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 <scsi/sg.h>
  30 #include <asm/unaligned.h>
  31
  32 #include "nvme.h"
  33
  34 #define NVME_MINORS             (1U << MINORBITS)
  35
  36 unsigned char admin_timeout = 60;
  37 module_param(admin_timeout, byte, 0644);
  38 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  39 EXPORT_SYMBOL_GPL(admin_timeout);
  40
  41 unsigned char nvme_io_timeout = 30;
  42 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
  43 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  44 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  45
  46 unsigned char shutdown_timeout = 5;
  47 module_param(shutdown_timeout, byte, 0644);
  48 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  49
  50 static int nvme_major;
  51 module_param(nvme_major, int, 0);
  52
  53 static int nvme_char_major;
  54 module_param(nvme_char_major, int, 0);
  55
  56 static LIST_HEAD(nvme_ctrl_list);
  57 static DEFINE_SPINLOCK(dev_list_lock);
  58
  59 static struct class *nvme_class;
  60
  61 void nvme_cancel_request(struct request *req, void *data, bool reserved)
  62 {
  63         int status;
  64
  65         if (!blk_mq_request_started(req))
  66                 return;
  67
  68         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
  69                                 "Cancelling I/O %d", req->tag);
  70
  71         status = NVME_SC_ABORT_REQ;
  72         if (blk_queue_dying(req->q))
  73                 status |= NVME_SC_DNR;
  74         blk_mq_complete_request(req, status);
  75 }
  76 EXPORT_SYMBOL_GPL(nvme_cancel_request);
  77
  78 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
  79                 enum nvme_ctrl_state new_state)
  80 {
  81         enum nvme_ctrl_state old_state = ctrl->state;
  82         bool changed = false;
  83
  84         spin_lock_irq(&ctrl->lock);
  85         switch (new_state) {
  86         case NVME_CTRL_LIVE:
  87                 switch (old_state) {
  88                 case NVME_CTRL_NEW:
  89                 case NVME_CTRL_RESETTING:
  90                         changed = true;
  91                         /* FALLTHRU */
  92                 default:
  93                         break;
  94                 }
  95                 break;
  96         case NVME_CTRL_RESETTING:
  97                 switch (old_state) {
  98                 case NVME_CTRL_NEW:
  99                 case NVME_CTRL_LIVE:
 100                         changed = true;
 101                         /* FALLTHRU */
 102                 default:
 103                         break;
 104                 }
 105                 break;
 106         case NVME_CTRL_DELETING:
 107                 switch (old_state) {
 108                 case NVME_CTRL_LIVE:
 109                 case NVME_CTRL_RESETTING:
 110                         changed = true;
 111                         /* FALLTHRU */
 112                 default:
 113                         break;
 114                 }
 115                 break;
 116         case NVME_CTRL_DEAD:
 117                 switch (old_state) {
 118                 case NVME_CTRL_DELETING:
 119                         changed = true;
 120                         /* FALLTHRU */
 121                 default:
 122                         break;
 123                 }
 124                 break;
 125         default:
 126                 break;
 127         }
 128         spin_unlock_irq(&ctrl->lock);
 129
 130         if (changed)
 131                 ctrl->state = new_state;
 132
 133         return changed;
 134 }
 135 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 136
 137 static void nvme_free_ns(struct kref *kref)
 138 {
 139         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 140
 141         if (ns->type == NVME_NS_LIGHTNVM)
 142                 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
 143
 144         spin_lock(&dev_list_lock);
 145         ns->disk->private_data = NULL;
 146         spin_unlock(&dev_list_lock);
 147
 148         put_disk(ns->disk);
 149         ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
 150         nvme_put_ctrl(ns->ctrl);
 151         kfree(ns);
 152 }
 153
 154 static void nvme_put_ns(struct nvme_ns *ns)
 155 {
 156         kref_put(&ns->kref, nvme_free_ns);
 157 }
 158
 159 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
 160 {
 161         struct nvme_ns *ns;
 162
 163         spin_lock(&dev_list_lock);
 164         ns = disk->private_data;
 165         if (ns) {
 166                 if (!kref_get_unless_zero(&ns->kref))
 167                         goto fail;
 168                 if (!try_module_get(ns->ctrl->ops->module))
 169                         goto fail_put_ns;
 170         }
 171         spin_unlock(&dev_list_lock);
 172
 173         return ns;
 174
 175 fail_put_ns:
 176         kref_put(&ns->kref, nvme_free_ns);
 177 fail:
 178         spin_unlock(&dev_list_lock);
 179         return NULL;
 180 }
 181
 182 void nvme_requeue_req(struct request *req)
 183 {
 184         unsigned long flags;
 185
 186         blk_mq_requeue_request(req);
 187         spin_lock_irqsave(req->q->queue_lock, flags);
 188         if (!blk_queue_stopped(req->q))
 189                 blk_mq_kick_requeue_list(req->q);
 190         spin_unlock_irqrestore(req->q->queue_lock, flags);
 191 }
 192 EXPORT_SYMBOL_GPL(nvme_requeue_req);
 193
 194 struct request *nvme_alloc_request(struct request_queue *q,
 195                 struct nvme_command *cmd, unsigned int flags)
 196 {
 197         struct request *req;
 198
 199         req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
 200         if (IS_ERR(req))
 201                 return req;
 202
 203         req->cmd_type = REQ_TYPE_DRV_PRIV;
 204         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 205         req->__data_len = 0;
 206         req->__sector = (sector_t) -1;
 207         req->bio = req->biotail = NULL;
 208
 209         req->cmd = (unsigned char *)cmd;
 210         req->cmd_len = sizeof(struct nvme_command);
 211
 212         return req;
 213 }
 214 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 215
 216 static inline void nvme_setup_flush(struct nvme_ns *ns,
 217                 struct nvme_command *cmnd)
 218 {
 219         memset(cmnd, 0, sizeof(*cmnd));
 220         cmnd->common.opcode = nvme_cmd_flush;
 221         cmnd->common.nsid = cpu_to_le32(ns->ns_id);
 222 }
 223
 224 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 225                 struct nvme_command *cmnd)
 226 {
 227         struct nvme_dsm_range *range;
 228         struct page *page;
 229         int offset;
 230         unsigned int nr_bytes = blk_rq_bytes(req);
 231
 232         range = kmalloc(sizeof(*range), GFP_ATOMIC);
 233         if (!range)
 234                 return BLK_MQ_RQ_QUEUE_BUSY;
 235
 236         range->cattr = cpu_to_le32(0);
 237         range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
 238         range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 239
 240         memset(cmnd, 0, sizeof(*cmnd));
 241         cmnd->dsm.opcode = nvme_cmd_dsm;
 242         cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
 243         cmnd->dsm.nr = 0;
 244         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 245
 246         req->completion_data = range;
 247         page = virt_to_page(range);
 248         offset = offset_in_page(range);
 249         blk_add_request_payload(req, page, offset, sizeof(*range));
 250
 251         /*
 252          * we set __data_len back to the size of the area to be discarded
 253          * on disk. This allows us to report completion on the full amount
 254          * of blocks described by the request.
 255          */
 256         req->__data_len = nr_bytes;
 257
 258         return 0;
 259 }
 260
 261 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
 262                 struct nvme_command *cmnd)
 263 {
 264         u16 control = 0;
 265         u32 dsmgmt = 0;
 266
 267         if (req->cmd_flags & REQ_FUA)
 268                 control |= NVME_RW_FUA;
 269         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 270                 control |= NVME_RW_LR;
 271
 272         if (req->cmd_flags & REQ_RAHEAD)
 273                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 274
 275         memset(cmnd, 0, sizeof(*cmnd));
 276         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 277         cmnd->rw.command_id = req->tag;
 278         cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
 279         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 280         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 281
 282         if (ns->ms) {
 283                 switch (ns->pi_type) {
 284                 case NVME_NS_DPS_PI_TYPE3:
 285                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 286                         break;
 287                 case NVME_NS_DPS_PI_TYPE1:
 288                 case NVME_NS_DPS_PI_TYPE2:
 289                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 290                                         NVME_RW_PRINFO_PRCHK_REF;
 291                         cmnd->rw.reftag = cpu_to_le32(
 292                                         nvme_block_nr(ns, blk_rq_pos(req)));
 293                         break;
 294                 }
 295                 if (!blk_integrity_rq(req))
 296                         control |= NVME_RW_PRINFO_PRACT;
 297         }
 298
 299         cmnd->rw.control = cpu_to_le16(control);
 300         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 301 }
 302
 303 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 304                 struct nvme_command *cmd)
 305 {
 306         int ret = 0;
 307
 308         if (req->cmd_type == REQ_TYPE_DRV_PRIV)
 309                 memcpy(cmd, req->cmd, sizeof(*cmd));
 310         else if (req_op(req) == REQ_OP_FLUSH)
 311                 nvme_setup_flush(ns, cmd);
 312         else if (req_op(req) == REQ_OP_DISCARD)
 313                 ret = nvme_setup_discard(ns, req, cmd);
 314         else
 315                 nvme_setup_rw(ns, req, cmd);
 316
 317         return ret;
 318 }
 319 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 320
 321 /*
 322  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 323  * if the result is positive, it's an NVM Express status code
 324  */
 325 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 326                 struct nvme_completion *cqe, void *buffer, unsigned bufflen,
 327                 unsigned timeout)
 328 {
 329         struct request *req;
 330         int ret;
 331
 332         req = nvme_alloc_request(q, cmd, 0);
 333         if (IS_ERR(req))
 334                 return PTR_ERR(req);
 335
 336         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 337         req->special = cqe;
 338
 339         if (buffer && bufflen) {
 340                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 341                 if (ret)
 342                         goto out;
 343         }
 344
 345         blk_execute_rq(req->q, NULL, req, 0);
 346         ret = req->errors;
 347  out:
 348         blk_mq_free_request(req);
 349         return ret;
 350 }
 351
 352 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 353                 void *buffer, unsigned bufflen)
 354 {
 355         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0);
 356 }
 357 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 358
 359 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 360                 void __user *ubuffer, unsigned bufflen,
 361                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 362                 u32 *result, unsigned timeout)
 363 {
 364         bool write = nvme_is_write(cmd);
 365         struct nvme_completion cqe;
 366         struct nvme_ns *ns = q->queuedata;
 367         struct gendisk *disk = ns ? ns->disk : NULL;
 368         struct request *req;
 369         struct bio *bio = NULL;
 370         void *meta = NULL;
 371         int ret;
 372
 373         req = nvme_alloc_request(q, cmd, 0);
 374         if (IS_ERR(req))
 375                 return PTR_ERR(req);
 376
 377         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 378         req->special = &cqe;
 379
 380         if (ubuffer && bufflen) {
 381                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 382                                 GFP_KERNEL);
 383                 if (ret)
 384                         goto out;
 385                 bio = req->bio;
 386
 387                 if (!disk)
 388                         goto submit;
 389                 bio->bi_bdev = bdget_disk(disk, 0);
 390                 if (!bio->bi_bdev) {
 391                         ret = -ENODEV;
 392                         goto out_unmap;
 393                 }
 394
 395                 if (meta_buffer && meta_len) {
 396                         struct bio_integrity_payload *bip;
 397
 398                         meta = kmalloc(meta_len, GFP_KERNEL);
 399                         if (!meta) {
 400                                 ret = -ENOMEM;
 401                                 goto out_unmap;
 402                         }
 403
 404                         if (write) {
 405                                 if (copy_from_user(meta, meta_buffer,
 406                                                 meta_len)) {
 407                                         ret = -EFAULT;
 408                                         goto out_free_meta;
 409                                 }
 410                         }
 411
 412                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 413                         if (IS_ERR(bip)) {
 414                                 ret = PTR_ERR(bip);
 415                                 goto out_free_meta;
 416                         }
 417
 418                         bip->bip_iter.bi_size = meta_len;
 419                         bip->bip_iter.bi_sector = meta_seed;
 420
 421                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 422                                         meta_len, offset_in_page(meta));
 423                         if (ret != meta_len) {
 424                                 ret = -ENOMEM;
 425                                 goto out_free_meta;
 426                         }
 427                 }
 428         }
 429  submit:
 430         blk_execute_rq(req->q, disk, req, 0);
 431         ret = req->errors;
 432         if (result)
 433                 *result = le32_to_cpu(cqe.result);
 434         if (meta && !ret && !write) {
 435                 if (copy_to_user(meta_buffer, meta, meta_len))
 436                         ret = -EFAULT;
 437         }
 438  out_free_meta:
 439         kfree(meta);
 440  out_unmap:
 441         if (bio) {
 442                 if (disk && bio->bi_bdev)
 443                         bdput(bio->bi_bdev);
 444                 blk_rq_unmap_user(bio);
 445         }
 446  out:
 447         blk_mq_free_request(req);
 448         return ret;
 449 }
 450
 451 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 452                 void __user *ubuffer, unsigned bufflen, u32 *result,
 453                 unsigned timeout)
 454 {
 455         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 456                         result, timeout);
 457 }
 458
 459 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 460 {
 461         struct nvme_command c = { };
 462         int error;
 463
 464         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 465         c.identify.opcode = nvme_admin_identify;
 466         c.identify.cns = cpu_to_le32(1);
 467
 468         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 469         if (!*id)
 470                 return -ENOMEM;
 471
 472         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 473                         sizeof(struct nvme_id_ctrl));
 474         if (error)
 475                 kfree(*id);
 476         return error;
 477 }
 478
 479 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 480 {
 481         struct nvme_command c = { };
 482
 483         c.identify.opcode = nvme_admin_identify;
 484         c.identify.cns = cpu_to_le32(2);
 485         c.identify.nsid = cpu_to_le32(nsid);
 486         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 487 }
 488
 489 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 490                 struct nvme_id_ns **id)
 491 {
 492         struct nvme_command c = { };
 493         int error;
 494
 495         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 496         c.identify.opcode = nvme_admin_identify,
 497         c.identify.nsid = cpu_to_le32(nsid),
 498
 499         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 500         if (!*id)
 501                 return -ENOMEM;
 502
 503         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 504                         sizeof(struct nvme_id_ns));
 505         if (error)
 506                 kfree(*id);
 507         return error;
 508 }
 509
 510 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 511                                         dma_addr_t dma_addr, u32 *result)
 512 {
 513         struct nvme_command c;
 514         struct nvme_completion cqe;
 515         int ret;
 516
 517         memset(&c, 0, sizeof(c));
 518         c.features.opcode = nvme_admin_get_features;
 519         c.features.nsid = cpu_to_le32(nsid);
 520         c.features.prp1 = cpu_to_le64(dma_addr);
 521         c.features.fid = cpu_to_le32(fid);
 522
 523         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
 524         if (ret >= 0)
 525                 *result = le32_to_cpu(cqe.result);
 526         return ret;
 527 }
 528
 529 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 530                                         dma_addr_t dma_addr, u32 *result)
 531 {
 532         struct nvme_command c;
 533         struct nvme_completion cqe;
 534         int ret;
 535
 536         memset(&c, 0, sizeof(c));
 537         c.features.opcode = nvme_admin_set_features;
 538         c.features.prp1 = cpu_to_le64(dma_addr);
 539         c.features.fid = cpu_to_le32(fid);
 540         c.features.dword11 = cpu_to_le32(dword11);
 541
 542         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
 543         if (ret >= 0)
 544                 *result = le32_to_cpu(cqe.result);
 545         return ret;
 546 }
 547
 548 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 549 {
 550         struct nvme_command c = { };
 551         int error;
 552
 553         c.common.opcode = nvme_admin_get_log_page,
 554         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 555         c.common.cdw10[0] = cpu_to_le32(
 556                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 557                          NVME_LOG_SMART),
 558
 559         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 560         if (!*log)
 561                 return -ENOMEM;
 562
 563         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 564                         sizeof(struct nvme_smart_log));
 565         if (error)
 566                 kfree(*log);
 567         return error;
 568 }
 569
 570 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 571 {
 572         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 573         u32 result;
 574         int status, nr_io_queues;
 575
 576         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
 577                         &result);
 578         if (status < 0)
 579                 return status;
 580
 581         /*
 582          * Degraded controllers might return an error when setting the queue
 583          * count.  We still want to be able to bring them online and offer
 584          * access to the admin queue, as that might be only way to fix them up.
 585          */
 586         if (status > 0) {
 587                 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
 588                 *count = 0;
 589         } else {
 590                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 591                 *count = min(*count, nr_io_queues);
 592         }
 593
 594         return 0;
 595 }
 596 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
 597
 598 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 599 {
 600         struct nvme_user_io io;
 601         struct nvme_command c;
 602         unsigned length, meta_len;
 603         void __user *metadata;
 604
 605         if (copy_from_user(&io, uio, sizeof(io)))
 606                 return -EFAULT;
 607         if (io.flags)
 608                 return -EINVAL;
 609
 610         switch (io.opcode) {
 611         case nvme_cmd_write:
 612         case nvme_cmd_read:
 613         case nvme_cmd_compare:
 614                 break;
 615         default:
 616                 return -EINVAL;
 617         }
 618
 619         length = (io.nblocks + 1) << ns->lba_shift;
 620         meta_len = (io.nblocks + 1) * ns->ms;
 621         metadata = (void __user *)(uintptr_t)io.metadata;
 622
 623         if (ns->ext) {
 624                 length += meta_len;
 625                 meta_len = 0;
 626         } else if (meta_len) {
 627                 if ((io.metadata & 3) || !io.metadata)
 628                         return -EINVAL;
 629         }
 630
 631         memset(&c, 0, sizeof(c));
 632         c.rw.opcode = io.opcode;
 633         c.rw.flags = io.flags;
 634         c.rw.nsid = cpu_to_le32(ns->ns_id);
 635         c.rw.slba = cpu_to_le64(io.slba);
 636         c.rw.length = cpu_to_le16(io.nblocks);
 637         c.rw.control = cpu_to_le16(io.control);
 638         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 639         c.rw.reftag = cpu_to_le32(io.reftag);
 640         c.rw.apptag = cpu_to_le16(io.apptag);
 641         c.rw.appmask = cpu_to_le16(io.appmask);
 642
 643         return __nvme_submit_user_cmd(ns->queue, &c,
 644                         (void __user *)(uintptr_t)io.addr, length,
 645                         metadata, meta_len, io.slba, NULL, 0);
 646 }
 647
 648 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 649                         struct nvme_passthru_cmd __user *ucmd)
 650 {
 651         struct nvme_passthru_cmd cmd;
 652         struct nvme_command c;
 653         unsigned timeout = 0;
 654         int status;
 655
 656         if (!capable(CAP_SYS_ADMIN))
 657                 return -EACCES;
 658         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 659                 return -EFAULT;
 660         if (cmd.flags)
 661                 return -EINVAL;
 662
 663         memset(&c, 0, sizeof(c));
 664         c.common.opcode = cmd.opcode;
 665         c.common.flags = cmd.flags;
 666         c.common.nsid = cpu_to_le32(cmd.nsid);
 667         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 668         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 669         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 670         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 671         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 672         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 673         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 674         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 675
 676         if (cmd.timeout_ms)
 677                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 678
 679         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 680                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 681                         &cmd.result, timeout);
 682         if (status >= 0) {
 683                 if (put_user(cmd.result, &ucmd->result))
 684                         return -EFAULT;
 685         }
 686
 687         return status;
 688 }
 689
 690 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 691                 unsigned int cmd, unsigned long arg)
 692 {
 693         struct nvme_ns *ns = bdev->bd_disk->private_data;
 694
 695         switch (cmd) {
 696         case NVME_IOCTL_ID:
 697                 force_successful_syscall_return();
 698                 return ns->ns_id;
 699         case NVME_IOCTL_ADMIN_CMD:
 700                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 701         case NVME_IOCTL_IO_CMD:
 702                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 703         case NVME_IOCTL_SUBMIT_IO:
 704                 return nvme_submit_io(ns, (void __user *)arg);
 705 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 706         case SG_GET_VERSION_NUM:
 707                 return nvme_sg_get_version_num((void __user *)arg);
 708         case SG_IO:
 709                 return nvme_sg_io(ns, (void __user *)arg);
 710 #endif
 711         default:
 712                 return -ENOTTY;
 713         }
 714 }
 715
 716 #ifdef CONFIG_COMPAT
 717 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 718                         unsigned int cmd, unsigned long arg)
 719 {
 720         switch (cmd) {
 721         case SG_IO:
 722                 return -ENOIOCTLCMD;
 723         }
 724         return nvme_ioctl(bdev, mode, cmd, arg);
 725 }
 726 #else
 727 #define nvme_compat_ioctl       NULL
 728 #endif
 729
 730 static int nvme_open(struct block_device *bdev, fmode_t mode)
 731 {
 732         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 733 }
 734
 735 static void nvme_release(struct gendisk *disk, fmode_t mode)
 736 {
 737         struct nvme_ns *ns = disk->private_data;
 738
 739         module_put(ns->ctrl->ops->module);
 740         nvme_put_ns(ns);
 741 }
 742
 743 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 744 {
 745         /* some standard values */
 746         geo->heads = 1 << 6;
 747         geo->sectors = 1 << 5;
 748         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 749         return 0;
 750 }
 751
 752 #ifdef CONFIG_BLK_DEV_INTEGRITY
 753 static void nvme_init_integrity(struct nvme_ns *ns)
 754 {
 755         struct blk_integrity integrity;
 756
 757         switch (ns->pi_type) {
 758         case NVME_NS_DPS_PI_TYPE3:
 759                 integrity.profile = &t10_pi_type3_crc;
 760                 integrity.tag_size = sizeof(u16) + sizeof(u32);
 761                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 762                 break;
 763         case NVME_NS_DPS_PI_TYPE1:
 764         case NVME_NS_DPS_PI_TYPE2:
 765                 integrity.profile = &t10_pi_type1_crc;
 766                 integrity.tag_size = sizeof(u16);
 767                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 768                 break;
 769         default:
 770                 integrity.profile = NULL;
 771                 break;
 772         }
 773         integrity.tuple_size = ns->ms;
 774         blk_integrity_register(ns->disk, &integrity);
 775         blk_queue_max_integrity_segments(ns->queue, 1);
 776 }
 777 #else
 778 static void nvme_init_integrity(struct nvme_ns *ns)
 779 {
 780 }
 781 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 782
 783 static void nvme_config_discard(struct nvme_ns *ns)
 784 {
 785         struct nvme_ctrl *ctrl = ns->ctrl;
 786         u32 logical_block_size = queue_logical_block_size(ns->queue);
 787
 788         if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
 789                 ns->queue->limits.discard_zeroes_data = 1;
 790         else
 791                 ns->queue->limits.discard_zeroes_data = 0;
 792
 793         ns->queue->limits.discard_alignment = logical_block_size;
 794         ns->queue->limits.discard_granularity = logical_block_size;
 795         blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
 796         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 797 }
 798
 799 static int nvme_revalidate_disk(struct gendisk *disk)
 800 {
 801         struct nvme_ns *ns = disk->private_data;
 802         struct nvme_id_ns *id;
 803         u8 lbaf, pi_type;
 804         u16 old_ms;
 805         unsigned short bs;
 806
 807         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
 808                 set_capacity(disk, 0);
 809                 return -ENODEV;
 810         }
 811         if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
 812                 dev_warn(disk_to_dev(ns->disk), "%s: Identify failure\n",
 813                                 __func__);
 814                 return -ENODEV;
 815         }
 816         if (id->ncap == 0) {
 817                 kfree(id);
 818                 return -ENODEV;
 819         }
 820
 821         if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
 822                 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
 823                         dev_warn(disk_to_dev(ns->disk),
 824                                 "%s: LightNVM init failure\n", __func__);
 825                         kfree(id);
 826                         return -ENODEV;
 827                 }
 828                 ns->type = NVME_NS_LIGHTNVM;
 829         }
 830
 831         if (ns->ctrl->vs >= NVME_VS(1, 1))
 832                 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
 833         if (ns->ctrl->vs >= NVME_VS(1, 2))
 834                 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
 835
 836         old_ms = ns->ms;
 837         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 838         ns->lba_shift = id->lbaf[lbaf].ds;
 839         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 840         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 841
 842         /*
 843          * If identify namespace failed, use default 512 byte block size so
 844          * block layer can use before failing read/write for 0 capacity.
 845          */
 846         if (ns->lba_shift == 0)
 847                 ns->lba_shift = 9;
 848         bs = 1 << ns->lba_shift;
 849         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 850         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 851                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 852
 853         blk_mq_freeze_queue(disk->queue);
 854         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 855                                 ns->ms != old_ms ||
 856                                 bs != queue_logical_block_size(disk->queue) ||
 857                                 (ns->ms && ns->ext)))
 858                 blk_integrity_unregister(disk);
 859
 860         ns->pi_type = pi_type;
 861         blk_queue_logical_block_size(ns->queue, bs);
 862
 863         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 864                 nvme_init_integrity(ns);
 865         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 866                 set_capacity(disk, 0);
 867         else
 868                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 869
 870         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 871                 nvme_config_discard(ns);
 872         blk_mq_unfreeze_queue(disk->queue);
 873
 874         kfree(id);
 875         return 0;
 876 }
 877
 878 static char nvme_pr_type(enum pr_type type)
 879 {
 880         switch (type) {
 881         case PR_WRITE_EXCLUSIVE:
 882                 return 1;
 883         case PR_EXCLUSIVE_ACCESS:
 884                 return 2;
 885         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 886                 return 3;
 887         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 888                 return 4;
 889         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 890                 return 5;
 891         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 892                 return 6;
 893         default:
 894                 return 0;
 895         }
 896 };
 897
 898 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 899                                 u64 key, u64 sa_key, u8 op)
 900 {
 901         struct nvme_ns *ns = bdev->bd_disk->private_data;
 902         struct nvme_command c;
 903         u8 data[16] = { 0, };
 904
 905         put_unaligned_le64(key, &data[0]);
 906         put_unaligned_le64(sa_key, &data[8]);
 907
 908         memset(&c, 0, sizeof(c));
 909         c.common.opcode = op;
 910         c.common.nsid = cpu_to_le32(ns->ns_id);
 911         c.common.cdw10[0] = cpu_to_le32(cdw10);
 912
 913         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
 914 }
 915
 916 static int nvme_pr_register(struct block_device *bdev, u64 old,
 917                 u64 new, unsigned flags)
 918 {
 919         u32 cdw10;
 920
 921         if (flags & ~PR_FL_IGNORE_KEY)
 922                 return -EOPNOTSUPP;
 923
 924         cdw10 = old ? 2 : 0;
 925         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
 926         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
 927         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
 928 }
 929
 930 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
 931                 enum pr_type type, unsigned flags)
 932 {
 933         u32 cdw10;
 934
 935         if (flags & ~PR_FL_IGNORE_KEY)
 936                 return -EOPNOTSUPP;
 937
 938         cdw10 = nvme_pr_type(type) << 8;
 939         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
 940         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
 941 }
 942
 943 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
 944                 enum pr_type type, bool abort)
 945 {
 946         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
 947         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
 948 }
 949
 950 static int nvme_pr_clear(struct block_device *bdev, u64 key)
 951 {
 952         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
 953         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
 954 }
 955
 956 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
 957 {
 958         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
 959         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
 960 }
 961
 962 static const struct pr_ops nvme_pr_ops = {
 963         .pr_register    = nvme_pr_register,
 964         .pr_reserve     = nvme_pr_reserve,
 965         .pr_release     = nvme_pr_release,
 966         .pr_preempt     = nvme_pr_preempt,
 967         .pr_clear       = nvme_pr_clear,
 968 };
 969
 970 static const struct block_device_operations nvme_fops = {
 971         .owner          = THIS_MODULE,
 972         .ioctl          = nvme_ioctl,
 973         .compat_ioctl   = nvme_compat_ioctl,
 974         .open           = nvme_open,
 975         .release        = nvme_release,
 976         .getgeo         = nvme_getgeo,
 977         .revalidate_disk= nvme_revalidate_disk,
 978         .pr_ops         = &nvme_pr_ops,
 979 };
 980
 981 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
 982 {
 983         unsigned long timeout =
 984                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
 985         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
 986         int ret;
 987
 988         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 989                 if ((csts & NVME_CSTS_RDY) == bit)
 990                         break;
 991
 992                 msleep(100);
 993                 if (fatal_signal_pending(current))
 994                         return -EINTR;
 995                 if (time_after(jiffies, timeout)) {
 996                         dev_err(ctrl->device,
 997                                 "Device not ready; aborting %s\n", enabled ?
 998                                                 "initialisation" : "reset");
 999                         return -ENODEV;
1000                 }
1001         }
1002
1003         return ret;
1004 }
1005
1006 /*
1007  * If the device has been passed off to us in an enabled state, just clear
1008  * the enabled bit.  The spec says we should set the 'shutdown notification
1009  * bits', but doing so may cause the device to complete commands to the
1010  * admin queue ... and we don't know what memory that might be pointing at!
1011  */
1012 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1013 {
1014         int ret;
1015
1016         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1017         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1018
1019         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1020         if (ret)
1021                 return ret;
1022         return nvme_wait_ready(ctrl, cap, false);
1023 }
1024 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1025
1026 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1027 {
1028         /*
1029          * Default to a 4K page size, with the intention to update this
1030          * path in the future to accomodate architectures with differing
1031          * kernel and IO page sizes.
1032          */
1033         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1034         int ret;
1035
1036         if (page_shift < dev_page_min) {
1037                 dev_err(ctrl->device,
1038                         "Minimum device page size %u too large for host (%u)\n",
1039                         1 << dev_page_min, 1 << page_shift);
1040                 return -ENODEV;
1041         }
1042
1043         ctrl->page_size = 1 << page_shift;
1044
1045         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1046         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1047         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1048         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1049         ctrl->ctrl_config |= NVME_CC_ENABLE;
1050
1051         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1052         if (ret)
1053                 return ret;
1054         return nvme_wait_ready(ctrl, cap, true);
1055 }
1056 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1057
1058 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1059 {
1060         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1061         u32 csts;
1062         int ret;
1063
1064         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1065         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1066
1067         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1068         if (ret)
1069                 return ret;
1070
1071         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1072                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1073                         break;
1074
1075                 msleep(100);
1076                 if (fatal_signal_pending(current))
1077                         return -EINTR;
1078                 if (time_after(jiffies, timeout)) {
1079                         dev_err(ctrl->device,
1080                                 "Device shutdown incomplete; abort shutdown\n");
1081                         return -ENODEV;
1082                 }
1083         }
1084
1085         return ret;
1086 }
1087 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1088
1089 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1090                 struct request_queue *q)
1091 {
1092         bool vwc = false;
1093
1094         if (ctrl->max_hw_sectors) {
1095                 u32 max_segments =
1096                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1097
1098                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1099                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1100         }
1101         if (ctrl->stripe_size)
1102                 blk_queue_chunk_sectors(q, ctrl->stripe_size >> 9);
1103         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1104         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1105                 vwc = true;
1106         blk_queue_write_cache(q, vwc, vwc);
1107 }
1108
1109 /*
1110  * Initialize the cached copies of the Identify data and various controller
1111  * register in our nvme_ctrl structure.  This should be called as soon as
1112  * the admin queue is fully up and running.
1113  */
1114 int nvme_init_identify(struct nvme_ctrl *ctrl)
1115 {
1116         struct nvme_id_ctrl *id;
1117         u64 cap;
1118         int ret, page_shift;
1119         u32 max_hw_sectors;
1120
1121         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1122         if (ret) {
1123                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1124                 return ret;
1125         }
1126
1127         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1128         if (ret) {
1129                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1130                 return ret;
1131         }
1132         page_shift = NVME_CAP_MPSMIN(cap) + 12;
1133
1134         if (ctrl->vs >= NVME_VS(1, 1))
1135                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1136
1137         ret = nvme_identify_ctrl(ctrl, &id);
1138         if (ret) {
1139                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1140                 return -EIO;
1141         }
1142
1143         ctrl->vid = le16_to_cpu(id->vid);
1144         ctrl->oncs = le16_to_cpup(&id->oncs);
1145         atomic_set(&ctrl->abort_limit, id->acl + 1);
1146         ctrl->vwc = id->vwc;
1147         ctrl->cntlid = le16_to_cpup(&id->cntlid);
1148         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1149         memcpy(ctrl->model, id->mn, sizeof(id->mn));
1150         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1151         if (id->mdts)
1152                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1153         else
1154                 max_hw_sectors = UINT_MAX;
1155         ctrl->max_hw_sectors =
1156                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1157
1158         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
1159                 unsigned int max_hw_sectors;
1160
1161                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
1162                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
1163                 if (ctrl->max_hw_sectors) {
1164                         ctrl->max_hw_sectors = min(max_hw_sectors,
1165                                                         ctrl->max_hw_sectors);
1166                 } else {
1167                         ctrl->max_hw_sectors = max_hw_sectors;
1168                 }
1169         }
1170
1171         nvme_set_queue_limits(ctrl, ctrl->admin_q);
1172
1173         kfree(id);
1174         return 0;
1175 }
1176 EXPORT_SYMBOL_GPL(nvme_init_identify);
1177
1178 static int nvme_dev_open(struct inode *inode, struct file *file)
1179 {
1180         struct nvme_ctrl *ctrl;
1181         int instance = iminor(inode);
1182         int ret = -ENODEV;
1183
1184         spin_lock(&dev_list_lock);
1185         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1186                 if (ctrl->instance != instance)
1187                         continue;
1188
1189                 if (!ctrl->admin_q) {
1190                         ret = -EWOULDBLOCK;
1191                         break;
1192                 }
1193                 if (!kref_get_unless_zero(&ctrl->kref))
1194                         break;
1195                 file->private_data = ctrl;
1196                 ret = 0;
1197                 break;
1198         }
1199         spin_unlock(&dev_list_lock);
1200
1201         return ret;
1202 }
1203
1204 static int nvme_dev_release(struct inode *inode, struct file *file)
1205 {
1206         nvme_put_ctrl(file->private_data);
1207         return 0;
1208 }
1209
1210 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1211 {
1212         struct nvme_ns *ns;
1213         int ret;
1214
1215         mutex_lock(&ctrl->namespaces_mutex);
1216         if (list_empty(&ctrl->namespaces)) {
1217                 ret = -ENOTTY;
1218                 goto out_unlock;
1219         }
1220
1221         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1222         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1223                 dev_warn(ctrl->device,
1224                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1225                 ret = -EINVAL;
1226                 goto out_unlock;
1227         }
1228
1229         dev_warn(ctrl->device,
1230                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1231         kref_get(&ns->kref);
1232         mutex_unlock(&ctrl->namespaces_mutex);
1233
1234         ret = nvme_user_cmd(ctrl, ns, argp);
1235         nvme_put_ns(ns);
1236         return ret;
1237
1238 out_unlock:
1239         mutex_unlock(&ctrl->namespaces_mutex);
1240         return ret;
1241 }
1242
1243 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1244                 unsigned long arg)
1245 {
1246         struct nvme_ctrl *ctrl = file->private_data;
1247         void __user *argp = (void __user *)arg;
1248
1249         switch (cmd) {
1250         case NVME_IOCTL_ADMIN_CMD:
1251                 return nvme_user_cmd(ctrl, NULL, argp);
1252         case NVME_IOCTL_IO_CMD:
1253                 return nvme_dev_user_cmd(ctrl, argp);
1254         case NVME_IOCTL_RESET:
1255                 dev_warn(ctrl->device, "resetting controller\n");
1256                 return ctrl->ops->reset_ctrl(ctrl);
1257         case NVME_IOCTL_SUBSYS_RESET:
1258                 return nvme_reset_subsystem(ctrl);
1259         case NVME_IOCTL_RESCAN:
1260                 nvme_queue_scan(ctrl);
1261                 return 0;
1262         default:
1263                 return -ENOTTY;
1264         }
1265 }
1266
1267 static const struct file_operations nvme_dev_fops = {
1268         .owner          = THIS_MODULE,
1269         .open           = nvme_dev_open,
1270         .release        = nvme_dev_release,
1271         .unlocked_ioctl = nvme_dev_ioctl,
1272         .compat_ioctl   = nvme_dev_ioctl,
1273 };
1274
1275 static ssize_t nvme_sysfs_reset(struct device *dev,
1276                                 struct device_attribute *attr, const char *buf,
1277                                 size_t count)
1278 {
1279         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1280         int ret;
1281
1282         ret = ctrl->ops->reset_ctrl(ctrl);
1283         if (ret < 0)
1284                 return ret;
1285         return count;
1286 }
1287 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1288
1289 static ssize_t nvme_sysfs_rescan(struct device *dev,
1290                                 struct device_attribute *attr, const char *buf,
1291                                 size_t count)
1292 {
1293         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1294
1295         nvme_queue_scan(ctrl);
1296         return count;
1297 }
1298 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1299
1300 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1301                                                                 char *buf)
1302 {
1303         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1304         struct nvme_ctrl *ctrl = ns->ctrl;
1305         int serial_len = sizeof(ctrl->serial);
1306         int model_len = sizeof(ctrl->model);
1307
1308         if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1309                 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1310
1311         if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1312                 return sprintf(buf, "eui.%8phN\n", ns->eui);
1313
1314         while (ctrl->serial[serial_len - 1] == ' ')
1315                 serial_len--;
1316         while (ctrl->model[model_len - 1] == ' ')
1317                 model_len--;
1318
1319         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1320                 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1321 }
1322 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1323
1324 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1325                                                                 char *buf)
1326 {
1327         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1328         return sprintf(buf, "%pU\n", ns->uuid);
1329 }
1330 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1331
1332 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1333                                                                 char *buf)
1334 {
1335         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1336         return sprintf(buf, "%8phd\n", ns->eui);
1337 }
1338 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1339
1340 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1341                                                                 char *buf)
1342 {
1343         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1344         return sprintf(buf, "%d\n", ns->ns_id);
1345 }
1346 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1347
1348 static struct attribute *nvme_ns_attrs[] = {
1349         &dev_attr_wwid.attr,
1350         &dev_attr_uuid.attr,
1351         &dev_attr_eui.attr,
1352         &dev_attr_nsid.attr,
1353         NULL,
1354 };
1355
1356 static umode_t nvme_attrs_are_visible(struct kobject *kobj,
1357                 struct attribute *a, int n)
1358 {
1359         struct device *dev = container_of(kobj, struct device, kobj);
1360         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1361
1362         if (a == &dev_attr_uuid.attr) {
1363                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1364                         return 0;
1365         }
1366         if (a == &dev_attr_eui.attr) {
1367                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1368                         return 0;
1369         }
1370         return a->mode;
1371 }
1372
1373 static const struct attribute_group nvme_ns_attr_group = {
1374         .attrs          = nvme_ns_attrs,
1375         .is_visible     = nvme_attrs_are_visible,
1376 };
1377
1378 #define nvme_show_str_function(field)                                           \
1379 static ssize_t  field##_show(struct device *dev,                                \
1380                             struct device_attribute *attr, char *buf)           \
1381 {                                                                               \
1382         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1383         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1384 }                                                                               \
1385 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1386
1387 #define nvme_show_int_function(field)                                           \
1388 static ssize_t  field##_show(struct device *dev,                                \
1389                             struct device_attribute *attr, char *buf)           \
1390 {                                                                               \
1391         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1392         return sprintf(buf, "%d\n", ctrl->field);       \
1393 }                                                                               \
1394 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1395
1396 nvme_show_str_function(model);
1397 nvme_show_str_function(serial);
1398 nvme_show_str_function(firmware_rev);
1399 nvme_show_int_function(cntlid);
1400
1401 static struct attribute *nvme_dev_attrs[] = {
1402         &dev_attr_reset_controller.attr,
1403         &dev_attr_rescan_controller.attr,
1404         &dev_attr_model.attr,
1405         &dev_attr_serial.attr,
1406         &dev_attr_firmware_rev.attr,
1407         &dev_attr_cntlid.attr,
1408         NULL
1409 };
1410
1411 static struct attribute_group nvme_dev_attrs_group = {
1412         .attrs = nvme_dev_attrs,
1413 };
1414
1415 static const struct attribute_group *nvme_dev_attr_groups[] = {
1416         &nvme_dev_attrs_group,
1417         NULL,
1418 };
1419
1420 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1421 {
1422         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1423         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1424
1425         return nsa->ns_id - nsb->ns_id;
1426 }
1427
1428 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1429 {
1430         struct nvme_ns *ns;
1431
1432         lockdep_assert_held(&ctrl->namespaces_mutex);
1433
1434         list_for_each_entry(ns, &ctrl->namespaces, list) {
1435                 if (ns->ns_id == nsid)
1436                         return ns;
1437                 if (ns->ns_id > nsid)
1438                         break;
1439         }
1440         return NULL;
1441 }
1442
1443 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1444 {
1445         struct nvme_ns *ns;
1446         struct gendisk *disk;
1447         int node = dev_to_node(ctrl->dev);
1448
1449         lockdep_assert_held(&ctrl->namespaces_mutex);
1450
1451         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1452         if (!ns)
1453                 return;
1454
1455         ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1456         if (ns->instance < 0)
1457                 goto out_free_ns;
1458
1459         ns->queue = blk_mq_init_queue(ctrl->tagset);
1460         if (IS_ERR(ns->queue))
1461                 goto out_release_instance;
1462         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1463         ns->queue->queuedata = ns;
1464         ns->ctrl = ctrl;
1465
1466         disk = alloc_disk_node(0, node);
1467         if (!disk)
1468                 goto out_free_queue;
1469
1470         kref_init(&ns->kref);
1471         ns->ns_id = nsid;
1472         ns->disk = disk;
1473         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1474
1475
1476         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1477         nvme_set_queue_limits(ctrl, ns->queue);
1478
1479         disk->major = nvme_major;
1480         disk->first_minor = 0;
1481         disk->fops = &nvme_fops;
1482         disk->private_data = ns;
1483         disk->queue = ns->queue;
1484         disk->driverfs_dev = ctrl->device;
1485         disk->flags = GENHD_FL_EXT_DEVT;
1486         sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1487
1488         if (nvme_revalidate_disk(ns->disk))
1489                 goto out_free_disk;
1490
1491         list_add_tail_rcu(&ns->list, &ctrl->namespaces);
1492         kref_get(&ctrl->kref);
1493         if (ns->type == NVME_NS_LIGHTNVM)
1494                 return;
1495
1496         add_disk(ns->disk);
1497         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1498                                         &nvme_ns_attr_group))
1499                 pr_warn("%s: failed to create sysfs group for identification\n",
1500                         ns->disk->disk_name);
1501         return;
1502  out_free_disk:
1503         kfree(disk);
1504  out_free_queue:
1505         blk_cleanup_queue(ns->queue);
1506  out_release_instance:
1507         ida_simple_remove(&ctrl->ns_ida, ns->instance);
1508  out_free_ns:
1509         kfree(ns);
1510 }
1511
1512 static void nvme_ns_remove(struct nvme_ns *ns)
1513 {
1514         lockdep_assert_held(&ns->ctrl->namespaces_mutex);
1515
1516         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1517                 return;
1518
1519         if (ns->disk->flags & GENHD_FL_UP) {
1520                 if (blk_get_integrity(ns->disk))
1521                         blk_integrity_unregister(ns->disk);
1522                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1523                                         &nvme_ns_attr_group);
1524                 del_gendisk(ns->disk);
1525                 blk_mq_abort_requeue_list(ns->queue);
1526                 blk_cleanup_queue(ns->queue);
1527         }
1528         list_del_init(&ns->list);
1529         synchronize_rcu();
1530         nvme_put_ns(ns);
1531 }
1532
1533 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1534 {
1535         struct nvme_ns *ns;
1536
1537         ns = nvme_find_ns(ctrl, nsid);
1538         if (ns) {
1539                 if (revalidate_disk(ns->disk))
1540                         nvme_ns_remove(ns);
1541         } else
1542                 nvme_alloc_ns(ctrl, nsid);
1543 }
1544
1545 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1546                                         unsigned nsid)
1547 {
1548         struct nvme_ns *ns, *next;
1549
1550         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1551                 if (ns->ns_id > nsid)
1552                         nvme_ns_remove(ns);
1553         }
1554 }
1555
1556 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1557 {
1558         struct nvme_ns *ns;
1559         __le32 *ns_list;
1560         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1561         int ret = 0;
1562
1563         ns_list = kzalloc(0x1000, GFP_KERNEL);
1564         if (!ns_list)
1565                 return -ENOMEM;
1566
1567         for (i = 0; i < num_lists; i++) {
1568                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1569                 if (ret)
1570                         goto free;
1571
1572                 for (j = 0; j < min(nn, 1024U); j++) {
1573                         nsid = le32_to_cpu(ns_list[j]);
1574                         if (!nsid)
1575                                 goto out;
1576
1577                         nvme_validate_ns(ctrl, nsid);
1578
1579                         while (++prev < nsid) {
1580                                 ns = nvme_find_ns(ctrl, prev);
1581                                 if (ns)
1582                                         nvme_ns_remove(ns);
1583                         }
1584                 }
1585                 nn -= j;
1586         }
1587  out:
1588         nvme_remove_invalid_namespaces(ctrl, prev);
1589  free:
1590         kfree(ns_list);
1591         return ret;
1592 }
1593
1594 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1595 {
1596         unsigned i;
1597
1598         lockdep_assert_held(&ctrl->namespaces_mutex);
1599
1600         for (i = 1; i <= nn; i++)
1601                 nvme_validate_ns(ctrl, i);
1602
1603         nvme_remove_invalid_namespaces(ctrl, nn);
1604 }
1605
1606 static void nvme_scan_work(struct work_struct *work)
1607 {
1608         struct nvme_ctrl *ctrl =
1609                 container_of(work, struct nvme_ctrl, scan_work);
1610         struct nvme_id_ctrl *id;
1611         unsigned nn;
1612
1613         if (ctrl->state != NVME_CTRL_LIVE)
1614                 return;
1615
1616         if (nvme_identify_ctrl(ctrl, &id))
1617                 return;
1618
1619         mutex_lock(&ctrl->namespaces_mutex);
1620         nn = le32_to_cpu(id->nn);
1621         if (ctrl->vs >= NVME_VS(1, 1) &&
1622             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1623                 if (!nvme_scan_ns_list(ctrl, nn))
1624                         goto done;
1625         }
1626         nvme_scan_ns_sequential(ctrl, nn);
1627  done:
1628         list_sort(NULL, &ctrl->namespaces, ns_cmp);
1629         mutex_unlock(&ctrl->namespaces_mutex);
1630         kfree(id);
1631
1632         if (ctrl->ops->post_scan)
1633                 ctrl->ops->post_scan(ctrl);
1634 }
1635
1636 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1637 {
1638         /*
1639          * Do not queue new scan work when a controller is reset during
1640          * removal.
1641          */
1642         if (ctrl->state == NVME_CTRL_LIVE)
1643                 schedule_work(&ctrl->scan_work);
1644 }
1645 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1646
1647 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1648 {
1649         struct nvme_ns *ns, *next;
1650
1651         /*
1652          * The dead states indicates the controller was not gracefully
1653          * disconnected. In that case, we won't be able to flush any data while
1654          * removing the namespaces' disks; fail all the queues now to avoid
1655          * potentially having to clean up the failed sync later.
1656          */
1657         if (ctrl->state == NVME_CTRL_DEAD)
1658                 nvme_kill_queues(ctrl);
1659
1660         mutex_lock(&ctrl->namespaces_mutex);
1661         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1662                 nvme_ns_remove(ns);
1663         mutex_unlock(&ctrl->namespaces_mutex);
1664 }
1665 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1666
1667 static void nvme_async_event_work(struct work_struct *work)
1668 {
1669         struct nvme_ctrl *ctrl =
1670                 container_of(work, struct nvme_ctrl, async_event_work);
1671
1672         spin_lock_irq(&ctrl->lock);
1673         while (ctrl->event_limit > 0) {
1674                 int aer_idx = --ctrl->event_limit;
1675
1676                 spin_unlock_irq(&ctrl->lock);
1677                 ctrl->ops->submit_async_event(ctrl, aer_idx);
1678                 spin_lock_irq(&ctrl->lock);
1679         }
1680         spin_unlock_irq(&ctrl->lock);
1681 }
1682
1683 void nvme_complete_async_event(struct nvme_ctrl *ctrl,
1684                 struct nvme_completion *cqe)
1685 {
1686         u16 status = le16_to_cpu(cqe->status) >> 1;
1687         u32 result = le32_to_cpu(cqe->result);
1688
1689         if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) {
1690                 ++ctrl->event_limit;
1691                 schedule_work(&ctrl->async_event_work);
1692         }
1693
1694         if (status != NVME_SC_SUCCESS)
1695                 return;
1696
1697         switch (result & 0xff07) {
1698         case NVME_AER_NOTICE_NS_CHANGED:
1699                 dev_info(ctrl->device, "rescanning\n");
1700                 nvme_queue_scan(ctrl);
1701                 break;
1702         default:
1703                 dev_warn(ctrl->device, "async event result %08x\n", result);
1704         }
1705 }
1706 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1707
1708 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1709 {
1710         ctrl->event_limit = NVME_NR_AERS;
1711         schedule_work(&ctrl->async_event_work);
1712 }
1713 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1714
1715 static DEFINE_IDA(nvme_instance_ida);
1716
1717 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1718 {
1719         int instance, error;
1720
1721         do {
1722                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1723                         return -ENODEV;
1724
1725                 spin_lock(&dev_list_lock);
1726                 error = ida_get_new(&nvme_instance_ida, &instance);
1727                 spin_unlock(&dev_list_lock);
1728         } while (error == -EAGAIN);
1729
1730         if (error)
1731                 return -ENODEV;
1732
1733         ctrl->instance = instance;
1734         return 0;
1735 }
1736
1737 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1738 {
1739         spin_lock(&dev_list_lock);
1740         ida_remove(&nvme_instance_ida, ctrl->instance);
1741         spin_unlock(&dev_list_lock);
1742 }
1743
1744 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1745 {
1746         flush_work(&ctrl->async_event_work);
1747         flush_work(&ctrl->scan_work);
1748         nvme_remove_namespaces(ctrl);
1749
1750         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1751
1752         spin_lock(&dev_list_lock);
1753         list_del(&ctrl->node);
1754         spin_unlock(&dev_list_lock);
1755 }
1756 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1757
1758 static void nvme_free_ctrl(struct kref *kref)
1759 {
1760         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1761
1762         put_device(ctrl->device);
1763         nvme_release_instance(ctrl);
1764         ida_destroy(&ctrl->ns_ida);
1765
1766         ctrl->ops->free_ctrl(ctrl);
1767 }
1768
1769 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1770 {
1771         kref_put(&ctrl->kref, nvme_free_ctrl);
1772 }
1773 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1774
1775 /*
1776  * Initialize a NVMe controller structures.  This needs to be called during
1777  * earliest initialization so that we have the initialized structured around
1778  * during probing.
1779  */
1780 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1781                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1782 {
1783         int ret;
1784
1785         ctrl->state = NVME_CTRL_NEW;
1786         spin_lock_init(&ctrl->lock);
1787         INIT_LIST_HEAD(&ctrl->namespaces);
1788         mutex_init(&ctrl->namespaces_mutex);
1789         kref_init(&ctrl->kref);
1790         ctrl->dev = dev;
1791         ctrl->ops = ops;
1792         ctrl->quirks = quirks;
1793         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
1794         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
1795
1796         ret = nvme_set_instance(ctrl);
1797         if (ret)
1798                 goto out;
1799
1800         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1801                                 MKDEV(nvme_char_major, ctrl->instance),
1802                                 ctrl, nvme_dev_attr_groups,
1803                                 "nvme%d", ctrl->instance);
1804         if (IS_ERR(ctrl->device)) {
1805                 ret = PTR_ERR(ctrl->device);
1806                 goto out_release_instance;
1807         }
1808         get_device(ctrl->device);
1809         ida_init(&ctrl->ns_ida);
1810
1811         spin_lock(&dev_list_lock);
1812         list_add_tail(&ctrl->node, &nvme_ctrl_list);
1813         spin_unlock(&dev_list_lock);
1814
1815         return 0;
1816 out_release_instance:
1817         nvme_release_instance(ctrl);
1818 out:
1819         return ret;
1820 }
1821 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
1822
1823 /**
1824  * nvme_kill_queues(): Ends all namespace queues
1825  * @ctrl: the dead controller that needs to end
1826  *
1827  * Call this function when the driver determines it is unable to get the
1828  * controller in a state capable of servicing IO.
1829  */
1830 void nvme_kill_queues(struct nvme_ctrl *ctrl)
1831 {
1832         struct nvme_ns *ns;
1833
1834         rcu_read_lock();
1835         list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
1836                 if (!kref_get_unless_zero(&ns->kref))
1837                         continue;
1838
1839                 /*
1840                  * Revalidating a dead namespace sets capacity to 0. This will
1841                  * end buffered writers dirtying pages that can't be synced.
1842                  */
1843                 if (!test_and_set_bit(NVME_NS_DEAD, &ns->flags))
1844                         revalidate_disk(ns->disk);
1845
1846                 blk_set_queue_dying(ns->queue);
1847                 blk_mq_abort_requeue_list(ns->queue);
1848                 blk_mq_start_stopped_hw_queues(ns->queue, true);
1849
1850                 nvme_put_ns(ns);
1851         }
1852         rcu_read_unlock();
1853 }
1854 EXPORT_SYMBOL_GPL(nvme_kill_queues);
1855
1856 void nvme_stop_queues(struct nvme_ctrl *ctrl)
1857 {
1858         struct nvme_ns *ns;
1859
1860         rcu_read_lock();
1861         list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
1862                 spin_lock_irq(ns->queue->queue_lock);
1863                 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1864                 spin_unlock_irq(ns->queue->queue_lock);
1865
1866                 blk_mq_cancel_requeue_work(ns->queue);
1867                 blk_mq_stop_hw_queues(ns->queue);
1868         }
1869         rcu_read_unlock();
1870 }
1871 EXPORT_SYMBOL_GPL(nvme_stop_queues);
1872
1873 void nvme_start_queues(struct nvme_ctrl *ctrl)
1874 {
1875         struct nvme_ns *ns;
1876
1877         rcu_read_lock();
1878         list_for_each_entry_rcu(ns, &ctrl->namespaces, list) {
1879                 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1880                 blk_mq_start_stopped_hw_queues(ns->queue, true);
1881                 blk_mq_kick_requeue_list(ns->queue);
1882         }
1883         rcu_read_unlock();
1884 }
1885 EXPORT_SYMBOL_GPL(nvme_start_queues);
1886
1887 int __init nvme_core_init(void)
1888 {
1889         int result;
1890
1891         result = register_blkdev(nvme_major, "nvme");
1892         if (result < 0)
1893                 return result;
1894         else if (result > 0)
1895                 nvme_major = result;
1896
1897         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1898                                                         &nvme_dev_fops);
1899         if (result < 0)
1900                 goto unregister_blkdev;
1901         else if (result > 0)
1902                 nvme_char_major = result;
1903
1904         nvme_class = class_create(THIS_MODULE, "nvme");
1905         if (IS_ERR(nvme_class)) {
1906                 result = PTR_ERR(nvme_class);
1907                 goto unregister_chrdev;
1908         }
1909
1910         return 0;
1911
1912  unregister_chrdev:
1913         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1914  unregister_blkdev:
1915         unregister_blkdev(nvme_major, "nvme");
1916         return result;
1917 }
1918
1919 void nvme_core_exit(void)
1920 {
1921         class_destroy(nvme_class);
1922         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1923         unregister_blkdev(nvme_major, "nvme");
1924 }
1925
1926 MODULE_LICENSE("GPL");
1927 MODULE_VERSION("1.0");
1928 module_init(nvme_core_init);
1929 module_exit(nvme_core_exit);