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                 if (ctrl->state != NVME_CTRL_LIVE)
 126                         ret = -ENETRESET;
 127         }
 128
 129         return ret;
 130 }
 131 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 132
 133 static void nvme_delete_ctrl_work(struct work_struct *work)
 134 {
 135         struct nvme_ctrl *ctrl =
 136                 container_of(work, struct nvme_ctrl, delete_work);
 137
 138         flush_work(&ctrl->reset_work);
 139         nvme_stop_ctrl(ctrl);
 140         nvme_remove_namespaces(ctrl);
 141         ctrl->ops->delete_ctrl(ctrl);
 142         nvme_uninit_ctrl(ctrl);
 143         nvme_put_ctrl(ctrl);
 144 }
 145
 146 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 147 {
 148         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 149                 return -EBUSY;
 150         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 151                 return -EBUSY;
 152         return 0;
 153 }
 154 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 155
 156 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 157 {
 158         int ret = 0;
 159
 160         /*
 161          * Keep a reference until the work is flushed since ->delete_ctrl
 162          * can free the controller.
 163          */
 164         nvme_get_ctrl(ctrl);
 165         ret = nvme_delete_ctrl(ctrl);
 166         if (!ret)
 167                 flush_work(&ctrl->delete_work);
 168         nvme_put_ctrl(ctrl);
 169         return ret;
 170 }
 171 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
 172
 173 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 174 {
 175         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 176 }
 177
 178 static blk_status_t nvme_error_status(struct request *req)
 179 {
 180         switch (nvme_req(req)->status & 0x7ff) {
 181         case NVME_SC_SUCCESS:
 182                 return BLK_STS_OK;
 183         case NVME_SC_CAP_EXCEEDED:
 184                 return BLK_STS_NOSPC;
 185         case NVME_SC_LBA_RANGE:
 186                 return BLK_STS_TARGET;
 187         case NVME_SC_BAD_ATTRIBUTES:
 188         case NVME_SC_ONCS_NOT_SUPPORTED:
 189         case NVME_SC_INVALID_OPCODE:
 190         case NVME_SC_INVALID_FIELD:
 191         case NVME_SC_INVALID_NS:
 192                 return BLK_STS_NOTSUPP;
 193         case NVME_SC_WRITE_FAULT:
 194         case NVME_SC_READ_ERROR:
 195         case NVME_SC_UNWRITTEN_BLOCK:
 196         case NVME_SC_ACCESS_DENIED:
 197         case NVME_SC_READ_ONLY:
 198         case NVME_SC_COMPARE_FAILED:
 199                 return BLK_STS_MEDIUM;
 200         case NVME_SC_GUARD_CHECK:
 201         case NVME_SC_APPTAG_CHECK:
 202         case NVME_SC_REFTAG_CHECK:
 203         case NVME_SC_INVALID_PI:
 204                 return BLK_STS_PROTECTION;
 205         case NVME_SC_RESERVATION_CONFLICT:
 206                 return BLK_STS_NEXUS;
 207         default:
 208                 return BLK_STS_IOERR;
 209         }
 210 }
 211
 212 static inline bool nvme_req_needs_retry(struct request *req)
 213 {
 214         if (blk_noretry_request(req))
 215                 return false;
 216         if (nvme_req(req)->status & NVME_SC_DNR)
 217                 return false;
 218         if (nvme_req(req)->retries >= nvme_max_retries)
 219                 return false;
 220         return true;
 221 }
 222
 223 void nvme_complete_rq(struct request *req)
 224 {
 225         blk_status_t status = nvme_error_status(req);
 226
 227         trace_nvme_complete_rq(req);
 228
 229         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 230                 if (nvme_req_needs_failover(req, status)) {
 231                         nvme_failover_req(req);
 232                         return;
 233                 }
 234
 235                 if (!blk_queue_dying(req->q)) {
 236                         nvme_req(req)->retries++;
 237                         blk_mq_requeue_request(req, true);
 238                         return;
 239                 }
 240         }
 241         blk_mq_end_request(req, status);
 242 }
 243 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 244
 245 void nvme_cancel_request(struct request *req, void *data, bool reserved)
 246 {
 247         if (!blk_mq_request_started(req))
 248                 return;
 249
 250         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 251                                 "Cancelling I/O %d", req->tag);
 252
 253         nvme_req(req)->status = NVME_SC_ABORT_REQ;
 254         blk_mq_complete_request(req);
 255
 256 }
 257 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 258
 259 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 260                 enum nvme_ctrl_state new_state)
 261 {
 262         enum nvme_ctrl_state old_state;
 263         unsigned long flags;
 264         bool changed = false;
 265
 266         spin_lock_irqsave(&ctrl->lock, flags);
 267
 268         old_state = ctrl->state;
 269         switch (new_state) {
 270         case NVME_CTRL_ADMIN_ONLY:
 271                 switch (old_state) {
 272                 case NVME_CTRL_CONNECTING:
 273                         changed = true;
 274                         /* FALLTHRU */
 275                 default:
 276                         break;
 277                 }
 278                 break;
 279         case NVME_CTRL_LIVE:
 280                 switch (old_state) {
 281                 case NVME_CTRL_NEW:
 282                 case NVME_CTRL_RESETTING:
 283                 case NVME_CTRL_CONNECTING:
 284                         changed = true;
 285                         /* FALLTHRU */
 286                 default:
 287                         break;
 288                 }
 289                 break;
 290         case NVME_CTRL_RESETTING:
 291                 switch (old_state) {
 292                 case NVME_CTRL_NEW:
 293                 case NVME_CTRL_LIVE:
 294                 case NVME_CTRL_ADMIN_ONLY:
 295                         changed = true;
 296                         /* FALLTHRU */
 297                 default:
 298                         break;
 299                 }
 300                 break;
 301         case NVME_CTRL_CONNECTING:
 302                 switch (old_state) {
 303                 case NVME_CTRL_NEW:
 304                 case NVME_CTRL_RESETTING:
 305                         changed = true;
 306                         /* FALLTHRU */
 307                 default:
 308                         break;
 309                 }
 310                 break;
 311         case NVME_CTRL_DELETING:
 312                 switch (old_state) {
 313                 case NVME_CTRL_LIVE:
 314                 case NVME_CTRL_ADMIN_ONLY:
 315                 case NVME_CTRL_RESETTING:
 316                 case NVME_CTRL_CONNECTING:
 317                         changed = true;
 318                         /* FALLTHRU */
 319                 default:
 320                         break;
 321                 }
 322                 break;
 323         case NVME_CTRL_DEAD:
 324                 switch (old_state) {
 325                 case NVME_CTRL_DELETING:
 326                         changed = true;
 327                         /* FALLTHRU */
 328                 default:
 329                         break;
 330                 }
 331                 break;
 332         default:
 333                 break;
 334         }
 335
 336         if (changed)
 337                 ctrl->state = new_state;
 338
 339         spin_unlock_irqrestore(&ctrl->lock, flags);
 340         if (changed && ctrl->state == NVME_CTRL_LIVE)
 341                 nvme_kick_requeue_lists(ctrl);
 342         return changed;
 343 }
 344 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 345
 346 static void nvme_free_ns_head(struct kref *ref)
 347 {
 348         struct nvme_ns_head *head =
 349                 container_of(ref, struct nvme_ns_head, ref);
 350
 351         nvme_mpath_remove_disk(head);
 352         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 353         list_del_init(&head->entry);
 354         cleanup_srcu_struct(&head->srcu);
 355         kfree(head);
 356 }
 357
 358 static void nvme_put_ns_head(struct nvme_ns_head *head)
 359 {
 360         kref_put(&head->ref, nvme_free_ns_head);
 361 }
 362
 363 static void nvme_free_ns(struct kref *kref)
 364 {
 365         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 366
 367         if (ns->ndev)
 368                 nvme_nvm_unregister(ns);
 369
 370         put_disk(ns->disk);
 371         nvme_put_ns_head(ns->head);
 372         nvme_put_ctrl(ns->ctrl);
 373         kfree(ns);
 374 }
 375
 376 static void nvme_put_ns(struct nvme_ns *ns)
 377 {
 378         kref_put(&ns->kref, nvme_free_ns);
 379 }
 380
 381 struct request *nvme_alloc_request(struct request_queue *q,
 382                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 383 {
 384         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 385         struct request *req;
 386
 387         if (qid == NVME_QID_ANY) {
 388                 req = blk_mq_alloc_request(q, op, flags);
 389         } else {
 390                 req = blk_mq_alloc_request_hctx(q, op, flags,
 391                                 qid ? qid - 1 : 0);
 392         }
 393         if (IS_ERR(req))
 394                 return req;
 395
 396         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 397         nvme_req(req)->cmd = cmd;
 398
 399         return req;
 400 }
 401 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 402
 403 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 404 {
 405         struct nvme_command c;
 406
 407         memset(&c, 0, sizeof(c));
 408
 409         c.directive.opcode = nvme_admin_directive_send;
 410         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 411         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 412         c.directive.dtype = NVME_DIR_IDENTIFY;
 413         c.directive.tdtype = NVME_DIR_STREAMS;
 414         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 415
 416         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 417 }
 418
 419 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 420 {
 421         return nvme_toggle_streams(ctrl, false);
 422 }
 423
 424 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 425 {
 426         return nvme_toggle_streams(ctrl, true);
 427 }
 428
 429 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 430                                   struct streams_directive_params *s, u32 nsid)
 431 {
 432         struct nvme_command c;
 433
 434         memset(&c, 0, sizeof(c));
 435         memset(s, 0, sizeof(*s));
 436
 437         c.directive.opcode = nvme_admin_directive_recv;
 438         c.directive.nsid = cpu_to_le32(nsid);
 439         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 440         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 441         c.directive.dtype = NVME_DIR_STREAMS;
 442
 443         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 444 }
 445
 446 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 447 {
 448         struct streams_directive_params s;
 449         int ret;
 450
 451         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 452                 return 0;
 453         if (!streams)
 454                 return 0;
 455
 456         ret = nvme_enable_streams(ctrl);
 457         if (ret)
 458                 return ret;
 459
 460         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 461         if (ret)
 462                 return ret;
 463
 464         ctrl->nssa = le16_to_cpu(s.nssa);
 465         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 466                 dev_info(ctrl->device, "too few streams (%u) available\n",
 467                                         ctrl->nssa);
 468                 nvme_disable_streams(ctrl);
 469                 return 0;
 470         }
 471
 472         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 473         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 474         return 0;
 475 }
 476
 477 /*
 478  * Check if 'req' has a write hint associated with it. If it does, assign
 479  * a valid namespace stream to the write.
 480  */
 481 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 482                                      struct request *req, u16 *control,
 483                                      u32 *dsmgmt)
 484 {
 485         enum rw_hint streamid = req->write_hint;
 486
 487         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 488                 streamid = 0;
 489         else {
 490                 streamid--;
 491                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 492                         return;
 493
 494                 *control |= NVME_RW_DTYPE_STREAMS;
 495                 *dsmgmt |= streamid << 16;
 496         }
 497
 498         if (streamid < ARRAY_SIZE(req->q->write_hints))
 499                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 500 }
 501
 502 static inline void nvme_setup_flush(struct nvme_ns *ns,
 503                 struct nvme_command *cmnd)
 504 {
 505         memset(cmnd, 0, sizeof(*cmnd));
 506         cmnd->common.opcode = nvme_cmd_flush;
 507         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 508 }
 509
 510 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 511                 struct nvme_command *cmnd)
 512 {
 513         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 514         struct nvme_dsm_range *range;
 515         struct bio *bio;
 516
 517         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 518         if (!range)
 519                 return BLK_STS_RESOURCE;
 520
 521         __rq_for_each_bio(bio, req) {
 522                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 523                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 524
 525                 if (n < segments) {
 526                         range[n].cattr = cpu_to_le32(0);
 527                         range[n].nlb = cpu_to_le32(nlb);
 528                         range[n].slba = cpu_to_le64(slba);
 529                 }
 530                 n++;
 531         }
 532
 533         if (WARN_ON_ONCE(n != segments)) {
 534                 kfree(range);
 535                 return BLK_STS_IOERR;
 536         }
 537
 538         memset(cmnd, 0, sizeof(*cmnd));
 539         cmnd->dsm.opcode = nvme_cmd_dsm;
 540         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 541         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 542         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 543
 544         req->special_vec.bv_page = virt_to_page(range);
 545         req->special_vec.bv_offset = offset_in_page(range);
 546         req->special_vec.bv_len = sizeof(*range) * segments;
 547         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 548
 549         return BLK_STS_OK;
 550 }
 551
 552 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 553                 struct request *req, struct nvme_command *cmnd)
 554 {
 555         struct nvme_ctrl *ctrl = ns->ctrl;
 556         u16 control = 0;
 557         u32 dsmgmt = 0;
 558
 559         if (req->cmd_flags & REQ_FUA)
 560                 control |= NVME_RW_FUA;
 561         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 562                 control |= NVME_RW_LR;
 563
 564         if (req->cmd_flags & REQ_RAHEAD)
 565                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 566
 567         memset(cmnd, 0, sizeof(*cmnd));
 568         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 569         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 570         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 571         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 572
 573         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 574                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 575
 576         if (ns->ms) {
 577                 /*
 578                  * If formated with metadata, the block layer always provides a
 579                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 580                  * we enable the PRACT bit for protection information or set the
 581                  * namespace capacity to zero to prevent any I/O.
 582                  */
 583                 if (!blk_integrity_rq(req)) {
 584                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 585                                 return BLK_STS_NOTSUPP;
 586                         control |= NVME_RW_PRINFO_PRACT;
 587                 }
 588
 589                 switch (ns->pi_type) {
 590                 case NVME_NS_DPS_PI_TYPE3:
 591                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 592                         break;
 593                 case NVME_NS_DPS_PI_TYPE1:
 594                 case NVME_NS_DPS_PI_TYPE2:
 595                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 596                                         NVME_RW_PRINFO_PRCHK_REF;
 597                         cmnd->rw.reftag = cpu_to_le32(
 598                                         nvme_block_nr(ns, blk_rq_pos(req)));
 599                         break;
 600                 }
 601         }
 602
 603         cmnd->rw.control = cpu_to_le16(control);
 604         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 605         return 0;
 606 }
 607
 608 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 609                 struct nvme_command *cmd)
 610 {
 611         blk_status_t ret = BLK_STS_OK;
 612
 613         if (!(req->rq_flags & RQF_DONTPREP)) {
 614                 nvme_req(req)->retries = 0;
 615                 nvme_req(req)->flags = 0;
 616                 req->rq_flags |= RQF_DONTPREP;
 617         }
 618
 619         switch (req_op(req)) {
 620         case REQ_OP_DRV_IN:
 621         case REQ_OP_DRV_OUT:
 622                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 623                 break;
 624         case REQ_OP_FLUSH:
 625                 nvme_setup_flush(ns, cmd);
 626                 break;
 627         case REQ_OP_WRITE_ZEROES:
 628                 /* currently only aliased to deallocate for a few ctrls: */
 629         case REQ_OP_DISCARD:
 630                 ret = nvme_setup_discard(ns, req, cmd);
 631                 break;
 632         case REQ_OP_READ:
 633         case REQ_OP_WRITE:
 634                 ret = nvme_setup_rw(ns, req, cmd);
 635                 break;
 636         default:
 637                 WARN_ON_ONCE(1);
 638                 return BLK_STS_IOERR;
 639         }
 640
 641         cmd->common.command_id = req->tag;
 642         if (ns)
 643                 trace_nvme_setup_nvm_cmd(req->q->id, cmd);
 644         else
 645                 trace_nvme_setup_admin_cmd(cmd);
 646         return ret;
 647 }
 648 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 649
 650 /*
 651  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 652  * if the result is positive, it's an NVM Express status code
 653  */
 654 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 655                 union nvme_result *result, void *buffer, unsigned bufflen,
 656                 unsigned timeout, int qid, int at_head,
 657                 blk_mq_req_flags_t flags)
 658 {
 659         struct request *req;
 660         int ret;
 661
 662         req = nvme_alloc_request(q, cmd, flags, qid);
 663         if (IS_ERR(req))
 664                 return PTR_ERR(req);
 665
 666         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 667
 668         if (buffer && bufflen) {
 669                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 670                 if (ret)
 671                         goto out;
 672         }
 673
 674         blk_execute_rq(req->q, NULL, req, at_head);
 675         if (result)
 676                 *result = nvme_req(req)->result;
 677         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 678                 ret = -EINTR;
 679         else
 680                 ret = nvme_req(req)->status;
 681  out:
 682         blk_mq_free_request(req);
 683         return ret;
 684 }
 685 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 686
 687 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 688                 void *buffer, unsigned bufflen)
 689 {
 690         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 691                         NVME_QID_ANY, 0, 0);
 692 }
 693 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 694
 695 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 696                 unsigned len, u32 seed, bool write)
 697 {
 698         struct bio_integrity_payload *bip;
 699         int ret = -ENOMEM;
 700         void *buf;
 701
 702         buf = kmalloc(len, GFP_KERNEL);
 703         if (!buf)
 704                 goto out;
 705
 706         ret = -EFAULT;
 707         if (write && copy_from_user(buf, ubuf, len))
 708                 goto out_free_meta;
 709
 710         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 711         if (IS_ERR(bip)) {
 712                 ret = PTR_ERR(bip);
 713                 goto out_free_meta;
 714         }
 715
 716         bip->bip_iter.bi_size = len;
 717         bip->bip_iter.bi_sector = seed;
 718         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 719                         offset_in_page(buf));
 720         if (ret == len)
 721                 return buf;
 722         ret = -ENOMEM;
 723 out_free_meta:
 724         kfree(buf);
 725 out:
 726         return ERR_PTR(ret);
 727 }
 728
 729 static int nvme_submit_user_cmd(struct request_queue *q,
 730                 struct nvme_command *cmd, void __user *ubuffer,
 731                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 732                 u32 meta_seed, u32 *result, unsigned timeout)
 733 {
 734         bool write = nvme_is_write(cmd);
 735         struct nvme_ns *ns = q->queuedata;
 736         struct gendisk *disk = ns ? ns->disk : NULL;
 737         struct request *req;
 738         struct bio *bio = NULL;
 739         void *meta = NULL;
 740         int ret;
 741
 742         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 743         if (IS_ERR(req))
 744                 return PTR_ERR(req);
 745
 746         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 747
 748         if (ubuffer && bufflen) {
 749                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 750                                 GFP_KERNEL);
 751                 if (ret)
 752                         goto out;
 753                 bio = req->bio;
 754                 bio->bi_disk = disk;
 755                 if (disk && meta_buffer && meta_len) {
 756                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 757                                         meta_seed, write);
 758                         if (IS_ERR(meta)) {
 759                                 ret = PTR_ERR(meta);
 760                                 goto out_unmap;
 761                         }
 762                 }
 763         }
 764
 765         blk_execute_rq(req->q, disk, req, 0);
 766         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 767                 ret = -EINTR;
 768         else
 769                 ret = nvme_req(req)->status;
 770         if (result)
 771                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 772         if (meta && !ret && !write) {
 773                 if (copy_to_user(meta_buffer, meta, meta_len))
 774                         ret = -EFAULT;
 775         }
 776         kfree(meta);
 777  out_unmap:
 778         if (bio)
 779                 blk_rq_unmap_user(bio);
 780  out:
 781         blk_mq_free_request(req);
 782         return ret;
 783 }
 784
 785 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 786 {
 787         struct nvme_ctrl *ctrl = rq->end_io_data;
 788
 789         blk_mq_free_request(rq);
 790
 791         if (status) {
 792                 dev_err(ctrl->device,
 793                         "failed nvme_keep_alive_end_io error=%d\n",
 794                                 status);
 795                 return;
 796         }
 797
 798         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 799 }
 800
 801 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 802 {
 803         struct request *rq;
 804
 805         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 806                         NVME_QID_ANY);
 807         if (IS_ERR(rq))
 808                 return PTR_ERR(rq);
 809
 810         rq->timeout = ctrl->kato * HZ;
 811         rq->end_io_data = ctrl;
 812
 813         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 814
 815         return 0;
 816 }
 817
 818 static void nvme_keep_alive_work(struct work_struct *work)
 819 {
 820         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 821                         struct nvme_ctrl, ka_work);
 822
 823         if (nvme_keep_alive(ctrl)) {
 824                 /* allocation failure, reset the controller */
 825                 dev_err(ctrl->device, "keep-alive failed\n");
 826                 nvme_reset_ctrl(ctrl);
 827                 return;
 828         }
 829 }
 830
 831 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 832 {
 833         if (unlikely(ctrl->kato == 0))
 834                 return;
 835
 836         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
 837         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
 838         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
 839         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 840 }
 841 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
 842
 843 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 844 {
 845         if (unlikely(ctrl->kato == 0))
 846                 return;
 847
 848         cancel_delayed_work_sync(&ctrl->ka_work);
 849 }
 850 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 851
 852 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 853 {
 854         struct nvme_command c = { };
 855         int error;
 856
 857         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 858         c.identify.opcode = nvme_admin_identify;
 859         c.identify.cns = NVME_ID_CNS_CTRL;
 860
 861         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 862         if (!*id)
 863                 return -ENOMEM;
 864
 865         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 866                         sizeof(struct nvme_id_ctrl));
 867         if (error)
 868                 kfree(*id);
 869         return error;
 870 }
 871
 872 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
 873                 struct nvme_ns_ids *ids)
 874 {
 875         struct nvme_command c = { };
 876         int status;
 877         void *data;
 878         int pos;
 879         int len;
 880
 881         c.identify.opcode = nvme_admin_identify;
 882         c.identify.nsid = cpu_to_le32(nsid);
 883         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
 884
 885         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
 886         if (!data)
 887                 return -ENOMEM;
 888
 889         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
 890                                       NVME_IDENTIFY_DATA_SIZE);
 891         if (status)
 892                 goto free_data;
 893
 894         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
 895                 struct nvme_ns_id_desc *cur = data + pos;
 896
 897                 if (cur->nidl == 0)
 898                         break;
 899
 900                 switch (cur->nidt) {
 901                 case NVME_NIDT_EUI64:
 902                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
 903                                 dev_warn(ctrl->device,
 904                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
 905                                          cur->nidl);
 906                                 goto free_data;
 907                         }
 908                         len = NVME_NIDT_EUI64_LEN;
 909                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
 910                         break;
 911                 case NVME_NIDT_NGUID:
 912                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
 913                                 dev_warn(ctrl->device,
 914                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
 915                                          cur->nidl);
 916                                 goto free_data;
 917                         }
 918                         len = NVME_NIDT_NGUID_LEN;
 919                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
 920                         break;
 921                 case NVME_NIDT_UUID:
 922                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
 923                                 dev_warn(ctrl->device,
 924                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
 925                                          cur->nidl);
 926                                 goto free_data;
 927                         }
 928                         len = NVME_NIDT_UUID_LEN;
 929                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
 930                         break;
 931                 default:
 932                         /* Skip unnkown types */
 933                         len = cur->nidl;
 934                         break;
 935                 }
 936
 937                 len += sizeof(*cur);
 938         }
 939 free_data:
 940         kfree(data);
 941         return status;
 942 }
 943
 944 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 945 {
 946         struct nvme_command c = { };
 947
 948         c.identify.opcode = nvme_admin_identify;
 949         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
 950         c.identify.nsid = cpu_to_le32(nsid);
 951         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 952 }
 953
 954 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
 955                 unsigned nsid)
 956 {
 957         struct nvme_id_ns *id;
 958         struct nvme_command c = { };
 959         int error;
 960
 961         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 962         c.identify.opcode = nvme_admin_identify;
 963         c.identify.nsid = cpu_to_le32(nsid);
 964         c.identify.cns = NVME_ID_CNS_NS;
 965
 966         id = kmalloc(sizeof(*id), GFP_KERNEL);
 967         if (!id)
 968                 return NULL;
 969
 970         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
 971         if (error) {
 972                 dev_warn(ctrl->device, "Identify namespace failed\n");
 973                 kfree(id);
 974                 return NULL;
 975         }
 976
 977         return id;
 978 }
 979
 980 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 981                       void *buffer, size_t buflen, u32 *result)
 982 {
 983         struct nvme_command c;
 984         union nvme_result res;
 985         int ret;
 986
 987         memset(&c, 0, sizeof(c));
 988         c.features.opcode = nvme_admin_set_features;
 989         c.features.fid = cpu_to_le32(fid);
 990         c.features.dword11 = cpu_to_le32(dword11);
 991
 992         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
 993                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
 994         if (ret >= 0 && result)
 995                 *result = le32_to_cpu(res.u32);
 996         return ret;
 997 }
 998
 999 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1000 {
1001         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1002         u32 result;
1003         int status, nr_io_queues;
1004
1005         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1006                         &result);
1007         if (status < 0)
1008                 return status;
1009
1010         /*
1011          * Degraded controllers might return an error when setting the queue
1012          * count.  We still want to be able to bring them online and offer
1013          * access to the admin queue, as that might be only way to fix them up.
1014          */
1015         if (status > 0) {
1016                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1017                 *count = 0;
1018         } else {
1019                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1020                 *count = min(*count, nr_io_queues);
1021         }
1022
1023         return 0;
1024 }
1025 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1026
1027 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1028 {
1029         struct nvme_user_io io;
1030         struct nvme_command c;
1031         unsigned length, meta_len;
1032         void __user *metadata;
1033
1034         if (copy_from_user(&io, uio, sizeof(io)))
1035                 return -EFAULT;
1036         if (io.flags)
1037                 return -EINVAL;
1038
1039         switch (io.opcode) {
1040         case nvme_cmd_write:
1041         case nvme_cmd_read:
1042         case nvme_cmd_compare:
1043                 break;
1044         default:
1045                 return -EINVAL;
1046         }
1047
1048         length = (io.nblocks + 1) << ns->lba_shift;
1049         meta_len = (io.nblocks + 1) * ns->ms;
1050         metadata = (void __user *)(uintptr_t)io.metadata;
1051
1052         if (ns->ext) {
1053                 length += meta_len;
1054                 meta_len = 0;
1055         } else if (meta_len) {
1056                 if ((io.metadata & 3) || !io.metadata)
1057                         return -EINVAL;
1058         }
1059
1060         memset(&c, 0, sizeof(c));
1061         c.rw.opcode = io.opcode;
1062         c.rw.flags = io.flags;
1063         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1064         c.rw.slba = cpu_to_le64(io.slba);
1065         c.rw.length = cpu_to_le16(io.nblocks);
1066         c.rw.control = cpu_to_le16(io.control);
1067         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1068         c.rw.reftag = cpu_to_le32(io.reftag);
1069         c.rw.apptag = cpu_to_le16(io.apptag);
1070         c.rw.appmask = cpu_to_le16(io.appmask);
1071
1072         return nvme_submit_user_cmd(ns->queue, &c,
1073                         (void __user *)(uintptr_t)io.addr, length,
1074                         metadata, meta_len, io.slba, NULL, 0);
1075 }
1076
1077 static u32 nvme_known_admin_effects(u8 opcode)
1078 {
1079         switch (opcode) {
1080         case nvme_admin_format_nvm:
1081                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1082                                         NVME_CMD_EFFECTS_CSE_MASK;
1083         case nvme_admin_sanitize_nvm:
1084                 return NVME_CMD_EFFECTS_CSE_MASK;
1085         default:
1086                 break;
1087         }
1088         return 0;
1089 }
1090
1091 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1092                                                                 u8 opcode)
1093 {
1094         u32 effects = 0;
1095
1096         if (ns) {
1097                 if (ctrl->effects)
1098                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1099                 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1100                         dev_warn(ctrl->device,
1101                                  "IO command:%02x has unhandled effects:%08x\n",
1102                                  opcode, effects);
1103                 return 0;
1104         }
1105
1106         if (ctrl->effects)
1107                 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1108         else
1109                 effects = nvme_known_admin_effects(opcode);
1110
1111         /*
1112          * For simplicity, IO to all namespaces is quiesced even if the command
1113          * effects say only one namespace is affected.
1114          */
1115         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1116                 nvme_start_freeze(ctrl);
1117                 nvme_wait_freeze(ctrl);
1118         }
1119         return effects;
1120 }
1121
1122 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1123 {
1124         struct nvme_ns *ns, *next;
1125         LIST_HEAD(rm_list);
1126
1127         mutex_lock(&ctrl->namespaces_mutex);
1128         list_for_each_entry(ns, &ctrl->namespaces, list) {
1129                 if (ns->disk && nvme_revalidate_disk(ns->disk)) {
1130                         list_move_tail(&ns->list, &rm_list);
1131                 }
1132         }
1133         mutex_unlock(&ctrl->namespaces_mutex);
1134
1135         list_for_each_entry_safe(ns, next, &rm_list, list)
1136                 nvme_ns_remove(ns);
1137 }
1138
1139 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1140 {
1141         /*
1142          * Revalidate LBA changes prior to unfreezing. This is necessary to
1143          * prevent memory corruption if a logical block size was changed by
1144          * this command.
1145          */
1146         if (effects & NVME_CMD_EFFECTS_LBCC)
1147                 nvme_update_formats(ctrl);
1148         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1149                 nvme_unfreeze(ctrl);
1150         if (effects & NVME_CMD_EFFECTS_CCC)
1151                 nvme_init_identify(ctrl);
1152         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1153                 nvme_queue_scan(ctrl);
1154 }
1155
1156 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1157                         struct nvme_passthru_cmd __user *ucmd)
1158 {
1159         struct nvme_passthru_cmd cmd;
1160         struct nvme_command c;
1161         unsigned timeout = 0;
1162         u32 effects;
1163         int status;
1164
1165         if (!capable(CAP_SYS_ADMIN))
1166                 return -EACCES;
1167         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1168                 return -EFAULT;
1169         if (cmd.flags)
1170                 return -EINVAL;
1171
1172         memset(&c, 0, sizeof(c));
1173         c.common.opcode = cmd.opcode;
1174         c.common.flags = cmd.flags;
1175         c.common.nsid = cpu_to_le32(cmd.nsid);
1176         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1177         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1178         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1179         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1180         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1181         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1182         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1183         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1184
1185         if (cmd.timeout_ms)
1186                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1187
1188         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1189         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1190                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1191                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata,
1192                         0, &cmd.result, timeout);
1193         nvme_passthru_end(ctrl, effects);
1194
1195         if (status >= 0) {
1196                 if (put_user(cmd.result, &ucmd->result))
1197                         return -EFAULT;
1198         }
1199
1200         return status;
1201 }
1202
1203 /*
1204  * Issue ioctl requests on the first available path.  Note that unlike normal
1205  * block layer requests we will not retry failed request on another controller.
1206  */
1207 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1208                 struct nvme_ns_head **head, int *srcu_idx)
1209 {
1210 #ifdef CONFIG_NVME_MULTIPATH
1211         if (disk->fops == &nvme_ns_head_ops) {
1212                 *head = disk->private_data;
1213                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1214                 return nvme_find_path(*head);
1215         }
1216 #endif
1217         *head = NULL;
1218         *srcu_idx = -1;
1219         return disk->private_data;
1220 }
1221
1222 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1223 {
1224         if (head)
1225                 srcu_read_unlock(&head->srcu, idx);
1226 }
1227
1228 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1229 {
1230         switch (cmd) {
1231         case NVME_IOCTL_ID:
1232                 force_successful_syscall_return();
1233                 return ns->head->ns_id;
1234         case NVME_IOCTL_ADMIN_CMD:
1235                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1236         case NVME_IOCTL_IO_CMD:
1237                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1238         case NVME_IOCTL_SUBMIT_IO:
1239                 return nvme_submit_io(ns, (void __user *)arg);
1240         default:
1241 #ifdef CONFIG_NVM
1242                 if (ns->ndev)
1243                         return nvme_nvm_ioctl(ns, cmd, arg);
1244 #endif
1245                 if (is_sed_ioctl(cmd))
1246                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1247                                          (void __user *) arg);
1248                 return -ENOTTY;
1249         }
1250 }
1251
1252 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1253                 unsigned int cmd, unsigned long arg)
1254 {
1255         struct nvme_ns_head *head = NULL;
1256         struct nvme_ns *ns;
1257         int srcu_idx, ret;
1258
1259         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1260         if (unlikely(!ns))
1261                 ret = -EWOULDBLOCK;
1262         else
1263                 ret = nvme_ns_ioctl(ns, cmd, arg);
1264         nvme_put_ns_from_disk(head, srcu_idx);
1265         return ret;
1266 }
1267
1268 static int nvme_open(struct block_device *bdev, fmode_t mode)
1269 {
1270         struct nvme_ns *ns = bdev->bd_disk->private_data;
1271
1272 #ifdef CONFIG_NVME_MULTIPATH
1273         /* should never be called due to GENHD_FL_HIDDEN */
1274         if (WARN_ON_ONCE(ns->head->disk))
1275                 goto fail;
1276 #endif
1277         if (!kref_get_unless_zero(&ns->kref))
1278                 goto fail;
1279         if (!try_module_get(ns->ctrl->ops->module))
1280                 goto fail_put_ns;
1281
1282         return 0;
1283
1284 fail_put_ns:
1285         nvme_put_ns(ns);
1286 fail:
1287         return -ENXIO;
1288 }
1289
1290 static void nvme_release(struct gendisk *disk, fmode_t mode)
1291 {
1292         struct nvme_ns *ns = disk->private_data;
1293
1294         module_put(ns->ctrl->ops->module);
1295         nvme_put_ns(ns);
1296 }
1297
1298 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1299 {
1300         /* some standard values */
1301         geo->heads = 1 << 6;
1302         geo->sectors = 1 << 5;
1303         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1304         return 0;
1305 }
1306
1307 #ifdef CONFIG_BLK_DEV_INTEGRITY
1308 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1309 {
1310         struct blk_integrity integrity;
1311
1312         memset(&integrity, 0, sizeof(integrity));
1313         switch (pi_type) {
1314         case NVME_NS_DPS_PI_TYPE3:
1315                 integrity.profile = &t10_pi_type3_crc;
1316                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1317                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1318                 break;
1319         case NVME_NS_DPS_PI_TYPE1:
1320         case NVME_NS_DPS_PI_TYPE2:
1321                 integrity.profile = &t10_pi_type1_crc;
1322                 integrity.tag_size = sizeof(u16);
1323                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1324                 break;
1325         default:
1326                 integrity.profile = NULL;
1327                 break;
1328         }
1329         integrity.tuple_size = ms;
1330         blk_integrity_register(disk, &integrity);
1331         blk_queue_max_integrity_segments(disk->queue, 1);
1332 }
1333 #else
1334 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1335 {
1336 }
1337 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1338
1339 static void nvme_set_chunk_size(struct nvme_ns *ns)
1340 {
1341         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1342         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1343 }
1344
1345 static void nvme_config_discard(struct nvme_ctrl *ctrl,
1346                 unsigned stream_alignment, struct request_queue *queue)
1347 {
1348         u32 size = queue_logical_block_size(queue);
1349
1350         if (stream_alignment)
1351                 size *= stream_alignment;
1352
1353         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1354                         NVME_DSM_MAX_RANGES);
1355
1356         queue->limits.discard_alignment = 0;
1357         queue->limits.discard_granularity = size;
1358
1359         blk_queue_max_discard_sectors(queue, UINT_MAX);
1360         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1361         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, queue);
1362
1363         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1364                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1365 }
1366
1367 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1368                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1369 {
1370         memset(ids, 0, sizeof(*ids));
1371
1372         if (ctrl->vs >= NVME_VS(1, 1, 0))
1373                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1374         if (ctrl->vs >= NVME_VS(1, 2, 0))
1375                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1376         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1377                  /* Don't treat error as fatal we potentially
1378                   * already have a NGUID or EUI-64
1379                   */
1380                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1381                         dev_warn(ctrl->device,
1382                                  "%s: Identify Descriptors failed\n", __func__);
1383         }
1384 }
1385
1386 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1387 {
1388         return !uuid_is_null(&ids->uuid) ||
1389                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1390                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1391 }
1392
1393 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1394 {
1395         return uuid_equal(&a->uuid, &b->uuid) &&
1396                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1397                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1398 }
1399
1400 static void nvme_update_disk_info(struct gendisk *disk,
1401                 struct nvme_ns *ns, struct nvme_id_ns *id)
1402 {
1403         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1404         unsigned short bs = 1 << ns->lba_shift;
1405         unsigned stream_alignment = 0;
1406
1407         if (ns->ctrl->nr_streams && ns->sws && ns->sgs)
1408                 stream_alignment = ns->sws * ns->sgs;
1409
1410         blk_mq_freeze_queue(disk->queue);
1411         blk_integrity_unregister(disk);
1412
1413         blk_queue_logical_block_size(disk->queue, bs);
1414         blk_queue_physical_block_size(disk->queue, bs);
1415         blk_queue_io_min(disk->queue, bs);
1416
1417         if (ns->ms && !ns->ext &&
1418             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1419                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1420         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1421                 capacity = 0;
1422         set_capacity(disk, capacity);
1423
1424         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1425                 nvme_config_discard(ns->ctrl, stream_alignment, disk->queue);
1426         blk_mq_unfreeze_queue(disk->queue);
1427 }
1428
1429 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1430 {
1431         struct nvme_ns *ns = disk->private_data;
1432
1433         /*
1434          * If identify namespace failed, use default 512 byte block size so
1435          * block layer can use before failing read/write for 0 capacity.
1436          */
1437         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1438         if (ns->lba_shift == 0)
1439                 ns->lba_shift = 9;
1440         ns->noiob = le16_to_cpu(id->noiob);
1441         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1442         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1443         /* the PI implementation requires metadata equal t10 pi tuple size */
1444         if (ns->ms == sizeof(struct t10_pi_tuple))
1445                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1446         else
1447                 ns->pi_type = 0;
1448
1449         if (ns->noiob)
1450                 nvme_set_chunk_size(ns);
1451         nvme_update_disk_info(disk, ns, id);
1452 #ifdef CONFIG_NVME_MULTIPATH
1453         if (ns->head->disk)
1454                 nvme_update_disk_info(ns->head->disk, ns, id);
1455 #endif
1456 }
1457
1458 static int nvme_revalidate_disk(struct gendisk *disk)
1459 {
1460         struct nvme_ns *ns = disk->private_data;
1461         struct nvme_ctrl *ctrl = ns->ctrl;
1462         struct nvme_id_ns *id;
1463         struct nvme_ns_ids ids;
1464         int ret = 0;
1465
1466         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1467                 set_capacity(disk, 0);
1468                 return -ENODEV;
1469         }
1470
1471         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1472         if (!id)
1473                 return -ENODEV;
1474
1475         if (id->ncap == 0) {
1476                 ret = -ENODEV;
1477                 goto out;
1478         }
1479
1480         __nvme_revalidate_disk(disk, id);
1481         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1482         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1483                 dev_err(ctrl->device,
1484                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1485                 ret = -ENODEV;
1486         }
1487
1488 out:
1489         kfree(id);
1490         return ret;
1491 }
1492
1493 static char nvme_pr_type(enum pr_type type)
1494 {
1495         switch (type) {
1496         case PR_WRITE_EXCLUSIVE:
1497                 return 1;
1498         case PR_EXCLUSIVE_ACCESS:
1499                 return 2;
1500         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1501                 return 3;
1502         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1503                 return 4;
1504         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1505                 return 5;
1506         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1507                 return 6;
1508         default:
1509                 return 0;
1510         }
1511 };
1512
1513 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1514                                 u64 key, u64 sa_key, u8 op)
1515 {
1516         struct nvme_ns_head *head = NULL;
1517         struct nvme_ns *ns;
1518         struct nvme_command c;
1519         int srcu_idx, ret;
1520         u8 data[16] = { 0, };
1521
1522         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1523         if (unlikely(!ns))
1524                 return -EWOULDBLOCK;
1525
1526         put_unaligned_le64(key, &data[0]);
1527         put_unaligned_le64(sa_key, &data[8]);
1528
1529         memset(&c, 0, sizeof(c));
1530         c.common.opcode = op;
1531         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1532         c.common.cdw10[0] = cpu_to_le32(cdw10);
1533
1534         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1535         nvme_put_ns_from_disk(head, srcu_idx);
1536         return ret;
1537 }
1538
1539 static int nvme_pr_register(struct block_device *bdev, u64 old,
1540                 u64 new, unsigned flags)
1541 {
1542         u32 cdw10;
1543
1544         if (flags & ~PR_FL_IGNORE_KEY)
1545                 return -EOPNOTSUPP;
1546
1547         cdw10 = old ? 2 : 0;
1548         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1549         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1550         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1551 }
1552
1553 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1554                 enum pr_type type, unsigned flags)
1555 {
1556         u32 cdw10;
1557
1558         if (flags & ~PR_FL_IGNORE_KEY)
1559                 return -EOPNOTSUPP;
1560
1561         cdw10 = nvme_pr_type(type) << 8;
1562         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1563         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1564 }
1565
1566 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1567                 enum pr_type type, bool abort)
1568 {
1569         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1570         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1571 }
1572
1573 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1574 {
1575         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1576         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1577 }
1578
1579 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1580 {
1581         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1582         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1583 }
1584
1585 static const struct pr_ops nvme_pr_ops = {
1586         .pr_register    = nvme_pr_register,
1587         .pr_reserve     = nvme_pr_reserve,
1588         .pr_release     = nvme_pr_release,
1589         .pr_preempt     = nvme_pr_preempt,
1590         .pr_clear       = nvme_pr_clear,
1591 };
1592
1593 #ifdef CONFIG_BLK_SED_OPAL
1594 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1595                 bool send)
1596 {
1597         struct nvme_ctrl *ctrl = data;
1598         struct nvme_command cmd;
1599
1600         memset(&cmd, 0, sizeof(cmd));
1601         if (send)
1602                 cmd.common.opcode = nvme_admin_security_send;
1603         else
1604                 cmd.common.opcode = nvme_admin_security_recv;
1605         cmd.common.nsid = 0;
1606         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1607         cmd.common.cdw10[1] = cpu_to_le32(len);
1608
1609         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1610                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1611 }
1612 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1613 #endif /* CONFIG_BLK_SED_OPAL */
1614
1615 static const struct block_device_operations nvme_fops = {
1616         .owner          = THIS_MODULE,
1617         .ioctl          = nvme_ioctl,
1618         .compat_ioctl   = nvme_ioctl,
1619         .open           = nvme_open,
1620         .release        = nvme_release,
1621         .getgeo         = nvme_getgeo,
1622         .revalidate_disk= nvme_revalidate_disk,
1623         .pr_ops         = &nvme_pr_ops,
1624 };
1625
1626 #ifdef CONFIG_NVME_MULTIPATH
1627 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1628 {
1629         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1630
1631         if (!kref_get_unless_zero(&head->ref))
1632                 return -ENXIO;
1633         return 0;
1634 }
1635
1636 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1637 {
1638         nvme_put_ns_head(disk->private_data);
1639 }
1640
1641 const struct block_device_operations nvme_ns_head_ops = {
1642         .owner          = THIS_MODULE,
1643         .open           = nvme_ns_head_open,
1644         .release        = nvme_ns_head_release,
1645         .ioctl          = nvme_ioctl,
1646         .compat_ioctl   = nvme_ioctl,
1647         .getgeo         = nvme_getgeo,
1648         .pr_ops         = &nvme_pr_ops,
1649 };
1650 #endif /* CONFIG_NVME_MULTIPATH */
1651
1652 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1653 {
1654         unsigned long timeout =
1655                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1656         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1657         int ret;
1658
1659         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1660                 if (csts == ~0)
1661                         return -ENODEV;
1662                 if ((csts & NVME_CSTS_RDY) == bit)
1663                         break;
1664
1665                 msleep(100);
1666                 if (fatal_signal_pending(current))
1667                         return -EINTR;
1668                 if (time_after(jiffies, timeout)) {
1669                         dev_err(ctrl->device,
1670                                 "Device not ready; aborting %s\n", enabled ?
1671                                                 "initialisation" : "reset");
1672                         return -ENODEV;
1673                 }
1674         }
1675
1676         return ret;
1677 }
1678
1679 /*
1680  * If the device has been passed off to us in an enabled state, just clear
1681  * the enabled bit.  The spec says we should set the 'shutdown notification
1682  * bits', but doing so may cause the device to complete commands to the
1683  * admin queue ... and we don't know what memory that might be pointing at!
1684  */
1685 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1686 {
1687         int ret;
1688
1689         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1690         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1691
1692         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1693         if (ret)
1694                 return ret;
1695
1696         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1697                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1698
1699         return nvme_wait_ready(ctrl, cap, false);
1700 }
1701 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1702
1703 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1704 {
1705         /*
1706          * Default to a 4K page size, with the intention to update this
1707          * path in the future to accomodate architectures with differing
1708          * kernel and IO page sizes.
1709          */
1710         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1711         int ret;
1712
1713         if (page_shift < dev_page_min) {
1714                 dev_err(ctrl->device,
1715                         "Minimum device page size %u too large for host (%u)\n",
1716                         1 << dev_page_min, 1 << page_shift);
1717                 return -ENODEV;
1718         }
1719
1720         ctrl->page_size = 1 << page_shift;
1721
1722         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1723         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1724         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1725         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1726         ctrl->ctrl_config |= NVME_CC_ENABLE;
1727
1728         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1729         if (ret)
1730                 return ret;
1731         return nvme_wait_ready(ctrl, cap, true);
1732 }
1733 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1734
1735 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1736 {
1737         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1738         u32 csts;
1739         int ret;
1740
1741         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1742         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1743
1744         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1745         if (ret)
1746                 return ret;
1747
1748         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1749                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1750                         break;
1751
1752                 msleep(100);
1753                 if (fatal_signal_pending(current))
1754                         return -EINTR;
1755                 if (time_after(jiffies, timeout)) {
1756                         dev_err(ctrl->device,
1757                                 "Device shutdown incomplete; abort shutdown\n");
1758                         return -ENODEV;
1759                 }
1760         }
1761
1762         return ret;
1763 }
1764 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1765
1766 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1767                 struct request_queue *q)
1768 {
1769         bool vwc = false;
1770
1771         if (ctrl->max_hw_sectors) {
1772                 u32 max_segments =
1773                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1774
1775                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1776                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1777         }
1778         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1779             is_power_of_2(ctrl->max_hw_sectors))
1780                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1781         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1782         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1783                 vwc = true;
1784         blk_queue_write_cache(q, vwc, vwc);
1785 }
1786
1787 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1788 {
1789         __le64 ts;
1790         int ret;
1791
1792         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1793                 return 0;
1794
1795         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1796         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1797                         NULL);
1798         if (ret)
1799                 dev_warn_once(ctrl->device,
1800                         "could not set timestamp (%d)\n", ret);
1801         return ret;
1802 }
1803
1804 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1805 {
1806         /*
1807          * APST (Autonomous Power State Transition) lets us program a
1808          * table of power state transitions that the controller will
1809          * perform automatically.  We configure it with a simple
1810          * heuristic: we are willing to spend at most 2% of the time
1811          * transitioning between power states.  Therefore, when running
1812          * in any given state, we will enter the next lower-power
1813          * non-operational state after waiting 50 * (enlat + exlat)
1814          * microseconds, as long as that state's exit latency is under
1815          * the requested maximum latency.
1816          *
1817          * We will not autonomously enter any non-operational state for
1818          * which the total latency exceeds ps_max_latency_us.  Users
1819          * can set ps_max_latency_us to zero to turn off APST.
1820          */
1821
1822         unsigned apste;
1823         struct nvme_feat_auto_pst *table;
1824         u64 max_lat_us = 0;
1825         int max_ps = -1;
1826         int ret;
1827
1828         /*
1829          * If APST isn't supported or if we haven't been initialized yet,
1830          * then don't do anything.
1831          */
1832         if (!ctrl->apsta)
1833                 return 0;
1834
1835         if (ctrl->npss > 31) {
1836                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1837                 return 0;
1838         }
1839
1840         table = kzalloc(sizeof(*table), GFP_KERNEL);
1841         if (!table)
1842                 return 0;
1843
1844         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1845                 /* Turn off APST. */
1846                 apste = 0;
1847                 dev_dbg(ctrl->device, "APST disabled\n");
1848         } else {
1849                 __le64 target = cpu_to_le64(0);
1850                 int state;
1851
1852                 /*
1853                  * Walk through all states from lowest- to highest-power.
1854                  * According to the spec, lower-numbered states use more
1855                  * power.  NPSS, despite the name, is the index of the
1856                  * lowest-power state, not the number of states.
1857                  */
1858                 for (state = (int)ctrl->npss; state >= 0; state--) {
1859                         u64 total_latency_us, exit_latency_us, transition_ms;
1860
1861                         if (target)
1862                                 table->entries[state] = target;
1863
1864                         /*
1865                          * Don't allow transitions to the deepest state
1866                          * if it's quirked off.
1867                          */
1868                         if (state == ctrl->npss &&
1869                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1870                                 continue;
1871
1872                         /*
1873                          * Is this state a useful non-operational state for
1874                          * higher-power states to autonomously transition to?
1875                          */
1876                         if (!(ctrl->psd[state].flags &
1877                               NVME_PS_FLAGS_NON_OP_STATE))
1878                                 continue;
1879
1880                         exit_latency_us =
1881                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1882                         if (exit_latency_us > ctrl->ps_max_latency_us)
1883                                 continue;
1884
1885                         total_latency_us =
1886                                 exit_latency_us +
1887                                 le32_to_cpu(ctrl->psd[state].entry_lat);
1888
1889                         /*
1890                          * This state is good.  Use it as the APST idle
1891                          * target for higher power states.
1892                          */
1893                         transition_ms = total_latency_us + 19;
1894                         do_div(transition_ms, 20);
1895                         if (transition_ms > (1 << 24) - 1)
1896                                 transition_ms = (1 << 24) - 1;
1897
1898                         target = cpu_to_le64((state << 3) |
1899                                              (transition_ms << 8));
1900
1901                         if (max_ps == -1)
1902                                 max_ps = state;
1903
1904                         if (total_latency_us > max_lat_us)
1905                                 max_lat_us = total_latency_us;
1906                 }
1907
1908                 apste = 1;
1909
1910                 if (max_ps == -1) {
1911                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1912                 } else {
1913                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1914                                 max_ps, max_lat_us, (int)sizeof(*table), table);
1915                 }
1916         }
1917
1918         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1919                                 table, sizeof(*table), NULL);
1920         if (ret)
1921                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1922
1923         kfree(table);
1924         return ret;
1925 }
1926
1927 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1928 {
1929         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1930         u64 latency;
1931
1932         switch (val) {
1933         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1934         case PM_QOS_LATENCY_ANY:
1935                 latency = U64_MAX;
1936                 break;
1937
1938         default:
1939                 latency = val;
1940         }
1941
1942         if (ctrl->ps_max_latency_us != latency) {
1943                 ctrl->ps_max_latency_us = latency;
1944                 nvme_configure_apst(ctrl);
1945         }
1946 }
1947
1948 struct nvme_core_quirk_entry {
1949         /*
1950          * NVMe model and firmware strings are padded with spaces.  For
1951          * simplicity, strings in the quirk table are padded with NULLs
1952          * instead.
1953          */
1954         u16 vid;
1955         const char *mn;
1956         const char *fr;
1957         unsigned long quirks;
1958 };
1959
1960 static const struct nvme_core_quirk_entry core_quirks[] = {
1961         {
1962                 /*
1963                  * This Toshiba device seems to die using any APST states.  See:
1964                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1965                  */
1966                 .vid = 0x1179,
1967                 .mn = "THNSF5256GPUK TOSHIBA",
1968                 .quirks = NVME_QUIRK_NO_APST,
1969         }
1970 };
1971
1972 /* match is null-terminated but idstr is space-padded. */
1973 static bool string_matches(const char *idstr, const char *match, size_t len)
1974 {
1975         size_t matchlen;
1976
1977         if (!match)
1978                 return true;
1979
1980         matchlen = strlen(match);
1981         WARN_ON_ONCE(matchlen > len);
1982
1983         if (memcmp(idstr, match, matchlen))
1984                 return false;
1985
1986         for (; matchlen < len; matchlen++)
1987                 if (idstr[matchlen] != ' ')
1988                         return false;
1989
1990         return true;
1991 }
1992
1993 static bool quirk_matches(const struct nvme_id_ctrl *id,
1994                           const struct nvme_core_quirk_entry *q)
1995 {
1996         return q->vid == le16_to_cpu(id->vid) &&
1997                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1998                 string_matches(id->fr, q->fr, sizeof(id->fr));
1999 }
2000
2001 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2002                 struct nvme_id_ctrl *id)
2003 {
2004         size_t nqnlen;
2005         int off;
2006
2007         nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2008         if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2009                 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2010                 return;
2011         }
2012
2013         if (ctrl->vs >= NVME_VS(1, 2, 1))
2014                 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2015
2016         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2017         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2018                         "nqn.2014.08.org.nvmexpress:%4x%4x",
2019                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2020         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2021         off += sizeof(id->sn);
2022         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2023         off += sizeof(id->mn);
2024         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2025 }
2026
2027 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2028 {
2029         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2030         kfree(subsys);
2031 }
2032
2033 static void nvme_release_subsystem(struct device *dev)
2034 {
2035         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2036 }
2037
2038 static void nvme_destroy_subsystem(struct kref *ref)
2039 {
2040         struct nvme_subsystem *subsys =
2041                         container_of(ref, struct nvme_subsystem, ref);
2042
2043         mutex_lock(&nvme_subsystems_lock);
2044         list_del(&subsys->entry);
2045         mutex_unlock(&nvme_subsystems_lock);
2046
2047         ida_destroy(&subsys->ns_ida);
2048         device_del(&subsys->dev);
2049         put_device(&subsys->dev);
2050 }
2051
2052 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2053 {
2054         kref_put(&subsys->ref, nvme_destroy_subsystem);
2055 }
2056
2057 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2058 {
2059         struct nvme_subsystem *subsys;
2060
2061         lockdep_assert_held(&nvme_subsystems_lock);
2062
2063         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2064                 if (strcmp(subsys->subnqn, subsysnqn))
2065                         continue;
2066                 if (!kref_get_unless_zero(&subsys->ref))
2067                         continue;
2068                 return subsys;
2069         }
2070
2071         return NULL;
2072 }
2073
2074 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2075         struct device_attribute subsys_attr_##_name = \
2076                 __ATTR(_name, _mode, _show, NULL)
2077
2078 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2079                                     struct device_attribute *attr,
2080                                     char *buf)
2081 {
2082         struct nvme_subsystem *subsys =
2083                 container_of(dev, struct nvme_subsystem, dev);
2084
2085         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2086 }
2087 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2088
2089 #define nvme_subsys_show_str_function(field)                            \
2090 static ssize_t subsys_##field##_show(struct device *dev,                \
2091                             struct device_attribute *attr, char *buf)   \
2092 {                                                                       \
2093         struct nvme_subsystem *subsys =                                 \
2094                 container_of(dev, struct nvme_subsystem, dev);          \
2095         return sprintf(buf, "%.*s\n",                                   \
2096                        (int)sizeof(subsys->field), subsys->field);      \
2097 }                                                                       \
2098 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2099
2100 nvme_subsys_show_str_function(model);
2101 nvme_subsys_show_str_function(serial);
2102 nvme_subsys_show_str_function(firmware_rev);
2103
2104 static struct attribute *nvme_subsys_attrs[] = {
2105         &subsys_attr_model.attr,
2106         &subsys_attr_serial.attr,
2107         &subsys_attr_firmware_rev.attr,
2108         &subsys_attr_subsysnqn.attr,
2109         NULL,
2110 };
2111
2112 static struct attribute_group nvme_subsys_attrs_group = {
2113         .attrs = nvme_subsys_attrs,
2114 };
2115
2116 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2117         &nvme_subsys_attrs_group,
2118         NULL,
2119 };
2120
2121 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2122 {
2123         int count = 0;
2124         struct nvme_ctrl *ctrl;
2125
2126         mutex_lock(&subsys->lock);
2127         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2128                 if (ctrl->state != NVME_CTRL_DELETING &&
2129                     ctrl->state != NVME_CTRL_DEAD)
2130                         count++;
2131         }
2132         mutex_unlock(&subsys->lock);
2133
2134         return count;
2135 }
2136
2137 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2138 {
2139         struct nvme_subsystem *subsys, *found;
2140         int ret;
2141
2142         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2143         if (!subsys)
2144                 return -ENOMEM;
2145         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2146         if (ret < 0) {
2147                 kfree(subsys);
2148                 return ret;
2149         }
2150         subsys->instance = ret;
2151         mutex_init(&subsys->lock);
2152         kref_init(&subsys->ref);
2153         INIT_LIST_HEAD(&subsys->ctrls);
2154         INIT_LIST_HEAD(&subsys->nsheads);
2155         nvme_init_subnqn(subsys, ctrl, id);
2156         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2157         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2158         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2159         subsys->vendor_id = le16_to_cpu(id->vid);
2160         subsys->cmic = id->cmic;
2161
2162         subsys->dev.class = nvme_subsys_class;
2163         subsys->dev.release = nvme_release_subsystem;
2164         subsys->dev.groups = nvme_subsys_attrs_groups;
2165         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2166         device_initialize(&subsys->dev);
2167
2168         mutex_lock(&nvme_subsystems_lock);
2169         found = __nvme_find_get_subsystem(subsys->subnqn);
2170         if (found) {
2171                 /*
2172                  * Verify that the subsystem actually supports multiple
2173                  * controllers, else bail out.
2174                  */
2175                 if (nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2176                         dev_err(ctrl->device,
2177                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2178                                 found->subnqn);
2179                         nvme_put_subsystem(found);
2180                         ret = -EINVAL;
2181                         goto out_unlock;
2182                 }
2183
2184                 __nvme_release_subsystem(subsys);
2185                 subsys = found;
2186         } else {
2187                 ret = device_add(&subsys->dev);
2188                 if (ret) {
2189                         dev_err(ctrl->device,
2190                                 "failed to register subsystem device.\n");
2191                         goto out_unlock;
2192                 }
2193                 ida_init(&subsys->ns_ida);
2194                 list_add_tail(&subsys->entry, &nvme_subsystems);
2195         }
2196
2197         ctrl->subsys = subsys;
2198         mutex_unlock(&nvme_subsystems_lock);
2199
2200         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2201                         dev_name(ctrl->device))) {
2202                 dev_err(ctrl->device,
2203                         "failed to create sysfs link from subsystem.\n");
2204                 /* the transport driver will eventually put the subsystem */
2205                 return -EINVAL;
2206         }
2207
2208         mutex_lock(&subsys->lock);
2209         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2210         mutex_unlock(&subsys->lock);
2211
2212         return 0;
2213
2214 out_unlock:
2215         mutex_unlock(&nvme_subsystems_lock);
2216         put_device(&subsys->dev);
2217         return ret;
2218 }
2219
2220 static int nvme_get_log(struct nvme_ctrl *ctrl, u8 log_page, void *log,
2221                         size_t size)
2222 {
2223         struct nvme_command c = { };
2224
2225         c.common.opcode = nvme_admin_get_log_page;
2226         c.common.nsid = cpu_to_le32(NVME_NSID_ALL);
2227         c.common.cdw10[0] = nvme_get_log_dw10(log_page, size);
2228
2229         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2230 }
2231
2232 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2233 {
2234         int ret;
2235
2236         if (!ctrl->effects)
2237                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2238
2239         if (!ctrl->effects)
2240                 return 0;
2241
2242         ret = nvme_get_log(ctrl, NVME_LOG_CMD_EFFECTS, ctrl->effects,
2243                                         sizeof(*ctrl->effects));
2244         if (ret) {
2245                 kfree(ctrl->effects);
2246                 ctrl->effects = NULL;
2247         }
2248         return ret;
2249 }
2250
2251 /*
2252  * Initialize the cached copies of the Identify data and various controller
2253  * register in our nvme_ctrl structure.  This should be called as soon as
2254  * the admin queue is fully up and running.
2255  */
2256 int nvme_init_identify(struct nvme_ctrl *ctrl)
2257 {
2258         struct nvme_id_ctrl *id;
2259         u64 cap;
2260         int ret, page_shift;
2261         u32 max_hw_sectors;
2262         bool prev_apst_enabled;
2263
2264         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2265         if (ret) {
2266                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2267                 return ret;
2268         }
2269
2270         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2271         if (ret) {
2272                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2273                 return ret;
2274         }
2275         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2276
2277         if (ctrl->vs >= NVME_VS(1, 1, 0))
2278                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2279
2280         ret = nvme_identify_ctrl(ctrl, &id);
2281         if (ret) {
2282                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2283                 return -EIO;
2284         }
2285
2286         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2287                 ret = nvme_get_effects_log(ctrl);
2288                 if (ret < 0)
2289                         return ret;
2290         }
2291
2292         if (!ctrl->identified) {
2293                 int i;
2294
2295                 ret = nvme_init_subsystem(ctrl, id);
2296                 if (ret)
2297                         goto out_free;
2298
2299                 /*
2300                  * Check for quirks.  Quirk can depend on firmware version,
2301                  * so, in principle, the set of quirks present can change
2302                  * across a reset.  As a possible future enhancement, we
2303                  * could re-scan for quirks every time we reinitialize
2304                  * the device, but we'd have to make sure that the driver
2305                  * behaves intelligently if the quirks change.
2306                  */
2307                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2308                         if (quirk_matches(id, &core_quirks[i]))
2309                                 ctrl->quirks |= core_quirks[i].quirks;
2310                 }
2311         }
2312
2313         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2314                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2315                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2316         }
2317
2318         ctrl->oacs = le16_to_cpu(id->oacs);
2319         ctrl->oncs = le16_to_cpup(&id->oncs);
2320         atomic_set(&ctrl->abort_limit, id->acl + 1);
2321         ctrl->vwc = id->vwc;
2322         ctrl->cntlid = le16_to_cpup(&id->cntlid);
2323         if (id->mdts)
2324                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2325         else
2326                 max_hw_sectors = UINT_MAX;
2327         ctrl->max_hw_sectors =
2328                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2329
2330         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2331         ctrl->sgls = le32_to_cpu(id->sgls);
2332         ctrl->kas = le16_to_cpu(id->kas);
2333
2334         if (id->rtd3e) {
2335                 /* us -> s */
2336                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2337
2338                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2339                                                  shutdown_timeout, 60);
2340
2341                 if (ctrl->shutdown_timeout != shutdown_timeout)
2342                         dev_info(ctrl->device,
2343                                  "Shutdown timeout set to %u seconds\n",
2344                                  ctrl->shutdown_timeout);
2345         } else
2346                 ctrl->shutdown_timeout = shutdown_timeout;
2347
2348         ctrl->npss = id->npss;
2349         ctrl->apsta = id->apsta;
2350         prev_apst_enabled = ctrl->apst_enabled;
2351         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2352                 if (force_apst && id->apsta) {
2353                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2354                         ctrl->apst_enabled = true;
2355                 } else {
2356                         ctrl->apst_enabled = false;
2357                 }
2358         } else {
2359                 ctrl->apst_enabled = id->apsta;
2360         }
2361         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2362
2363         if (ctrl->ops->flags & NVME_F_FABRICS) {
2364                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2365                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2366                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2367                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2368
2369                 /*
2370                  * In fabrics we need to verify the cntlid matches the
2371                  * admin connect
2372                  */
2373                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2374                         ret = -EINVAL;
2375                         goto out_free;
2376                 }
2377
2378                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2379                         dev_err(ctrl->device,
2380                                 "keep-alive support is mandatory for fabrics\n");
2381                         ret = -EINVAL;
2382                         goto out_free;
2383                 }
2384         } else {
2385                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2386                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2387                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2388                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2389                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2390         }
2391
2392         kfree(id);
2393
2394         if (ctrl->apst_enabled && !prev_apst_enabled)
2395                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2396         else if (!ctrl->apst_enabled && prev_apst_enabled)
2397                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2398
2399         ret = nvme_configure_apst(ctrl);
2400         if (ret < 0)
2401                 return ret;
2402
2403         ret = nvme_configure_timestamp(ctrl);
2404         if (ret < 0)
2405                 return ret;
2406
2407         ret = nvme_configure_directives(ctrl);
2408         if (ret < 0)
2409                 return ret;
2410
2411         ctrl->identified = true;
2412
2413         return 0;
2414
2415 out_free:
2416         kfree(id);
2417         return ret;
2418 }
2419 EXPORT_SYMBOL_GPL(nvme_init_identify);
2420
2421 static int nvme_dev_open(struct inode *inode, struct file *file)
2422 {
2423         struct nvme_ctrl *ctrl =
2424                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2425
2426         switch (ctrl->state) {
2427         case NVME_CTRL_LIVE:
2428         case NVME_CTRL_ADMIN_ONLY:
2429                 break;
2430         default:
2431                 return -EWOULDBLOCK;
2432         }
2433
2434         file->private_data = ctrl;
2435         return 0;
2436 }
2437
2438 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2439 {
2440         struct nvme_ns *ns;
2441         int ret;
2442
2443         mutex_lock(&ctrl->namespaces_mutex);
2444         if (list_empty(&ctrl->namespaces)) {
2445                 ret = -ENOTTY;
2446                 goto out_unlock;
2447         }
2448
2449         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2450         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2451                 dev_warn(ctrl->device,
2452                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2453                 ret = -EINVAL;
2454                 goto out_unlock;
2455         }
2456
2457         dev_warn(ctrl->device,
2458                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2459         kref_get(&ns->kref);
2460         mutex_unlock(&ctrl->namespaces_mutex);
2461
2462         ret = nvme_user_cmd(ctrl, ns, argp);
2463         nvme_put_ns(ns);
2464         return ret;
2465
2466 out_unlock:
2467         mutex_unlock(&ctrl->namespaces_mutex);
2468         return ret;
2469 }
2470
2471 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2472                 unsigned long arg)
2473 {
2474         struct nvme_ctrl *ctrl = file->private_data;
2475         void __user *argp = (void __user *)arg;
2476
2477         switch (cmd) {
2478         case NVME_IOCTL_ADMIN_CMD:
2479                 return nvme_user_cmd(ctrl, NULL, argp);
2480         case NVME_IOCTL_IO_CMD:
2481                 return nvme_dev_user_cmd(ctrl, argp);
2482         case NVME_IOCTL_RESET:
2483                 dev_warn(ctrl->device, "resetting controller\n");
2484                 return nvme_reset_ctrl_sync(ctrl);
2485         case NVME_IOCTL_SUBSYS_RESET:
2486                 return nvme_reset_subsystem(ctrl);
2487         case NVME_IOCTL_RESCAN:
2488                 nvme_queue_scan(ctrl);
2489                 return 0;
2490         default:
2491                 return -ENOTTY;
2492         }
2493 }
2494
2495 static const struct file_operations nvme_dev_fops = {
2496         .owner          = THIS_MODULE,
2497         .open           = nvme_dev_open,
2498         .unlocked_ioctl = nvme_dev_ioctl,
2499         .compat_ioctl   = nvme_dev_ioctl,
2500 };
2501
2502 static ssize_t nvme_sysfs_reset(struct device *dev,
2503                                 struct device_attribute *attr, const char *buf,
2504                                 size_t count)
2505 {
2506         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2507         int ret;
2508
2509         ret = nvme_reset_ctrl_sync(ctrl);
2510         if (ret < 0)
2511                 return ret;
2512         return count;
2513 }
2514 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2515
2516 static ssize_t nvme_sysfs_rescan(struct device *dev,
2517                                 struct device_attribute *attr, const char *buf,
2518                                 size_t count)
2519 {
2520         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2521
2522         nvme_queue_scan(ctrl);
2523         return count;
2524 }
2525 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2526
2527 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2528 {
2529         struct gendisk *disk = dev_to_disk(dev);
2530
2531         if (disk->fops == &nvme_fops)
2532                 return nvme_get_ns_from_dev(dev)->head;
2533         else
2534                 return disk->private_data;
2535 }
2536
2537 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2538                 char *buf)
2539 {
2540         struct nvme_ns_head *head = dev_to_ns_head(dev);
2541         struct nvme_ns_ids *ids = &head->ids;
2542         struct nvme_subsystem *subsys = head->subsys;
2543         int serial_len = sizeof(subsys->serial);
2544         int model_len = sizeof(subsys->model);
2545
2546         if (!uuid_is_null(&ids->uuid))
2547                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2548
2549         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2550                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2551
2552         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2553                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2554
2555         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2556                                   subsys->serial[serial_len - 1] == '\0'))
2557                 serial_len--;
2558         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2559                                  subsys->model[model_len - 1] == '\0'))
2560                 model_len--;
2561
2562         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2563                 serial_len, subsys->serial, model_len, subsys->model,
2564                 head->ns_id);
2565 }
2566 static DEVICE_ATTR_RO(wwid);
2567
2568 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2569                 char *buf)
2570 {
2571         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2572 }
2573 static DEVICE_ATTR_RO(nguid);
2574
2575 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2576                 char *buf)
2577 {
2578         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2579
2580         /* For backward compatibility expose the NGUID to userspace if
2581          * we have no UUID set
2582          */
2583         if (uuid_is_null(&ids->uuid)) {
2584                 printk_ratelimited(KERN_WARNING
2585                                    "No UUID available providing old NGUID\n");
2586                 return sprintf(buf, "%pU\n", ids->nguid);
2587         }
2588         return sprintf(buf, "%pU\n", &ids->uuid);
2589 }
2590 static DEVICE_ATTR_RO(uuid);
2591
2592 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2593                 char *buf)
2594 {
2595         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2596 }
2597 static DEVICE_ATTR_RO(eui);
2598
2599 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2600                 char *buf)
2601 {
2602         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2603 }
2604 static DEVICE_ATTR_RO(nsid);
2605
2606 static struct attribute *nvme_ns_id_attrs[] = {
2607         &dev_attr_wwid.attr,
2608         &dev_attr_uuid.attr,
2609         &dev_attr_nguid.attr,
2610         &dev_attr_eui.attr,
2611         &dev_attr_nsid.attr,
2612         NULL,
2613 };
2614
2615 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2616                 struct attribute *a, int n)
2617 {
2618         struct device *dev = container_of(kobj, struct device, kobj);
2619         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2620
2621         if (a == &dev_attr_uuid.attr) {
2622                 if (uuid_is_null(&ids->uuid) &&
2623                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2624                         return 0;
2625         }
2626         if (a == &dev_attr_nguid.attr) {
2627                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2628                         return 0;
2629         }
2630         if (a == &dev_attr_eui.attr) {
2631                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2632                         return 0;
2633         }
2634         return a->mode;
2635 }
2636
2637 const struct attribute_group nvme_ns_id_attr_group = {
2638         .attrs          = nvme_ns_id_attrs,
2639         .is_visible     = nvme_ns_id_attrs_are_visible,
2640 };
2641
2642 #define nvme_show_str_function(field)                                           \
2643 static ssize_t  field##_show(struct device *dev,                                \
2644                             struct device_attribute *attr, char *buf)           \
2645 {                                                                               \
2646         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2647         return sprintf(buf, "%.*s\n",                                           \
2648                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2649 }                                                                               \
2650 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2651
2652 nvme_show_str_function(model);
2653 nvme_show_str_function(serial);
2654 nvme_show_str_function(firmware_rev);
2655
2656 #define nvme_show_int_function(field)                                           \
2657 static ssize_t  field##_show(struct device *dev,                                \
2658                             struct device_attribute *attr, char *buf)           \
2659 {                                                                               \
2660         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2661         return sprintf(buf, "%d\n", ctrl->field);       \
2662 }                                                                               \
2663 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2664
2665 nvme_show_int_function(cntlid);
2666
2667 static ssize_t nvme_sysfs_delete(struct device *dev,
2668                                 struct device_attribute *attr, const char *buf,
2669                                 size_t count)
2670 {
2671         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2672
2673         if (device_remove_file_self(dev, attr))
2674                 nvme_delete_ctrl_sync(ctrl);
2675         return count;
2676 }
2677 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2678
2679 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2680                                          struct device_attribute *attr,
2681                                          char *buf)
2682 {
2683         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2684
2685         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2686 }
2687 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2688
2689 static ssize_t nvme_sysfs_show_state(struct device *dev,
2690                                      struct device_attribute *attr,
2691                                      char *buf)
2692 {
2693         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2694         static const char *const state_name[] = {
2695                 [NVME_CTRL_NEW]         = "new",
2696                 [NVME_CTRL_LIVE]        = "live",
2697                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2698                 [NVME_CTRL_RESETTING]   = "resetting",
2699                 [NVME_CTRL_CONNECTING]  = "connecting",
2700                 [NVME_CTRL_DELETING]    = "deleting",
2701                 [NVME_CTRL_DEAD]        = "dead",
2702         };
2703
2704         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2705             state_name[ctrl->state])
2706                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2707
2708         return sprintf(buf, "unknown state\n");
2709 }
2710
2711 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2712
2713 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2714                                          struct device_attribute *attr,
2715                                          char *buf)
2716 {
2717         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2718
2719         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2720 }
2721 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2722
2723 static ssize_t nvme_sysfs_show_address(struct device *dev,
2724                                          struct device_attribute *attr,
2725                                          char *buf)
2726 {
2727         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2728
2729         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2730 }
2731 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2732
2733 static struct attribute *nvme_dev_attrs[] = {
2734         &dev_attr_reset_controller.attr,
2735         &dev_attr_rescan_controller.attr,
2736         &dev_attr_model.attr,
2737         &dev_attr_serial.attr,
2738         &dev_attr_firmware_rev.attr,
2739         &dev_attr_cntlid.attr,
2740         &dev_attr_delete_controller.attr,
2741         &dev_attr_transport.attr,
2742         &dev_attr_subsysnqn.attr,
2743         &dev_attr_address.attr,
2744         &dev_attr_state.attr,
2745         NULL
2746 };
2747
2748 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2749                 struct attribute *a, int n)
2750 {
2751         struct device *dev = container_of(kobj, struct device, kobj);
2752         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2753
2754         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2755                 return 0;
2756         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2757                 return 0;
2758
2759         return a->mode;
2760 }
2761
2762 static struct attribute_group nvme_dev_attrs_group = {
2763         .attrs          = nvme_dev_attrs,
2764         .is_visible     = nvme_dev_attrs_are_visible,
2765 };
2766
2767 static const struct attribute_group *nvme_dev_attr_groups[] = {
2768         &nvme_dev_attrs_group,
2769         NULL,
2770 };
2771
2772 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2773                 unsigned nsid)
2774 {
2775         struct nvme_ns_head *h;
2776
2777         lockdep_assert_held(&subsys->lock);
2778
2779         list_for_each_entry(h, &subsys->nsheads, entry) {
2780                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2781                         return h;
2782         }
2783
2784         return NULL;
2785 }
2786
2787 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2788                 struct nvme_ns_head *new)
2789 {
2790         struct nvme_ns_head *h;
2791
2792         lockdep_assert_held(&subsys->lock);
2793
2794         list_for_each_entry(h, &subsys->nsheads, entry) {
2795                 if (nvme_ns_ids_valid(&new->ids) &&
2796                     nvme_ns_ids_equal(&new->ids, &h->ids))
2797                         return -EINVAL;
2798         }
2799
2800         return 0;
2801 }
2802
2803 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2804                 unsigned nsid, struct nvme_id_ns *id)
2805 {
2806         struct nvme_ns_head *head;
2807         int ret = -ENOMEM;
2808
2809         head = kzalloc(sizeof(*head), GFP_KERNEL);
2810         if (!head)
2811                 goto out;
2812         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2813         if (ret < 0)
2814                 goto out_free_head;
2815         head->instance = ret;
2816         INIT_LIST_HEAD(&head->list);
2817         init_srcu_struct(&head->srcu);
2818         head->subsys = ctrl->subsys;
2819         head->ns_id = nsid;
2820         kref_init(&head->ref);
2821
2822         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2823
2824         ret = __nvme_check_ids(ctrl->subsys, head);
2825         if (ret) {
2826                 dev_err(ctrl->device,
2827                         "duplicate IDs for nsid %d\n", nsid);
2828                 goto out_cleanup_srcu;
2829         }
2830
2831         ret = nvme_mpath_alloc_disk(ctrl, head);
2832         if (ret)
2833                 goto out_cleanup_srcu;
2834
2835         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2836         return head;
2837 out_cleanup_srcu:
2838         cleanup_srcu_struct(&head->srcu);
2839         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2840 out_free_head:
2841         kfree(head);
2842 out:
2843         return ERR_PTR(ret);
2844 }
2845
2846 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2847                 struct nvme_id_ns *id)
2848 {
2849         struct nvme_ctrl *ctrl = ns->ctrl;
2850         bool is_shared = id->nmic & (1 << 0);
2851         struct nvme_ns_head *head = NULL;
2852         int ret = 0;
2853
2854         mutex_lock(&ctrl->subsys->lock);
2855         if (is_shared)
2856                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2857         if (!head) {
2858                 head = nvme_alloc_ns_head(ctrl, nsid, id);
2859                 if (IS_ERR(head)) {
2860                         ret = PTR_ERR(head);
2861                         goto out_unlock;
2862                 }
2863         } else {
2864                 struct nvme_ns_ids ids;
2865
2866                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
2867                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2868                         dev_err(ctrl->device,
2869                                 "IDs don't match for shared namespace %d\n",
2870                                         nsid);
2871                         ret = -EINVAL;
2872                         goto out_unlock;
2873                 }
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
2945         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2946         if (!ns)
2947                 return;
2948
2949         ns->queue = blk_mq_init_queue(ctrl->tagset);
2950         if (IS_ERR(ns->queue))
2951                 goto out_free_ns;
2952         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2953         ns->queue->queuedata = ns;
2954         ns->ctrl = ctrl;
2955
2956         kref_init(&ns->kref);
2957         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2958
2959         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2960         nvme_set_queue_limits(ctrl, ns->queue);
2961
2962         id = nvme_identify_ns(ctrl, nsid);
2963         if (!id)
2964                 goto out_free_queue;
2965
2966         if (id->ncap == 0)
2967                 goto out_free_id;
2968
2969         if (nvme_init_ns_head(ns, nsid, id))
2970                 goto out_free_id;
2971         nvme_setup_streams_ns(ctrl, ns);
2972
2973 #ifdef CONFIG_NVME_MULTIPATH
2974         /*
2975          * If multipathing is enabled we need to always use the subsystem
2976          * instance number for numbering our devices to avoid conflicts
2977          * between subsystems that have multiple controllers and thus use
2978          * the multipath-aware subsystem node and those that have a single
2979          * controller and use the controller node directly.
2980          */
2981         if (ns->head->disk) {
2982                 sprintf(disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
2983                                 ctrl->cntlid, ns->head->instance);
2984                 flags = GENHD_FL_HIDDEN;
2985         } else {
2986                 sprintf(disk_name, "nvme%dn%d", ctrl->subsys->instance,
2987                                 ns->head->instance);
2988         }
2989 #else
2990         /*
2991          * But without the multipath code enabled, multiple controller per
2992          * subsystems are visible as devices and thus we cannot use the
2993          * subsystem instance.
2994          */
2995         sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
2996 #endif
2997
2998         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
2999                 if (nvme_nvm_register(ns, disk_name, node)) {
3000                         dev_warn(ctrl->device, "LightNVM init failure\n");
3001                         goto out_unlink_ns;
3002                 }
3003         }
3004
3005         disk = alloc_disk_node(0, node);
3006         if (!disk)
3007                 goto out_unlink_ns;
3008
3009         disk->fops = &nvme_fops;
3010         disk->private_data = ns;
3011         disk->queue = ns->queue;
3012         disk->flags = flags;
3013         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3014         ns->disk = disk;
3015
3016         __nvme_revalidate_disk(disk, id);
3017
3018         mutex_lock(&ctrl->namespaces_mutex);
3019         list_add_tail(&ns->list, &ctrl->namespaces);
3020         mutex_unlock(&ctrl->namespaces_mutex);
3021
3022         nvme_get_ctrl(ctrl);
3023
3024         kfree(id);
3025
3026         device_add_disk(ctrl->device, ns->disk);
3027         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
3028                                         &nvme_ns_id_attr_group))
3029                 pr_warn("%s: failed to create sysfs group for identification\n",
3030                         ns->disk->disk_name);
3031         if (ns->ndev && nvme_nvm_register_sysfs(ns))
3032                 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
3033                         ns->disk->disk_name);
3034
3035         nvme_mpath_add_disk(ns->head);
3036         nvme_mpath_add_disk_links(ns);
3037         return;
3038  out_unlink_ns:
3039         mutex_lock(&ctrl->subsys->lock);
3040         list_del_rcu(&ns->siblings);
3041         mutex_unlock(&ctrl->subsys->lock);
3042  out_free_id:
3043         kfree(id);
3044  out_free_queue:
3045         blk_cleanup_queue(ns->queue);
3046  out_free_ns:
3047         kfree(ns);
3048 }
3049
3050 static void nvme_ns_remove(struct nvme_ns *ns)
3051 {
3052         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3053                 return;
3054
3055         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3056                 nvme_mpath_remove_disk_links(ns);
3057                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
3058                                         &nvme_ns_id_attr_group);
3059                 if (ns->ndev)
3060                         nvme_nvm_unregister_sysfs(ns);
3061                 del_gendisk(ns->disk);
3062                 blk_cleanup_queue(ns->queue);
3063                 if (blk_get_integrity(ns->disk))
3064                         blk_integrity_unregister(ns->disk);
3065         }
3066
3067         mutex_lock(&ns->ctrl->subsys->lock);
3068         nvme_mpath_clear_current_path(ns);
3069         list_del_rcu(&ns->siblings);
3070         mutex_unlock(&ns->ctrl->subsys->lock);
3071
3072         mutex_lock(&ns->ctrl->namespaces_mutex);
3073         list_del_init(&ns->list);
3074         mutex_unlock(&ns->ctrl->namespaces_mutex);
3075
3076         synchronize_srcu(&ns->head->srcu);
3077         nvme_mpath_check_last_path(ns);
3078         nvme_put_ns(ns);
3079 }
3080
3081 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3082 {
3083         struct nvme_ns *ns;
3084
3085         ns = nvme_find_get_ns(ctrl, nsid);
3086         if (ns) {
3087                 if (ns->disk && revalidate_disk(ns->disk))
3088                         nvme_ns_remove(ns);
3089                 nvme_put_ns(ns);
3090         } else
3091                 nvme_alloc_ns(ctrl, nsid);
3092 }
3093
3094 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3095                                         unsigned nsid)
3096 {
3097         struct nvme_ns *ns, *next;
3098
3099         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3100                 if (ns->head->ns_id > nsid)
3101                         nvme_ns_remove(ns);
3102         }
3103 }
3104
3105 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3106 {
3107         struct nvme_ns *ns;
3108         __le32 *ns_list;
3109         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3110         int ret = 0;
3111
3112         ns_list = kzalloc(0x1000, GFP_KERNEL);
3113         if (!ns_list)
3114                 return -ENOMEM;
3115
3116         for (i = 0; i < num_lists; i++) {
3117                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3118                 if (ret)
3119                         goto free;
3120
3121                 for (j = 0; j < min(nn, 1024U); j++) {
3122                         nsid = le32_to_cpu(ns_list[j]);
3123                         if (!nsid)
3124                                 goto out;
3125
3126                         nvme_validate_ns(ctrl, nsid);
3127
3128                         while (++prev < nsid) {
3129                                 ns = nvme_find_get_ns(ctrl, prev);
3130                                 if (ns) {
3131                                         nvme_ns_remove(ns);
3132                                         nvme_put_ns(ns);
3133                                 }
3134                         }
3135                 }
3136                 nn -= j;
3137         }
3138  out:
3139         nvme_remove_invalid_namespaces(ctrl, prev);
3140  free:
3141         kfree(ns_list);
3142         return ret;
3143 }
3144
3145 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3146 {
3147         unsigned i;
3148
3149         for (i = 1; i <= nn; i++)
3150                 nvme_validate_ns(ctrl, i);
3151
3152         nvme_remove_invalid_namespaces(ctrl, nn);
3153 }
3154
3155 static void nvme_scan_work(struct work_struct *work)
3156 {
3157         struct nvme_ctrl *ctrl =
3158                 container_of(work, struct nvme_ctrl, scan_work);
3159         struct nvme_id_ctrl *id;
3160         unsigned nn;
3161
3162         if (ctrl->state != NVME_CTRL_LIVE)
3163                 return;
3164
3165         WARN_ON_ONCE(!ctrl->tagset);
3166
3167         if (nvme_identify_ctrl(ctrl, &id))
3168                 return;
3169
3170         nn = le32_to_cpu(id->nn);
3171         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3172             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3173                 if (!nvme_scan_ns_list(ctrl, nn))
3174                         goto done;
3175         }
3176         nvme_scan_ns_sequential(ctrl, nn);
3177  done:
3178         mutex_lock(&ctrl->namespaces_mutex);
3179         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3180         mutex_unlock(&ctrl->namespaces_mutex);
3181         kfree(id);
3182 }
3183
3184 void nvme_queue_scan(struct nvme_ctrl *ctrl)
3185 {
3186         /*
3187          * Only new queue scan work when admin and IO queues are both alive
3188          */
3189         if (ctrl->state == NVME_CTRL_LIVE)
3190                 queue_work(nvme_wq, &ctrl->scan_work);
3191 }
3192 EXPORT_SYMBOL_GPL(nvme_queue_scan);
3193
3194 /*
3195  * This function iterates the namespace list unlocked to allow recovery from
3196  * controller failure. It is up to the caller to ensure the namespace list is
3197  * not modified by scan work while this function is executing.
3198  */
3199 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3200 {
3201         struct nvme_ns *ns, *next;
3202
3203         /*
3204          * The dead states indicates the controller was not gracefully
3205          * disconnected. In that case, we won't be able to flush any data while
3206          * removing the namespaces' disks; fail all the queues now to avoid
3207          * potentially having to clean up the failed sync later.
3208          */
3209         if (ctrl->state == NVME_CTRL_DEAD)
3210                 nvme_kill_queues(ctrl);
3211
3212         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
3213                 nvme_ns_remove(ns);
3214 }
3215 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3216
3217 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3218 {
3219         char *envp[2] = { NULL, NULL };
3220         u32 aen_result = ctrl->aen_result;
3221
3222         ctrl->aen_result = 0;
3223         if (!aen_result)
3224                 return;
3225
3226         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3227         if (!envp[0])
3228                 return;
3229         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3230         kfree(envp[0]);
3231 }
3232
3233 static void nvme_async_event_work(struct work_struct *work)
3234 {
3235         struct nvme_ctrl *ctrl =
3236                 container_of(work, struct nvme_ctrl, async_event_work);
3237
3238         nvme_aen_uevent(ctrl);
3239         ctrl->ops->submit_async_event(ctrl);
3240 }
3241
3242 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3243 {
3244
3245         u32 csts;
3246
3247         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3248                 return false;
3249
3250         if (csts == ~0)
3251                 return false;
3252
3253         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3254 }
3255
3256 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3257 {
3258         struct nvme_fw_slot_info_log *log;
3259
3260         log = kmalloc(sizeof(*log), GFP_KERNEL);
3261         if (!log)
3262                 return;
3263
3264         if (nvme_get_log(ctrl, NVME_LOG_FW_SLOT, log, sizeof(*log)))
3265                 dev_warn(ctrl->device,
3266                                 "Get FW SLOT INFO log error\n");
3267         kfree(log);
3268 }
3269
3270 static void nvme_fw_act_work(struct work_struct *work)
3271 {
3272         struct nvme_ctrl *ctrl = container_of(work,
3273                                 struct nvme_ctrl, fw_act_work);
3274         unsigned long fw_act_timeout;
3275
3276         if (ctrl->mtfa)
3277                 fw_act_timeout = jiffies +
3278                                 msecs_to_jiffies(ctrl->mtfa * 100);
3279         else
3280                 fw_act_timeout = jiffies +
3281                                 msecs_to_jiffies(admin_timeout * 1000);
3282
3283         nvme_stop_queues(ctrl);
3284         while (nvme_ctrl_pp_status(ctrl)) {
3285                 if (time_after(jiffies, fw_act_timeout)) {
3286                         dev_warn(ctrl->device,
3287                                 "Fw activation timeout, reset controller\n");
3288                         nvme_reset_ctrl(ctrl);
3289                         break;
3290                 }
3291                 msleep(100);
3292         }
3293
3294         if (ctrl->state != NVME_CTRL_LIVE)
3295                 return;
3296
3297         nvme_start_queues(ctrl);
3298         /* read FW slot information to clear the AER */
3299         nvme_get_fw_slot_info(ctrl);
3300 }
3301
3302 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3303                 union nvme_result *res)
3304 {
3305         u32 result = le32_to_cpu(res->u32);
3306
3307         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3308                 return;
3309
3310         switch (result & 0x7) {
3311         case NVME_AER_ERROR:
3312         case NVME_AER_SMART:
3313         case NVME_AER_CSS:
3314         case NVME_AER_VS:
3315                 ctrl->aen_result = result;
3316                 break;
3317         default:
3318                 break;
3319         }
3320
3321         switch (result & 0xff07) {
3322         case NVME_AER_NOTICE_NS_CHANGED:
3323                 dev_info(ctrl->device, "rescanning\n");
3324                 nvme_queue_scan(ctrl);
3325                 break;
3326         case NVME_AER_NOTICE_FW_ACT_STARTING:
3327                 queue_work(nvme_wq, &ctrl->fw_act_work);
3328                 break;
3329         default:
3330                 dev_warn(ctrl->device, "async event result %08x\n", result);
3331         }
3332         queue_work(nvme_wq, &ctrl->async_event_work);
3333 }
3334 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3335
3336 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3337 {
3338         nvme_stop_keep_alive(ctrl);
3339         flush_work(&ctrl->async_event_work);
3340         flush_work(&ctrl->scan_work);
3341         cancel_work_sync(&ctrl->fw_act_work);
3342 }
3343 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3344
3345 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3346 {
3347         if (ctrl->kato)
3348                 nvme_start_keep_alive(ctrl);
3349
3350         if (ctrl->queue_count > 1) {
3351                 nvme_queue_scan(ctrl);
3352                 queue_work(nvme_wq, &ctrl->async_event_work);
3353                 nvme_start_queues(ctrl);
3354         }
3355 }
3356 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3357
3358 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3359 {
3360         cdev_device_del(&ctrl->cdev, ctrl->device);
3361 }
3362 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3363
3364 static void nvme_free_ctrl(struct device *dev)
3365 {
3366         struct nvme_ctrl *ctrl =
3367                 container_of(dev, struct nvme_ctrl, ctrl_device);
3368         struct nvme_subsystem *subsys = ctrl->subsys;
3369
3370         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3371         kfree(ctrl->effects);
3372
3373         if (subsys) {
3374                 mutex_lock(&subsys->lock);
3375                 list_del(&ctrl->subsys_entry);
3376                 mutex_unlock(&subsys->lock);
3377                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3378         }
3379
3380         ctrl->ops->free_ctrl(ctrl);
3381
3382         if (subsys)
3383                 nvme_put_subsystem(subsys);
3384 }
3385
3386 /*
3387  * Initialize a NVMe controller structures.  This needs to be called during
3388  * earliest initialization so that we have the initialized structured around
3389  * during probing.
3390  */
3391 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3392                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3393 {
3394         int ret;
3395
3396         ctrl->state = NVME_CTRL_NEW;
3397         spin_lock_init(&ctrl->lock);
3398         INIT_LIST_HEAD(&ctrl->namespaces);
3399         mutex_init(&ctrl->namespaces_mutex);
3400         ctrl->dev = dev;
3401         ctrl->ops = ops;
3402         ctrl->quirks = quirks;
3403         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3404         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3405         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3406         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3407
3408         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3409         if (ret < 0)
3410                 goto out;
3411         ctrl->instance = ret;
3412
3413         device_initialize(&ctrl->ctrl_device);
3414         ctrl->device = &ctrl->ctrl_device;
3415         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3416         ctrl->device->class = nvme_class;
3417         ctrl->device->parent = ctrl->dev;
3418         ctrl->device->groups = nvme_dev_attr_groups;
3419         ctrl->device->release = nvme_free_ctrl;
3420         dev_set_drvdata(ctrl->device, ctrl);
3421         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3422         if (ret)
3423                 goto out_release_instance;
3424
3425         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3426         ctrl->cdev.owner = ops->module;
3427         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3428         if (ret)
3429                 goto out_free_name;
3430
3431         /*
3432          * Initialize latency tolerance controls.  The sysfs files won't
3433          * be visible to userspace unless the device actually supports APST.
3434          */
3435         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3436         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3437                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3438
3439         return 0;
3440 out_free_name:
3441         kfree_const(dev->kobj.name);
3442 out_release_instance:
3443         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3444 out:
3445         return ret;
3446 }
3447 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3448
3449 /**
3450  * nvme_kill_queues(): Ends all namespace queues
3451  * @ctrl: the dead controller that needs to end
3452  *
3453  * Call this function when the driver determines it is unable to get the
3454  * controller in a state capable of servicing IO.
3455  */
3456 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3457 {
3458         struct nvme_ns *ns;
3459
3460         mutex_lock(&ctrl->namespaces_mutex);
3461
3462         /* Forcibly unquiesce queues to avoid blocking dispatch */
3463         if (ctrl->admin_q)
3464                 blk_mq_unquiesce_queue(ctrl->admin_q);
3465
3466         list_for_each_entry(ns, &ctrl->namespaces, list) {
3467                 /*
3468                  * Revalidating a dead namespace sets capacity to 0. This will
3469                  * end buffered writers dirtying pages that can't be synced.
3470                  */
3471                 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
3472                         continue;
3473                 revalidate_disk(ns->disk);
3474                 blk_set_queue_dying(ns->queue);
3475
3476                 /* Forcibly unquiesce queues to avoid blocking dispatch */
3477                 blk_mq_unquiesce_queue(ns->queue);
3478         }
3479         mutex_unlock(&ctrl->namespaces_mutex);
3480 }
3481 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3482
3483 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3484 {
3485         struct nvme_ns *ns;
3486
3487         mutex_lock(&ctrl->namespaces_mutex);
3488         list_for_each_entry(ns, &ctrl->namespaces, list)
3489                 blk_mq_unfreeze_queue(ns->queue);
3490         mutex_unlock(&ctrl->namespaces_mutex);
3491 }
3492 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3493
3494 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3495 {
3496         struct nvme_ns *ns;
3497
3498         mutex_lock(&ctrl->namespaces_mutex);
3499         list_for_each_entry(ns, &ctrl->namespaces, list) {
3500                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3501                 if (timeout <= 0)
3502                         break;
3503         }
3504         mutex_unlock(&ctrl->namespaces_mutex);
3505 }
3506 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3507
3508 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3509 {
3510         struct nvme_ns *ns;
3511
3512         mutex_lock(&ctrl->namespaces_mutex);
3513         list_for_each_entry(ns, &ctrl->namespaces, list)
3514                 blk_mq_freeze_queue_wait(ns->queue);
3515         mutex_unlock(&ctrl->namespaces_mutex);
3516 }
3517 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3518
3519 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3520 {
3521         struct nvme_ns *ns;
3522
3523         mutex_lock(&ctrl->namespaces_mutex);
3524         list_for_each_entry(ns, &ctrl->namespaces, list)
3525                 blk_freeze_queue_start(ns->queue);
3526         mutex_unlock(&ctrl->namespaces_mutex);
3527 }
3528 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3529
3530 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3531 {
3532         struct nvme_ns *ns;
3533
3534         mutex_lock(&ctrl->namespaces_mutex);
3535         list_for_each_entry(ns, &ctrl->namespaces, list)
3536                 blk_mq_quiesce_queue(ns->queue);
3537         mutex_unlock(&ctrl->namespaces_mutex);
3538 }
3539 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3540
3541 void nvme_start_queues(struct nvme_ctrl *ctrl)
3542 {
3543         struct nvme_ns *ns;
3544
3545         mutex_lock(&ctrl->namespaces_mutex);
3546         list_for_each_entry(ns, &ctrl->namespaces, list)
3547                 blk_mq_unquiesce_queue(ns->queue);
3548         mutex_unlock(&ctrl->namespaces_mutex);
3549 }
3550 EXPORT_SYMBOL_GPL(nvme_start_queues);
3551
3552 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set)
3553 {
3554         if (!ctrl->ops->reinit_request)
3555                 return 0;
3556
3557         return blk_mq_tagset_iter(set, set->driver_data,
3558                         ctrl->ops->reinit_request);
3559 }
3560 EXPORT_SYMBOL_GPL(nvme_reinit_tagset);
3561
3562 int __init nvme_core_init(void)
3563 {
3564         int result = -ENOMEM;
3565
3566         nvme_wq = alloc_workqueue("nvme-wq",
3567                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3568         if (!nvme_wq)
3569                 goto out;
3570
3571         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3572                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3573         if (!nvme_reset_wq)
3574                 goto destroy_wq;
3575
3576         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3577                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3578         if (!nvme_delete_wq)
3579                 goto destroy_reset_wq;
3580
3581         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3582         if (result < 0)
3583                 goto destroy_delete_wq;
3584
3585         nvme_class = class_create(THIS_MODULE, "nvme");
3586         if (IS_ERR(nvme_class)) {
3587                 result = PTR_ERR(nvme_class);
3588                 goto unregister_chrdev;
3589         }
3590
3591         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3592         if (IS_ERR(nvme_subsys_class)) {
3593                 result = PTR_ERR(nvme_subsys_class);
3594                 goto destroy_class;
3595         }
3596         return 0;
3597
3598 destroy_class:
3599         class_destroy(nvme_class);
3600 unregister_chrdev:
3601         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3602 destroy_delete_wq:
3603         destroy_workqueue(nvme_delete_wq);
3604 destroy_reset_wq:
3605         destroy_workqueue(nvme_reset_wq);
3606 destroy_wq:
3607         destroy_workqueue(nvme_wq);
3608 out:
3609         return result;
3610 }
3611
3612 void nvme_core_exit(void)
3613 {
3614         ida_destroy(&nvme_subsystems_ida);
3615         class_destroy(nvme_subsys_class);
3616         class_destroy(nvme_class);
3617         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3618         destroy_workqueue(nvme_delete_wq);
3619         destroy_workqueue(nvme_reset_wq);
3620         destroy_workqueue(nvme_wq);
3621 }
3622
3623 MODULE_LICENSE("GPL");
3624 MODULE_VERSION("1.0");
3625 module_init(nvme_core_init);
3626 module_exit(nvme_core_exit);