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 static int nvme_major;
  37 module_param(nvme_major, int, 0);
  38
  39 static int nvme_char_major;
  40 module_param(nvme_char_major, int, 0);
  41
  42 static LIST_HEAD(nvme_ctrl_list);
  43 DEFINE_SPINLOCK(dev_list_lock);
  44
  45 static struct class *nvme_class;
  46
  47 static void nvme_free_ns(struct kref *kref)
  48 {
  49         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
  50
  51         if (ns->type == NVME_NS_LIGHTNVM)
  52                 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
  53
  54         spin_lock(&dev_list_lock);
  55         ns->disk->private_data = NULL;
  56         spin_unlock(&dev_list_lock);
  57
  58         nvme_put_ctrl(ns->ctrl);
  59         put_disk(ns->disk);
  60         kfree(ns);
  61 }
  62
  63 static void nvme_put_ns(struct nvme_ns *ns)
  64 {
  65         kref_put(&ns->kref, nvme_free_ns);
  66 }
  67
  68 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
  69 {
  70         struct nvme_ns *ns;
  71
  72         spin_lock(&dev_list_lock);
  73         ns = disk->private_data;
  74         if (ns && !kref_get_unless_zero(&ns->kref))
  75                 ns = NULL;
  76         spin_unlock(&dev_list_lock);
  77
  78         return ns;
  79 }
  80
  81 void nvme_requeue_req(struct request *req)
  82 {
  83         unsigned long flags;
  84
  85         blk_mq_requeue_request(req);
  86         spin_lock_irqsave(req->q->queue_lock, flags);
  87         if (!blk_queue_stopped(req->q))
  88                 blk_mq_kick_requeue_list(req->q);
  89         spin_unlock_irqrestore(req->q->queue_lock, flags);
  90 }
  91
  92 struct request *nvme_alloc_request(struct request_queue *q,
  93                 struct nvme_command *cmd, unsigned int flags)
  94 {
  95         bool write = cmd->common.opcode & 1;
  96         struct request *req;
  97
  98         req = blk_mq_alloc_request(q, write, flags);
  99         if (IS_ERR(req))
 100                 return req;
 101
 102         req->cmd_type = REQ_TYPE_DRV_PRIV;
 103         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 104         req->__data_len = 0;
 105         req->__sector = (sector_t) -1;
 106         req->bio = req->biotail = NULL;
 107
 108         req->cmd = (unsigned char *)cmd;
 109         req->cmd_len = sizeof(struct nvme_command);
 110         req->special = (void *)0;
 111
 112         return req;
 113 }
 114
 115 /*
 116  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 117  * if the result is positive, it's an NVM Express status code
 118  */
 119 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 120                 void *buffer, unsigned bufflen, u32 *result, unsigned timeout)
 121 {
 122         struct request *req;
 123         int ret;
 124
 125         req = nvme_alloc_request(q, cmd, 0);
 126         if (IS_ERR(req))
 127                 return PTR_ERR(req);
 128
 129         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 130
 131         if (buffer && bufflen) {
 132                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 133                 if (ret)
 134                         goto out;
 135         }
 136
 137         blk_execute_rq(req->q, NULL, req, 0);
 138         if (result)
 139                 *result = (u32)(uintptr_t)req->special;
 140         ret = req->errors;
 141  out:
 142         blk_mq_free_request(req);
 143         return ret;
 144 }
 145
 146 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 147                 void *buffer, unsigned bufflen)
 148 {
 149         return __nvme_submit_sync_cmd(q, cmd, buffer, bufflen, NULL, 0);
 150 }
 151
 152 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 153                 void __user *ubuffer, unsigned bufflen,
 154                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 155                 u32 *result, unsigned timeout)
 156 {
 157         bool write = cmd->common.opcode & 1;
 158         struct nvme_ns *ns = q->queuedata;
 159         struct gendisk *disk = ns ? ns->disk : NULL;
 160         struct request *req;
 161         struct bio *bio = NULL;
 162         void *meta = NULL;
 163         int ret;
 164
 165         req = nvme_alloc_request(q, cmd, 0);
 166         if (IS_ERR(req))
 167                 return PTR_ERR(req);
 168
 169         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 170
 171         if (ubuffer && bufflen) {
 172                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 173                                 GFP_KERNEL);
 174                 if (ret)
 175                         goto out;
 176                 bio = req->bio;
 177
 178                 if (!disk)
 179                         goto submit;
 180                 bio->bi_bdev = bdget_disk(disk, 0);
 181                 if (!bio->bi_bdev) {
 182                         ret = -ENODEV;
 183                         goto out_unmap;
 184                 }
 185
 186                 if (meta_buffer) {
 187                         struct bio_integrity_payload *bip;
 188
 189                         meta = kmalloc(meta_len, GFP_KERNEL);
 190                         if (!meta) {
 191                                 ret = -ENOMEM;
 192                                 goto out_unmap;
 193                         }
 194
 195                         if (write) {
 196                                 if (copy_from_user(meta, meta_buffer,
 197                                                 meta_len)) {
 198                                         ret = -EFAULT;
 199                                         goto out_free_meta;
 200                                 }
 201                         }
 202
 203                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 204                         if (IS_ERR(bip)) {
 205                                 ret = PTR_ERR(bip);
 206                                 goto out_free_meta;
 207                         }
 208
 209                         bip->bip_iter.bi_size = meta_len;
 210                         bip->bip_iter.bi_sector = meta_seed;
 211
 212                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 213                                         meta_len, offset_in_page(meta));
 214                         if (ret != meta_len) {
 215                                 ret = -ENOMEM;
 216                                 goto out_free_meta;
 217                         }
 218                 }
 219         }
 220  submit:
 221         blk_execute_rq(req->q, disk, req, 0);
 222         ret = req->errors;
 223         if (result)
 224                 *result = (u32)(uintptr_t)req->special;
 225         if (meta && !ret && !write) {
 226                 if (copy_to_user(meta_buffer, meta, meta_len))
 227                         ret = -EFAULT;
 228         }
 229  out_free_meta:
 230         kfree(meta);
 231  out_unmap:
 232         if (bio) {
 233                 if (disk && bio->bi_bdev)
 234                         bdput(bio->bi_bdev);
 235                 blk_rq_unmap_user(bio);
 236         }
 237  out:
 238         blk_mq_free_request(req);
 239         return ret;
 240 }
 241
 242 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 243                 void __user *ubuffer, unsigned bufflen, u32 *result,
 244                 unsigned timeout)
 245 {
 246         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 247                         result, timeout);
 248 }
 249
 250 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 251 {
 252         struct nvme_command c = { };
 253         int error;
 254
 255         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 256         c.identify.opcode = nvme_admin_identify;
 257         c.identify.cns = cpu_to_le32(1);
 258
 259         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 260         if (!*id)
 261                 return -ENOMEM;
 262
 263         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 264                         sizeof(struct nvme_id_ctrl));
 265         if (error)
 266                 kfree(*id);
 267         return error;
 268 }
 269
 270 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 271 {
 272         struct nvme_command c = { };
 273
 274         c.identify.opcode = nvme_admin_identify;
 275         c.identify.cns = cpu_to_le32(2);
 276         c.identify.nsid = cpu_to_le32(nsid);
 277         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 278 }
 279
 280 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 281                 struct nvme_id_ns **id)
 282 {
 283         struct nvme_command c = { };
 284         int error;
 285
 286         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 287         c.identify.opcode = nvme_admin_identify,
 288         c.identify.nsid = cpu_to_le32(nsid),
 289
 290         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 291         if (!*id)
 292                 return -ENOMEM;
 293
 294         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 295                         sizeof(struct nvme_id_ns));
 296         if (error)
 297                 kfree(*id);
 298         return error;
 299 }
 300
 301 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 302                                         dma_addr_t dma_addr, u32 *result)
 303 {
 304         struct nvme_command c;
 305
 306         memset(&c, 0, sizeof(c));
 307         c.features.opcode = nvme_admin_get_features;
 308         c.features.nsid = cpu_to_le32(nsid);
 309         c.features.prp1 = cpu_to_le64(dma_addr);
 310         c.features.fid = cpu_to_le32(fid);
 311
 312         return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
 313 }
 314
 315 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 316                                         dma_addr_t dma_addr, u32 *result)
 317 {
 318         struct nvme_command c;
 319
 320         memset(&c, 0, sizeof(c));
 321         c.features.opcode = nvme_admin_set_features;
 322         c.features.prp1 = cpu_to_le64(dma_addr);
 323         c.features.fid = cpu_to_le32(fid);
 324         c.features.dword11 = cpu_to_le32(dword11);
 325
 326         return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
 327 }
 328
 329 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 330 {
 331         struct nvme_command c = { };
 332         int error;
 333
 334         c.common.opcode = nvme_admin_get_log_page,
 335         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 336         c.common.cdw10[0] = cpu_to_le32(
 337                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 338                          NVME_LOG_SMART),
 339
 340         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 341         if (!*log)
 342                 return -ENOMEM;
 343
 344         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 345                         sizeof(struct nvme_smart_log));
 346         if (error)
 347                 kfree(*log);
 348         return error;
 349 }
 350
 351 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 352 {
 353         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 354         u32 result;
 355         int status, nr_io_queues;
 356
 357         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
 358                         &result);
 359         if (status)
 360                 return status;
 361
 362         nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 363         *count = min(*count, nr_io_queues);
 364         return 0;
 365 }
 366
 367 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 368 {
 369         struct nvme_user_io io;
 370         struct nvme_command c;
 371         unsigned length, meta_len;
 372         void __user *metadata;
 373
 374         if (copy_from_user(&io, uio, sizeof(io)))
 375                 return -EFAULT;
 376
 377         switch (io.opcode) {
 378         case nvme_cmd_write:
 379         case nvme_cmd_read:
 380         case nvme_cmd_compare:
 381                 break;
 382         default:
 383                 return -EINVAL;
 384         }
 385
 386         length = (io.nblocks + 1) << ns->lba_shift;
 387         meta_len = (io.nblocks + 1) * ns->ms;
 388         metadata = (void __user *)(uintptr_t)io.metadata;
 389
 390         if (ns->ext) {
 391                 length += meta_len;
 392                 meta_len = 0;
 393         } else if (meta_len) {
 394                 if ((io.metadata & 3) || !io.metadata)
 395                         return -EINVAL;
 396         }
 397
 398         memset(&c, 0, sizeof(c));
 399         c.rw.opcode = io.opcode;
 400         c.rw.flags = io.flags;
 401         c.rw.nsid = cpu_to_le32(ns->ns_id);
 402         c.rw.slba = cpu_to_le64(io.slba);
 403         c.rw.length = cpu_to_le16(io.nblocks);
 404         c.rw.control = cpu_to_le16(io.control);
 405         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 406         c.rw.reftag = cpu_to_le32(io.reftag);
 407         c.rw.apptag = cpu_to_le16(io.apptag);
 408         c.rw.appmask = cpu_to_le16(io.appmask);
 409
 410         return __nvme_submit_user_cmd(ns->queue, &c,
 411                         (void __user *)(uintptr_t)io.addr, length,
 412                         metadata, meta_len, io.slba, NULL, 0);
 413 }
 414
 415 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 416                         struct nvme_passthru_cmd __user *ucmd)
 417 {
 418         struct nvme_passthru_cmd cmd;
 419         struct nvme_command c;
 420         unsigned timeout = 0;
 421         int status;
 422
 423         if (!capable(CAP_SYS_ADMIN))
 424                 return -EACCES;
 425         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 426                 return -EFAULT;
 427
 428         memset(&c, 0, sizeof(c));
 429         c.common.opcode = cmd.opcode;
 430         c.common.flags = cmd.flags;
 431         c.common.nsid = cpu_to_le32(cmd.nsid);
 432         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 433         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 434         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 435         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 436         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 437         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 438         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 439         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 440
 441         if (cmd.timeout_ms)
 442                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 443
 444         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 445                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 446                         &cmd.result, timeout);
 447         if (status >= 0) {
 448                 if (put_user(cmd.result, &ucmd->result))
 449                         return -EFAULT;
 450         }
 451
 452         return status;
 453 }
 454
 455 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 456                 unsigned int cmd, unsigned long arg)
 457 {
 458         struct nvme_ns *ns = bdev->bd_disk->private_data;
 459
 460         switch (cmd) {
 461         case NVME_IOCTL_ID:
 462                 force_successful_syscall_return();
 463                 return ns->ns_id;
 464         case NVME_IOCTL_ADMIN_CMD:
 465                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 466         case NVME_IOCTL_IO_CMD:
 467                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 468         case NVME_IOCTL_SUBMIT_IO:
 469                 return nvme_submit_io(ns, (void __user *)arg);
 470         case SG_GET_VERSION_NUM:
 471                 return nvme_sg_get_version_num((void __user *)arg);
 472         case SG_IO:
 473                 return nvme_sg_io(ns, (void __user *)arg);
 474         default:
 475                 return -ENOTTY;
 476         }
 477 }
 478
 479 #ifdef CONFIG_COMPAT
 480 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 481                         unsigned int cmd, unsigned long arg)
 482 {
 483         switch (cmd) {
 484         case SG_IO:
 485                 return -ENOIOCTLCMD;
 486         }
 487         return nvme_ioctl(bdev, mode, cmd, arg);
 488 }
 489 #else
 490 #define nvme_compat_ioctl       NULL
 491 #endif
 492
 493 static int nvme_open(struct block_device *bdev, fmode_t mode)
 494 {
 495         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 496 }
 497
 498 static void nvme_release(struct gendisk *disk, fmode_t mode)
 499 {
 500         nvme_put_ns(disk->private_data);
 501 }
 502
 503 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 504 {
 505         /* some standard values */
 506         geo->heads = 1 << 6;
 507         geo->sectors = 1 << 5;
 508         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 509         return 0;
 510 }
 511
 512 #ifdef CONFIG_BLK_DEV_INTEGRITY
 513 static void nvme_init_integrity(struct nvme_ns *ns)
 514 {
 515         struct blk_integrity integrity;
 516
 517         switch (ns->pi_type) {
 518         case NVME_NS_DPS_PI_TYPE3:
 519                 integrity.profile = &t10_pi_type3_crc;
 520                 break;
 521         case NVME_NS_DPS_PI_TYPE1:
 522         case NVME_NS_DPS_PI_TYPE2:
 523                 integrity.profile = &t10_pi_type1_crc;
 524                 break;
 525         default:
 526                 integrity.profile = NULL;
 527                 break;
 528         }
 529         integrity.tuple_size = ns->ms;
 530         blk_integrity_register(ns->disk, &integrity);
 531         blk_queue_max_integrity_segments(ns->queue, 1);
 532 }
 533 #else
 534 static void nvme_init_integrity(struct nvme_ns *ns)
 535 {
 536 }
 537 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 538
 539 static void nvme_config_discard(struct nvme_ns *ns)
 540 {
 541         u32 logical_block_size = queue_logical_block_size(ns->queue);
 542         ns->queue->limits.discard_zeroes_data = 0;
 543         ns->queue->limits.discard_alignment = logical_block_size;
 544         ns->queue->limits.discard_granularity = logical_block_size;
 545         blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
 546         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 547 }
 548
 549 static int nvme_revalidate_disk(struct gendisk *disk)
 550 {
 551         struct nvme_ns *ns = disk->private_data;
 552         struct nvme_id_ns *id;
 553         u8 lbaf, pi_type;
 554         u16 old_ms;
 555         unsigned short bs;
 556
 557         if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
 558                 dev_warn(ns->ctrl->dev, "%s: Identify failure nvme%dn%d\n",
 559                                 __func__, ns->ctrl->instance, ns->ns_id);
 560                 return -ENODEV;
 561         }
 562         if (id->ncap == 0) {
 563                 kfree(id);
 564                 return -ENODEV;
 565         }
 566
 567         if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
 568                 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
 569                         dev_warn(ns->ctrl->dev,
 570                                 "%s: LightNVM init failure\n", __func__);
 571                         kfree(id);
 572                         return -ENODEV;
 573                 }
 574                 ns->type = NVME_NS_LIGHTNVM;
 575         }
 576
 577         if (ns->ctrl->vs >= NVME_VS(1, 1))
 578                 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
 579         if (ns->ctrl->vs >= NVME_VS(1, 2))
 580                 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
 581
 582         old_ms = ns->ms;
 583         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 584         ns->lba_shift = id->lbaf[lbaf].ds;
 585         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 586         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 587
 588         /*
 589          * If identify namespace failed, use default 512 byte block size so
 590          * block layer can use before failing read/write for 0 capacity.
 591          */
 592         if (ns->lba_shift == 0)
 593                 ns->lba_shift = 9;
 594         bs = 1 << ns->lba_shift;
 595         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 596         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 597                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 598
 599         blk_mq_freeze_queue(disk->queue);
 600         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 601                                 ns->ms != old_ms ||
 602                                 bs != queue_logical_block_size(disk->queue) ||
 603                                 (ns->ms && ns->ext)))
 604                 blk_integrity_unregister(disk);
 605
 606         ns->pi_type = pi_type;
 607         blk_queue_logical_block_size(ns->queue, bs);
 608
 609         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 610                 nvme_init_integrity(ns);
 611         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 612                 set_capacity(disk, 0);
 613         else
 614                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 615
 616         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 617                 nvme_config_discard(ns);
 618         blk_mq_unfreeze_queue(disk->queue);
 619
 620         kfree(id);
 621         return 0;
 622 }
 623
 624 static char nvme_pr_type(enum pr_type type)
 625 {
 626         switch (type) {
 627         case PR_WRITE_EXCLUSIVE:
 628                 return 1;
 629         case PR_EXCLUSIVE_ACCESS:
 630                 return 2;
 631         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 632                 return 3;
 633         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 634                 return 4;
 635         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 636                 return 5;
 637         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 638                 return 6;
 639         default:
 640                 return 0;
 641         }
 642 };
 643
 644 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 645                                 u64 key, u64 sa_key, u8 op)
 646 {
 647         struct nvme_ns *ns = bdev->bd_disk->private_data;
 648         struct nvme_command c;
 649         u8 data[16] = { 0, };
 650
 651         put_unaligned_le64(key, &data[0]);
 652         put_unaligned_le64(sa_key, &data[8]);
 653
 654         memset(&c, 0, sizeof(c));
 655         c.common.opcode = op;
 656         c.common.nsid = cpu_to_le32(ns->ns_id);
 657         c.common.cdw10[0] = cpu_to_le32(cdw10);
 658
 659         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
 660 }
 661
 662 static int nvme_pr_register(struct block_device *bdev, u64 old,
 663                 u64 new, unsigned flags)
 664 {
 665         u32 cdw10;
 666
 667         if (flags & ~PR_FL_IGNORE_KEY)
 668                 return -EOPNOTSUPP;
 669
 670         cdw10 = old ? 2 : 0;
 671         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
 672         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
 673         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
 674 }
 675
 676 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
 677                 enum pr_type type, unsigned flags)
 678 {
 679         u32 cdw10;
 680
 681         if (flags & ~PR_FL_IGNORE_KEY)
 682                 return -EOPNOTSUPP;
 683
 684         cdw10 = nvme_pr_type(type) << 8;
 685         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
 686         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
 687 }
 688
 689 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
 690                 enum pr_type type, bool abort)
 691 {
 692         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
 693         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
 694 }
 695
 696 static int nvme_pr_clear(struct block_device *bdev, u64 key)
 697 {
 698         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
 699         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
 700 }
 701
 702 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
 703 {
 704         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
 705         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
 706 }
 707
 708 static const struct pr_ops nvme_pr_ops = {
 709         .pr_register    = nvme_pr_register,
 710         .pr_reserve     = nvme_pr_reserve,
 711         .pr_release     = nvme_pr_release,
 712         .pr_preempt     = nvme_pr_preempt,
 713         .pr_clear       = nvme_pr_clear,
 714 };
 715
 716 static const struct block_device_operations nvme_fops = {
 717         .owner          = THIS_MODULE,
 718         .ioctl          = nvme_ioctl,
 719         .compat_ioctl   = nvme_compat_ioctl,
 720         .open           = nvme_open,
 721         .release        = nvme_release,
 722         .getgeo         = nvme_getgeo,
 723         .revalidate_disk= nvme_revalidate_disk,
 724         .pr_ops         = &nvme_pr_ops,
 725 };
 726
 727 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
 728 {
 729         unsigned long timeout =
 730                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
 731         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
 732         int ret;
 733
 734         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 735                 if ((csts & NVME_CSTS_RDY) == bit)
 736                         break;
 737
 738                 msleep(100);
 739                 if (fatal_signal_pending(current))
 740                         return -EINTR;
 741                 if (time_after(jiffies, timeout)) {
 742                         dev_err(ctrl->dev,
 743                                 "Device not ready; aborting %s\n", enabled ?
 744                                                 "initialisation" : "reset");
 745                         return -ENODEV;
 746                 }
 747         }
 748
 749         return ret;
 750 }
 751
 752 /*
 753  * If the device has been passed off to us in an enabled state, just clear
 754  * the enabled bit.  The spec says we should set the 'shutdown notification
 755  * bits', but doing so may cause the device to complete commands to the
 756  * admin queue ... and we don't know what memory that might be pointing at!
 757  */
 758 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 759 {
 760         int ret;
 761
 762         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 763         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
 764
 765         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 766         if (ret)
 767                 return ret;
 768         return nvme_wait_ready(ctrl, cap, false);
 769 }
 770
 771 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 772 {
 773         /*
 774          * Default to a 4K page size, with the intention to update this
 775          * path in the future to accomodate architectures with differing
 776          * kernel and IO page sizes.
 777          */
 778         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
 779         int ret;
 780
 781         if (page_shift < dev_page_min) {
 782                 dev_err(ctrl->dev,
 783                         "Minimum device page size %u too large for host (%u)\n",
 784                         1 << dev_page_min, 1 << page_shift);
 785                 return -ENODEV;
 786         }
 787
 788         ctrl->page_size = 1 << page_shift;
 789
 790         ctrl->ctrl_config = NVME_CC_CSS_NVM;
 791         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
 792         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
 793         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
 794         ctrl->ctrl_config |= NVME_CC_ENABLE;
 795
 796         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 797         if (ret)
 798                 return ret;
 799         return nvme_wait_ready(ctrl, cap, true);
 800 }
 801
 802 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
 803 {
 804         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
 805         u32 csts;
 806         int ret;
 807
 808         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 809         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
 810
 811         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 812         if (ret)
 813                 return ret;
 814
 815         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 816                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
 817                         break;
 818
 819                 msleep(100);
 820                 if (fatal_signal_pending(current))
 821                         return -EINTR;
 822                 if (time_after(jiffies, timeout)) {
 823                         dev_err(ctrl->dev,
 824                                 "Device shutdown incomplete; abort shutdown\n");
 825                         return -ENODEV;
 826                 }
 827         }
 828
 829         return ret;
 830 }
 831
 832 /*
 833  * Initialize the cached copies of the Identify data and various controller
 834  * register in our nvme_ctrl structure.  This should be called as soon as
 835  * the admin queue is fully up and running.
 836  */
 837 int nvme_init_identify(struct nvme_ctrl *ctrl)
 838 {
 839         struct nvme_id_ctrl *id;
 840         u64 cap;
 841         int ret, page_shift;
 842
 843         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
 844         if (ret) {
 845                 dev_err(ctrl->dev, "Reading VS failed (%d)\n", ret);
 846                 return ret;
 847         }
 848
 849         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
 850         if (ret) {
 851                 dev_err(ctrl->dev, "Reading CAP failed (%d)\n", ret);
 852                 return ret;
 853         }
 854         page_shift = NVME_CAP_MPSMIN(cap) + 12;
 855
 856         if (ctrl->vs >= NVME_VS(1, 1))
 857                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
 858
 859         ret = nvme_identify_ctrl(ctrl, &id);
 860         if (ret) {
 861                 dev_err(ctrl->dev, "Identify Controller failed (%d)\n", ret);
 862                 return -EIO;
 863         }
 864
 865         ctrl->oncs = le16_to_cpup(&id->oncs);
 866         atomic_set(&ctrl->abort_limit, id->acl + 1);
 867         ctrl->vwc = id->vwc;
 868         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
 869         memcpy(ctrl->model, id->mn, sizeof(id->mn));
 870         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
 871         if (id->mdts)
 872                 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
 873         else
 874                 ctrl->max_hw_sectors = UINT_MAX;
 875
 876         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
 877                 unsigned int max_hw_sectors;
 878
 879                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
 880                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
 881                 if (ctrl->max_hw_sectors) {
 882                         ctrl->max_hw_sectors = min(max_hw_sectors,
 883                                                         ctrl->max_hw_sectors);
 884                 } else {
 885                         ctrl->max_hw_sectors = max_hw_sectors;
 886                 }
 887         }
 888
 889         kfree(id);
 890         return 0;
 891 }
 892
 893 static int nvme_dev_open(struct inode *inode, struct file *file)
 894 {
 895         struct nvme_ctrl *ctrl;
 896         int instance = iminor(inode);
 897         int ret = -ENODEV;
 898
 899         spin_lock(&dev_list_lock);
 900         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
 901                 if (ctrl->instance != instance)
 902                         continue;
 903
 904                 if (!ctrl->admin_q) {
 905                         ret = -EWOULDBLOCK;
 906                         break;
 907                 }
 908                 if (!kref_get_unless_zero(&ctrl->kref))
 909                         break;
 910                 file->private_data = ctrl;
 911                 ret = 0;
 912                 break;
 913         }
 914         spin_unlock(&dev_list_lock);
 915
 916         return ret;
 917 }
 918
 919 static int nvme_dev_release(struct inode *inode, struct file *file)
 920 {
 921         nvme_put_ctrl(file->private_data);
 922         return 0;
 923 }
 924
 925 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
 926 {
 927         struct nvme_ns *ns;
 928         int ret;
 929
 930         mutex_lock(&ctrl->namespaces_mutex);
 931         if (list_empty(&ctrl->namespaces)) {
 932                 ret = -ENOTTY;
 933                 goto out_unlock;
 934         }
 935
 936         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
 937         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
 938                 dev_warn(ctrl->dev,
 939                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
 940                 ret = -EINVAL;
 941                 goto out_unlock;
 942         }
 943
 944         dev_warn(ctrl->dev,
 945                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
 946         kref_get(&ns->kref);
 947         mutex_unlock(&ctrl->namespaces_mutex);
 948
 949         ret = nvme_user_cmd(ctrl, ns, argp);
 950         nvme_put_ns(ns);
 951         return ret;
 952
 953 out_unlock:
 954         mutex_unlock(&ctrl->namespaces_mutex);
 955         return ret;
 956 }
 957
 958 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
 959                 unsigned long arg)
 960 {
 961         struct nvme_ctrl *ctrl = file->private_data;
 962         void __user *argp = (void __user *)arg;
 963
 964         switch (cmd) {
 965         case NVME_IOCTL_ADMIN_CMD:
 966                 return nvme_user_cmd(ctrl, NULL, argp);
 967         case NVME_IOCTL_IO_CMD:
 968                 return nvme_dev_user_cmd(ctrl, argp);
 969         case NVME_IOCTL_RESET:
 970                 dev_warn(ctrl->dev, "resetting controller\n");
 971                 return ctrl->ops->reset_ctrl(ctrl);
 972         case NVME_IOCTL_SUBSYS_RESET:
 973                 return nvme_reset_subsystem(ctrl);
 974         default:
 975                 return -ENOTTY;
 976         }
 977 }
 978
 979 static const struct file_operations nvme_dev_fops = {
 980         .owner          = THIS_MODULE,
 981         .open           = nvme_dev_open,
 982         .release        = nvme_dev_release,
 983         .unlocked_ioctl = nvme_dev_ioctl,
 984         .compat_ioctl   = nvme_dev_ioctl,
 985 };
 986
 987 static ssize_t nvme_sysfs_reset(struct device *dev,
 988                                 struct device_attribute *attr, const char *buf,
 989                                 size_t count)
 990 {
 991         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
 992         int ret;
 993
 994         ret = ctrl->ops->reset_ctrl(ctrl);
 995         if (ret < 0)
 996                 return ret;
 997         return count;
 998 }
 999 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1000
1001 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1002                                                                 char *buf)
1003 {
1004         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1005         return sprintf(buf, "%pU\n", ns->uuid);
1006 }
1007 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1008
1009 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1010                                                                 char *buf)
1011 {
1012         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1013         return sprintf(buf, "%8phd\n", ns->eui);
1014 }
1015 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1016
1017 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1018                                                                 char *buf)
1019 {
1020         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1021         return sprintf(buf, "%d\n", ns->ns_id);
1022 }
1023 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1024
1025 static struct attribute *nvme_ns_attrs[] = {
1026         &dev_attr_uuid.attr,
1027         &dev_attr_eui.attr,
1028         &dev_attr_nsid.attr,
1029         NULL,
1030 };
1031
1032 static umode_t nvme_attrs_are_visible(struct kobject *kobj,
1033                 struct attribute *a, int n)
1034 {
1035         struct device *dev = container_of(kobj, struct device, kobj);
1036         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1037
1038         if (a == &dev_attr_uuid.attr) {
1039                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1040                         return 0;
1041         }
1042         if (a == &dev_attr_eui.attr) {
1043                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1044                         return 0;
1045         }
1046         return a->mode;
1047 }
1048
1049 static const struct attribute_group nvme_ns_attr_group = {
1050         .attrs          = nvme_ns_attrs,
1051         .is_visible     = nvme_attrs_are_visible,
1052 };
1053
1054 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1055 {
1056         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1057         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1058
1059         return nsa->ns_id - nsb->ns_id;
1060 }
1061
1062 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1063 {
1064         struct nvme_ns *ns;
1065
1066         lockdep_assert_held(&ctrl->namespaces_mutex);
1067
1068         list_for_each_entry(ns, &ctrl->namespaces, list) {
1069                 if (ns->ns_id == nsid)
1070                         return ns;
1071                 if (ns->ns_id > nsid)
1072                         break;
1073         }
1074         return NULL;
1075 }
1076
1077 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1078 {
1079         struct nvme_ns *ns;
1080         struct gendisk *disk;
1081         int node = dev_to_node(ctrl->dev);
1082
1083         lockdep_assert_held(&ctrl->namespaces_mutex);
1084
1085         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1086         if (!ns)
1087                 return;
1088
1089         ns->queue = blk_mq_init_queue(ctrl->tagset);
1090         if (IS_ERR(ns->queue))
1091                 goto out_free_ns;
1092         queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
1093         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1094         ns->queue->queuedata = ns;
1095         ns->ctrl = ctrl;
1096
1097         disk = alloc_disk_node(0, node);
1098         if (!disk)
1099                 goto out_free_queue;
1100
1101         kref_init(&ns->kref);
1102         ns->ns_id = nsid;
1103         ns->disk = disk;
1104         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1105
1106         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1107         if (ctrl->max_hw_sectors) {
1108                 blk_queue_max_hw_sectors(ns->queue, ctrl->max_hw_sectors);
1109                 blk_queue_max_segments(ns->queue,
1110                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1);
1111         }
1112         if (ctrl->stripe_size)
1113                 blk_queue_chunk_sectors(ns->queue, ctrl->stripe_size >> 9);
1114         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1115                 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
1116         blk_queue_virt_boundary(ns->queue, ctrl->page_size - 1);
1117
1118         disk->major = nvme_major;
1119         disk->first_minor = 0;
1120         disk->fops = &nvme_fops;
1121         disk->private_data = ns;
1122         disk->queue = ns->queue;
1123         disk->driverfs_dev = ctrl->device;
1124         disk->flags = GENHD_FL_EXT_DEVT;
1125         sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, nsid);
1126
1127         if (nvme_revalidate_disk(ns->disk))
1128                 goto out_free_disk;
1129
1130         list_add_tail(&ns->list, &ctrl->namespaces);
1131         kref_get(&ctrl->kref);
1132         if (ns->type == NVME_NS_LIGHTNVM)
1133                 return;
1134
1135         add_disk(ns->disk);
1136         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1137                                         &nvme_ns_attr_group))
1138                 pr_warn("%s: failed to create sysfs group for identification\n",
1139                         ns->disk->disk_name);
1140         return;
1141  out_free_disk:
1142         kfree(disk);
1143  out_free_queue:
1144         blk_cleanup_queue(ns->queue);
1145  out_free_ns:
1146         kfree(ns);
1147 }
1148
1149 static void nvme_ns_remove(struct nvme_ns *ns)
1150 {
1151         bool kill = nvme_io_incapable(ns->ctrl) &&
1152                         !blk_queue_dying(ns->queue);
1153
1154         lockdep_assert_held(&ns->ctrl->namespaces_mutex);
1155
1156         if (kill)
1157                 blk_set_queue_dying(ns->queue);
1158         if (ns->disk->flags & GENHD_FL_UP) {
1159                 if (blk_get_integrity(ns->disk))
1160                         blk_integrity_unregister(ns->disk);
1161                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1162                                         &nvme_ns_attr_group);
1163                 del_gendisk(ns->disk);
1164         }
1165         if (kill || !blk_queue_dying(ns->queue)) {
1166                 blk_mq_abort_requeue_list(ns->queue);
1167                 blk_cleanup_queue(ns->queue);
1168         }
1169         list_del_init(&ns->list);
1170         nvme_put_ns(ns);
1171 }
1172
1173 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1174 {
1175         struct nvme_ns *ns;
1176
1177         ns = nvme_find_ns(ctrl, nsid);
1178         if (ns) {
1179                 if (revalidate_disk(ns->disk))
1180                         nvme_ns_remove(ns);
1181         } else
1182                 nvme_alloc_ns(ctrl, nsid);
1183 }
1184
1185 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1186 {
1187         struct nvme_ns *ns;
1188         __le32 *ns_list;
1189         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1190         int ret = 0;
1191
1192         ns_list = kzalloc(0x1000, GFP_KERNEL);
1193         if (!ns_list)
1194                 return -ENOMEM;
1195
1196         for (i = 0; i < num_lists; i++) {
1197                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1198                 if (ret)
1199                         goto out;
1200
1201                 for (j = 0; j < min(nn, 1024U); j++) {
1202                         nsid = le32_to_cpu(ns_list[j]);
1203                         if (!nsid)
1204                                 goto out;
1205
1206                         nvme_validate_ns(ctrl, nsid);
1207
1208                         while (++prev < nsid) {
1209                                 ns = nvme_find_ns(ctrl, prev);
1210                                 if (ns)
1211                                         nvme_ns_remove(ns);
1212                         }
1213                 }
1214                 nn -= j;
1215         }
1216  out:
1217         kfree(ns_list);
1218         return ret;
1219 }
1220
1221 static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn)
1222 {
1223         struct nvme_ns *ns, *next;
1224         unsigned i;
1225
1226         lockdep_assert_held(&ctrl->namespaces_mutex);
1227
1228         for (i = 1; i <= nn; i++)
1229                 nvme_validate_ns(ctrl, i);
1230
1231         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1232                 if (ns->ns_id > nn)
1233                         nvme_ns_remove(ns);
1234         }
1235 }
1236
1237 void nvme_scan_namespaces(struct nvme_ctrl *ctrl)
1238 {
1239         struct nvme_id_ctrl *id;
1240         unsigned nn;
1241
1242         if (nvme_identify_ctrl(ctrl, &id))
1243                 return;
1244
1245         mutex_lock(&ctrl->namespaces_mutex);
1246         nn = le32_to_cpu(id->nn);
1247         if (ctrl->vs >= NVME_VS(1, 1) &&
1248             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1249                 if (!nvme_scan_ns_list(ctrl, nn))
1250                         goto done;
1251         }
1252         __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn));
1253  done:
1254         list_sort(NULL, &ctrl->namespaces, ns_cmp);
1255         mutex_unlock(&ctrl->namespaces_mutex);
1256         kfree(id);
1257 }
1258
1259 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1260 {
1261         struct nvme_ns *ns, *next;
1262
1263         mutex_lock(&ctrl->namespaces_mutex);
1264         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1265                 nvme_ns_remove(ns);
1266         mutex_unlock(&ctrl->namespaces_mutex);
1267 }
1268
1269 static DEFINE_IDA(nvme_instance_ida);
1270
1271 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1272 {
1273         int instance, error;
1274
1275         do {
1276                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1277                         return -ENODEV;
1278
1279                 spin_lock(&dev_list_lock);
1280                 error = ida_get_new(&nvme_instance_ida, &instance);
1281                 spin_unlock(&dev_list_lock);
1282         } while (error == -EAGAIN);
1283
1284         if (error)
1285                 return -ENODEV;
1286
1287         ctrl->instance = instance;
1288         return 0;
1289 }
1290
1291 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1292 {
1293         spin_lock(&dev_list_lock);
1294         ida_remove(&nvme_instance_ida, ctrl->instance);
1295         spin_unlock(&dev_list_lock);
1296 }
1297
1298 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1299  {
1300         device_remove_file(ctrl->device, &dev_attr_reset_controller);
1301         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1302
1303         spin_lock(&dev_list_lock);
1304         list_del(&ctrl->node);
1305         spin_unlock(&dev_list_lock);
1306 }
1307
1308 static void nvme_free_ctrl(struct kref *kref)
1309 {
1310         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1311
1312         put_device(ctrl->device);
1313         nvme_release_instance(ctrl);
1314
1315         ctrl->ops->free_ctrl(ctrl);
1316 }
1317
1318 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1319 {
1320         kref_put(&ctrl->kref, nvme_free_ctrl);
1321 }
1322
1323 /*
1324  * Initialize a NVMe controller structures.  This needs to be called during
1325  * earliest initialization so that we have the initialized structured around
1326  * during probing.
1327  */
1328 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1329                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1330 {
1331         int ret;
1332
1333         INIT_LIST_HEAD(&ctrl->namespaces);
1334         mutex_init(&ctrl->namespaces_mutex);
1335         kref_init(&ctrl->kref);
1336         ctrl->dev = dev;
1337         ctrl->ops = ops;
1338         ctrl->quirks = quirks;
1339
1340         ret = nvme_set_instance(ctrl);
1341         if (ret)
1342                 goto out;
1343
1344         ctrl->device = device_create(nvme_class, ctrl->dev,
1345                                 MKDEV(nvme_char_major, ctrl->instance),
1346                                 dev, "nvme%d", ctrl->instance);
1347         if (IS_ERR(ctrl->device)) {
1348                 ret = PTR_ERR(ctrl->device);
1349                 goto out_release_instance;
1350         }
1351         get_device(ctrl->device);
1352         dev_set_drvdata(ctrl->device, ctrl);
1353
1354         ret = device_create_file(ctrl->device, &dev_attr_reset_controller);
1355         if (ret)
1356                 goto out_put_device;
1357
1358         spin_lock(&dev_list_lock);
1359         list_add_tail(&ctrl->node, &nvme_ctrl_list);
1360         spin_unlock(&dev_list_lock);
1361
1362         return 0;
1363
1364 out_put_device:
1365         put_device(ctrl->device);
1366         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1367 out_release_instance:
1368         nvme_release_instance(ctrl);
1369 out:
1370         return ret;
1371 }
1372
1373 void nvme_freeze_queues(struct nvme_ctrl *ctrl)
1374 {
1375         struct nvme_ns *ns;
1376
1377         mutex_lock(&ctrl->namespaces_mutex);
1378         list_for_each_entry(ns, &ctrl->namespaces, list) {
1379                 blk_mq_freeze_queue_start(ns->queue);
1380
1381                 spin_lock_irq(ns->queue->queue_lock);
1382                 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1383                 spin_unlock_irq(ns->queue->queue_lock);
1384
1385                 blk_mq_cancel_requeue_work(ns->queue);
1386                 blk_mq_stop_hw_queues(ns->queue);
1387         }
1388         mutex_unlock(&ctrl->namespaces_mutex);
1389 }
1390
1391 void nvme_unfreeze_queues(struct nvme_ctrl *ctrl)
1392 {
1393         struct nvme_ns *ns;
1394
1395         mutex_lock(&ctrl->namespaces_mutex);
1396         list_for_each_entry(ns, &ctrl->namespaces, list) {
1397                 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1398                 blk_mq_unfreeze_queue(ns->queue);
1399                 blk_mq_start_stopped_hw_queues(ns->queue, true);
1400                 blk_mq_kick_requeue_list(ns->queue);
1401         }
1402         mutex_unlock(&ctrl->namespaces_mutex);
1403 }
1404
1405 int __init nvme_core_init(void)
1406 {
1407         int result;
1408
1409         result = register_blkdev(nvme_major, "nvme");
1410         if (result < 0)
1411                 return result;
1412         else if (result > 0)
1413                 nvme_major = result;
1414
1415         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1416                                                         &nvme_dev_fops);
1417         if (result < 0)
1418                 goto unregister_blkdev;
1419         else if (result > 0)
1420                 nvme_char_major = result;
1421
1422         nvme_class = class_create(THIS_MODULE, "nvme");
1423         if (IS_ERR(nvme_class)) {
1424                 result = PTR_ERR(nvme_class);
1425                 goto unregister_chrdev;
1426         }
1427
1428         return 0;
1429
1430  unregister_chrdev:
1431         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1432  unregister_blkdev:
1433         unregister_blkdev(nvme_major, "nvme");
1434         return result;
1435 }
1436
1437 void nvme_core_exit(void)
1438 {
1439         unregister_blkdev(nvme_major, "nvme");
1440         class_destroy(nvme_class);
1441         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1442 }