blk_execute_rq(req->q, disk, req, 0);
 760         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 761                 ret = -EINTR;
 762         else
 763                 ret = nvme_req(req)->status;
 764         if (result)
 765                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 766         if (meta && !ret && !write) {
 767                 if (copy_to_user(meta_buffer, meta, meta_len))
 768                         ret = -EFAULT;
 769         }
 770         kfree(meta);
 771  out_unmap:
 772         if (bio)
 773                 blk_rq_unmap_user(bio);
 774  out:
 775         blk_mq_free_request(req);
 776         return ret;
 777 }
 778
 779 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 780 {
 781         struct nvme_ctrl *ctrl = rq->end_io_data;
 782
 783         blk_mq_free_request(rq);
 784
 785         if (status) {
 786                 dev_err(ctrl->device,
 787                         "failed nvme_keep_alive_end_io error=%d\n",
 788                                 status);
 789                 return;
 790         }
 791
 792         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 793 }
 794
 795 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 796 {
 797         struct nvme_command c;
 798         struct request *rq;
 799
 800         memset(&c, 0, sizeof(c));
 801         c.common.opcode = nvme_admin_keep_alive;
 802
 803         rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
 804                         NVME_QID_ANY);
 805         if (IS_ERR(rq))
 806                 return PTR_ERR(rq);
 807
 808         rq->timeout = ctrl->kato * HZ;
 809         rq->end_io_data = ctrl;
 810
 811         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 812
 813         return 0;
 814 }
 815
 816 static void nvme_keep_alive_work(struct work_struct *work)
 817 {
 818         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 819                         struct nvme_ctrl, ka_work);
 820
 821         if (nvme_keep_alive(ctrl)) {
 822                 /* allocation failure, reset the controller */
 823                 dev_err(ctrl->device, "keep-alive failed\n");
 824                 nvme_reset_ctrl(ctrl);
 825                 return;
 826         }
 827 }
 828
 829 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 830 {
 831         if (unlikely(ctrl->kato == 0))
 832                 return;
 833
 834         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
 835         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 836 }
 837 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
 838
 839 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 840 {
 841         if (unlikely(ctrl->kato == 0))
 842                 return;
 843
 844         cancel_delayed_work_sync(&ctrl->ka_work);
 845 }
 846 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 847
 848 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 849 {
 850         struct nvme_command c = { };
 851         int error;
 852
 853         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 854         c.identify.opcode = nvme_admin_identify;
 855         c.identify.cns = NVME_ID_CNS_CTRL;
 856
 857         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 858         if (!*id)
 859                 return -ENOMEM;
 860
 861         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 862                         sizeof(struct nvme_id_ctrl));
 863         if (error)
 864                 kfree(*id);
 865         return error;
 866 }
 867
 868 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
 869                 struct nvme_ns_ids *ids)
 870 {
 871         struct nvme_command c = { };
 872         int status;
 873         void *data;
 874         int pos;
 875         int len;
 876
 877         c.identify.opcode = nvme_admin_identify;
 878         c.identify.nsid = cpu_to_le32(nsid);
 879         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
 880
 881         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
 882         if (!data)
 883                 return -ENOMEM;
 884
 885         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
 886                                       NVME_IDENTIFY_DATA_SIZE);
 887         if (status)
 888                 goto free_data;
 889
 890         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
 891                 struct nvme_ns_id_desc *cur = data + pos;
 892
 893                 if (cur->nidl == 0)
 894                         break;
 895
 896                 switch (cur->nidt) {
 897                 case NVME_NIDT_EUI64:
 898                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
 899                                 dev_warn(ctrl->device,
 900                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
 901                                          cur->nidl);
 902                                 goto free_data;
 903                         }
 904                         len = NVME_NIDT_EUI64_LEN;
 905                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
 906                         break;
 907                 case NVME_NIDT_NGUID:
 908                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
 909                                 dev_warn(ctrl->device,
 910                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
 911                                          cur->nidl);
 912                                 goto free_data;
 913                         }
 914                         len = NVME_NIDT_NGUID_LEN;
 915                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
 916                         break;
 917                 case NVME_NIDT_UUID:
 918                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
 919                                 dev_warn(ctrl->device,
 920                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
 921                                          cur->nidl);
 922                                 goto free_data;
 923                         }
 924                         len = NVME_NIDT_UUID_LEN;
 925                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
 926                         break;
 927                 default:
 928                         /* Skip unnkown types */
 929                         len = cur->nidl;
 930                         break;
 931                 }
 932
 933                 len += sizeof(*cur);
 934         }
 935 free_data:
 936         kfree(data);
 937         return status;
 938 }
 939
 940 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 941 {
 942         struct nvme_command c = { };
 943
 944         c.identify.opcode = nvme_admin_identify;
 945         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
 946         c.identify.nsid = cpu_to_le32(nsid);
 947         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 948 }
 949
 950 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
 951                 unsigned nsid)
 952 {
 953         struct nvme_id_ns *id;
 954         struct nvme_command c = { };
 955         int error;
 956
 957         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 958         c.identify.opcode = nvme_admin_identify;
 959         c.identify.nsid = cpu_to_le32(nsid);
 960         c.identify.cns = NVME_ID_CNS_NS;
 961
 962         id = kmalloc(sizeof(*id), GFP_KERNEL);
 963         if (!id)
 964                 return NULL;
 965
 966         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
 967         if (error) {
 968                 dev_warn(ctrl->device, "Identify namespace failed\n");
 969                 kfree(id);
 970                 return NULL;
 971         }
 972
 973         return id;
 974 }
 975
 976 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 977                       void *buffer, size_t buflen, u32 *result)
 978 {
 979         struct nvme_command c;
 980         union nvme_result res;
 981         int ret;
 982
 983         memset(&c, 0, sizeof(c));
 984         c.features.opcode = nvme_admin_set_features;
 985         c.features.fid = cpu_to_le32(fid);
 986         c.features.dword11 = cpu_to_le32(dword11);
 987
 988         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
 989                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
 990         if (ret >= 0 && result)
 991                 *result = le32_to_cpu(res.u32);
 992         return ret;
 993 }
 994
 995 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 996 {
 997         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 998         u32 result;
 999         int status, nr_io_queues;
1000
1001         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1002                         &result);
1003         if (status < 0)
1004                 return status;
1005
1006         /*
1007          * Degraded controllers might return an error when setting the queue
1008          * count.  We still want to be able to bring them online and offer
1009          * access to the admin queue, as that might be only way to fix them up.
1010          */
1011         if (status > 0) {
1012                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1013                 *count = 0;
1014         } else {
1015                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1016                 *count = min(*count, nr_io_queues);
1017         }
1018
1019         return 0;
1020 }
1021 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1022
1023 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1024 {
1025         struct nvme_user_io io;
1026         struct nvme_command c;
1027         unsigned length, meta_len;
1028         void __user *metadata;
1029
1030         if (copy_from_user(&io, uio, sizeof(io)))
1031                 return -EFAULT;
1032         if (io.flags)
1033                 return -EINVAL;
1034
1035         switch (io.opcode) {
1036         case nvme_cmd_write:
1037         case nvme_cmd_read:
1038         case nvme_cmd_compare:
1039                 break;
1040         default:
1041                 return -EINVAL;
1042         }
1043
1044         length = (io.nblocks + 1) << ns->lba_shift;
1045         meta_len = (io.nblocks + 1) * ns->ms;
1046         metadata = (void __user *)(uintptr_t)io.metadata;
1047
1048         if (ns->ext) {
1049                 length += meta_len;
1050                 meta_len = 0;
1051         } else if (meta_len) {
1052                 if ((io.metadata & 3) || !io.metadata)
1053                         return -EINVAL;
1054         }
1055
1056         memset(&c, 0, sizeof(c));
1057         c.rw.opcode = io.opcode;
1058         c.rw.flags = io.flags;
1059         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1060         c.rw.slba = cpu_to_le64(io.slba);
1061         c.rw.length = cpu_to_le16(io.nblocks);
1062         c.rw.control = cpu_to_le16(io.control);
1063         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1064         c.rw.reftag = cpu_to_le32(io.reftag);
1065         c.rw.apptag = cpu_to_le16(io.apptag);
1066         c.rw.appmask = cpu_to_le16(io.appmask);
1067
1068         return nvme_submit_user_cmd(ns->queue, &c,
1069                         (void __user *)(uintptr_t)io.addr, length,
1070                         metadata, meta_len, io.slba, NULL, 0);
1071 }
1072
1073 static u32 nvme_known_admin_effects(u8 opcode)
1074 {
1075         switch (opcode) {
1076         case nvme_admin_format_nvm:
1077                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1078                                         NVME_CMD_EFFECTS_CSE_MASK;
1079         case nvme_admin_sanitize_nvm:
1080                 return NVME_CMD_EFFECTS_CSE_MASK;
1081         default:
1082                 break;
1083         }
1084         return 0;
1085 }
1086
1087 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1088                                                                 u8 opcode)
1089 {
1090         u32 effects = 0;
1091
1092         if (ns) {
1093                 if (ctrl->effects)
1094                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1095                 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1096                         dev_warn(ctrl->device,
1097                                  "IO command:%02x has unhandled effects:%08x\n",
1098                                  opcode, effects);
1099                 return 0;
1100         }
1101
1102         if (ctrl->effects)
1103                 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1104         else
1105                 effects = nvme_known_admin_effects(opcode);
1106
1107         /*
1108          * For simplicity, IO to all namespaces is quiesced even if the command
1109          * effects say only one namespace is affected.
1110          */
1111         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1112                 nvme_start_freeze(ctrl);
1113                 nvme_wait_freeze(ctrl);
1114         }
1115         return effects;
1116 }
1117
1118 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1119 {
1120         struct nvme_ns *ns;
1121
1122         mutex_lock(&ctrl->namespaces_mutex);
1123         list_for_each_entry(ns, &ctrl->namespaces, list) {
1124                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1125                         nvme_ns_remove(ns);
1126         }
1127         mutex_unlock(&ctrl->namespaces_mutex);
1128 }
1129
1130 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1131 {
1132         /*
1133          * Revalidate LBA changes prior to unfreezing. This is necessary to
1134          * prevent memory corruption if a logical block size was changed by
1135          * this command.
1136          */
1137         if (effects & NVME_CMD_EFFECTS_LBCC)
1138                 nvme_update_formats(ctrl);
1139         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1140                 nvme_unfreeze(ctrl);
1141         if (effects & NVME_CMD_EFFECTS_CCC)
1142                 nvme_init_identify(ctrl);
1143         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1144                 nvme_queue_scan(ctrl);
1145 }
1146
1147 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1148                         struct nvme_passthru_cmd __user *ucmd)
1149 {
1150         struct nvme_passthru_cmd cmd;
1151         struct nvme_command c;
1152         unsigned timeout = 0;
1153         u32 effects;
1154         int status;
1155
1156         if (!capable(CAP_SYS_ADMIN))
1157                 return -EACCES;
1158         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1159                 return -EFAULT;
1160         if (cmd.flags)
1161                 return -EINVAL;
1162
1163         memset(&c, 0, sizeof(c));
1164         c.common.opcode = cmd.opcode;
1165         c.common.flags = cmd.flags;
1166         c.common.nsid = cpu_to_le32(cmd.nsid);
1167         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1168         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1169         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1170         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1171         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1172         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1173         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1174         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1175
1176         if (cmd.timeout_ms)
1177                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1178
1179         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1180         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1181                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1182                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1183                         0, &cmd.result, timeout);
1184         nvme_passthru_end(ctrl, effects);
1185
1186         if (status >= 0) {
1187                 if (put_user(cmd.result, &ucmd->result))
1188                         return -EFAULT;
1189         }
1190
1191         return status;
1192 }
1193
1194 /*
1195  * Issue ioctl requests on the first available path.  Note that unlike normal
1196  * block layer requests we will not retry failed request on another controller.
1197  */
1198 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1199                 struct nvme_ns_head **head, int *srcu_idx)
1200 {
1201 #ifdef CONFIG_NVME_MULTIPATH
1202         if (disk->fops == &nvme_ns_head_ops) {
1203                 *head = disk->private_data;
1204                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1205                 return nvme_find_path(*head);
1206         }
1207 #endif
1208         *head = NULL;
1209         *srcu_idx = -1;
1210         return disk->private_data;
1211 }
1212
1213 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1214 {
1215         if (head)
1216                 srcu_read_unlock(&head->srcu, idx);
1217 }
1218
1219 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1220 {
1221         switch (cmd) {
1222         case NVME_IOCTL_ID:
1223                 force_successful_syscall_return();
1224                 return ns->head->ns_id;
1225         case NVME_IOCTL_ADMIN_CMD:
1226                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1227         case NVME_IOCTL_IO_CMD:
1228                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1229         case NVME_IOCTL_SUBMIT_IO:
1230                 return nvme_submit_io(ns, (void __user *)arg);
1231         default:
1232 #ifdef CONFIG_NVM
1233                 if (ns->ndev)
1234                         return nvme_nvm_ioctl(ns, cmd, arg);
1235 #endif
1236                 if (is_sed_ioctl(cmd))
1237                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1238                                          (void __user *) arg);
1239                 return -ENOTTY;
1240         }
1241 }
1242
1243 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1244                 unsigned int cmd, unsigned long arg)
1245 {
1246         struct nvme_ns_head *head = NULL;
1247         struct nvme_ns *ns;
1248         int srcu_idx, ret;
1249
1250         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1251         if (unlikely(!ns))
1252                 ret = -EWOULDBLOCK;
1253         else
1254                 ret = nvme_ns_ioctl(ns, cmd, arg);
1255         nvme_put_ns_from_disk(head, srcu_idx);
1256         return ret;
1257 }
1258
1259 static int nvme_open(struct block_device *bdev, fmode_t mode)
1260 {
1261         struct nvme_ns *ns = bdev->bd_disk->private_data;
1262
1263 #ifdef CONFIG_NVME_MULTIPATH
1264         /* should never be called due to GENHD_FL_HIDDEN */
1265         if (WARN_ON_ONCE(ns->head->disk))
1266                 goto fail;
1267 #endif
1268         if (!kref_get_unless_zero(&ns->kref))
1269                 goto fail;
1270         if (!try_module_get(ns->ctrl->ops->module))
1271                 goto fail_put_ns;
1272
1273         return 0;
1274
1275 fail_put_ns:
1276         nvme_put_ns(ns);
1277 fail:
1278         return -ENXIO;
1279 }
1280
1281 static void nvme_release(struct gendisk *disk, fmode_t mode)
1282 {
1283         struct nvme_ns *ns = disk->private_data;
1284
1285         module_put(ns->ctrl->ops->module);
1286         nvme_put_ns(ns);
1287 }
1288
1289 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1290 {
1291         /* some standard values */
1292         geo->heads = 1 << 6;
1293         geo->sectors = 1 << 5;
1294         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1295         return 0;
1296 }
1297
1298 #ifdef CONFIG_BLK_DEV_INTEGRITY
1299 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1300 {
1301         struct blk_integrity integrity;
1302
1303         memset(&integrity, 0, sizeof(integrity));
1304         switch (pi_type) {
1305         case NVME_NS_DPS_PI_TYPE3:
1306                 integrity.profile = &t10_pi_type3_crc;
1307                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1308                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1309                 break;
1310         case NVME_NS_DPS_PI_TYPE1:
1311         case NVME_NS_DPS_PI_TYPE2:
1312                 integrity.profile = &t10_pi_type1_crc;
1313                 integrity.tag_size = sizeof(u16);
1314                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1315                 break;
1316         default:
1317                 integrity.profile = NULL;
1318                 break;
1319         }
1320         integrity.tuple_size = ms;
1321         blk_integrity_register(disk, &integrity);
1322         blk_queue_max_integrity_segments(disk->queue, 1);
1323 }
1324 #else
1325 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1326 {
1327 }
1328 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1329
1330 static void nvme_set_chunk_size(struct nvme_ns *ns)
1331 {
1332         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1333         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1334 }
1335
1336 static void nvme_config_discard(struct nvme_ctrl *ctrl,
1337                 unsigned stream_alignment, struct request_queue *queue)
1338 {
1339         u32 size = queue_logical_block_size(queue);
1340
1341         if (stream_alignment)
1342                 size *= stream_alignment;
1343
1344         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1345                         NVME_DSM_MAX_RANGES);
1346
1347         queue->limits.discard_alignment = 0;
1348         queue->limits.discard_granularity = size;
1349
1350         blk_queue_max_discard_sectors(queue, UINT_MAX);
1351         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1352         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, queue);
1353
1354         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1355                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1356 }
1357
1358 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1359                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1360 {
1361         memset(ids, 0, sizeof(*ids));
1362
1363         if (ctrl->vs >= NVME_VS(1, 1, 0))
1364                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1365         if (ctrl->vs >= NVME_VS(1, 2, 0))
1366                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1367         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1368                  /* Don't treat error as fatal we potentially
1369                   * already have a NGUID or EUI-64
1370                   */
1371                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1372                         dev_warn(ctrl->device,
1373                                  "%s: Identify Descriptors failed\n", __func__);
1374         }
1375 }
1376
1377 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1378 {
1379         return !uuid_is_null(&ids->uuid) ||
1380                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1381                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1382 }
1383
1384 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1385 {
1386         return uuid_equal(&a->uuid, &b->uuid) &&
1387                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1388                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1389 }
1390
1391 static void nvme_update_disk_info(struct gendisk *disk,
1392                 struct nvme_ns *ns, struct nvme_id_ns *id)
1393 {
1394         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1395         unsigned short bs = 1 << ns->lba_shift;
1396         unsigned stream_alignment = 0;
1397
1398         if (ns->ctrl->nr_streams && ns->sws && ns->sgs)
1399                 stream_alignment = ns->sws * ns->sgs;
1400
1401         blk_mq_freeze_queue(disk->queue);
1402         blk_integrity_unregister(disk);
1403
1404         blk_queue_logical_block_size(disk->queue, bs);
1405         blk_queue_physical_block_size(disk->queue, bs);
1406         blk_queue_io_min(disk->queue, bs);
1407
1408         if (ns->ms && !ns->ext &&
1409             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1410                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1411         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1412                 capacity = 0;
1413         set_capacity(disk, capacity);
1414
1415         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1416                 nvme_config_discard(ns->ctrl, stream_alignment, disk->queue);
1417         blk_mq_unfreeze_queue(disk->queue);
1418 }
1419
1420 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1421 {
1422         struct nvme_ns *ns = disk->private_data;
1423
1424         /*
1425          * If identify namespace failed, use default 512 byte block size so
1426          * block layer can use before failing read/write for 0 capacity.
1427          */
1428         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1429         if (ns->lba_shift == 0)
1430                 ns->lba_shift = 9;
1431         ns->noiob = le16_to_cpu(id->noiob);
1432         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1433         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1434         /* the PI implementation requires metadata equal t10 pi tuple size */
1435         if (ns->ms == sizeof(struct t10_pi_tuple))
1436                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1437         else
1438                 ns->pi_type = 0;
1439
1440         if (ns->noiob)
1441                 nvme_set_chunk_size(ns);
1442         nvme_update_disk_info(disk, ns, id);
1443 #ifdef CONFIG_NVME_MULTIPATH
1444         if (ns->head->disk)
1445                 nvme_update_disk_info(ns->head->disk, ns, id);
1446 #endif
1447 }
1448
1449 static int nvme_revalidate_disk(struct gendisk *disk)
1450 {
1451         struct nvme_ns *ns = disk->private_data;
1452         struct nvme_ctrl *ctrl = ns->ctrl;
1453         struct nvme_id_ns *id;
1454         struct nvme_ns_ids ids;
1455         int ret = 0;
1456
1457         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1458                 set_capacity(disk, 0);
1459                 return -ENODEV;
1460         }
1461
1462         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1463         if (!id)
1464                 return -ENODEV;
1465
1466         if (id->ncap == 0) {
1467                 ret = -ENODEV;
1468                 goto out;
1469         }
1470
1471         __nvme_revalidate_disk(disk, id);
1472         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1473         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1474                 dev_err(ctrl->device,
1475                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1476                 ret = -ENODEV;
1477         }
1478
1479 out:
1480         kfree(id);
1481         return ret;
1482 }
1483
1484 static char nvme_pr_type(enum pr_type type)
1485 {
1486         switch (type) {
1487         case PR_WRITE_EXCLUSIVE:
1488                 return 1;
1489         case PR_EXCLUSIVE_ACCESS:
1490                 return 2;
1491         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1492                 return 3;
1493         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1494                 return 4;
1495         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1496                 return 5;
1497         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1498                 return 6;
1499         default:
1500                 return 0;
1501         }
1502 };
1503
1504 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1505                                 u64 key, u64 sa_key, u8 op)
1506 {
1507         struct nvme_ns_head *head = NULL;
1508         struct nvme_ns *ns;
1509         struct nvme_command c;
1510         int srcu_idx, ret;
1511         u8 data[16] = { 0, };
1512
1513         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1514         if (unlikely(!ns))
1515                 return -EWOULDBLOCK;
1516
1517         put_unaligned_le64(key, &data[0]);
1518         put_unaligned_le64(sa_key, &data[8]);
1519
1520         memset(&c, 0, sizeof(c));
1521         c.common.opcode = op;
1522         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1523         c.common.cdw10[0] = cpu_to_le32(cdw10);
1524
1525         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1526         nvme_put_ns_from_disk(head, srcu_idx);
1527         return ret;
1528 }
1529
1530 static int nvme_pr_register(struct block_device *bdev, u64 old,
1531                 u64 new, unsigned flags)
1532 {
1533         u32 cdw10;
1534
1535         if (flags & ~PR_FL_IGNORE_KEY)
1536                 return -EOPNOTSUPP;
1537
1538         cdw10 = old ? 2 : 0;
1539         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1540         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1541         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1542 }
1543
1544 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1545                 enum pr_type type, unsigned flags)
1546 {
1547         u32 cdw10;
1548
1549         if (flags & ~PR_FL_IGNORE_KEY)
1550                 return -EOPNOTSUPP;
1551
1552         cdw10 = nvme_pr_type(type) << 8;
1553         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1554         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1555 }
1556
1557 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1558                 enum pr_type type, bool abort)
1559 {
1560         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1561         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1562 }
1563
1564 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1565 {
1566         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1567         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1568 }
1569
1570 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1571 {
1572         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1573         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1574 }
1575
1576 static const struct pr_ops nvme_pr_ops = {
1577         .pr_register    = nvme_pr_register,
1578         .pr_reserve     = nvme_pr_reserve,
1579         .pr_release     = nvme_pr_release,
1580         .pr_preempt     = nvme_pr_preempt,
1581         .pr_clear       = nvme_pr_clear,
1582 };
1583
1584 #ifdef CONFIG_BLK_SED_OPAL
1585 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1586                 bool send)
1587 {
1588         struct nvme_ctrl *ctrl = data;
1589         struct nvme_command cmd;
1590
1591         memset(&cmd, 0, sizeof(cmd));
1592         if (send)
1593                 cmd.common.opcode = nvme_admin_security_send;
1594         else
1595                 cmd.common.opcode = nvme_admin_security_recv;
1596         cmd.common.nsid = 0;
1597         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1598         cmd.common.cdw10[1] = cpu_to_le32(len);
1599
1600         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1601                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1602 }
1603 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1604 #endif /* CONFIG_BLK_SED_OPAL */
1605
1606 static const struct block_device_operations nvme_fops = {
1607         .owner          = THIS_MODULE,
1608         .ioctl          = nvme_ioctl,
1609         .compat_ioctl   = nvme_ioctl,
1610         .open           = nvme_open,
1611         .release        = nvme_release,
1612         .getgeo         = nvme_getgeo,
1613         .revalidate_disk= nvme_revalidate_disk,
1614         .pr_ops         = &nvme_pr_ops,
1615 };
1616
1617 #ifdef CONFIG_NVME_MULTIPATH
1618 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1619 {
1620         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1621
1622         if (!kref_get_unless_zero(&head->ref))
1623                 return -ENXIO;
1624         return 0;
1625 }
1626
1627 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1628 {
1629         nvme_put_ns_head(disk->private_data);
1630 }
1631
1632 const struct block_device_operations nvme_ns_head_ops = {
1633         .owner          = THIS_MODULE,
1634         .open           = nvme_ns_head_open,
1635         .release        = nvme_ns_head_release,
1636         .ioctl          = nvme_ioctl,
1637         .compat_ioctl   = nvme_ioctl,
1638         .getgeo         = nvme_getgeo,
1639         .pr_ops         = &nvme_pr_ops,
1640 };
1641 #endif /* CONFIG_NVME_MULTIPATH */
1642
1643 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1644 {
1645         unsigned long timeout =
1646                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1647         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1648         int ret;
1649
1650         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1651                 if (csts == ~0)
1652                         return -ENODEV;
1653                 if ((csts & NVME_CSTS_RDY) == bit)
1654                         break;
1655
1656                 msleep(100);
1657                 if (fatal_signal_pending(current))
1658                         return -EINTR;
1659                 if (time_after(jiffies, timeout)) {
1660                         dev_err(ctrl->device,
1661                                 "Device not ready; aborting %s\n", enabled ?
1662                                                 "initialisation" : "reset");
1663                         return -ENODEV;
1664                 }
1665         }
1666
1667         return ret;
1668 }
1669
1670 /*
1671  * If the device has been passed off to us in an enabled state, just clear
1672  * the enabled bit.  The spec says we should set the 'shutdown notification
1673  * bits', but doing so may cause the device to complete commands to the
1674  * admin queue ... and we don't know what memory that might be pointing at!
1675  */
1676 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1677 {
1678         int ret;
1679
1680         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1681         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1682
1683         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1684         if (ret)
1685                 return ret;
1686
1687         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1688                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1689
1690         return nvme_wait_ready(ctrl, cap, false);
1691 }
1692 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1693
1694 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1695 {
1696         /*
1697          * Default to a 4K page size, with the intention to update this
1698          * path in the future to accomodate architectures with differing
1699          * kernel and IO page sizes.
1700          */
1701         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1702         int ret;
1703
1704         if (page_shift < dev_page_min) {
1705                 dev_err(ctrl->device,
1706                         "Minimum device page size %u too large for host (%u)\n",
1707                         1 << dev_page_min, 1 << page_shift);
1708                 return -ENODEV;
1709         }
1710
1711         ctrl->page_size = 1 << page_shift;
1712
1713         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1714         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1715         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1716         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1717         ctrl->ctrl_config |= NVME_CC_ENABLE;
1718
1719         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1720         if (ret)
1721                 return ret;
1722         return nvme_wait_ready(ctrl, cap, true);
1723 }
1724 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1725
1726 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1727 {
1728         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1729         u32 csts;
1730         int ret;
1731
1732         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1733         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1734
1735         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1736         if (ret)
1737                 return ret;
1738
1739         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1740                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1741                         break;
1742
1743                 msleep(100);
1744                 if (fatal_signal_pending(current))
1745                         return -EINTR;
1746                 if (time_after(jiffies, timeout)) {
1747                         dev_err(ctrl->device,
1748                                 "Device shutdown incomplete; abort shutdown\n");
1749                         return -ENODEV;
1750                 }
1751         }
1752
1753         return ret;
1754 }
1755 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1756
1757 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1758                 struct request_queue *q)
1759 {
1760         bool vwc = false;
1761
1762         if (ctrl->max_hw_sectors) {
1763                 u32 max_segments =
1764                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1765
1766                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1767                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1768         }
1769         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1770             is_power_of_2(ctrl->max_hw_sectors))
1771                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1772         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1773         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1774                 vwc = true;
1775         blk_queue_write_cache(q, vwc, vwc);
1776 }
1777
1778 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1779 {
1780         __le64 ts;
1781         int ret;
1782
1783         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1784                 return 0;
1785
1786         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1787         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1788                         NULL);
1789         if (ret)
1790                 dev_warn_once(ctrl->device,
1791                         "could not set timestamp (%d)\n", ret);
1792         return ret;
1793 }
1794
1795 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1796 {
1797         /*
1798          * APST (Autonomous Power State Transition) lets us program a
1799          * table of power state transitions that the controller will
1800          * perform automatically.  We configure it with a simple
1801          * heuristic: we are willing to spend at most 2% of the time
1802          * transitioning between power states.  Therefore, when running
1803          * in any given state, we will enter the next lower-power
1804          * non-operational state after waiting 50 * (enlat + exlat)
1805          * microseconds, as long as that state's exit latency is under
1806          * the requested maximum latency.
1807          *
1808          * We will not autonomously enter any non-operational state for
1809          * which the total latency exceeds ps_max_latency_us.  Users
1810          * can set ps_max_latency_us to zero to turn off APST.
1811          */
1812
1813         unsigned apste;
1814         struct nvme_feat_auto_pst *table;
1815         u64 max_lat_us = 0;
1816         int max_ps = -1;
1817         int ret;
1818
1819         /*
1820          * If APST isn't supported or if we haven't been initialized yet,
1821          * then don't do anything.
1822          */
1823         if (!ctrl->apsta)
1824                 return 0;
1825
1826         if (ctrl->npss > 31) {
1827                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1828                 return 0;
1829         }
1830
1831         table = kzalloc(sizeof(*table), GFP_KERNEL);
1832         if (!table)
1833                 return 0;
1834
1835         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1836                 /* Turn off APST. */
1837                 apste = 0;
1838                 dev_dbg(ctrl->device, "APST disabled\n");
1839         } else {
1840                 __le64 target = cpu_to_le64(0);
1841                 int state;
1842
1843                 /*
1844                  * Walk through all states from lowest- to highest-power.
1845                  * According to the spec, lower-numbered states use more
1846                  * power.  NPSS, despite the name, is the index of the
1847                  * lowest-power state, not the number of states.
1848                  */
1849                 for (state = (int)ctrl->npss; state >= 0; state--) {
1850                         u64 total_latency_us, exit_latency_us, transition_ms;
1851
1852                         if (target)
1853                                 table->entries[state] = target;
1854
1855                         /*
1856                          * Don't allow transitions to the deepest state
1857                          * if it's quirked off.
1858                          */
1859                         if (state == ctrl->npss &&
1860                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1861                                 continue;
1862
1863                         /*
1864                          * Is this state a useful non-operational state for
1865                          * higher-power states to autonomously transition to?
1866                          */
1867                         if (!(ctrl->psd[state].flags &
1868                               NVME_PS_FLAGS_NON_OP_STATE))
1869                                 continue;
1870
1871                         exit_latency_us =
1872                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1873                         if (exit_latency_us > ctrl->ps_max_latency_us)
1874                                 continue;
1875
1876                         total_latency_us =
1877                                 exit_latency_us +
1878                                 le32_to_cpu(ctrl->psd[state].entry_lat);
1879
1880                         /*
1881                          * This state is good.  Use it as the APST idle
1882                          * target for higher power states.
1883                          */
1884                         transition_ms = total_latency_us + 19;
1885                         do_div(transition_ms, 20);
1886                         if (transition_ms > (1 << 24) - 1)
1887                                 transition_ms = (1 << 24) - 1;
1888
1889                         target = cpu_to_le64((state << 3) |
1890                                              (transition_ms << 8));
1891
1892                         if (max_ps == -1)
1893                                 max_ps = state;
1894
1895                         if (total_latency_us > max_lat_us)
1896                                 max_lat_us = total_latency_us;
1897                 }
1898
1899                 apste = 1;
1900
1901                 if (max_ps == -1) {
1902                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1903                 } else {
1904                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1905                                 max_ps, max_lat_us, (int)sizeof(*table), table);
1906                 }
1907         }
1908
1909         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1910                                 table, sizeof(*table), NULL);
1911         if (ret)
1912                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1913
1914         kfree(table);
1915         return ret;
1916 }
1917
1918 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1919 {
1920         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1921         u64 latency;
1922
1923         switch (val) {
1924         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1925         case PM_QOS_LATENCY_ANY:
1926                 latency = U64_MAX;
1927                 break;
1928
1929         default:
1930                 latency = val;
1931         }
1932
1933         if (ctrl->ps_max_latency_us != latency) {
1934                 ctrl->ps_max_latency_us = latency;
1935                 nvme_configure_apst(ctrl);
1936         }
1937 }
1938
1939 struct nvme_core_quirk_entry {
1940         /*
1941          * NVMe model and firmware strings are padded with spaces.  For
1942          * simplicity, strings in the quirk table are padded with NULLs
1943          * instead.
1944          */
1945         u16 vid;
1946         const char *mn;
1947         const char *fr;
1948         unsigned long quirks;
1949 };
1950
1951 static const struct nvme_core_quirk_entry core_quirks[] = {
1952         {
1953                 /*
1954                  * This Toshiba device seems to die using any APST states.  See:
1955                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1956                  */
1957                 .vid = 0x1179,
1958                 .mn = "THNSF5256GPUK TOSHIBA",
1959                 .quirks = NVME_QUIRK_NO_APST,
1960         }
1961 };
1962
1963 /* match is null-terminated but idstr is space-padded. */
1964 static bool string_matches(const char *idstr, const char *match, size_t len)
1965 {
1966         size_t matchlen;
1967
1968         if (!match)
1969                 return true;
1970
1971         matchlen = strlen(match);
1972         WARN_ON_ONCE(matchlen > len);
1973
1974         if (memcmp(idstr, match, matchlen))
1975                 return false;
1976
1977         for (; matchlen < len; matchlen++)
1978                 if (idstr[matchlen] != ' ')
1979                         return false;
1980
1981         return true;
1982 }
1983
1984 static bool quirk_matches(const struct nvme_id_ctrl *id,
1985                           const struct nvme_core_quirk_entry *q)
1986 {
1987         return q->vid == le16_to_cpu(id->vid) &&
1988                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1989                 string_matches(id->fr, q->fr, sizeof(id->fr));
1990 }
1991
1992 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
1993                 struct nvme_id_ctrl *id)
1994 {
1995         size_t nqnlen;
1996         int off;
1997
1998         nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
1999         if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2000                 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2001                 return;
2002         }
2003
2004         if (ctrl->vs >= NVME_VS(1, 2, 1))
2005                 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2006
2007         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2008         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2009                         "nqn.2014.08.org.nvmexpress:%4x%4x",
2010                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2011         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2012         off += sizeof(id->sn);
2013         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2014         off += sizeof(id->mn);
2015         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2016 }
2017
2018 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2019 {
2020         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2021         kfree(subsys);
2022 }
2023
2024 static void nvme_release_subsystem(struct device *dev)
2025 {
2026         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2027 }
2028
2029 static void nvme_destroy_subsystem(struct kref *ref)
2030 {
2031         struct nvme_subsystem *subsys =
2032                         container_of(ref, struct nvme_subsystem, ref);
2033
2034         mutex_lock(&nvme_subsystems_lock);
2035         list_del(&subsys->entry);
2036         mutex_unlock(&nvme_subsystems_lock);
2037
2038         ida_destroy(&subsys->ns_ida);
2039         device_del(&subsys->dev);
2040         put_device(&subsys->dev);
2041 }
2042
2043 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2044 {
2045         kref_put(&subsys->ref, nvme_destroy_subsystem);
2046 }
2047
2048 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2049 {
2050         struct nvme_subsystem *subsys;
2051
2052         lockdep_assert_held(&nvme_subsystems_lock);
2053
2054         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2055                 if (strcmp(subsys->subnqn, subsysnqn))
2056                         continue;
2057                 if (!kref_get_unless_zero(&subsys->ref))
2058                         continue;
2059                 return subsys;
2060         }
2061
2062         return NULL;
2063 }
2064
2065 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2066         struct device_attribute subsys_attr_##_name = \
2067                 __ATTR(_name, _mode, _show, NULL)
2068
2069 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2070                                     struct device_attribute *attr,
2071                                     char *buf)
2072 {
2073         struct nvme_subsystem *subsys =
2074                 container_of(dev, struct nvme_subsystem, dev);
2075
2076         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2077 }
2078 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2079
2080 #define nvme_subsys_show_str_function(field)                            \
2081 static ssize_t subsys_##field##_show(struct device *dev,                \
2082                             struct device_attribute *attr, char *buf)   \
2083 {                                                                       \
2084         struct nvme_subsystem *subsys =                                 \
2085                 container_of(dev, struct nvme_subsystem, dev);          \
2086         return sprintf(buf, "%.*s\n",                                   \
2087                        (int)sizeof(subsys->field), subsys->field);      \
2088 }                                                                       \
2089 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2090
2091 nvme_subsys_show_str_function(model);
2092 nvme_subsys_show_str_function(serial);
2093 nvme_subsys_show_str_function(firmware_rev);
2094
2095 static struct attribute *nvme_subsys_attrs[] = {
2096         &subsys_attr_model.attr,
2097         &subsys_attr_serial.attr,
2098         &subsys_attr_firmware_rev.attr,
2099         &subsys_attr_subsysnqn.attr,
2100         NULL,
2101 };
2102
2103 static struct attribute_group nvme_subsys_attrs_group = {
2104         .attrs = nvme_subsys_attrs,
2105 };
2106
2107 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2108         &nvme_subsys_attrs_group,
2109         NULL,
2110 };
2111
2112 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2113 {
2114         int count = 0;
2115         struct nvme_ctrl *ctrl;
2116
2117         mutex_lock(&subsys->lock);
2118         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2119                 if (ctrl->state != NVME_CTRL_DELETING &&
2120                     ctrl->state != NVME_CTRL_DEAD)
2121                         count++;
2122         }
2123         mutex_unlock(&subsys->lock);
2124
2125         return count;
2126 }
2127
2128 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2129 {
2130         struct nvme_subsystem *subsys, *found;
2131         int ret;
2132
2133         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2134         if (!subsys)
2135                 return -ENOMEM;
2136         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2137         if (ret < 0) {
2138                 kfree(subsys);
2139                 return ret;
2140         }
2141         subsys->instance = ret;
2142         mutex_init(&subsys->lock);
2143         kref_init(&subsys->ref);
2144         INIT_LIST_HEAD(&subsys->ctrls);
2145         INIT_LIST_HEAD(&subsys->nsheads);
2146         nvme_init_subnqn(subsys, ctrl, id);
2147         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2148         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2149         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2150         subsys->vendor_id = le16_to_cpu(id->vid);
2151         subsys->cmic = id->cmic;
2152
2153         subsys->dev.class = nvme_subsys_class;
2154         subsys->dev.release = nvme_release_subsystem;
2155         subsys->dev.groups = nvme_subsys_attrs_groups;
2156         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2157         device_initialize(&subsys->dev);
2158
2159         mutex_lock(&nvme_subsystems_lock);
2160         found = __nvme_find_get_subsystem(subsys->subnqn);
2161         if (found) {
2162                 /*
2163                  * Verify that the subsystem actually supports multiple
2164                  * controllers, else bail out.
2165                  */
2166                 if (nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2167                         dev_err(ctrl->device,
2168                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2169                                 found->subnqn);
2170                         nvme_put_subsystem(found);
2171                         ret = -EINVAL;
2172                         goto out_unlock;
2173                 }
2174
2175                 __nvme_release_subsystem(subsys);
2176                 subsys = found;
2177         } else {
2178                 ret = device_add(&subsys->dev);
2179                 if (ret) {
2180                         dev_err(ctrl->device,
2181                                 "failed to register subsystem device.\n");
2182                         goto out_unlock;
2183                 }
2184                 ida_init(&subsys->ns_ida);
2185                 list_add_tail(&subsys->entry, &nvme_subsystems);
2186         }
2187
2188         ctrl->subsys = subsys;
2189         mutex_unlock(&nvme_subsystems_lock);
2190
2191         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2192                         dev_name(ctrl->device))) {
2193                 dev_err(ctrl->device,
2194                         "failed to create sysfs link from subsystem.\n");
2195                 /* the transport driver will eventually put the subsystem */
2196                 return -EINVAL;
2197         }
2198
2199         mutex_lock(&subsys->lock);
2200         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2201         mutex_unlock(&subsys->lock);
2202
2203         return 0;
2204
2205 out_unlock:
2206         mutex_unlock(&nvme_subsystems_lock);
2207         put_device(&subsys->dev);
2208         return ret;
2209 }
2210
2211 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
2212                         size_t size)
2213 {
2214         struct nvme_command c = { };
2215
2216         c.common.opcode = nvme_admin_get_log_page;
2217         c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
2218         c.common.cdw10[0] = nvme_get_log_dw10(log_page, size);
2219
2220         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2221 }
2222
2223 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2224 {
2225         int ret;
2226
2227         if (!ctrl->effects)
2228                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2229
2230         if (!ctrl->effects)
2231                 return 0;
2232
2233         ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects,
2234                                         sizeof(*ctrl->effects));
2235         if (ret) {
2236                 kfree(ctrl->effects);
2237                 ctrl->effects = NULL;
2238         }
2239         return ret;
2240 }
2241
2242 /*
2243  * Initialize the cached copies of the Identify data and various controller
2244  * register in our nvme_ctrl structure.  This should be called as soon as
2245  * the admin queue is fully up and running.
2246  */
2247 int nvme_init_identify(struct nvme_ctrl *ctrl)
2248 {
2249         struct nvme_id_ctrl *id;
2250         u64 cap;
2251         int ret, page_shift;
2252         u32 max_hw_sectors;
2253         bool prev_apst_enabled;
2254
2255         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2256         if (ret) {
2257                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2258                 return ret;
2259         }
2260
2261         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2262         if (ret) {
2263                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2264                 return ret;
2265         }
2266         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2267
2268         if (ctrl->vs >= NVME_VS(1, 1, 0))
2269                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2270
2271         ret = nvme_identify_ctrl(ctrl, &id);
2272         if (ret) {
2273                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2274                 return -EIO;
2275         }
2276
2277         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2278                 ret = nvme_get_effects_log(ctrl);
2279                 if (ret < 0)
2280                         return ret;
2281         }
2282
2283         if (!ctrl->identified) {
2284                 int i;
2285
2286                 ret = nvme_init_subsystem(ctrl, id);
2287                 if (ret)
2288                         goto out_free;
2289
2290                 /*
2291                  * Check for quirks.  Quirk can depend on firmware version,
2292                  * so, in principle, the set of quirks present can change
2293                  * across a reset.  As a possible future enhancement, we
2294                  * could re-scan for quirks every time we reinitialize
2295                  * the device, but we'd have to make sure that the driver
2296                  * behaves intelligently if the quirks change.
2297                  */
2298                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2299                         if (quirk_matches(id, &core_quirks[i]))
2300                                 ctrl->quirks |= core_quirks[i].quirks;
2301                 }
2302         }
2303
2304         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2305                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2306                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2307         }
2308
2309         ctrl->oacs = le16_to_cpu(id->oacs);
2310         ctrl->oncs = le16_to_cpup(&id->oncs);
2311         atomic_set(&ctrl->abort_limit, id->acl + 1);
2312         ctrl->vwc = id->vwc;
2313         ctrl->cntlid = le16_to_cpup(&id->cntlid);
2314         if (id->mdts)
2315                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2316         else
2317                 max_hw_sectors = UINT_MAX;
2318         ctrl->max_hw_sectors =
2319                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2320
2321         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2322         ctrl->sgls = le32_to_cpu(id->sgls);
2323         ctrl->kas = le16_to_cpu(id->kas);
2324
2325         if (id->rtd3e) {
2326                 /* us -> s */
2327                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2328
2329                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2330                                                  shutdown_timeout, 60);
2331
2332                 if (ctrl->shutdown_timeout != shutdown_timeout)
2333                         dev_info(ctrl->device,
2334                                  "Shutdown timeout set to %u seconds\n",
2335                                  ctrl->shutdown_timeout);
2336         } else
2337                 ctrl->shutdown_timeout = shutdown_timeout;
2338
2339         ctrl->npss = id->npss;
2340         ctrl->apsta = id->apsta;
2341         prev_apst_enabled = ctrl->apst_enabled;
2342         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2343                 if (force_apst && id->apsta) {
2344                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2345                         ctrl->apst_enabled = true;
2346                 } else {
2347                         ctrl->apst_enabled = false;
2348                 }
2349         } else {
2350                 ctrl->apst_enabled = id->apsta;
2351         }
2352         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2353
2354         if (ctrl->ops->flags & NVME_F_FABRICS) {
2355                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2356                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2357                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2358                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2359
2360                 /*
2361                  * In fabrics we need to verify the cntlid matches the
2362                  * admin connect
2363                  */
2364                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2365                         ret = -EINVAL;
2366                         goto out_free;
2367                 }
2368
2369                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2370                         dev_err(ctrl->device,
2371                                 "keep-alive support is mandatory for fabrics\n");
2372                         ret = -EINVAL;
2373                         goto out_free;
2374                 }
2375         } else {
2376                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2377                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2378                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2379                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2380                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2381         }
2382
2383         kfree(id);
2384
2385         if (ctrl->apst_enabled && !prev_apst_enabled)
2386                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2387         else if (!ctrl->apst_enabled && prev_apst_enabled)
2388                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2389
2390         ret = nvme_configure_apst(ctrl);
2391         if (ret < 0)
2392                 return ret;
2393
2394         ret = nvme_configure_timestamp(ctrl);
2395         if (ret < 0)
2396                 return ret;
2397
2398         ret = nvme_configure_directives(ctrl);
2399         if (ret < 0)
2400                 return ret;
2401
2402         ctrl->identified = true;
2403
2404         return 0;
2405
2406 out_free:
2407         kfree(id);
2408         return ret;
2409 }
2410 EXPORT_SYMBOL_GPL(nvme_init_identify);
2411
2412 static int nvme_dev_open(struct inode *inode, struct file *file)
2413 {
2414         struct nvme_ctrl *ctrl =
2415                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2416
2417         switch (ctrl->state) {
2418         case NVME_CTRL_LIVE:
2419         case NVME_CTRL_ADMIN_ONLY:
2420                 break;
2421         default:
2422                 return -EWOULDBLOCK;
2423         }
2424
2425         file->private_data = ctrl;
2426         return 0;
2427 }
2428
2429 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2430 {
2431         struct nvme_ns *ns;
2432         int ret;
2433
2434         mutex_lock(&ctrl->namespaces_mutex);
2435         if (list_empty(&ctrl->namespaces)) {
2436                 ret = -ENOTTY;
2437                 goto out_unlock;
2438         }
2439
2440         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2441         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2442                 dev_warn(ctrl->device,
2443                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2444                 ret = -EINVAL;
2445                 goto out_unlock;
2446         }
2447
2448         dev_warn(ctrl->device,
2449                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2450         kref_get(&ns->kref);
2451         mutex_unlock(&ctrl->namespaces_mutex);
2452
2453         ret = nvme_user_cmd(ctrl, ns, argp);
2454         nvme_put_ns(ns);
2455         return ret;
2456
2457 out_unlock:
2458         mutex_unlock(&ctrl->namespaces_mutex);
2459         return ret;
2460 }
2461
2462 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2463                 unsigned long arg)
2464 {
2465         struct nvme_ctrl *ctrl = file->private_data;
2466         void __user *argp = (void __user *)arg;
2467
2468         switch (cmd) {
2469         case NVME_IOCTL_ADMIN_CMD:
2470                 return nvme_user_cmd(ctrl, NULL, argp);
2471         case NVME_IOCTL_IO_CMD:
2472                 return nvme_dev_user_cmd(ctrl, argp);
2473         case NVME_IOCTL_RESET:
2474                 dev_warn(ctrl->device, "resetting controller\n");
2475                 return nvme_reset_ctrl_sync(ctrl);
2476         case NVME_IOCTL_SUBSYS_RESET:
2477                 return nvme_reset_subsystem(ctrl);
2478         case NVME_IOCTL_RESCAN:
2479                 nvme_queue_scan(ctrl);
2480                 return 0;
2481         default:
2482                 return -ENOTTY;
2483         }
2484 }
2485
2486 static const struct file_operations nvme_dev_fops = {
2487         .owner          = THIS_MODULE,
2488         .open           = nvme_dev_open,
2489         .unlocked_ioctl = nvme_dev_ioctl,
2490         .compat_ioctl   = nvme_dev_ioctl,
2491 };
2492
2493 static ssize_t nvme_sysfs_reset(struct device *dev,
2494                                 struct device_attribute *attr, const char *buf,
2495                                 size_t count)
2496 {
2497         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2498         int ret;
2499
2500         ret = nvme_reset_ctrl_sync(ctrl);
2501         if (ret < 0)
2502                 return ret;
2503         return count;
2504 }
2505 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2506
2507 static ssize_t nvme_sysfs_rescan(struct device *dev,
2508                                 struct device_attribute *attr, const char *buf,
2509                                 size_t count)
2510 {
2511         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2512
2513         nvme_queue_scan(ctrl);
2514         return count;
2515 }
2516 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2517
2518 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2519 {
2520         struct gendisk *disk = dev_to_disk(dev);
2521
2522         if (disk->fops == &nvme_fops)
2523                 return nvme_get_ns_from_dev(dev)->head;
2524         else
2525                 return disk->private_data;
2526 }
2527
2528 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2529                 char *buf)
2530 {
2531         struct nvme_ns_head *head = dev_to_ns_head(dev);
2532         struct nvme_ns_ids *ids = &head->ids;
2533         struct nvme_subsystem *subsys = head->subsys;
2534         int serial_len = sizeof(subsys->serial);
2535         int model_len = sizeof(subsys->model);
2536
2537         if (!uuid_is_null(&ids->uuid))
2538                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2539
2540         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2541                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2542
2543         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2544                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2545
2546         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2547                                   subsys->serial[serial_len - 1] == '\0'))
2548                 serial_len--;
2549         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2550                                  subsys->model[model_len - 1] == '\0'))
2551                 model_len--;
2552
2553         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2554                 serial_len, subsys->serial, model_len, subsys->model,
2555                 head->ns_id);
2556 }
2557 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
2558
2559 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2560                 char *buf)
2561 {
2562         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2563 }
2564 static DEVICE_ATTR(nguid, S_IRUGO, nguid_show, NULL);
2565
2566 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2567                 char *buf)
2568 {
2569         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2570
2571         /* For backward compatibility expose the NGUID to userspace if
2572          * we have no UUID set
2573          */
2574         if (uuid_is_null(&ids->uuid)) {
2575                 printk_ratelimited(KERN_WARNING
2576                                    "No UUID available providing old NGUID\n");
2577                 return sprintf(buf, "%pU\n", ids->nguid);
2578         }
2579         return sprintf(buf, "%pU\n", &ids->uuid);
2580 }
2581 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
2582
2583 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2584                 char *buf)
2585 {
2586         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2587 }
2588 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
2589
2590 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2591                 char *buf)
2592 {
2593         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2594 }
2595 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
2596
2597 static struct attribute *nvme_ns_id_attrs[] = {
2598         &dev_attr_wwid.attr,
2599         &dev_attr_uuid.attr,
2600         &dev_attr_nguid.attr,
2601         &dev_attr_eui.attr,
2602         &dev_attr_nsid.attr,
2603         NULL,
2604 };
2605
2606 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2607                 struct attribute *a, int n)
2608 {
2609         struct device *dev = container_of(kobj, struct device, kobj);
2610         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2611
2612         if (a == &dev_attr_uuid.attr) {
2613                 if (uuid_is_null(&ids->uuid) &&
2614                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2615                         return 0;
2616         }
2617         if (a == &dev_attr_nguid.attr) {
2618                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2619                         return 0;
2620         }
2621         if (a == &dev_attr_eui.attr) {
2622                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2623                         return 0;
2624         }
2625         return a->mode;
2626 }
2627
2628 const struct attribute_group nvme_ns_id_attr_group = {
2629         .attrs          = nvme_ns_id_attrs,
2630         .is_visible     = nvme_ns_id_attrs_are_visible,
2631 };
2632
2633 #define nvme_show_str_function(field)                                           \
2634 static ssize_t  field##_show(struct device *dev,                                \
2635                             struct device_attribute *attr, char *buf)           \
2636 {                                                                               \
2637         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2638         return sprintf(buf, "%.*s\n",                                           \
2639                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2640 }                                                                               \
2641 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2642
2643 nvme_show_str_function(model);
2644 nvme_show_str_function(serial);
2645 nvme_show_str_function(firmware_rev);
2646
2647 #define nvme_show_int_function(field)                                           \
2648 static ssize_t  field##_show(struct device *dev,                                \
2649                             struct device_attribute *attr, char *buf)           \
2650 {                                                                               \
2651         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2652         return sprintf(buf, "%d\n", ctrl->field);       \
2653 }                                                                               \
2654 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2655
2656 nvme_show_int_function(cntlid);
2657
2658 static ssize_t nvme_sysfs_delete(struct device *dev,
2659                                 struct device_attribute *attr, const char *buf,
2660                                 size_t count)
2661 {
2662         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2663
2664         if (device_remove_file_self(dev, attr))
2665                 nvme_delete_ctrl_sync(ctrl);
2666         return count;
2667 }
2668 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2669
2670 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2671                                          struct device_attribute *attr,
2672                                          char *buf)
2673 {
2674         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2675
2676         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2677 }
2678 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2679
2680 static ssize_t nvme_sysfs_show_state(struct device *dev,
2681                                      struct device_attribute *attr,
2682                                      char *buf)
2683 {
2684         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2685         static const char *const state_name[] = {
2686                 [NVME_CTRL_NEW]         = "new",
2687                 [NVME_CTRL_LIVE]        = "live",
2688                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2689                 [NVME_CTRL_RESETTING]   = "resetting",
2690                 [NVME_CTRL_RECONNECTING]= "reconnecting",
2691                 [NVME_CTRL_DELETING]    = "deleting",
2692                 [NVME_CTRL_DEAD]        = "dead",
2693         };
2694
2695         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2696             state_name[ctrl->state])
2697                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2698
2699         return sprintf(buf, "unknown state\n");
2700 }
2701
2702 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2703
2704 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2705                                          struct device_attribute *attr,
2706                                          char *buf)
2707 {
2708         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2709
2710         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2711 }
2712 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2713
2714 static ssize_t nvme_sysfs_show_address(struct device *dev,
2715                                          struct device_attribute *attr,
2716                                          char *buf)
2717 {
2718         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2719
2720         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2721 }
2722 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2723
2724 static struct attribute *nvme_dev_attrs[] = {
2725         &dev_attr_reset_controller.attr,
2726         &dev_attr_rescan_controller.attr,
2727         &dev_attr_model.attr,
2728         &dev_attr_serial.attr,
2729         &dev_attr_firmware_rev.attr,
2730         &dev_attr_cntlid.attr,
2731         &dev_attr_delete_controller.attr,
2732         &dev_attr_transport.attr,
2733         &dev_attr_subsysnqn.attr,
2734         &dev_attr_address.attr,
2735         &dev_attr_state.attr,
2736         NULL
2737 };
2738
2739 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2740                 struct attribute *a, int n)
2741 {
2742         struct device *dev = container_of(kobj, struct device, kobj);
2743         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2744
2745         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2746                 return 0;
2747         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2748                 return 0;
2749
2750         return a->mode;
2751 }
2752
2753 static struct attribute_group nvme_dev_attrs_group = {
2754         .attrs          = nvme_dev_attrs,
2755         .is_visible     = nvme_dev_attrs_are_visible,
2756 };
2757
2758 static const struct attribute_group *nvme_dev_attr_groups[] = {
2759         &nvme_dev_attrs_group,
2760         NULL,
2761 };
2762
2763 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2764                 unsigned nsid)
2765 {
2766         struct nvme_ns_head *h;
2767
2768         lockdep_assert_held(&subsys->lock);
2769
2770         list_for_each_entry(h, &subsys->nsheads, entry) {
2771                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2772                         return h;
2773         }
2774
2775         return NULL;
2776 }
2777
2778 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2779                 struct nvme_ns_head *new)
2780 {
2781         struct nvme_ns_head *h;
2782
2783         lockdep_assert_held(&subsys->lock);
2784
2785         list_for_each_entry(h, &subsys->nsheads, entry) {
2786                 if (nvme_ns_ids_valid(&new->ids) &&
2787                     nvme_ns_ids_equal(&new->ids, &h->ids))
2788                         return -EINVAL;
2789         }
2790
2791         return 0;
2792 }
2793
2794 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2795                 unsigned nsid, struct nvme_id_ns *id)
2796 {
2797         struct nvme_ns_head *head;
2798         int ret = -ENOMEM;
2799
2800         head = kzalloc(sizeof(*head), GFP_KERNEL);
2801         if (!head)
2802                 goto out;
2803         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2804         if (ret < 0)
2805                 goto out_free_head;
2806         head->instance = ret;
2807         INIT_LIST_HEAD(&head->list);
2808         init_srcu_struct(&head->srcu);
2809         head->subsys = ctrl->subsys;
2810         head->ns_id = nsid;
2811         kref_init(&head->ref);
2812
2813         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2814
2815         ret = __nvme_check_ids(ctrl->subsys, head);
2816         if (ret) {
2817                 dev_err(ctrl->device,
2818                         "duplicate IDs for nsid %d\n", nsid);
2819                 goto out_cleanup_srcu;
2820         }
2821
2822         ret = nvme_mpath_alloc_disk(ctrl, head);
2823         if (ret)
2824                 goto out_cleanup_srcu;
2825
2826         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2827         return head;
2828 out_cleanup_srcu:
2829         cleanup_srcu_struct(&head->srcu);
2830         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2831 out_free_head:
2832         kfree(head);
2833 out:
2834         return ERR_PTR(ret);
2835 }
2836
2837 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2838                 struct nvme_id_ns *id, bool *new)
2839 {
2840         struct nvme_ctrl *ctrl = ns->ctrl;
2841         bool is_shared = id->nmic & (1 << 0);
2842         struct nvme_ns_head *head = NULL;
2843         int ret = 0;
2844
2845         mutex_lock(&ctrl->subsys->lock);
2846         if (is_shared)
2847                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2848         if (!head) {
2849                 head = nvme_alloc_ns_head(ctrl, nsid, id);
2850                 if (IS_ERR(head)) {
2851                         ret = PTR_ERR(head);
2852                         goto out_unlock;
2853                 }
2854
2855                 *new = true;
2856         } else {
2857                 struct nvme_ns_ids ids;
2858
2859                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
2860                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2861                         dev_err(ctrl->device,
2862                                 "IDs don't match for shared namespace %d\n",
2863                                         nsid);
2864                         ret = -EINVAL;
2865                         goto out_unlock;
2866                 }
2867
2868                 *new = false;
2869         }
2870
2871         list_add_tail(&ns->siblings, &head->list);
2872         ns->head = head;
2873
2874 out_unlock:
2875         mutex_unlock(&ctrl->subsys->lock);
2876         return ret;
2877 }
2878
2879 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2880 {
2881         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2882         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2883
2884         return nsa->head->ns_id - nsb->head->ns_id;
2885 }
2886
2887 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2888 {
2889         struct nvme_ns *ns, *ret = NULL;
2890
2891         mutex_lock(&ctrl->namespaces_mutex);
2892         list_for_each_entry(ns, &ctrl->namespaces, list) {
2893                 if (ns->head->ns_id == nsid) {
2894                         if (!kref_get_unless_zero(&ns->kref))
2895                                 continue;
2896                         ret = ns;
2897                         break;
2898                 }
2899                 if (ns->head->ns_id > nsid)
2900                         break;
2901         }
2902         mutex_unlock(&ctrl->namespaces_mutex);
2903         return ret;
2904 }
2905
2906 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2907 {
2908         struct streams_directive_params s;
2909         int ret;
2910
2911         if (!ctrl->nr_streams)
2912                 return 0;
2913
2914         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2915         if (ret)
2916                 return ret;
2917
2918         ns->sws = le32_to_cpu(s.sws);
2919         ns->sgs = le16_to_cpu(s.sgs);
2920
2921         if (ns->sws) {
2922                 unsigned int bs = 1 << ns->lba_shift;
2923
2924                 blk_queue_io_min(ns->queue, bs * ns->sws);
2925                 if (ns->sgs)
2926                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2927         }
2928
2929         return 0;
2930 }
2931
2932 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2933 {
2934         struct nvme_ns *ns;
2935         struct gendisk *disk;
2936         struct nvme_id_ns *id;
2937         char disk_name[DISK_NAME_LEN];
2938         int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
2939         bool new = true;
2940
2941         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2942         if (!ns)
2943                 return;
2944
2945         ns->queue = blk_mq_init_queue(ctrl->tagset);
2946         if (IS_ERR(ns->queue))
2947                 goto out_free_ns;
2948         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2949         ns->queue->queuedata = ns;
2950         ns->ctrl = ctrl;
2951
2952         kref_init(&ns->kref);
2953         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2954
2955         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2956         nvme_set_queue_limits(ctrl, ns->queue);
2957
2958         id = nvme_identify_ns(ctrl, nsid);
2959         if (!id)
2960                 goto out_free_queue;
2961
2962         if (id->ncap == 0)
2963                 goto out_free_id;
2964
2965         if (nvme_init_ns_head(ns, nsid, id, &new))
2966                 goto out_free_id;
2967         nvme_setup_streams_ns(ctrl, ns);
2968
2969 #ifdef CONFIG_NVME_MULTIPATH
2970         /*
2971          * If multipathing is enabled we need to always use the subsystem
2972          * instance number for numbering our devices to avoid conflicts
2973          * between subsystems that have multiple controllers and thus use
2974          * the multipath-aware subsystem node and those that have a single
2975          * controller and use the controller node directly.
2976          */
2977         if (ns->head->disk) {
2978                 sprintf(disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
2979                                 ctrl->cntlid, ns->head->instance);
2980                 flags = GENHD_FL_HIDDEN;
2981         } else {
2982                 sprintf(disk_name, "nvme%dn%d", ctrl->subsys->instance,
2983                                 ns->head->instance);
2984         }
2985 #else
2986         /*
2987          * But without the multipath code enabled, multiple controller per
2988          * subsystems are visible as devices and thus we cannot use the
2989          * subsystem instance.
2990          */
2991         sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
2992 #endif
2993
2994         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
2995                 if (nvme_nvm_register(ns, disk_name, node)) {
2996                         dev_warn(ctrl->device, "LightNVM init failure\n");
2997                         goto out_unlink_ns;
2998                 }
2999         }
3000
3001         disk = alloc_disk_node(0, node);
3002         if (!disk)
3003                 goto out_unlink_ns;
3004
3005         disk->fops = &nvme_fops;
3006         disk->private_data = ns;
3007         disk->queue = ns->queue;
3008         disk->flags = flags;
3009         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3010         ns->disk = disk;
3011
3012         __nvme_revalidate_disk(disk, id);
3013
3014         mutex_lock(&ctrl->namespaces_mutex);
3015         list_add_tail(&ns->list, &ctrl->namespaces);
3016         mutex_unlock(&ctrl->namespaces_mutex);
3017
3018         nvme_get_ctrl(ctrl);
3019
3020         kfree(id);
3021
3022         device_add_disk(ctrl->device, ns->disk);
3023         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
3024                                         &nvme_ns_id_attr_group))
3025                 pr_warn("%s: failed to create sysfs group for identification\n",
3026                         ns->disk->disk_name);
3027         if (ns->ndev && nvme_nvm_register_sysfs(ns))
3028                 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
3029                         ns->disk->disk_name);
3030
3031         if (new)
3032                 nvme_mpath_add_disk(ns->head);
3033         nvme_mpath_add_disk_links(ns);
3034         return;
3035  out_unlink_ns:
3036         mutex_lock(&ctrl->subsys->lock);
3037         list_del_rcu(&ns->siblings);
3038         mutex_unlock(&ctrl->subsys->lock);
3039  out_free_id:
3040         kfree(id);
3041  out_free_queue:
3042         blk_cleanup_queue(ns->queue);
3043  out_free_ns:
3044         kfree(ns);
3045 }
3046
3047 static void nvme_ns_remove(struct nvme_ns *ns)
3048 {
3049         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3050                 return;
3051
3052         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3053                 nvme_mpath_remove_disk_links(ns);
3054                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
3055                                         &nvme_ns_id_attr_group);
3056                 if (ns->ndev)
3057                         nvme_nvm_unregister_sysfs(ns);
3058                 del_gendisk(ns->disk);
3059                 blk_cleanup_queue(ns->queue);
3060                 if (blk_get_integrity(ns->disk))
3061                         blk_integrity_unregister(ns->disk);
3062         }
3063
3064         mutex_lock(&ns->ctrl->subsys->lock);
3065         nvme_mpath_clear_current_path(ns);
3066         list_del_rcu(&ns->siblings);
3067         mutex_unlock(&ns->ctrl->subsys->lock);
3068
3069         mutex_lock(&ns->ctrl->namespaces_mutex);
3070         list_del_init(&ns->list);
3071         mutex_unlock(&ns->ctrl->namespaces_mutex);
3072
3073         synchronize_srcu(&ns->head->srcu);
3074         nvme_mpath_check_last_path(ns);
3075         nvme_put_ns(ns);
3076 }
3077
3078 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3079 {
3080         struct nvme_ns *ns;
3081
3082         ns = nvme_find_get_ns(ctrl, nsid);
3083         if (ns) {
3084                 if (ns->disk && revalidate_disk(ns->disk))
3085                         nvme_ns_remove(ns);
3086                 nvme_put_ns(ns);
3087         } else
3088                 nvme_alloc_ns(ctrl, nsid);
3089 }
3090
3091 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3092                                         unsigned nsid)
3093 {
3094         struct nvme_ns *ns, *next;
3095
3096         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3097                 if (ns->head->ns_id > nsid)
3098                         nvme_ns_remove(ns);
3099         }
3100 }
3101
3102 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3103 {
3104         struct nvme_ns *ns;
3105         __le32 *ns_list;
3106         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3107         int ret = 0;
3108
3109         ns_list = kzalloc(0x1000, GFP_KERNEL);
3110         if (!ns_list)
3111                 return -ENOMEM;
3112
3113         for (i = 0; i < num_lists; i++) {
3114                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3115                 if (ret)
3116                         goto free;
3117
3118                 for (j = 0; j < min(nn, 1024U); j++) {
3119                         nsid = le32_to_cpu(ns_list[j]);
3120                         if (!nsid)
3121                                 goto out;
3122
3123                         nvme_validate_ns(ctrl, nsid);
3124
3125                         while (++prev < nsid) {
3126                                 ns = nvme_find_get_ns(ctrl, prev);
3127                                 if (ns) {
3128                                         nvme_ns_remove(ns);
3129                                         nvme_put_ns(ns);
3130                                 }
3131                         }
3132                 }
3133                 nn -= j;
3134         }
3135  out:
3136         nvme_remove_invalid_namespaces(ctrl, prev);
3137  free:
3138         kfree(ns_list);
3139         return ret;
3140 }
3141
3142 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3143 {
3144         unsigned i;
3145
3146         for (i = 1; i <= nn; i++)
3147                 nvme_validate_ns(ctrl, i);
3148
3149         nvme_remove_invalid_namespaces(ctrl, nn);
3150 }
3151
3152 static void nvme_scan_work(struct work_struct *work)
3153 {
3154         struct nvme_ctrl *ctrl =
3155                 container_of(work, struct nvme_ctrl, scan_work);
3156         struct nvme_id_ctrl *id;
3157         unsigned nn;
3158
3159         if (ctrl->state != NVME_CTRL_LIVE)
3160                 return;
3161
3162         WARN_ON_ONCE(!ctrl->tagset);
3163
3164         if (nvme_identify_ctrl(ctrl, &id))
3165                 return;
3166
3167         nn = le32_to_cpu(id->nn);
3168         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3169             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3170                 if (!nvme_scan_ns_list(ctrl, nn))
3171                         goto done;
3172         }
3173         nvme_scan_ns_sequential(ctrl, nn);
3174  done:
3175         mutex_lock(&ctrl->namespaces_mutex);
3176         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3177         mutex_unlock(&ctrl->namespaces_mutex);
3178         kfree(id);
3179 }
3180
3181 void nvme_queue_scan(struct nvme_ctrl *ctrl)
3182 {
3183         /*
3184          * Only new queue scan work when admin and IO queues are both alive
3185          */
3186         if (ctrl->state == NVME_CTRL_LIVE)
3187                 queue_work(nvme_wq, &ctrl->scan_work);
3188 }
3189 EXPORT_SYMBOL_GPL(nvme_queue_scan);
3190
3191 /*
3192  * This function iterates the namespace list unlocked to allow recovery from
3193  * controller failure. It is up to the caller to ensure the namespace list is
3194  * not modified by scan work while this function is executing.
3195  */
3196 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3197 {
3198         struct nvme_ns *ns, *next;
3199
3200         /*
3201          * The dead states indicates the controller was not gracefully
3202          * disconnected. In that case, we won't be able to flush any data while
3203          * removing the namespaces' disks; fail all the queues now to avoid
3204          * potentially having to clean up the failed sync later.
3205          */
3206         if (ctrl->state == NVME_CTRL_DEAD)
3207                 nvme_kill_queues(ctrl);
3208
3209         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
3210                 nvme_ns_remove(ns);
3211 }
3212 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3213
3214 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3215 {
3216         char *envp[2] = { NULL, NULL };
3217         u32 aen_result = ctrl->aen_result;
3218
3219         ctrl->aen_result = 0;
3220         if (!aen_result)
3221                 return;
3222
3223         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3224         if (!envp[0])
3225                 return;
3226         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3227         kfree(envp[0]);
3228 }
3229
3230 static void nvme_async_event_work(struct work_struct *work)
3231 {
3232         struct nvme_ctrl *ctrl =
3233                 container_of(work, struct nvme_ctrl, async_event_work);
3234
3235         nvme_aen_uevent(ctrl);
3236         ctrl->ops->submit_async_event(ctrl);
3237 }
3238
3239 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3240 {
3241
3242         u32 csts;
3243
3244         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3245                 return false;
3246
3247         if (csts == ~0)
3248                 return false;
3249
3250         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3251 }
3252
3253 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3254 {
3255         struct nvme_fw_slot_info_log *log;
3256
3257         log = kmalloc(sizeof(*log), GFP_KERNEL);
3258         if (!log)
3259                 return;
3260
3261         if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
3262                 dev_warn(ctrl->device,
3263                                 "Get FW SLOT INFO log error\n");
3264         kfree(log);
3265 }
3266
3267 static void nvme_fw_act_work(struct work_struct *work)
3268 {
3269         struct nvme_ctrl *ctrl = container_of(work,
3270                                 struct nvme_ctrl, fw_act_work);
3271         unsigned long fw_act_timeout;
3272
3273         if (ctrl->mtfa)
3274                 fw_act_timeout = jiffies +
3275                                 msecs_to_jiffies(ctrl->mtfa * 100);
3276         else
3277                 fw_act_timeout = jiffies +
3278                                 msecs_to_jiffies(admin_timeout * 1000);
3279
3280         nvme_stop_queues(ctrl);
3281         while (nvme_ctrl_pp_status(ctrl)) {
3282                 if (time_after(jiffies, fw_act_timeout)) {
3283                         dev_warn(ctrl->device,
3284                                 "Fw activation timeout, reset controller\n");
3285                         nvme_reset_ctrl(ctrl);
3286                         break;
3287                 }
3288                 msleep(100);
3289         }
3290
3291         if (ctrl->state != NVME_CTRL_LIVE)
3292                 return;
3293
3294         nvme_start_queues(ctrl);
3295         /* read FW slot information to clear the AER */
3296         nvme_get_fw_slot_info(ctrl);
3297 }
3298
3299 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3300                 union nvme_result *res)
3301 {
3302         u32 result = le32_to_cpu(res->u32);
3303
3304         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3305                 return;
3306
3307         switch (result & 0x7) {
3308         case NVME_AER_ERROR:
3309         case NVME_AER_SMART:
3310         case NVME_AER_CSS:
3311         case NVME_AER_VS:
3312                 ctrl->aen_result = result;
3313                 break;
3314         default:
3315                 break;
3316         }
3317
3318         switch (result & 0xff07) {
3319         case NVME_AER_NOTICE_NS_CHANGED:
3320                 dev_info(ctrl->device, "rescanning\n");
3321                 nvme_queue_scan(ctrl);
3322                 break;
3323         case NVME_AER_NOTICE_FW_ACT_STARTING:
3324                 queue_work(nvme_wq, &ctrl->fw_act_work);
3325                 break;
3326         default:
3327                 dev_warn(ctrl->device, "async event result %08x\n", result);
3328         }
3329         queue_work(nvme_wq, &ctrl->async_event_work);
3330 }
3331 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3332
3333 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3334 {
3335         nvme_stop_keep_alive(ctrl);
3336         flush_work(&ctrl->async_event_work);
3337         flush_work(&ctrl->scan_work);
3338         cancel_work_sync(&ctrl->fw_act_work);
3339 }
3340 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3341
3342 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3343 {
3344         if (ctrl->kato)
3345                 nvme_start_keep_alive(ctrl);
3346
3347         if (ctrl->queue_count > 1) {
3348                 nvme_queue_scan(ctrl);
3349                 queue_work(nvme_wq, &ctrl->async_event_work);
3350                 nvme_start_queues(ctrl);
3351         }
3352 }
3353 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3354
3355 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3356 {
3357         cdev_device_del(&ctrl->cdev, ctrl->device);
3358 }
3359 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3360
3361 static void nvme_free_ctrl(struct device *dev)
3362 {
3363         struct nvme_ctrl *ctrl =
3364                 container_of(dev, struct nvme_ctrl, ctrl_device);
3365         struct nvme_subsystem *subsys = ctrl->subsys;
3366
3367         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3368         kfree(ctrl->effects);
3369
3370         if (subsys) {
3371                 mutex_lock(&subsys->lock);
3372                 list_del(&ctrl->subsys_entry);
3373                 mutex_unlock(&subsys->lock);
3374                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3375         }
3376
3377         ctrl->ops->free_ctrl(ctrl);
3378
3379         if (subsys)
3380                 nvme_put_subsystem(subsys);
3381 }
3382
3383 /*
3384  * Initialize a NVMe controller structures.  This needs to be called during
3385  * earliest initialization so that we have the initialized structured around
3386  * during probing.
3387  */
3388 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3389                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3390 {
3391         int ret;
3392
3393         ctrl->state = NVME_CTRL_NEW;
3394         spin_lock_init(&ctrl->lock);
3395         INIT_LIST_HEAD(&ctrl->namespaces);
3396         mutex_init(&ctrl->namespaces_mutex);
3397         ctrl->dev = dev;
3398         ctrl->ops = ops;
3399         ctrl->quirks = quirks;
3400         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3401         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3402         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3403         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3404
3405         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3406         if (ret < 0)
3407                 goto out;
3408         ctrl->instance = ret;
3409
3410         device_initialize(&ctrl->ctrl_device);
3411         ctrl->device = &ctrl->ctrl_device;
3412         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3413         ctrl->device->class = nvme_class;
3414         ctrl->device->parent = ctrl->dev;
3415         ctrl->device->groups = nvme_dev_attr_groups;
3416         ctrl->device->release = nvme_free_ctrl;
3417         dev_set_drvdata(ctrl->device, ctrl);
3418         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3419         if (ret)
3420                 goto out_release_instance;
3421
3422         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3423         ctrl->cdev.owner = ops->module;
3424         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3425         if (ret)
3426                 goto out_free_name;
3427
3428         /*
3429          * Initialize latency tolerance controls.  The sysfs files won't
3430          * be visible to userspace unless the device actually supports APST.
3431          */
3432         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3433         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3434                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3435
3436         return 0;
3437 out_free_name:
3438         kfree_const(dev->kobj.name);
3439 out_release_instance:
3440         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3441 out:
3442         return ret;
3443 }
3444 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3445
3446 /**
3447  * nvme_kill_queues(): Ends all namespace queues
3448  * @ctrl: the dead controller that needs to end
3449  *
3450  * Call this function when the driver determines it is unable to get the
3451  * controller in a state capable of servicing IO.
3452  */
3453 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3454 {
3455         struct nvme_ns *ns;
3456
3457         mutex_lock(&ctrl->namespaces_mutex);
3458
3459         /* Forcibly unquiesce queues to avoid blocking dispatch */
3460         if (ctrl->admin_q)
3461                 blk_mq_unquiesce_queue(ctrl->admin_q);
3462
3463         list_for_each_entry(ns, &ctrl->namespaces, list) {
3464                 /*
3465                  * Revalidating a dead namespace sets capacity to 0. This will
3466                  * end buffered writers dirtying pages that can't be synced.
3467                  */
3468                 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
3469                         continue;
3470                 revalidate_disk(ns->disk);
3471                 blk_set_queue_dying(ns->queue);
3472
3473                 /* Forcibly unquiesce queues to avoid blocking dispatch */
3474                 blk_mq_unquiesce_queue(ns->queue);
3475         }
3476         mutex_unlock(&ctrl->namespaces_mutex);
3477 }
3478 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3479
3480 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3481 {
3482         struct nvme_ns *ns;
3483
3484         mutex_lock(&ctrl->namespaces_mutex);
3485         list_for_each_entry(ns, &ctrl->namespaces, list)
3486                 blk_mq_unfreeze_queue(ns->queue);
3487         mutex_unlock(&ctrl->namespaces_mutex);
3488 }
3489 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3490
3491 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3492 {
3493         struct nvme_ns *ns;
3494
3495         mutex_lock(&ctrl->namespaces_mutex);
3496         list_for_each_entry(ns, &ctrl->namespaces, list) {
3497                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3498                 if (timeout <= 0)
3499                         break;
3500         }
3501         mutex_unlock(&ctrl->namespaces_mutex);
3502 }
3503 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3504
3505 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3506 {
3507         struct nvme_ns *ns;
3508
3509         mutex_lock(&ctrl->namespaces_mutex);
3510         list_for_each_entry(ns, &ctrl->namespaces, list)
3511                 blk_mq_freeze_queue_wait(ns->queue);
3512         mutex_unlock(&ctrl->namespaces_mutex);
3513 }
3514 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3515
3516 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3517 {
3518         struct nvme_ns *ns;
3519
3520         mutex_lock(&ctrl->namespaces_mutex);
3521         list_for_each_entry(ns, &ctrl->namespaces, list)
3522                 blk_freeze_queue_start(ns->queue);
3523         mutex_unlock(&ctrl->namespaces_mutex);
3524 }
3525 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3526
3527 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3528 {
3529         struct nvme_ns *ns;
3530
3531         mutex_lock(&ctrl->namespaces_mutex);
3532         list_for_each_entry(ns, &ctrl->namespaces, list)
3533                 blk_mq_quiesce_queue(ns->queue);
3534         mutex_unlock(&ctrl->namespaces_mutex);
3535 }
3536 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3537
3538 void nvme_start_queues(struct nvme_ctrl *ctrl)
3539 {
3540         struct nvme_ns *ns;
3541
3542         mutex_lock(&ctrl->namespaces_mutex);
3543         list_for_each_entry(ns, &ctrl->namespaces, list)
3544                 blk_mq_unquiesce_queue(ns->queue);
3545         mutex_unlock(&ctrl->namespaces_mutex);
3546 }
3547 EXPORT_SYMBOL_GPL(nvme_start_queues);
3548
3549 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set)
3550 {
3551         if (!ctrl->ops->reinit_request)
3552                 return 0;
3553
3554         return blk_mq_tagset_iter(set, set->driver_data,
3555                         ctrl->ops->reinit_request);
3556 }
3557 EXPORT_SYMBOL_GPL(nvme_reinit_tagset);
3558
3559 int __init nvme_core_init(void)
3560 {
3561         int result = -ENOMEM;
3562
3563         nvme_wq = alloc_workqueue("nvme-wq",
3564                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3565         if (!nvme_wq)
3566                 goto out;
3567
3568         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3569                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3570         if (!nvme_reset_wq)
3571                 goto destroy_wq;
3572
3573         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3574                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3575         if (!nvme_delete_wq)
3576                 goto destroy_reset_wq;
3577
3578         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3579         if (result < 0)
3580                 goto destroy_delete_wq;
3581
3582         nvme_class = class_create(THIS_MODULE, "nvme");
3583         if (IS_ERR(nvme_class)) {
3584                 result = PTR_ERR(nvme_class);
3585                 goto unregister_chrdev;
3586         }
3587
3588         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3589         if (IS_ERR(nvme_subsys_class)) {
3590                 result = PTR_ERR(nvme_subsys_class);
3591                 goto destroy_class;
3592         }
3593         return 0;
3594
3595 destroy_class:
3596         class_destroy(nvme_class);
3597 unregister_chrdev:
3598         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3599 destroy_delete_wq:
3600         destroy_workqueue(nvme_delete_wq);
3601 destroy_reset_wq:
3602         destroy_workqueue(nvme_reset_wq);
3603 destroy_wq:
3604         destroy_workqueue(nvme_wq);
3605 out:
3606         return result;
3607 }
3608
3609 void nvme_core_exit(void)
3610 {
3611         ida_destroy(&nvme_subsystems_ida);
3612         class_destroy(nvme_subsys_class);
3613         class_destroy(nvme_class);
3614         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3615         destroy_workqueue(nvme_delete_wq);
3616         destroy_workqueue(nvme_reset_wq);
3617         destroy_workqueue(nvme_wq);
3618 }
3619
3620 MODULE_LICENSE("GPL");
3621 MODULE_VERSION("1.0");
3622 module_init(nvme_core_init);
3623 module_exit(nvme_core_exit);