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_CONNECTING:
 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_CONNECTING:
 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_CONNECTING:
 298                 switch (old_state) {
 299                 case NVME_CTRL_NEW:
 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_CONNECTING:
 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                 if (n < segments) {
 522                         range[n].cattr = cpu_to_le32(0);
 523                         range[n].nlb = cpu_to_le32(nlb);
 524                         range[n].slba = cpu_to_le64(slba);
 525                 }
 526                 n++;
 527         }
 528
 529         if (WARN_ON_ONCE(n != segments)) {
 530                 kfree(range);
 531                 return BLK_STS_IOERR;
 532         }
 533
 534         memset(cmnd, 0, sizeof(*cmnd));
 535         cmnd->dsm.opcode = nvme_cmd_dsm;
 536         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 537         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 538         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 539
 540         req->special_vec.bv_page = virt_to_page(range);
 541         req->special_vec.bv_offset = offset_in_page(range);
 542         req->special_vec.bv_len = sizeof(*range) * segments;
 543         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 544
 545         return BLK_STS_OK;
 546 }
 547
 548 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 549                 struct request *req, struct nvme_command *cmnd)
 550 {
 551         struct nvme_ctrl *ctrl = ns->ctrl;
 552         u16 control = 0;
 553         u32 dsmgmt = 0;
 554
 555         if (req->cmd_flags & REQ_FUA)
 556                 control |= NVME_RW_FUA;
 557         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 558                 control |= NVME_RW_LR;
 559
 560         if (req->cmd_flags & REQ_RAHEAD)
 561                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 562
 563         memset(cmnd, 0, sizeof(*cmnd));
 564         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 565         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 566         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 567         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 568
 569         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 570                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 571
 572         if (ns->ms) {
 573                 /*
 574                  * If formated with metadata, the block layer always provides a
 575                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 576                  * we enable the PRACT bit for protection information or set the
 577                  * namespace capacity to zero to prevent any I/O.
 578                  */
 579                 if (!blk_integrity_rq(req)) {
 580                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 581                                 return BLK_STS_NOTSUPP;
 582                         control |= NVME_RW_PRINFO_PRACT;
 583                 }
 584
 585                 switch (ns->pi_type) {
 586                 case NVME_NS_DPS_PI_TYPE3:
 587                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 588                         break;
 589                 case NVME_NS_DPS_PI_TYPE1:
 590                 case NVME_NS_DPS_PI_TYPE2:
 591                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 592                                         NVME_RW_PRINFO_PRCHK_REF;
 593                         cmnd->rw.reftag = cpu_to_le32(
 594                                         nvme_block_nr(ns, blk_rq_pos(req)));
 595                         break;
 596                 }
 597         }
 598
 599         cmnd->rw.control = cpu_to_le16(control);
 600         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 601         return 0;
 602 }
 603
 604 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 605                 struct nvme_command *cmd)
 606 {
 607         blk_status_t ret = BLK_STS_OK;
 608
 609         if (!(req->rq_flags & RQF_DONTPREP)) {
 610                 nvme_req(req)->retries = 0;
 611                 nvme_req(req)->flags = 0;
 612                 req->rq_flags |= RQF_DONTPREP;
 613         }
 614
 615         switch (req_op(req)) {
 616         case REQ_OP_DRV_IN:
 617         case REQ_OP_DRV_OUT:
 618                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 619                 break;
 620         case REQ_OP_FLUSH:
 621                 nvme_setup_flush(ns, cmd);
 622                 break;
 623         case REQ_OP_WRITE_ZEROES:
 624                 /* currently only aliased to deallocate for a few ctrls: */
 625         case REQ_OP_DISCARD:
 626                 ret = nvme_setup_discard(ns, req, cmd);
 627                 break;
 628         case REQ_OP_READ:
 629         case REQ_OP_WRITE:
 630                 ret = nvme_setup_rw(ns, req, cmd);
 631                 break;
 632         default:
 633                 WARN_ON_ONCE(1);
 634                 return BLK_STS_IOERR;
 635         }
 636
 637         cmd->common.command_id = req->tag;
 638         if (ns)
 639                 trace_nvme_setup_nvm_cmd(req->q->id, cmd);
 640         else
 641                 trace_nvme_setup_admin_cmd(cmd);
 642         return ret;
 643 }
 644 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 645
 646 /*
 647  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 648  * if the result is positive, it's an NVM Express status code
 649  */
 650 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 651                 union nvme_result *result, void *buffer, unsigned bufflen,
 652                 unsigned timeout, int qid, int at_head,
 653                 blk_mq_req_flags_t flags)
 654 {
 655         struct request *req;
 656         int ret;
 657
 658         req = nvme_alloc_request(q, cmd, flags, qid);
 659         if (IS_ERR(req))
 660                 return PTR_ERR(req);
 661
 662         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 663
 664         if (buffer && bufflen) {
 665                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 666                 if (ret)
 667                         goto out;
 668         }
 669
 670         blk_execute_rq(req->q, NULL, req, at_head);
 671         if (result)
 672                 *result = nvme_req(req)->result;
 673         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 674                 ret = -EINTR;
 675         else
 676                 ret = nvme_req(req)->status;
 677  out:
 678         blk_mq_free_request(req);
 679         return ret;
 680 }
 681 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 682
 683 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 684                 void *buffer, unsigned bufflen)
 685 {
 686         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 687                         NVME_QID_ANY, 0, 0);
 688 }
 689 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 690
 691 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 692                 unsigned len, u32 seed, bool write)
 693 {
 694         struct bio_integrity_payload *bip;
 695         int ret = -ENOMEM;
 696         void *buf;
 697
 698         buf = kmalloc(len, GFP_KERNEL);
 699         if (!buf)
 700                 goto out;
 701
 702         ret = -EFAULT;
 703         if (write && copy_from_user(buf, ubuf, len))
 704                 goto out_free_meta;
 705
 706         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 707         if (IS_ERR(bip)) {
 708                 ret = PTR_ERR(bip);
 709                 goto out_free_meta;
 710         }
 711
 712         bip->bip_iter.bi_size = len;
 713         bip->bip_iter.bi_sector = seed;
 714         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 715                         offset_in_page(buf));
 716         if (ret == len)
 717                 return buf;
 718         ret = -ENOMEM;
 719 out_free_meta:
 720         kfree(buf);
 721 out:
 722         return ERR_PTR(ret);
 723 }
 724
 725 static int nvme_submit_user_cmd(struct request_queue *q,
 726                 struct nvme_command *cmd, void __user *ubuffer,
 727                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 728                 u32 meta_seed, u32 *result, unsigned timeout)
 729 {
 730         bool write = nvme_is_write(cmd);
 731         struct nvme_ns *ns = q->queuedata;
 732         struct gendisk *disk = ns ? ns->disk : NULL;
 733         struct request *req;
 734         struct bio *bio = NULL;
 735         void *meta = NULL;
 736         int ret;
 737
 738         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 739         if (IS_ERR(req))
 740                 return PTR_ERR(req);
 741
 742         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 743
 744         if (ubuffer && bufflen) {
 745                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 746                                 GFP_KERNEL);
 747                 if (ret)
 748                         goto out;
 749                 bio = req->bio;
 750                 bio->bi_disk = disk;
 751                 if (disk && meta_buffer && meta_len) {
 752                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 753                                         meta_seed, write);
 754                         if (IS_ERR(meta)) {
 755                                 ret = PTR_ERR(meta);
 756                                 goto out_unmap;
 757                         }
 758                 }
 759         }
 760
 761         blk_execute_rq(req->q, disk, req, 0);
 762         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 763                 ret = -EINTR;
 764         else
 765                 ret = nvme_req(req)->status;
 766         if (result)
 767                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 768         if (meta && !ret && !write) {
 769                 if (copy_to_user(meta_buffer, meta, meta_len))
 770                         ret = -EFAULT;
 771         }
 772         kfree(meta);
 773  out_unmap:
 774         if (bio)
 775                 blk_rq_unmap_user(bio);
 776  out:
 777         blk_mq_free_request(req);
 778         return ret;
 779 }
 780
 781 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 782 {
 783         struct nvme_ctrl *ctrl = rq->end_io_data;
 784
 785         blk_mq_free_request(rq);
 786
 787         if (status) {
 788                 dev_err(ctrl->device,
 789                         "failed nvme_keep_alive_end_io error=%d\n",
 790                                 status);
 791                 return;
 792         }
 793
 794         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 795 }
 796
 797 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 798 {
 799         struct request *rq;
 800
 801         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 802                         NVME_QID_ANY);
 803         if (IS_ERR(rq))
 804                 return PTR_ERR(rq);
 805
 806         rq->timeout = ctrl->kato * HZ;
 807         rq->end_io_data = ctrl;
 808
 809         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 810
 811         return 0;
 812 }
 813
 814 static void nvme_keep_alive_work(struct work_struct *work)
 815 {
 816         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 817                         struct nvme_ctrl, ka_work);
 818
 819         if (nvme_keep_alive(ctrl)) {
 820                 /* allocation failure, reset the controller */
 821                 dev_err(ctrl->device, "keep-alive failed\n");
 822                 nvme_reset_ctrl(ctrl);
 823                 return;
 824         }
 825 }
 826
 827 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 828 {
 829         if (unlikely(ctrl->kato == 0))
 830                 return;
 831
 832         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
 833         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
 834         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
 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, *next;
1121         LIST_HEAD(rm_list);
1122
1123         mutex_lock(&ctrl->namespaces_mutex);
1124         list_for_each_entry(ns, &ctrl->namespaces, list) {
1125                 if (ns->disk && nvme_revalidate_disk(ns->disk)) {
1126                         list_move_tail(&ns->list, &rm_list);
1127                 }
1128         }
1129         mutex_unlock(&ctrl->namespaces_mutex);
1130
1131         list_for_each_entry_safe(ns, next, &rm_list, list)
1132                 nvme_ns_remove(ns);
1133 }
1134
1135 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1136 {
1137         /*
1138          * Revalidate LBA changes prior to unfreezing. This is necessary to
1139          * prevent memory corruption if a logical block size was changed by
1140          * this command.
1141          */
1142         if (effects & NVME_CMD_EFFECTS_LBCC)
1143                 nvme_update_formats(ctrl);
1144         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1145                 nvme_unfreeze(ctrl);
1146         if (effects & NVME_CMD_EFFECTS_CCC)
1147                 nvme_init_identify(ctrl);
1148         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1149                 nvme_queue_scan(ctrl);
1150 }
1151
1152 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1153                         struct nvme_passthru_cmd __user *ucmd)
1154 {
1155         struct nvme_passthru_cmd cmd;
1156         struct nvme_command c;
1157         unsigned timeout = 0;
1158         u32 effects;
1159         int status;
1160
1161         if (!capable(CAP_SYS_ADMIN))
1162                 return -EACCES;
1163         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1164                 return -EFAULT;
1165         if (cmd.flags)
1166                 return -EINVAL;
1167
1168         memset(&c, 0, sizeof(c));
1169         c.common.opcode = cmd.opcode;
1170         c.common.flags = cmd.flags;
1171         c.common.nsid = cpu_to_le32(cmd.nsid);
1172         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1173         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1174         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1175         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1176         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1177         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1178         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1179         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1180
1181         if (cmd.timeout_ms)
1182                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1183
1184         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1185         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1186                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1187                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1188                         0, &cmd.result, timeout);
1189         nvme_passthru_end(ctrl, effects);
1190
1191         if (status >= 0) {
1192                 if (put_user(cmd.result, &ucmd->result))
1193                         return -EFAULT;
1194         }
1195
1196         return status;
1197 }
1198
1199 /*
1200  * Issue ioctl requests on the first available path.  Note that unlike normal
1201  * block layer requests we will not retry failed request on another controller.
1202  */
1203 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1204                 struct nvme_ns_head **head, int *srcu_idx)
1205 {
1206 #ifdef CONFIG_NVME_MULTIPATH
1207         if (disk->fops == &nvme_ns_head_ops) {
1208                 *head = disk->private_data;
1209                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1210                 return nvme_find_path(*head);
1211         }
1212 #endif
1213         *head = NULL;
1214         *srcu_idx = -1;
1215         return disk->private_data;
1216 }
1217
1218 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1219 {
1220         if (head)
1221                 srcu_read_unlock(&head->srcu, idx);
1222 }
1223
1224 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1225 {
1226         switch (cmd) {
1227         case NVME_IOCTL_ID:
1228                 force_successful_syscall_return();
1229                 return ns->head->ns_id;
1230         case NVME_IOCTL_ADMIN_CMD:
1231                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1232         case NVME_IOCTL_IO_CMD:
1233                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1234         case NVME_IOCTL_SUBMIT_IO:
1235                 return nvme_submit_io(ns, (void __user *)arg);
1236         default:
1237 #ifdef CONFIG_NVM
1238                 if (ns->ndev)
1239                         return nvme_nvm_ioctl(ns, cmd, arg);
1240 #endif
1241                 if (is_sed_ioctl(cmd))
1242                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1243                                          (void __user *) arg);
1244                 return -ENOTTY;
1245         }
1246 }
1247
1248 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1249                 unsigned int cmd, unsigned long arg)
1250 {
1251         struct nvme_ns_head *head = NULL;
1252         struct nvme_ns *ns;
1253         int srcu_idx, ret;
1254
1255         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1256         if (unlikely(!ns))
1257                 ret = -EWOULDBLOCK;
1258         else
1259                 ret = nvme_ns_ioctl(ns, cmd, arg);
1260         nvme_put_ns_from_disk(head, srcu_idx);
1261         return ret;
1262 }
1263
1264 static int nvme_open(struct block_device *bdev, fmode_t mode)
1265 {
1266         struct nvme_ns *ns = bdev->bd_disk->private_data;
1267
1268 #ifdef CONFIG_NVME_MULTIPATH
1269         /* should never be called due to GENHD_FL_HIDDEN */
1270         if (WARN_ON_ONCE(ns->head->disk))
1271                 goto fail;
1272 #endif
1273         if (!kref_get_unless_zero(&ns->kref))
1274                 goto fail;
1275         if (!try_module_get(ns->ctrl->ops->module))
1276                 goto fail_put_ns;
1277
1278         return 0;
1279
1280 fail_put_ns:
1281         nvme_put_ns(ns);
1282 fail:
1283         return -ENXIO;
1284 }
1285
1286 static void nvme_release(struct gendisk *disk, fmode_t mode)
1287 {
1288         struct nvme_ns *ns = disk->private_data;
1289
1290         module_put(ns->ctrl->ops->module);
1291         nvme_put_ns(ns);
1292 }
1293
1294 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1295 {
1296         /* some standard values */
1297         geo->heads = 1 << 6;
1298         geo->sectors = 1 << 5;
1299         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1300         return 0;
1301 }
1302
1303 #ifdef CONFIG_BLK_DEV_INTEGRITY
1304 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1305 {
1306         struct blk_integrity integrity;
1307
1308         memset(&integrity, 0, sizeof(integrity));
1309         switch (pi_type) {
1310         case NVME_NS_DPS_PI_TYPE3:
1311                 integrity.profile = &t10_pi_type3_crc;
1312                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1313                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1314                 break;
1315         case NVME_NS_DPS_PI_TYPE1:
1316         case NVME_NS_DPS_PI_TYPE2:
1317                 integrity.profile = &t10_pi_type1_crc;
1318                 integrity.tag_size = sizeof(u16);
1319                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1320                 break;
1321         default:
1322                 integrity.profile = NULL;
1323                 break;
1324         }
1325         integrity.tuple_size = ms;
1326         blk_integrity_register(disk, &integrity);
1327         blk_queue_max_integrity_segments(disk->queue, 1);
1328 }
1329 #else
1330 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1331 {
1332 }
1333 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1334
1335 static void nvme_set_chunk_size(struct nvme_ns *ns)
1336 {
1337         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1338         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1339 }
1340
1341 static void nvme_config_discard(struct nvme_ctrl *ctrl,
1342                 unsigned stream_alignment, struct request_queue *queue)
1343 {
1344         u32 size = queue_logical_block_size(queue);
1345
1346         if (stream_alignment)
1347                 size *= stream_alignment;
1348
1349         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1350                         NVME_DSM_MAX_RANGES);
1351
1352         queue->limits.discard_alignment = 0;
1353         queue->limits.discard_granularity = size;
1354
1355         blk_queue_max_discard_sectors(queue, UINT_MAX);
1356         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1357         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, queue);
1358
1359         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1360                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1361 }
1362
1363 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1364                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1365 {
1366         memset(ids, 0, sizeof(*ids));
1367
1368         if (ctrl->vs >= NVME_VS(1, 1, 0))
1369                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1370         if (ctrl->vs >= NVME_VS(1, 2, 0))
1371                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1372         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1373                  /* Don't treat error as fatal we potentially
1374                   * already have a NGUID or EUI-64
1375                   */
1376                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1377                         dev_warn(ctrl->device,
1378                                  "%s: Identify Descriptors failed\n", __func__);
1379         }
1380 }
1381
1382 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1383 {
1384         return !uuid_is_null(&ids->uuid) ||
1385                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1386                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1387 }
1388
1389 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1390 {
1391         return uuid_equal(&a->uuid, &b->uuid) &&
1392                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1393                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1394 }
1395
1396 static void nvme_update_disk_info(struct gendisk *disk,
1397                 struct nvme_ns *ns, struct nvme_id_ns *id)
1398 {
1399         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1400         unsigned short bs = 1 << ns->lba_shift;
1401         unsigned stream_alignment = 0;
1402
1403         if (ns->ctrl->nr_streams && ns->sws && ns->sgs)
1404                 stream_alignment = ns->sws * ns->sgs;
1405
1406         blk_mq_freeze_queue(disk->queue);
1407         blk_integrity_unregister(disk);
1408
1409         blk_queue_logical_block_size(disk->queue, bs);
1410         blk_queue_physical_block_size(disk->queue, bs);
1411         blk_queue_io_min(disk->queue, bs);
1412
1413         if (ns->ms && !ns->ext &&
1414             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1415                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1416         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1417                 capacity = 0;
1418         set_capacity(disk, capacity);
1419
1420         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1421                 nvme_config_discard(ns->ctrl, stream_alignment, disk->queue);
1422         blk_mq_unfreeze_queue(disk->queue);
1423 }
1424
1425 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1426 {
1427         struct nvme_ns *ns = disk->private_data;
1428
1429         /*
1430          * If identify namespace failed, use default 512 byte block size so
1431          * block layer can use before failing read/write for 0 capacity.
1432          */
1433         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1434         if (ns->lba_shift == 0)
1435                 ns->lba_shift = 9;
1436         ns->noiob = le16_to_cpu(id->noiob);
1437         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1438         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1439         /* the PI implementation requires metadata equal t10 pi tuple size */
1440         if (ns->ms == sizeof(struct t10_pi_tuple))
1441                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1442         else
1443                 ns->pi_type = 0;
1444
1445         if (ns->noiob)
1446                 nvme_set_chunk_size(ns);
1447         nvme_update_disk_info(disk, ns, id);
1448 #ifdef CONFIG_NVME_MULTIPATH
1449         if (ns->head->disk)
1450                 nvme_update_disk_info(ns->head->disk, ns, id);
1451 #endif
1452 }
1453
1454 static int nvme_revalidate_disk(struct gendisk *disk)
1455 {
1456         struct nvme_ns *ns = disk->private_data;
1457         struct nvme_ctrl *ctrl = ns->ctrl;
1458         struct nvme_id_ns *id;
1459         struct nvme_ns_ids ids;
1460         int ret = 0;
1461
1462         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1463                 set_capacity(disk, 0);
1464                 return -ENODEV;
1465         }
1466
1467         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1468         if (!id)
1469                 return -ENODEV;
1470
1471         if (id->ncap == 0) {
1472                 ret = -ENODEV;
1473                 goto out;
1474         }
1475
1476         __nvme_revalidate_disk(disk, id);
1477         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1478         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1479                 dev_err(ctrl->device,
1480                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1481                 ret = -ENODEV;
1482         }
1483
1484 out:
1485         kfree(id);
1486         return ret;
1487 }
1488
1489 static char nvme_pr_type(enum pr_type type)
1490 {
1491         switch (type) {
1492         case PR_WRITE_EXCLUSIVE:
1493                 return 1;
1494         case PR_EXCLUSIVE_ACCESS:
1495                 return 2;
1496         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1497                 return 3;
1498         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1499                 return 4;
1500         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1501                 return 5;
1502         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1503                 return 6;
1504         default:
1505                 return 0;
1506         }
1507 };
1508
1509 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1510                                 u64 key, u64 sa_key, u8 op)
1511 {
1512         struct nvme_ns_head *head = NULL;
1513         struct nvme_ns *ns;
1514         struct nvme_command c;
1515         int srcu_idx, ret;
1516         u8 data[16] = { 0, };
1517
1518         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1519         if (unlikely(!ns))
1520                 return -EWOULDBLOCK;
1521
1522         put_unaligned_le64(key, &data[0]);
1523         put_unaligned_le64(sa_key, &data[8]);
1524
1525         memset(&c, 0, sizeof(c));
1526         c.common.opcode = op;
1527         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1528         c.common.cdw10[0] = cpu_to_le32(cdw10);
1529
1530         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1531         nvme_put_ns_from_disk(head, srcu_idx);
1532         return ret;
1533 }
1534
1535 static int nvme_pr_register(struct block_device *bdev, u64 old,
1536                 u64 new, unsigned flags)
1537 {
1538         u32 cdw10;
1539
1540         if (flags & ~PR_FL_IGNORE_KEY)
1541                 return -EOPNOTSUPP;
1542
1543         cdw10 = old ? 2 : 0;
1544         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1545         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1546         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1547 }
1548
1549 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1550                 enum pr_type type, unsigned flags)
1551 {
1552         u32 cdw10;
1553
1554         if (flags & ~PR_FL_IGNORE_KEY)
1555                 return -EOPNOTSUPP;
1556
1557         cdw10 = nvme_pr_type(type) << 8;
1558         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1559         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1560 }
1561
1562 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1563                 enum pr_type type, bool abort)
1564 {
1565         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1566         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1567 }
1568
1569 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1570 {
1571         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1572         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1573 }
1574
1575 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1576 {
1577         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1578         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1579 }
1580
1581 static const struct pr_ops nvme_pr_ops = {
1582         .pr_register    = nvme_pr_register,
1583         .pr_reserve     = nvme_pr_reserve,
1584         .pr_release     = nvme_pr_release,
1585         .pr_preempt     = nvme_pr_preempt,
1586         .pr_clear       = nvme_pr_clear,
1587 };
1588
1589 #ifdef CONFIG_BLK_SED_OPAL
1590 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1591                 bool send)
1592 {
1593         struct nvme_ctrl *ctrl = data;
1594         struct nvme_command cmd;
1595
1596         memset(&cmd, 0, sizeof(cmd));
1597         if (send)
1598                 cmd.common.opcode = nvme_admin_security_send;
1599         else
1600                 cmd.common.opcode = nvme_admin_security_recv;
1601         cmd.common.nsid = 0;
1602         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1603         cmd.common.cdw10[1] = cpu_to_le32(len);
1604
1605         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1606                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1607 }
1608 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1609 #endif /* CONFIG_BLK_SED_OPAL */
1610
1611 static const struct block_device_operations nvme_fops = {
1612         .owner          = THIS_MODULE,
1613         .ioctl          = nvme_ioctl,
1614         .compat_ioctl   = nvme_ioctl,
1615         .open           = nvme_open,
1616         .release        = nvme_release,
1617         .getgeo         = nvme_getgeo,
1618         .revalidate_disk= nvme_revalidate_disk,
1619         .pr_ops         = &nvme_pr_ops,
1620 };
1621
1622 #ifdef CONFIG_NVME_MULTIPATH
1623 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1624 {
1625         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1626
1627         if (!kref_get_unless_zero(&head->ref))
1628                 return -ENXIO;
1629         return 0;
1630 }
1631
1632 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1633 {
1634         nvme_put_ns_head(disk->private_data);
1635 }
1636
1637 const struct block_device_operations nvme_ns_head_ops = {
1638         .owner          = THIS_MODULE,
1639         .open           = nvme_ns_head_open,
1640         .release        = nvme_ns_head_release,
1641         .ioctl          = nvme_ioctl,
1642         .compat_ioctl   = nvme_ioctl,
1643         .getgeo         = nvme_getgeo,
1644         .pr_ops         = &nvme_pr_ops,
1645 };
1646 #endif /* CONFIG_NVME_MULTIPATH */
1647
1648 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1649 {
1650         unsigned long timeout =
1651                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1652         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1653         int ret;
1654
1655         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1656                 if (csts == ~0)
1657                         return -ENODEV;
1658                 if ((csts & NVME_CSTS_RDY) == bit)
1659                         break;
1660
1661                 msleep(100);
1662                 if (fatal_signal_pending(current))
1663                         return -EINTR;
1664                 if (time_after(jiffies, timeout)) {
1665                         dev_err(ctrl->device,
1666                                 "Device not ready; aborting %s\n", enabled ?
1667                                                 "initialisation" : "reset");
1668                         return -ENODEV;
1669                 }
1670         }
1671
1672         return ret;
1673 }
1674
1675 /*
1676  * If the device has been passed off to us in an enabled state, just clear
1677  * the enabled bit.  The spec says we should set the 'shutdown notification
1678  * bits', but doing so may cause the device to complete commands to the
1679  * admin queue ... and we don't know what memory that might be pointing at!
1680  */
1681 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1682 {
1683         int ret;
1684
1685         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1686         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1687
1688         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1689         if (ret)
1690                 return ret;
1691
1692         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1693                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1694
1695         return nvme_wait_ready(ctrl, cap, false);
1696 }
1697 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1698
1699 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1700 {
1701         /*
1702          * Default to a 4K page size, with the intention to update this
1703          * path in the future to accomodate architectures with differing
1704          * kernel and IO page sizes.
1705          */
1706         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1707         int ret;
1708
1709         if (page_shift < dev_page_min) {
1710                 dev_err(ctrl->device,
1711                         "Minimum device page size %u too large for host (%u)\n",
1712                         1 << dev_page_min, 1 << page_shift);
1713                 return -ENODEV;
1714         }
1715
1716         ctrl->page_size = 1 << page_shift;
1717
1718         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1719         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1720         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1721         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1722         ctrl->ctrl_config |= NVME_CC_ENABLE;
1723
1724         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1725         if (ret)
1726                 return ret;
1727         return nvme_wait_ready(ctrl, cap, true);
1728 }
1729 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1730
1731 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1732 {
1733         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1734         u32 csts;
1735         int ret;
1736
1737         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1738         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1739
1740         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1741         if (ret)
1742                 return ret;
1743
1744         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1745                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1746                         break;
1747
1748                 msleep(100);
1749                 if (fatal_signal_pending(current))
1750                         return -EINTR;
1751                 if (time_after(jiffies, timeout)) {
1752                         dev_err(ctrl->device,
1753                                 "Device shutdown incomplete; abort shutdown\n");
1754                         return -ENODEV;
1755                 }
1756         }
1757
1758         return ret;
1759 }
1760 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1761
1762 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1763                 struct request_queue *q)
1764 {
1765         bool vwc = false;
1766
1767         if (ctrl->max_hw_sectors) {
1768                 u32 max_segments =
1769                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1770
1771                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1772                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1773         }
1774         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1775             is_power_of_2(ctrl->max_hw_sectors))
1776                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1777         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1778         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1779                 vwc = true;
1780         blk_queue_write_cache(q, vwc, vwc);
1781 }
1782
1783 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1784 {
1785         __le64 ts;
1786         int ret;
1787
1788         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1789                 return 0;
1790
1791         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1792         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1793                         NULL);
1794         if (ret)
1795                 dev_warn_once(ctrl->device,
1796                         "could not set timestamp (%d)\n", ret);
1797         return ret;
1798 }
1799
1800 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1801 {
1802         /*
1803          * APST (Autonomous Power State Transition) lets us program a
1804          * table of power state transitions that the controller will
1805          * perform automatically.  We configure it with a simple
1806          * heuristic: we are willing to spend at most 2% of the time
1807          * transitioning between power states.  Therefore, when running
1808          * in any given state, we will enter the next lower-power
1809          * non-operational state after waiting 50 * (enlat + exlat)
1810          * microseconds, as long as that state's exit latency is under
1811          * the requested maximum latency.
1812          *
1813          * We will not autonomously enter any non-operational state for
1814          * which the total latency exceeds ps_max_latency_us.  Users
1815          * can set ps_max_latency_us to zero to turn off APST.
1816          */
1817
1818         unsigned apste;
1819         struct nvme_feat_auto_pst *table;
1820         u64 max_lat_us = 0;
1821         int max_ps = -1;
1822         int ret;
1823
1824         /*
1825          * If APST isn't supported or if we haven't been initialized yet,
1826          * then don't do anything.
1827          */
1828         if (!ctrl->apsta)
1829                 return 0;
1830
1831         if (ctrl->npss > 31) {
1832                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1833                 return 0;
1834         }
1835
1836         table = kzalloc(sizeof(*table), GFP_KERNEL);
1837         if (!table)
1838                 return 0;
1839
1840         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1841                 /* Turn off APST. */
1842                 apste = 0;
1843                 dev_dbg(ctrl->device, "APST disabled\n");
1844         } else {
1845                 __le64 target = cpu_to_le64(0);
1846                 int state;
1847
1848                 /*
1849                  * Walk through all states from lowest- to highest-power.
1850                  * According to the spec, lower-numbered states use more
1851                  * power.  NPSS, despite the name, is the index of the
1852                  * lowest-power state, not the number of states.
1853                  */
1854                 for (state = (int)ctrl->npss; state >= 0; state--) {
1855                         u64 total_latency_us, exit_latency_us, transition_ms;
1856
1857                         if (target)
1858                                 table->entries[state] = target;
1859
1860                         /*
1861                          * Don't allow transitions to the deepest state
1862                          * if it's quirked off.
1863                          */
1864                         if (state == ctrl->npss &&
1865                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1866                                 continue;
1867
1868                         /*
1869                          * Is this state a useful non-operational state for
1870                          * higher-power states to autonomously transition to?
1871                          */
1872                         if (!(ctrl->psd[state].flags &
1873                               NVME_PS_FLAGS_NON_OP_STATE))
1874                                 continue;
1875
1876                         exit_latency_us =
1877                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1878                         if (exit_latency_us > ctrl->ps_max_latency_us)
1879                                 continue;
1880
1881                         total_latency_us =
1882                                 exit_latency_us +
1883                                 le32_to_cpu(ctrl->psd[state].entry_lat);
1884
1885                         /*
1886                          * This state is good.  Use it as the APST idle
1887                          * target for higher power states.
1888                          */
1889                         transition_ms = total_latency_us + 19;
1890                         do_div(transition_ms, 20);
1891                         if (transition_ms > (1 << 24) - 1)
1892                                 transition_ms = (1 << 24) - 1;
1893
1894                         target = cpu_to_le64((state << 3) |
1895                                              (transition_ms << 8));
1896
1897                         if (max_ps == -1)
1898                                 max_ps = state;
1899
1900                         if (total_latency_us > max_lat_us)
1901                                 max_lat_us = total_latency_us;
1902                 }
1903
1904                 apste = 1;
1905
1906                 if (max_ps == -1) {
1907                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1908                 } else {
1909                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1910                                 max_ps, max_lat_us, (int)sizeof(*table), table);
1911                 }
1912         }
1913
1914         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1915                                 table, sizeof(*table), NULL);
1916         if (ret)
1917                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1918
1919         kfree(table);
1920         return ret;
1921 }
1922
1923 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1924 {
1925         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1926         u64 latency;
1927
1928         switch (val) {
1929         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1930         case PM_QOS_LATENCY_ANY:
1931                 latency = U64_MAX;
1932                 break;
1933
1934         default:
1935                 latency = val;
1936         }
1937
1938         if (ctrl->ps_max_latency_us != latency) {
1939                 ctrl->ps_max_latency_us = latency;
1940                 nvme_configure_apst(ctrl);
1941         }
1942 }
1943
1944 struct nvme_core_quirk_entry {
1945         /*
1946          * NVMe model and firmware strings are padded with spaces.  For
1947          * simplicity, strings in the quirk table are padded with NULLs
1948          * instead.
1949          */
1950         u16 vid;
1951         const char *mn;
1952         const char *fr;
1953         unsigned long quirks;
1954 };
1955
1956 static const struct nvme_core_quirk_entry core_quirks[] = {
1957         {
1958                 /*
1959                  * This Toshiba device seems to die using any APST states.  See:
1960                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1961                  */
1962                 .vid = 0x1179,
1963                 .mn = "THNSF5256GPUK TOSHIBA",
1964                 .quirks = NVME_QUIRK_NO_APST,
1965         }
1966 };
1967
1968 /* match is null-terminated but idstr is space-padded. */
1969 static bool string_matches(const char *idstr, const char *match, size_t len)
1970 {
1971         size_t matchlen;
1972
1973         if (!match)
1974                 return true;
1975
1976         matchlen = strlen(match);
1977         WARN_ON_ONCE(matchlen > len);
1978
1979         if (memcmp(idstr, match, matchlen))
1980                 return false;
1981
1982         for (; matchlen < len; matchlen++)
1983                 if (idstr[matchlen] != ' ')
1984                         return false;
1985
1986         return true;
1987 }
1988
1989 static bool quirk_matches(const struct nvme_id_ctrl *id,
1990                           const struct nvme_core_quirk_entry *q)
1991 {
1992         return q->vid == le16_to_cpu(id->vid) &&
1993                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1994                 string_matches(id->fr, q->fr, sizeof(id->fr));
1995 }
1996
1997 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
1998                 struct nvme_id_ctrl *id)
1999 {
2000         size_t nqnlen;
2001         int off;
2002
2003         nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2004         if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2005                 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2006                 return;
2007         }
2008
2009         if (ctrl->vs >= NVME_VS(1, 2, 1))
2010                 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2011
2012         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2013         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2014                         "nqn.2014.08.org.nvmexpress:%4x%4x",
2015                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2016         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2017         off += sizeof(id->sn);
2018         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2019         off += sizeof(id->mn);
2020         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2021 }
2022
2023 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2024 {
2025         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2026         kfree(subsys);
2027 }
2028
2029 static void nvme_release_subsystem(struct device *dev)
2030 {
2031         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2032 }
2033
2034 static void nvme_destroy_subsystem(struct kref *ref)
2035 {
2036         struct nvme_subsystem *subsys =
2037                         container_of(ref, struct nvme_subsystem, ref);
2038
2039         mutex_lock(&nvme_subsystems_lock);
2040         list_del(&subsys->entry);
2041         mutex_unlock(&nvme_subsystems_lock);
2042
2043         ida_destroy(&subsys->ns_ida);
2044         device_del(&subsys->dev);
2045         put_device(&subsys->dev);
2046 }
2047
2048 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2049 {
2050         kref_put(&subsys->ref, nvme_destroy_subsystem);
2051 }
2052
2053 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2054 {
2055         struct nvme_subsystem *subsys;
2056
2057         lockdep_assert_held(&nvme_subsystems_lock);
2058
2059         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2060                 if (strcmp(subsys->subnqn, subsysnqn))
2061                         continue;
2062                 if (!kref_get_unless_zero(&subsys->ref))
2063                         continue;
2064                 return subsys;
2065         }
2066
2067         return NULL;
2068 }
2069
2070 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2071         struct device_attribute subsys_attr_##_name = \
2072                 __ATTR(_name, _mode, _show, NULL)
2073
2074 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2075                                     struct device_attribute *attr,
2076                                     char *buf)
2077 {
2078         struct nvme_subsystem *subsys =
2079                 container_of(dev, struct nvme_subsystem, dev);
2080
2081         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2082 }
2083 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2084
2085 #define nvme_subsys_show_str_function(field)                            \
2086 static ssize_t subsys_##field##_show(struct device *dev,                \
2087                             struct device_attribute *attr, char *buf)   \
2088 {                                                                       \
2089         struct nvme_subsystem *subsys =                                 \
2090                 container_of(dev, struct nvme_subsystem, dev);          \
2091         return sprintf(buf, "%.*s\n",                                   \
2092                        (int)sizeof(subsys->field), subsys->field);      \
2093 }                                                                       \
2094 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2095
2096 nvme_subsys_show_str_function(model);
2097 nvme_subsys_show_str_function(serial);
2098 nvme_subsys_show_str_function(firmware_rev);
2099
2100 static struct attribute *nvme_subsys_attrs[] = {
2101         &subsys_attr_model.attr,
2102         &subsys_attr_serial.attr,
2103         &subsys_attr_firmware_rev.attr,
2104         &subsys_attr_subsysnqn.attr,
2105         NULL,
2106 };
2107
2108 static struct attribute_group nvme_subsys_attrs_group = {
2109         .attrs = nvme_subsys_attrs,
2110 };
2111
2112 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2113         &nvme_subsys_attrs_group,
2114         NULL,
2115 };
2116
2117 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2118 {
2119         int count = 0;
2120         struct nvme_ctrl *ctrl;
2121
2122         mutex_lock(&subsys->lock);
2123         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2124                 if (ctrl->state != NVME_CTRL_DELETING &&
2125                     ctrl->state != NVME_CTRL_DEAD)
2126                         count++;
2127         }
2128         mutex_unlock(&subsys->lock);
2129
2130         return count;
2131 }
2132
2133 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2134 {
2135         struct nvme_subsystem *subsys, *found;
2136         int ret;
2137
2138         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2139         if (!subsys)
2140                 return -ENOMEM;
2141         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2142         if (ret < 0) {
2143                 kfree(subsys);
2144                 return ret;
2145         }
2146         subsys->instance = ret;
2147         mutex_init(&subsys->lock);
2148         kref_init(&subsys->ref);
2149         INIT_LIST_HEAD(&subsys->ctrls);
2150         INIT_LIST_HEAD(&subsys->nsheads);
2151         nvme_init_subnqn(subsys, ctrl, id);
2152         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2153         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2154         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2155         subsys->vendor_id = le16_to_cpu(id->vid);
2156         subsys->cmic = id->cmic;
2157
2158         subsys->dev.class = nvme_subsys_class;
2159         subsys->dev.release = nvme_release_subsystem;
2160         subsys->dev.groups = nvme_subsys_attrs_groups;
2161         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2162         device_initialize(&subsys->dev);
2163
2164         mutex_lock(&nvme_subsystems_lock);
2165         found = __nvme_find_get_subsystem(subsys->subnqn);
2166         if (found) {
2167                 /*
2168                  * Verify that the subsystem actually supports multiple
2169                  * controllers, else bail out.
2170                  */
2171                 if (nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2172                         dev_err(ctrl->device,
2173                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2174                                 found->subnqn);
2175                         nvme_put_subsystem(found);
2176                         ret = -EINVAL;
2177                         goto out_unlock;
2178                 }
2179
2180                 __nvme_release_subsystem(subsys);
2181                 subsys = found;
2182         } else {
2183                 ret = device_add(&subsys->dev);
2184                 if (ret) {
2185                         dev_err(ctrl->device,
2186                                 "failed to register subsystem device.\n");
2187                         goto out_unlock;
2188                 }
2189                 ida_init(&subsys->ns_ida);
2190                 list_add_tail(&subsys->entry, &nvme_subsystems);
2191         }
2192
2193         ctrl->subsys = subsys;
2194         mutex_unlock(&nvme_subsystems_lock);
2195
2196         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2197                         dev_name(ctrl->device))) {
2198                 dev_err(ctrl->device,
2199                         "failed to create sysfs link from subsystem.\n");
2200                 /* the transport driver will eventually put the subsystem */
2201                 return -EINVAL;
2202         }
2203
2204         mutex_lock(&subsys->lock);
2205         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2206         mutex_unlock(&subsys->lock);
2207
2208         return 0;
2209
2210 out_unlock:
2211         mutex_unlock(&nvme_subsystems_lock);
2212         put_device(&subsys->dev);
2213         return ret;
2214 }
2215
2216 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
2217                         size_t size)
2218 {
2219         struct nvme_command c = { };
2220
2221         c.common.opcode = nvme_admin_get_log_page;
2222         c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
2223         c.common.cdw10[0] = nvme_get_log_dw10(log_page, size);
2224
2225         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2226 }
2227
2228 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2229 {
2230         int ret;
2231
2232         if (!ctrl->effects)
2233                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2234
2235         if (!ctrl->effects)
2236                 return 0;
2237
2238         ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects,
2239                                         sizeof(*ctrl->effects));
2240         if (ret) {
2241                 kfree(ctrl->effects);
2242                 ctrl->effects = NULL;
2243         }
2244         return ret;
2245 }
2246
2247 /*
2248  * Initialize the cached copies of the Identify data and various controller
2249  * register in our nvme_ctrl structure.  This should be called as soon as
2250  * the admin queue is fully up and running.
2251  */
2252 int nvme_init_identify(struct nvme_ctrl *ctrl)
2253 {
2254         struct nvme_id_ctrl *id;
2255         u64 cap;
2256         int ret, page_shift;
2257         u32 max_hw_sectors;
2258         bool prev_apst_enabled;
2259
2260         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2261         if (ret) {
2262                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2263                 return ret;
2264         }
2265
2266         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2267         if (ret) {
2268                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2269                 return ret;
2270         }
2271         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2272
2273         if (ctrl->vs >= NVME_VS(1, 1, 0))
2274                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2275
2276         ret = nvme_identify_ctrl(ctrl, &id);
2277         if (ret) {
2278                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2279                 return -EIO;
2280         }
2281
2282         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2283                 ret = nvme_get_effects_log(ctrl);
2284                 if (ret < 0)
2285                         return ret;
2286         }
2287
2288         if (!ctrl->identified) {
2289                 int i;
2290
2291                 ret = nvme_init_subsystem(ctrl, id);
2292                 if (ret)
2293                         goto out_free;
2294
2295                 /*
2296                  * Check for quirks.  Quirk can depend on firmware version,
2297                  * so, in principle, the set of quirks present can change
2298                  * across a reset.  As a possible future enhancement, we
2299                  * could re-scan for quirks every time we reinitialize
2300                  * the device, but we'd have to make sure that the driver
2301                  * behaves intelligently if the quirks change.
2302                  */
2303                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2304                         if (quirk_matches(id, &core_quirks[i]))
2305                                 ctrl->quirks |= core_quirks[i].quirks;
2306                 }
2307         }
2308
2309         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2310                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2311                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2312         }
2313
2314         ctrl->oacs = le16_to_cpu(id->oacs);
2315         ctrl->oncs = le16_to_cpup(&id->oncs);
2316         atomic_set(&ctrl->abort_limit, id->acl + 1);
2317         ctrl->vwc = id->vwc;
2318         ctrl->cntlid = le16_to_cpup(&id->cntlid);
2319         if (id->mdts)
2320                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2321         else
2322                 max_hw_sectors = UINT_MAX;
2323         ctrl->max_hw_sectors =
2324                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2325
2326         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2327         ctrl->sgls = le32_to_cpu(id->sgls);
2328         ctrl->kas = le16_to_cpu(id->kas);
2329
2330         if (id->rtd3e) {
2331                 /* us -> s */
2332                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2333
2334                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2335                                                  shutdown_timeout, 60);
2336
2337                 if (ctrl->shutdown_timeout != shutdown_timeout)
2338                         dev_info(ctrl->device,
2339                                  "Shutdown timeout set to %u seconds\n",
2340                                  ctrl->shutdown_timeout);
2341         } else
2342                 ctrl->shutdown_timeout = shutdown_timeout;
2343
2344         ctrl->npss = id->npss;
2345         ctrl->apsta = id->apsta;
2346         prev_apst_enabled = ctrl->apst_enabled;
2347         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2348                 if (force_apst && id->apsta) {
2349                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2350                         ctrl->apst_enabled = true;
2351                 } else {
2352                         ctrl->apst_enabled = false;
2353                 }
2354         } else {
2355                 ctrl->apst_enabled = id->apsta;
2356         }
2357         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2358
2359         if (ctrl->ops->flags & NVME_F_FABRICS) {
2360                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2361                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2362                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2363                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2364
2365                 /*
2366                  * In fabrics we need to verify the cntlid matches the
2367                  * admin connect
2368                  */
2369                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2370                         ret = -EINVAL;
2371                         goto out_free;
2372                 }
2373
2374                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2375                         dev_err(ctrl->device,
2376                                 "keep-alive support is mandatory for fabrics\n");
2377                         ret = -EINVAL;
2378                         goto out_free;
2379                 }
2380         } else {
2381                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2382                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2383                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2384                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2385                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2386         }
2387
2388         kfree(id);
2389
2390         if (ctrl->apst_enabled && !prev_apst_enabled)
2391                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2392         else if (!ctrl->apst_enabled && prev_apst_enabled)
2393                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2394
2395         ret = nvme_configure_apst(ctrl);
2396         if (ret < 0)
2397                 return ret;
2398
2399         ret = nvme_configure_timestamp(ctrl);
2400         if (ret < 0)
2401                 return ret;
2402
2403         ret = nvme_configure_directives(ctrl);
2404         if (ret < 0)
2405                 return ret;
2406
2407         ctrl->identified = true;
2408
2409         return 0;
2410
2411 out_free:
2412         kfree(id);
2413         return ret;
2414 }
2415 EXPORT_SYMBOL_GPL(nvme_init_identify);
2416
2417 static int nvme_dev_open(struct inode *inode, struct file *file)
2418 {
2419         struct nvme_ctrl *ctrl =
2420                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2421
2422         switch (ctrl->state) {
2423         case NVME_CTRL_LIVE:
2424         case NVME_CTRL_ADMIN_ONLY:
2425                 break;
2426         default:
2427                 return -EWOULDBLOCK;
2428         }
2429
2430         file->private_data = ctrl;
2431         return 0;
2432 }
2433
2434 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2435 {
2436         struct nvme_ns *ns;
2437         int ret;
2438
2439         mutex_lock(&ctrl->namespaces_mutex);
2440         if (list_empty(&ctrl->namespaces)) {
2441                 ret = -ENOTTY;
2442                 goto out_unlock;
2443         }
2444
2445         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2446         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2447                 dev_warn(ctrl->device,
2448                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2449                 ret = -EINVAL;
2450                 goto out_unlock;
2451         }
2452
2453         dev_warn(ctrl->device,
2454                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2455         kref_get(&ns->kref);
2456         mutex_unlock(&ctrl->namespaces_mutex);
2457
2458         ret = nvme_user_cmd(ctrl, ns, argp);
2459         nvme_put_ns(ns);
2460         return ret;
2461
2462 out_unlock:
2463         mutex_unlock(&ctrl->namespaces_mutex);
2464         return ret;
2465 }
2466
2467 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2468                 unsigned long arg)
2469 {
2470         struct nvme_ctrl *ctrl = file->private_data;
2471         void __user *argp = (void __user *)arg;
2472
2473         switch (cmd) {
2474         case NVME_IOCTL_ADMIN_CMD:
2475                 return nvme_user_cmd(ctrl, NULL, argp);
2476         case NVME_IOCTL_IO_CMD:
2477                 return nvme_dev_user_cmd(ctrl, argp);
2478         case NVME_IOCTL_RESET:
2479                 dev_warn(ctrl->device, "resetting controller\n");
2480                 return nvme_reset_ctrl_sync(ctrl);
2481         case NVME_IOCTL_SUBSYS_RESET:
2482                 return nvme_reset_subsystem(ctrl);
2483         case NVME_IOCTL_RESCAN:
2484                 nvme_queue_scan(ctrl);
2485                 return 0;
2486         default:
2487                 return -ENOTTY;
2488         }
2489 }
2490
2491 static const struct file_operations nvme_dev_fops = {
2492         .owner          = THIS_MODULE,
2493         .open           = nvme_dev_open,
2494         .unlocked_ioctl = nvme_dev_ioctl,
2495         .compat_ioctl   = nvme_dev_ioctl,
2496 };
2497
2498 static ssize_t nvme_sysfs_reset(struct device *dev,
2499                                 struct device_attribute *attr, const char *buf,
2500                                 size_t count)
2501 {
2502         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2503         int ret;
2504
2505         ret = nvme_reset_ctrl_sync(ctrl);
2506         if (ret < 0)
2507                 return ret;
2508         return count;
2509 }
2510 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2511
2512 static ssize_t nvme_sysfs_rescan(struct device *dev,
2513                                 struct device_attribute *attr, const char *buf,
2514                                 size_t count)
2515 {
2516         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2517
2518         nvme_queue_scan(ctrl);
2519         return count;
2520 }
2521 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2522
2523 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2524 {
2525         struct gendisk *disk = dev_to_disk(dev);
2526
2527         if (disk->fops == &nvme_fops)
2528                 return nvme_get_ns_from_dev(dev)->head;
2529         else
2530                 return disk->private_data;
2531 }
2532
2533 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2534                 char *buf)
2535 {
2536         struct nvme_ns_head *head = dev_to_ns_head(dev);
2537         struct nvme_ns_ids *ids = &head->ids;
2538         struct nvme_subsystem *subsys = head->subsys;
2539         int serial_len = sizeof(subsys->serial);
2540         int model_len = sizeof(subsys->model);
2541
2542         if (!uuid_is_null(&ids->uuid))
2543                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2544
2545         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2546                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2547
2548         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2549                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2550
2551         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2552                                   subsys->serial[serial_len - 1] == '\0'))
2553                 serial_len--;
2554         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2555                                  subsys->model[model_len - 1] == '\0'))
2556                 model_len--;
2557
2558         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2559                 serial_len, subsys->serial, model_len, subsys->model,
2560                 head->ns_id);
2561 }
2562 static DEVICE_ATTR_RO(wwid);
2563
2564 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2565                 char *buf)
2566 {
2567         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2568 }
2569 static DEVICE_ATTR_RO(nguid);
2570
2571 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2572                 char *buf)
2573 {
2574         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2575
2576         /* For backward compatibility expose the NGUID to userspace if
2577          * we have no UUID set
2578          */
2579         if (uuid_is_null(&ids->uuid)) {
2580                 printk_ratelimited(KERN_WARNING
2581                                    "No UUID available providing old NGUID\n");
2582                 return sprintf(buf, "%pU\n", ids->nguid);
2583         }
2584         return sprintf(buf, "%pU\n", &ids->uuid);
2585 }
2586 static DEVICE_ATTR_RO(uuid);
2587
2588 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2589                 char *buf)
2590 {
2591         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2592 }
2593 static DEVICE_ATTR_RO(eui);
2594
2595 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2596                 char *buf)
2597 {
2598         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2599 }
2600 static DEVICE_ATTR_RO(nsid);
2601
2602 static struct attribute *nvme_ns_id_attrs[] = {
2603         &dev_attr_wwid.attr,
2604         &dev_attr_uuid.attr,
2605         &dev_attr_nguid.attr,
2606         &dev_attr_eui.attr,
2607         &dev_attr_nsid.attr,
2608         NULL,
2609 };
2610
2611 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2612                 struct attribute *a, int n)
2613 {
2614         struct device *dev = container_of(kobj, struct device, kobj);
2615         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2616
2617         if (a == &dev_attr_uuid.attr) {
2618                 if (uuid_is_null(&ids->uuid) &&
2619                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2620                         return 0;
2621         }
2622         if (a == &dev_attr_nguid.attr) {
2623                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2624                         return 0;
2625         }
2626         if (a == &dev_attr_eui.attr) {
2627                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2628                         return 0;
2629         }
2630         return a->mode;
2631 }
2632
2633 const struct attribute_group nvme_ns_id_attr_group = {
2634         .attrs          = nvme_ns_id_attrs,
2635         .is_visible     = nvme_ns_id_attrs_are_visible,
2636 };
2637
2638 #define nvme_show_str_function(field)                                           \
2639 static ssize_t  field##_show(struct device *dev,                                \
2640                             struct device_attribute *attr, char *buf)           \
2641 {                                                                               \
2642         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2643         return sprintf(buf, "%.*s\n",                                           \
2644                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2645 }                                                                               \
2646 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2647
2648 nvme_show_str_function(model);
2649 nvme_show_str_function(serial);
2650 nvme_show_str_function(firmware_rev);
2651
2652 #define nvme_show_int_function(field)                                           \
2653 static ssize_t  field##_show(struct device *dev,                                \
2654                             struct device_attribute *attr, char *buf)           \
2655 {                                                                               \
2656         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2657         return sprintf(buf, "%d\n", ctrl->field);       \
2658 }                                                                               \
2659 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2660
2661 nvme_show_int_function(cntlid);
2662
2663 static ssize_t nvme_sysfs_delete(struct device *dev,
2664                                 struct device_attribute *attr, const char *buf,
2665                                 size_t count)
2666 {
2667         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2668
2669         if (device_remove_file_self(dev, attr))
2670                 nvme_delete_ctrl_sync(ctrl);
2671         return count;
2672 }
2673 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2674
2675 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2676                                          struct device_attribute *attr,
2677                                          char *buf)
2678 {
2679         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2680
2681         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2682 }
2683 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2684
2685 static ssize_t nvme_sysfs_show_state(struct device *dev,
2686                                      struct device_attribute *attr,
2687                                      char *buf)
2688 {
2689         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2690         static const char *const state_name[] = {
2691                 [NVME_CTRL_NEW]         = "new",
2692                 [NVME_CTRL_LIVE]        = "live",
2693                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2694                 [NVME_CTRL_RESETTING]   = "resetting",
2695                 [NVME_CTRL_CONNECTING]  = "connecting",
2696                 [NVME_CTRL_DELETING]    = "deleting",
2697                 [NVME_CTRL_DEAD]        = "dead",
2698         };
2699
2700         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2701             state_name[ctrl->state])
2702                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2703
2704         return sprintf(buf, "unknown state\n");
2705 }
2706
2707 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2708
2709 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2710                                          struct device_attribute *attr,
2711                                          char *buf)
2712 {
2713         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2714
2715         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2716 }
2717 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2718
2719 static ssize_t nvme_sysfs_show_address(struct device *dev,
2720                                          struct device_attribute *attr,
2721                                          char *buf)
2722 {
2723         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2724
2725         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2726 }
2727 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2728
2729 static struct attribute *nvme_dev_attrs[] = {
2730         &dev_attr_reset_controller.attr,
2731         &dev_attr_rescan_controller.attr,
2732         &dev_attr_model.attr,
2733         &dev_attr_serial.attr,
2734         &dev_attr_firmware_rev.attr,
2735         &dev_attr_cntlid.attr,
2736         &dev_attr_delete_controller.attr,
2737         &dev_attr_transport.attr,
2738         &dev_attr_subsysnqn.attr,
2739         &dev_attr_address.attr,
2740         &dev_attr_state.attr,
2741         NULL
2742 };
2743
2744 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2745                 struct attribute *a, int n)
2746 {
2747         struct device *dev = container_of(kobj, struct device, kobj);
2748         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2749
2750         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2751                 return 0;
2752         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2753                 return 0;
2754
2755         return a->mode;
2756 }
2757
2758 static struct attribute_group nvme_dev_attrs_group = {
2759         .attrs          = nvme_dev_attrs,
2760         .is_visible     = nvme_dev_attrs_are_visible,
2761 };
2762
2763 static const struct attribute_group *nvme_dev_attr_groups[] = {
2764         &nvme_dev_attrs_group,
2765         NULL,
2766 };
2767
2768 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2769                 unsigned nsid)
2770 {
2771         struct nvme_ns_head *h;
2772
2773         lockdep_assert_held(&subsys->lock);
2774
2775         list_for_each_entry(h, &subsys->nsheads, entry) {
2776                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2777                         return h;
2778         }
2779
2780         return NULL;
2781 }
2782
2783 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2784                 struct nvme_ns_head *new)
2785 {
2786         struct nvme_ns_head *h;
2787
2788         lockdep_assert_held(&subsys->lock);
2789
2790         list_for_each_entry(h, &subsys->nsheads, entry) {
2791                 if (nvme_ns_ids_valid(&new->ids) &&
2792                     nvme_ns_ids_equal(&new->ids, &h->ids))
2793                         return -EINVAL;
2794         }
2795
2796         return 0;
2797 }
2798
2799 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2800                 unsigned nsid, struct nvme_id_ns *id)
2801 {
2802         struct nvme_ns_head *head;
2803         int ret = -ENOMEM;
2804
2805         head = kzalloc(sizeof(*head), GFP_KERNEL);
2806         if (!head)
2807                 goto out;
2808         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2809         if (ret < 0)
2810                 goto out_free_head;
2811         head->instance = ret;
2812         INIT_LIST_HEAD(&head->list);
2813         init_srcu_struct(&head->srcu);
2814         head->subsys = ctrl->subsys;
2815         head->ns_id = nsid;
2816         kref_init(&head->ref);
2817
2818         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2819
2820         ret = __nvme_check_ids(ctrl->subsys, head);
2821         if (ret) {
2822                 dev_err(ctrl->device,
2823                         "duplicate IDs for nsid %d\n", nsid);
2824                 goto out_cleanup_srcu;
2825         }
2826
2827         ret = nvme_mpath_alloc_disk(ctrl, head);
2828         if (ret)
2829                 goto out_cleanup_srcu;
2830
2831         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2832         return head;
2833 out_cleanup_srcu:
2834         cleanup_srcu_struct(&head->srcu);
2835         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2836 out_free_head:
2837         kfree(head);
2838 out:
2839         return ERR_PTR(ret);
2840 }
2841
2842 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2843                 struct nvme_id_ns *id, bool *new)
2844 {
2845         struct nvme_ctrl *ctrl = ns->ctrl;
2846         bool is_shared = id->nmic & (1 << 0);
2847         struct nvme_ns_head *head = NULL;
2848         int ret = 0;
2849
2850         mutex_lock(&ctrl->subsys->lock);
2851         if (is_shared)
2852                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2853         if (!head) {
2854                 head = nvme_alloc_ns_head(ctrl, nsid, id);
2855                 if (IS_ERR(head)) {
2856                         ret = PTR_ERR(head);
2857                         goto out_unlock;
2858                 }
2859
2860                 *new = true;
2861         } else {
2862                 struct nvme_ns_ids ids;
2863
2864                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
2865                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2866                         dev_err(ctrl->device,
2867                                 "IDs don't match for shared namespace %d\n",
2868                                         nsid);
2869                         ret = -EINVAL;
2870                         goto out_unlock;
2871                 }
2872
2873                 *new = false;
2874         }
2875
2876         list_add_tail(&ns->siblings, &head->list);
2877         ns->head = head;
2878
2879 out_unlock:
2880         mutex_unlock(&ctrl->subsys->lock);
2881         return ret;
2882 }
2883
2884 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
2885 {
2886         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
2887         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
2888
2889         return nsa->head->ns_id - nsb->head->ns_id;
2890 }
2891
2892 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2893 {
2894         struct nvme_ns *ns, *ret = NULL;
2895
2896         mutex_lock(&ctrl->namespaces_mutex);
2897         list_for_each_entry(ns, &ctrl->namespaces, list) {
2898                 if (ns->head->ns_id == nsid) {
2899                         if (!kref_get_unless_zero(&ns->kref))
2900                                 continue;
2901                         ret = ns;
2902                         break;
2903                 }
2904                 if (ns->head->ns_id > nsid)
2905                         break;
2906         }
2907         mutex_unlock(&ctrl->namespaces_mutex);
2908         return ret;
2909 }
2910
2911 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
2912 {
2913         struct streams_directive_params s;
2914         int ret;
2915
2916         if (!ctrl->nr_streams)
2917                 return 0;
2918
2919         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2920         if (ret)
2921                 return ret;
2922
2923         ns->sws = le32_to_cpu(s.sws);
2924         ns->sgs = le16_to_cpu(s.sgs);
2925
2926         if (ns->sws) {
2927                 unsigned int bs = 1 << ns->lba_shift;
2928
2929                 blk_queue_io_min(ns->queue, bs * ns->sws);
2930                 if (ns->sgs)
2931                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
2932         }
2933
2934         return 0;
2935 }
2936
2937 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2938 {
2939         struct nvme_ns *ns;
2940         struct gendisk *disk;
2941         struct nvme_id_ns *id;
2942         char disk_name[DISK_NAME_LEN];
2943         int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
2944         bool new = true;
2945
2946         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2947         if (!ns)
2948                 return;
2949
2950         ns->queue = blk_mq_init_queue(ctrl->tagset);
2951         if (IS_ERR(ns->queue))
2952                 goto out_free_ns;
2953         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2954         ns->queue->queuedata = ns;
2955         ns->ctrl = ctrl;
2956
2957         kref_init(&ns->kref);
2958         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2959
2960         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2961         nvme_set_queue_limits(ctrl, ns->queue);
2962
2963         id = nvme_identify_ns(ctrl, nsid);
2964         if (!id)
2965                 goto out_free_queue;
2966
2967         if (id->ncap == 0)
2968                 goto out_free_id;
2969
2970         if (nvme_init_ns_head(ns, nsid, id, &new))
2971                 goto out_free_id;
2972         nvme_setup_streams_ns(ctrl, ns);
2973
2974 #ifdef CONFIG_NVME_MULTIPATH
2975         /*
2976          * If multipathing is enabled we need to always use the subsystem
2977          * instance number for numbering our devices to avoid conflicts
2978          * between subsystems that have multiple controllers and thus use
2979          * the multipath-aware subsystem node and those that have a single
2980          * controller and use the controller node directly.
2981          */
2982         if (ns->head->disk) {
2983                 sprintf(disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
2984                                 ctrl->cntlid, ns->head->instance);
2985                 flags = GENHD_FL_HIDDEN;
2986         } else {
2987                 sprintf(disk_name, "nvme%dn%d", ctrl->subsys->instance,
2988                                 ns->head->instance);
2989         }
2990 #else
2991         /*
2992          * But without the multipath code enabled, multiple controller per
2993          * subsystems are visible as devices and thus we cannot use the
2994          * subsystem instance.
2995          */
2996         sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
2997 #endif
2998
2999         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3000                 if (nvme_nvm_register(ns, disk_name, node)) {
3001                         dev_warn(ctrl->device, "LightNVM init failure\n");
3002                         goto out_unlink_ns;
3003                 }
3004         }
3005
3006         disk = alloc_disk_node(0, node);
3007         if (!disk)
3008                 goto out_unlink_ns;
3009
3010         disk->fops = &nvme_fops;
3011         disk->private_data = ns;
3012         disk->queue = ns->queue;
3013         disk->flags = flags;
3014         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3015         ns->disk = disk;
3016
3017         __nvme_revalidate_disk(disk, id);
3018
3019         mutex_lock(&ctrl->namespaces_mutex);
3020         list_add_tail(&ns->list, &ctrl->namespaces);
3021         mutex_unlock(&ctrl->namespaces_mutex);
3022
3023         nvme_get_ctrl(ctrl);
3024
3025         kfree(id);
3026
3027         device_add_disk(ctrl->device, ns->disk);
3028         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
3029                                         &nvme_ns_id_attr_group))
3030                 pr_warn("%s: failed to create sysfs group for identification\n",
3031                         ns->disk->disk_name);
3032         if (ns->ndev && nvme_nvm_register_sysfs(ns))
3033                 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
3034                         ns->disk->disk_name);
3035
3036         if (new)
3037                 nvme_mpath_add_disk(ns->head);
3038         nvme_mpath_add_disk_links(ns);
3039         return;
3040  out_unlink_ns:
3041         mutex_lock(&ctrl->subsys->lock);
3042         list_del_rcu(&ns->siblings);
3043         mutex_unlock(&ctrl->subsys->lock);
3044  out_free_id:
3045         kfree(id);
3046  out_free_queue:
3047         blk_cleanup_queue(ns->queue);
3048  out_free_ns:
3049         kfree(ns);
3050 }
3051
3052 static void nvme_ns_remove(struct nvme_ns *ns)
3053 {
3054         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3055                 return;
3056
3057         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3058                 nvme_mpath_remove_disk_links(ns);
3059                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
3060                                         &nvme_ns_id_attr_group);
3061                 if (ns->ndev)
3062                         nvme_nvm_unregister_sysfs(ns);
3063                 del_gendisk(ns->disk);
3064                 blk_cleanup_queue(ns->queue);
3065                 if (blk_get_integrity(ns->disk))
3066                         blk_integrity_unregister(ns->disk);
3067         }
3068
3069         mutex_lock(&ns->ctrl->subsys->lock);
3070         nvme_mpath_clear_current_path(ns);
3071         list_del_rcu(&ns->siblings);
3072         mutex_unlock(&ns->ctrl->subsys->lock);
3073
3074         mutex_lock(&ns->ctrl->namespaces_mutex);
3075         list_del_init(&ns->list);
3076         mutex_unlock(&ns->ctrl->namespaces_mutex);
3077
3078         synchronize_srcu(&ns->head->srcu);
3079         nvme_mpath_check_last_path(ns);
3080         nvme_put_ns(ns);
3081 }
3082
3083 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3084 {
3085         struct nvme_ns *ns;
3086
3087         ns = nvme_find_get_ns(ctrl, nsid);
3088         if (ns) {
3089                 if (ns->disk && revalidate_disk(ns->disk))
3090                         nvme_ns_remove(ns);
3091                 nvme_put_ns(ns);
3092         } else
3093                 nvme_alloc_ns(ctrl, nsid);
3094 }
3095
3096 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3097                                         unsigned nsid)
3098 {
3099         struct nvme_ns *ns, *next;
3100
3101         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3102                 if (ns->head->ns_id > nsid)
3103                         nvme_ns_remove(ns);
3104         }
3105 }
3106
3107 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3108 {
3109         struct nvme_ns *ns;
3110         __le32 *ns_list;
3111         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3112         int ret = 0;
3113
3114         ns_list = kzalloc(0x1000, GFP_KERNEL);
3115         if (!ns_list)
3116                 return -ENOMEM;
3117
3118         for (i = 0; i < num_lists; i++) {
3119                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3120                 if (ret)
3121                         goto free;
3122
3123                 for (j = 0; j < min(nn, 1024U); j++) {
3124                         nsid = le32_to_cpu(ns_list[j]);
3125                         if (!nsid)
3126                                 goto out;
3127
3128                         nvme_validate_ns(ctrl, nsid);
3129
3130                         while (++prev < nsid) {
3131                                 ns = nvme_find_get_ns(ctrl, prev);
3132                                 if (ns) {
3133                                         nvme_ns_remove(ns);
3134                                         nvme_put_ns(ns);
3135                                 }
3136                         }
3137                 }
3138                 nn -= j;
3139         }
3140  out:
3141         nvme_remove_invalid_namespaces(ctrl, prev);
3142  free:
3143         kfree(ns_list);
3144         return ret;
3145 }
3146
3147 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3148 {
3149         unsigned i;
3150
3151         for (i = 1; i <= nn; i++)
3152                 nvme_validate_ns(ctrl, i);
3153
3154         nvme_remove_invalid_namespaces(ctrl, nn);
3155 }
3156
3157 static void nvme_scan_work(struct work_struct *work)
3158 {
3159         struct nvme_ctrl *ctrl =
3160                 container_of(work, struct nvme_ctrl, scan_work);
3161         struct nvme_id_ctrl *id;
3162         unsigned nn;
3163
3164         if (ctrl->state != NVME_CTRL_LIVE)
3165                 return;
3166
3167         WARN_ON_ONCE(!ctrl->tagset);
3168
3169         if (nvme_identify_ctrl(ctrl, &id))
3170                 return;
3171
3172         nn = le32_to_cpu(id->nn);
3173         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3174             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3175                 if (!nvme_scan_ns_list(ctrl, nn))
3176                         goto done;
3177         }
3178         nvme_scan_ns_sequential(ctrl, nn);
3179  done:
3180         mutex_lock(&ctrl->namespaces_mutex);
3181         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3182         mutex_unlock(&ctrl->namespaces_mutex);
3183         kfree(id);
3184 }
3185
3186 void nvme_queue_scan(struct nvme_ctrl *ctrl)
3187 {
3188         /*
3189          * Only new queue scan work when admin and IO queues are both alive
3190          */
3191         if (ctrl->state == NVME_CTRL_LIVE)
3192                 queue_work(nvme_wq, &ctrl->scan_work);
3193 }
3194 EXPORT_SYMBOL_GPL(nvme_queue_scan);
3195
3196 /*
3197  * This function iterates the namespace list unlocked to allow recovery from
3198  * controller failure. It is up to the caller to ensure the namespace list is
3199  * not modified by scan work while this function is executing.
3200  */
3201 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3202 {
3203         struct nvme_ns *ns, *next;
3204
3205         /*
3206          * The dead states indicates the controller was not gracefully
3207          * disconnected. In that case, we won't be able to flush any data while
3208          * removing the namespaces' disks; fail all the queues now to avoid
3209          * potentially having to clean up the failed sync later.
3210          */
3211         if (ctrl->state == NVME_CTRL_DEAD)
3212                 nvme_kill_queues(ctrl);
3213
3214         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
3215                 nvme_ns_remove(ns);
3216 }
3217 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3218
3219 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3220 {
3221         char *envp[2] = { NULL, NULL };
3222         u32 aen_result = ctrl->aen_result;
3223
3224         ctrl->aen_result = 0;
3225         if (!aen_result)
3226                 return;
3227
3228         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3229         if (!envp[0])
3230                 return;
3231         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3232         kfree(envp[0]);
3233 }
3234
3235 static void nvme_async_event_work(struct work_struct *work)
3236 {
3237         struct nvme_ctrl *ctrl =
3238                 container_of(work, struct nvme_ctrl, async_event_work);
3239
3240         nvme_aen_uevent(ctrl);
3241         ctrl->ops->submit_async_event(ctrl);
3242 }
3243
3244 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3245 {
3246
3247         u32 csts;
3248
3249         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3250                 return false;
3251
3252         if (csts == ~0)
3253                 return false;
3254
3255         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3256 }
3257
3258 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3259 {
3260         struct nvme_fw_slot_info_log *log;
3261
3262         log = kmalloc(sizeof(*log), GFP_KERNEL);
3263         if (!log)
3264                 return;
3265
3266         if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
3267                 dev_warn(ctrl->device,
3268                                 "Get FW SLOT INFO log error\n");
3269         kfree(log);
3270 }
3271
3272 static void nvme_fw_act_work(struct work_struct *work)
3273 {
3274         struct nvme_ctrl *ctrl = container_of(work,
3275                                 struct nvme_ctrl, fw_act_work);
3276         unsigned long fw_act_timeout;
3277
3278         if (ctrl->mtfa)
3279                 fw_act_timeout = jiffies +
3280                                 msecs_to_jiffies(ctrl->mtfa * 100);
3281         else
3282                 fw_act_timeout = jiffies +
3283                                 msecs_to_jiffies(admin_timeout * 1000);
3284
3285         nvme_stop_queues(ctrl);
3286         while (nvme_ctrl_pp_status(ctrl)) {
3287                 if (time_after(jiffies, fw_act_timeout)) {
3288                         dev_warn(ctrl->device,
3289                                 "Fw activation timeout, reset controller\n");
3290                         nvme_reset_ctrl(ctrl);
3291                         break;
3292                 }
3293                 msleep(100);
3294         }
3295
3296         if (ctrl->state != NVME_CTRL_LIVE)
3297                 return;
3298
3299         nvme_start_queues(ctrl);
3300         /* read FW slot information to clear the AER */
3301         nvme_get_fw_slot_info(ctrl);
3302 }
3303
3304 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3305                 union nvme_result *res)
3306 {
3307         u32 result = le32_to_cpu(res->u32);
3308
3309         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3310                 return;
3311
3312         switch (result & 0x7) {
3313         case NVME_AER_ERROR:
3314         case NVME_AER_SMART:
3315         case NVME_AER_CSS:
3316         case NVME_AER_VS:
3317                 ctrl->aen_result = result;
3318                 break;
3319         default:
3320                 break;
3321         }
3322
3323         switch (result & 0xff07) {
3324         case NVME_AER_NOTICE_NS_CHANGED:
3325                 dev_info(ctrl->device, "rescanning\n");
3326                 nvme_queue_scan(ctrl);
3327                 break;
3328         case NVME_AER_NOTICE_FW_ACT_STARTING:
3329                 queue_work(nvme_wq, &ctrl->fw_act_work);
3330                 break;
3331         default:
3332                 dev_warn(ctrl->device, "async event result %08x\n", result);
3333         }
3334         queue_work(nvme_wq, &ctrl->async_event_work);
3335 }
3336 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3337
3338 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3339 {
3340         nvme_stop_keep_alive(ctrl);
3341         flush_work(&ctrl->async_event_work);
3342         flush_work(&ctrl->scan_work);
3343         cancel_work_sync(&ctrl->fw_act_work);
3344 }
3345 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3346
3347 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3348 {
3349         if (ctrl->kato)
3350                 nvme_start_keep_alive(ctrl);
3351
3352         if (ctrl->queue_count > 1) {
3353                 nvme_queue_scan(ctrl);
3354                 queue_work(nvme_wq, &ctrl->async_event_work);
3355                 nvme_start_queues(ctrl);
3356         }
3357 }
3358 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3359
3360 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3361 {
3362         cdev_device_del(&ctrl->cdev, ctrl->device);
3363 }
3364 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3365
3366 static void nvme_free_ctrl(struct device *dev)
3367 {
3368         struct nvme_ctrl *ctrl =
3369                 container_of(dev, struct nvme_ctrl, ctrl_device);
3370         struct nvme_subsystem *subsys = ctrl->subsys;
3371
3372         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3373         kfree(ctrl->effects);
3374
3375         if (subsys) {
3376                 mutex_lock(&subsys->lock);
3377                 list_del(&ctrl->subsys_entry);
3378                 mutex_unlock(&subsys->lock);
3379                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3380         }
3381
3382         ctrl->ops->free_ctrl(ctrl);
3383
3384         if (subsys)
3385                 nvme_put_subsystem(subsys);
3386 }
3387
3388 /*
3389  * Initialize a NVMe controller structures.  This needs to be called during
3390  * earliest initialization so that we have the initialized structured around
3391  * during probing.
3392  */
3393 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3394                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3395 {
3396         int ret;
3397
3398         ctrl->state = NVME_CTRL_NEW;
3399         spin_lock_init(&ctrl->lock);
3400         INIT_LIST_HEAD(&ctrl->namespaces);
3401         mutex_init(&ctrl->namespaces_mutex);
3402         ctrl->dev = dev;
3403         ctrl->ops = ops;
3404         ctrl->quirks = quirks;
3405         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3406         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3407         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3408         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3409
3410         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3411         if (ret < 0)
3412                 goto out;
3413         ctrl->instance = ret;
3414
3415         device_initialize(&ctrl->ctrl_device);
3416         ctrl->device = &ctrl->ctrl_device;
3417         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3418         ctrl->device->class = nvme_class;
3419         ctrl->device->parent = ctrl->dev;
3420         ctrl->device->groups = nvme_dev_attr_groups;
3421         ctrl->device->release = nvme_free_ctrl;
3422         dev_set_drvdata(ctrl->device, ctrl);
3423         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3424         if (ret)
3425                 goto out_release_instance;
3426
3427         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3428         ctrl->cdev.owner = ops->module;
3429         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3430         if (ret)
3431                 goto out_free_name;
3432
3433         /*
3434          * Initialize latency tolerance controls.  The sysfs files won't
3435          * be visible to userspace unless the device actually supports APST.
3436          */
3437         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3438         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3439                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3440
3441         return 0;
3442 out_free_name:
3443         kfree_const(dev->kobj.name);
3444 out_release_instance:
3445         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3446 out:
3447         return ret;
3448 }
3449 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3450
3451 /**
3452  * nvme_kill_queues(): Ends all namespace queues
3453  * @ctrl: the dead controller that needs to end
3454  *
3455  * Call this function when the driver determines it is unable to get the
3456  * controller in a state capable of servicing IO.
3457  */
3458 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3459 {
3460         struct nvme_ns *ns;
3461
3462         mutex_lock(&ctrl->namespaces_mutex);
3463
3464         /* Forcibly unquiesce queues to avoid blocking dispatch */
3465         if (ctrl->admin_q)
3466                 blk_mq_unquiesce_queue(ctrl->admin_q);
3467
3468         list_for_each_entry(ns, &ctrl->namespaces, list) {
3469                 /*
3470                  * Revalidating a dead namespace sets capacity to 0. This will
3471                  * end buffered writers dirtying pages that can't be synced.
3472                  */
3473                 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
3474                         continue;
3475                 revalidate_disk(ns->disk);
3476                 blk_set_queue_dying(ns->queue);
3477
3478                 /* Forcibly unquiesce queues to avoid blocking dispatch */
3479                 blk_mq_unquiesce_queue(ns->queue);
3480         }
3481         mutex_unlock(&ctrl->namespaces_mutex);
3482 }
3483 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3484
3485 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3486 {
3487         struct nvme_ns *ns;
3488
3489         mutex_lock(&ctrl->namespaces_mutex);
3490         list_for_each_entry(ns, &ctrl->namespaces, list)
3491                 blk_mq_unfreeze_queue(ns->queue);
3492         mutex_unlock(&ctrl->namespaces_mutex);
3493 }
3494 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3495
3496 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3497 {
3498         struct nvme_ns *ns;
3499
3500         mutex_lock(&ctrl->namespaces_mutex);
3501         list_for_each_entry(ns, &ctrl->namespaces, list) {
3502                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3503                 if (timeout <= 0)
3504                         break;
3505         }
3506         mutex_unlock(&ctrl->namespaces_mutex);
3507 }
3508 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3509
3510 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3511 {
3512         struct nvme_ns *ns;
3513
3514         mutex_lock(&ctrl->namespaces_mutex);
3515         list_for_each_entry(ns, &ctrl->namespaces, list)
3516                 blk_mq_freeze_queue_wait(ns->queue);
3517         mutex_unlock(&ctrl->namespaces_mutex);
3518 }
3519 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3520
3521 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3522 {
3523         struct nvme_ns *ns;
3524
3525         mutex_lock(&ctrl->namespaces_mutex);
3526         list_for_each_entry(ns, &ctrl->namespaces, list)
3527                 blk_freeze_queue_start(ns->queue);
3528         mutex_unlock(&ctrl->namespaces_mutex);
3529 }
3530 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3531
3532 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3533 {
3534         struct nvme_ns *ns;
3535
3536         mutex_lock(&ctrl->namespaces_mutex);
3537         list_for_each_entry(ns, &ctrl->namespaces, list)
3538                 blk_mq_quiesce_queue(ns->queue);
3539         mutex_unlock(&ctrl->namespaces_mutex);
3540 }
3541 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3542
3543 void nvme_start_queues(struct nvme_ctrl *ctrl)
3544 {
3545         struct nvme_ns *ns;
3546
3547         mutex_lock(&ctrl->namespaces_mutex);
3548         list_for_each_entry(ns, &ctrl->namespaces, list)
3549                 blk_mq_unquiesce_queue(ns->queue);
3550         mutex_unlock(&ctrl->namespaces_mutex);
3551 }
3552 EXPORT_SYMBOL_GPL(nvme_start_queues);
3553
3554 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set)
3555 {
3556         if (!ctrl->ops->reinit_request)
3557                 return 0;
3558
3559         return blk_mq_tagset_iter(set, set->driver_data,
3560                         ctrl->ops->reinit_request);
3561 }
3562 EXPORT_SYMBOL_GPL(nvme_reinit_tagset);
3563
3564 int __init nvme_core_init(void)
3565 {
3566         int result = -ENOMEM;
3567
3568         nvme_wq = alloc_workqueue("nvme-wq",
3569                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3570         if (!nvme_wq)
3571                 goto out;
3572
3573         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3574                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3575         if (!nvme_reset_wq)
3576                 goto destroy_wq;
3577
3578         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3579                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3580         if (!nvme_delete_wq)
3581                 goto destroy_reset_wq;
3582
3583         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3584         if (result < 0)
3585                 goto destroy_delete_wq;
3586
3587         nvme_class = class_create(THIS_MODULE, "nvme");
3588         if (IS_ERR(nvme_class)) {
3589                 result = PTR_ERR(nvme_class);
3590                 goto unregister_chrdev;
3591         }
3592
3593         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3594         if (IS_ERR(nvme_subsys_class)) {
3595                 result = PTR_ERR(nvme_subsys_class);
3596                 goto destroy_class;
3597         }
3598         return 0;
3599
3600 destroy_class:
3601         class_destroy(nvme_class);
3602 unregister_chrdev:
3603         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3604 destroy_delete_wq:
3605         destroy_workqueue(nvme_delete_wq);
3606 destroy_reset_wq:
3607         destroy_workqueue(nvme_reset_wq);
3608 destroy_wq:
3609         destroy_workqueue(nvme_wq);
3610 out:
3611         return result;
3612 }
3613
3614 void nvme_core_exit(void)
3615 {
3616         ida_destroy(&nvme_subsystems_ida);
3617         class_destroy(nvme_subsys_class);
3618         class_destroy(nvme_class);
3619         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3620         destroy_workqueue(nvme_delete_wq);
3621         destroy_workqueue(nvme_reset_wq);
3622         destroy_workqueue(nvme_wq);
3623 }
3624
3625 MODULE_LICENSE("GPL");
3626 MODULE_VERSION("1.0");
3627 module_init(nvme_core_init);
3628 module_exit(nvme_core_exit);