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 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 471         case SG_GET_VERSION_NUM:
 472                 return nvme_sg_get_version_num((void __user *)arg);
 473         case SG_IO:
 474                 return nvme_sg_io(ns, (void __user *)arg);
 475 #endif
 476         default:
 477                 return -ENOTTY;
 478         }
 479 }
 480
 481 #ifdef CONFIG_COMPAT
 482 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 483                         unsigned int cmd, unsigned long arg)
 484 {
 485         switch (cmd) {
 486         case SG_IO:
 487                 return -ENOIOCTLCMD;
 488         }
 489         return nvme_ioctl(bdev, mode, cmd, arg);
 490 }
 491 #else
 492 #define nvme_compat_ioctl       NULL
 493 #endif
 494
 495 static int nvme_open(struct block_device *bdev, fmode_t mode)
 496 {
 497         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 498 }
 499
 500 static void nvme_release(struct gendisk *disk, fmode_t mode)
 501 {
 502         nvme_put_ns(disk->private_data);
 503 }
 504
 505 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 506 {
 507         /* some standard values */
 508         geo->heads = 1 << 6;
 509         geo->sectors = 1 << 5;
 510         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 511         return 0;
 512 }
 513
 514 #ifdef CONFIG_BLK_DEV_INTEGRITY
 515 static void nvme_init_integrity(struct nvme_ns *ns)
 516 {
 517         struct blk_integrity integrity;
 518
 519         switch (ns->pi_type) {
 520         case NVME_NS_DPS_PI_TYPE3:
 521                 integrity.profile = &t10_pi_type3_crc;
 522                 break;
 523         case NVME_NS_DPS_PI_TYPE1:
 524         case NVME_NS_DPS_PI_TYPE2:
 525                 integrity.profile = &t10_pi_type1_crc;
 526                 break;
 527         default:
 528                 integrity.profile = NULL;
 529                 break;
 530         }
 531         integrity.tuple_size = ns->ms;
 532         blk_integrity_register(ns->disk, &integrity);
 533         blk_queue_max_integrity_segments(ns->queue, 1);
 534 }
 535 #else
 536 static void nvme_init_integrity(struct nvme_ns *ns)
 537 {
 538 }
 539 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 540
 541 static void nvme_config_discard(struct nvme_ns *ns)
 542 {
 543         u32 logical_block_size = queue_logical_block_size(ns->queue);
 544         ns->queue->limits.discard_zeroes_data = 0;
 545         ns->queue->limits.discard_alignment = logical_block_size;
 546         ns->queue->limits.discard_granularity = logical_block_size;
 547         blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
 548         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 549 }
 550
 551 static int nvme_revalidate_disk(struct gendisk *disk)
 552 {
 553         struct nvme_ns *ns = disk->private_data;
 554         struct nvme_id_ns *id;
 555         u8 lbaf, pi_type;
 556         u16 old_ms;
 557         unsigned short bs;
 558
 559         if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
 560                 dev_warn(ns->ctrl->dev, "%s: Identify failure nvme%dn%d\n",
 561                                 __func__, ns->ctrl->instance, ns->ns_id);
 562                 return -ENODEV;
 563         }
 564         if (id->ncap == 0) {
 565                 kfree(id);
 566                 return -ENODEV;
 567         }
 568
 569         if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
 570                 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
 571                         dev_warn(ns->ctrl->dev,
 572                                 "%s: LightNVM init failure\n", __func__);
 573                         kfree(id);
 574                         return -ENODEV;
 575                 }
 576                 ns->type = NVME_NS_LIGHTNVM;
 577         }
 578
 579         if (ns->ctrl->vs >= NVME_VS(1, 1))
 580                 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
 581         if (ns->ctrl->vs >= NVME_VS(1, 2))
 582                 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
 583
 584         old_ms = ns->ms;
 585         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 586         ns->lba_shift = id->lbaf[lbaf].ds;
 587         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 588         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 589
 590         /*
 591          * If identify namespace failed, use default 512 byte block size so
 592          * block layer can use before failing read/write for 0 capacity.
 593          */
 594         if (ns->lba_shift == 0)
 595                 ns->lba_shift = 9;
 596         bs = 1 << ns->lba_shift;
 597         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 598         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 599                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 600
 601         blk_mq_freeze_queue(disk->queue);
 602         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 603                                 ns->ms != old_ms ||
 604                                 bs != queue_logical_block_size(disk->queue) ||
 605                                 (ns->ms && ns->ext)))
 606                 blk_integrity_unregister(disk);
 607
 608         ns->pi_type = pi_type;
 609         blk_queue_logical_block_size(ns->queue, bs);
 610
 611         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 612                 nvme_init_integrity(ns);
 613         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 614                 set_capacity(disk, 0);
 615         else
 616                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 617
 618         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 619                 nvme_config_discard(ns);
 620         blk_mq_unfreeze_queue(disk->queue);
 621
 622         kfree(id);
 623         return 0;
 624 }
 625
 626 static char nvme_pr_type(enum pr_type type)
 627 {
 628         switch (type) {
 629         case PR_WRITE_EXCLUSIVE:
 630                 return 1;
 631         case PR_EXCLUSIVE_ACCESS:
 632                 return 2;
 633         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 634                 return 3;
 635         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 636                 return 4;
 637         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 638                 return 5;
 639         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 640                 return 6;
 641         default:
 642                 return 0;
 643         }
 644 };
 645
 646 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 647                                 u64 key, u64 sa_key, u8 op)
 648 {
 649         struct nvme_ns *ns = bdev->bd_disk->private_data;
 650         struct nvme_command c;
 651         u8 data[16] = { 0, };
 652
 653         put_unaligned_le64(key, &data[0]);
 654         put_unaligned_le64(sa_key, &data[8]);
 655
 656         memset(&c, 0, sizeof(c));
 657         c.common.opcode = op;
 658         c.common.nsid = cpu_to_le32(ns->ns_id);
 659         c.common.cdw10[0] = cpu_to_le32(cdw10);
 660
 661         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
 662 }
 663
 664 static int nvme_pr_register(struct block_device *bdev, u64 old,
 665                 u64 new, unsigned flags)
 666 {
 667         u32 cdw10;
 668
 669         if (flags & ~PR_FL_IGNORE_KEY)
 670                 return -EOPNOTSUPP;
 671
 672         cdw10 = old ? 2 : 0;
 673         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
 674         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
 675         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
 676 }
 677
 678 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
 679                 enum pr_type type, unsigned flags)
 680 {
 681         u32 cdw10;
 682
 683         if (flags & ~PR_FL_IGNORE_KEY)
 684                 return -EOPNOTSUPP;
 685
 686         cdw10 = nvme_pr_type(type) << 8;
 687         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
 688         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
 689 }
 690
 691 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
 692                 enum pr_type type, bool abort)
 693 {
 694         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
 695         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
 696 }
 697
 698 static int nvme_pr_clear(struct block_device *bdev, u64 key)
 699 {
 700         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
 701         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
 702 }
 703
 704 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
 705 {
 706         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
 707         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
 708 }
 709
 710 static const struct pr_ops nvme_pr_ops = {
 711         .pr_register    = nvme_pr_register,
 712         .pr_reserve     = nvme_pr_reserve,
 713         .pr_release     = nvme_pr_release,
 714         .pr_preempt     = nvme_pr_preempt,
 715         .pr_clear       = nvme_pr_clear,
 716 };
 717
 718 static const struct block_device_operations nvme_fops = {
 719         .owner          = THIS_MODULE,
 720         .ioctl          = nvme_ioctl,
 721         .compat_ioctl   = nvme_compat_ioctl,
 722         .open           = nvme_open,
 723         .release        = nvme_release,
 724         .getgeo         = nvme_getgeo,
 725         .revalidate_disk= nvme_revalidate_disk,
 726         .pr_ops         = &nvme_pr_ops,
 727 };
 728
 729 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
 730 {
 731         unsigned long timeout =
 732                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
 733         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
 734         int ret;
 735
 736         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 737                 if ((csts & NVME_CSTS_RDY) == bit)
 738                         break;
 739
 740                 msleep(100);
 741                 if (fatal_signal_pending(current))
 742                         return -EINTR;
 743                 if (time_after(jiffies, timeout)) {
 744                         dev_err(ctrl->dev,
 745                                 "Device not ready; aborting %s\n", enabled ?
 746                                                 "initialisation" : "reset");
 747                         return -ENODEV;
 748                 }
 749         }
 750
 751         return ret;
 752 }
 753
 754 /*
 755  * If the device has been passed off to us in an enabled state, just clear
 756  * the enabled bit.  The spec says we should set the 'shutdown notification
 757  * bits', but doing so may cause the device to complete commands to the
 758  * admin queue ... and we don't know what memory that might be pointing at!
 759  */
 760 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 761 {
 762         int ret;
 763
 764         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 765         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
 766
 767         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 768         if (ret)
 769                 return ret;
 770         return nvme_wait_ready(ctrl, cap, false);
 771 }
 772
 773 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 774 {
 775         /*
 776          * Default to a 4K page size, with the intention to update this
 777          * path in the future to accomodate architectures with differing
 778          * kernel and IO page sizes.
 779          */
 780         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
 781         int ret;
 782
 783         if (page_shift < dev_page_min) {
 784                 dev_err(ctrl->dev,
 785                         "Minimum device page size %u too large for host (%u)\n",
 786                         1 << dev_page_min, 1 << page_shift);
 787                 return -ENODEV;
 788         }
 789
 790         ctrl->page_size = 1 << page_shift;
 791
 792         ctrl->ctrl_config = NVME_CC_CSS_NVM;
 793         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
 794         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
 795         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
 796         ctrl->ctrl_config |= NVME_CC_ENABLE;
 797
 798         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 799         if (ret)
 800                 return ret;
 801         return nvme_wait_ready(ctrl, cap, true);
 802 }
 803
 804 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
 805 {
 806         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
 807         u32 csts;
 808         int ret;
 809
 810         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 811         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
 812
 813         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 814         if (ret)
 815                 return ret;
 816
 817         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 818                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
 819                         break;
 820
 821                 msleep(100);
 822                 if (fatal_signal_pending(current))
 823                         return -EINTR;
 824                 if (time_after(jiffies, timeout)) {
 825                         dev_err(ctrl->dev,
 826                                 "Device shutdown incomplete; abort shutdown\n");
 827                         return -ENODEV;
 828                 }
 829         }
 830
 831         return ret;
 832 }
 833
 834 /*
 835  * Initialize the cached copies of the Identify data and various controller
 836  * register in our nvme_ctrl structure.  This should be called as soon as
 837  * the admin queue is fully up and running.
 838  */
 839 int nvme_init_identify(struct nvme_ctrl *ctrl)
 840 {
 841         struct nvme_id_ctrl *id;
 842         u64 cap;
 843         int ret, page_shift;
 844
 845         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
 846         if (ret) {
 847                 dev_err(ctrl->dev, "Reading VS failed (%d)\n", ret);
 848                 return ret;
 849         }
 850
 851         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
 852         if (ret) {
 853                 dev_err(ctrl->dev, "Reading CAP failed (%d)\n", ret);
 854                 return ret;
 855         }
 856         page_shift = NVME_CAP_MPSMIN(cap) + 12;
 857
 858         if (ctrl->vs >= NVME_VS(1, 1))
 859                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
 860
 861         ret = nvme_identify_ctrl(ctrl, &id);
 862         if (ret) {
 863                 dev_err(ctrl->dev, "Identify Controller failed (%d)\n", ret);
 864                 return -EIO;
 865         }
 866
 867         ctrl->oncs = le16_to_cpup(&id->oncs);
 868         atomic_set(&ctrl->abort_limit, id->acl + 1);
 869         ctrl->vwc = id->vwc;
 870         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
 871         memcpy(ctrl->model, id->mn, sizeof(id->mn));
 872         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
 873         if (id->mdts)
 874                 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
 875         else
 876                 ctrl->max_hw_sectors = UINT_MAX;
 877
 878         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
 879                 unsigned int max_hw_sectors;
 880
 881                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
 882                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
 883                 if (ctrl->max_hw_sectors) {
 884                         ctrl->max_hw_sectors = min(max_hw_sectors,
 885                                                         ctrl->max_hw_sectors);
 886                 } else {
 887                         ctrl->max_hw_sectors = max_hw_sectors;
 888                 }
 889         }
 890
 891         kfree(id);
 892         return 0;
 893 }
 894
 895 static int nvme_dev_open(struct inode *inode, struct file *file)
 896 {
 897         struct nvme_ctrl *ctrl;
 898         int instance = iminor(inode);
 899         int ret = -ENODEV;
 900
 901         spin_lock(&dev_list_lock);
 902         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
 903                 if (ctrl->instance != instance)
 904                         continue;
 905
 906                 if (!ctrl->admin_q) {
 907                         ret = -EWOULDBLOCK;
 908                         break;
 909                 }
 910                 if (!kref_get_unless_zero(&ctrl->kref))
 911                         break;
 912                 file->private_data = ctrl;
 913                 ret = 0;
 914                 break;
 915         }
 916         spin_unlock(&dev_list_lock);
 917
 918         return ret;
 919 }
 920
 921 static int nvme_dev_release(struct inode *inode, struct file *file)
 922 {
 923         nvme_put_ctrl(file->private_data);
 924         return 0;
 925 }
 926
 927 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
 928 {
 929         struct nvme_ns *ns;
 930         int ret;
 931
 932         mutex_lock(&ctrl->namespaces_mutex);
 933         if (list_empty(&ctrl->namespaces)) {
 934                 ret = -ENOTTY;
 935                 goto out_unlock;
 936         }
 937
 938         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
 939         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
 940                 dev_warn(ctrl->dev,
 941                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
 942                 ret = -EINVAL;
 943                 goto out_unlock;
 944         }
 945
 946         dev_warn(ctrl->dev,
 947                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
 948         kref_get(&ns->kref);
 949         mutex_unlock(&ctrl->namespaces_mutex);
 950
 951         ret = nvme_user_cmd(ctrl, ns, argp);
 952         nvme_put_ns(ns);
 953         return ret;
 954
 955 out_unlock:
 956         mutex_unlock(&ctrl->namespaces_mutex);
 957         return ret;
 958 }
 959
 960 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
 961                 unsigned long arg)
 962 {
 963         struct nvme_ctrl *ctrl = file->private_data;
 964         void __user *argp = (void __user *)arg;
 965
 966         switch (cmd) {
 967         case NVME_IOCTL_ADMIN_CMD:
 968                 return nvme_user_cmd(ctrl, NULL, argp);
 969         case NVME_IOCTL_IO_CMD:
 970                 return nvme_dev_user_cmd(ctrl, argp);
 971         case NVME_IOCTL_RESET:
 972                 dev_warn(ctrl->dev, "resetting controller\n");
 973                 return ctrl->ops->reset_ctrl(ctrl);
 974         case NVME_IOCTL_SUBSYS_RESET:
 975                 return nvme_reset_subsystem(ctrl);
 976         default:
 977                 return -ENOTTY;
 978         }
 979 }
 980
 981 static const struct file_operations nvme_dev_fops = {
 982         .owner          = THIS_MODULE,
 983         .open           = nvme_dev_open,
 984         .release        = nvme_dev_release,
 985         .unlocked_ioctl = nvme_dev_ioctl,
 986         .compat_ioctl   = nvme_dev_ioctl,
 987 };
 988
 989 static ssize_t nvme_sysfs_reset(struct device *dev,
 990                                 struct device_attribute *attr, const char *buf,
 991                                 size_t count)
 992 {
 993         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
 994         int ret;
 995
 996         ret = ctrl->ops->reset_ctrl(ctrl);
 997         if (ret < 0)
 998                 return ret;
 999         return count;
1000 }
1001 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1002
1003 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1004                                                                 char *buf)
1005 {
1006         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1007         return sprintf(buf, "%pU\n", ns->uuid);
1008 }
1009 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1010
1011 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1012                                                                 char *buf)
1013 {
1014         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1015         return sprintf(buf, "%8phd\n", ns->eui);
1016 }
1017 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1018
1019 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1020                                                                 char *buf)
1021 {
1022         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1023         return sprintf(buf, "%d\n", ns->ns_id);
1024 }
1025 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1026
1027 static struct attribute *nvme_ns_attrs[] = {
1028         &dev_attr_uuid.attr,
1029         &dev_attr_eui.attr,
1030         &dev_attr_nsid.attr,
1031         NULL,
1032 };
1033
1034 static umode_t nvme_attrs_are_visible(struct kobject *kobj,
1035                 struct attribute *a, int n)
1036 {
1037         struct device *dev = container_of(kobj, struct device, kobj);
1038         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1039
1040         if (a == &dev_attr_uuid.attr) {
1041                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1042                         return 0;
1043         }
1044         if (a == &dev_attr_eui.attr) {
1045                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1046                         return 0;
1047         }
1048         return a->mode;
1049 }
1050
1051 static const struct attribute_group nvme_ns_attr_group = {
1052         .attrs          = nvme_ns_attrs,
1053         .is_visible     = nvme_attrs_are_visible,
1054 };
1055
1056 #define nvme_show_function(field)                                               \
1057 static ssize_t  field##_show(struct device *dev,                                \
1058                             struct device_attribute *attr, char *buf)           \
1059 {                                                                               \
1060         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1061         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1062 }                                                                               \
1063 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1064
1065 nvme_show_function(model);
1066 nvme_show_function(serial);
1067 nvme_show_function(firmware_rev);
1068
1069 static struct attribute *nvme_dev_attrs[] = {
1070         &dev_attr_reset_controller.attr,
1071         &dev_attr_model.attr,
1072         &dev_attr_serial.attr,
1073         &dev_attr_firmware_rev.attr,
1074         NULL
1075 };
1076
1077 static struct attribute_group nvme_dev_attrs_group = {
1078         .attrs = nvme_dev_attrs,
1079 };
1080
1081 static const struct attribute_group *nvme_dev_attr_groups[] = {
1082         &nvme_dev_attrs_group,
1083         NULL,
1084 };
1085
1086 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1087 {
1088         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1089         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1090
1091         return nsa->ns_id - nsb->ns_id;
1092 }
1093
1094 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1095 {
1096         struct nvme_ns *ns;
1097
1098         lockdep_assert_held(&ctrl->namespaces_mutex);
1099
1100         list_for_each_entry(ns, &ctrl->namespaces, list) {
1101                 if (ns->ns_id == nsid)
1102                         return ns;
1103                 if (ns->ns_id > nsid)
1104                         break;
1105         }
1106         return NULL;
1107 }
1108
1109 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1110 {
1111         struct nvme_ns *ns;
1112         struct gendisk *disk;
1113         int node = dev_to_node(ctrl->dev);
1114
1115         lockdep_assert_held(&ctrl->namespaces_mutex);
1116
1117         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1118         if (!ns)
1119                 return;
1120
1121         ns->queue = blk_mq_init_queue(ctrl->tagset);
1122         if (IS_ERR(ns->queue))
1123                 goto out_free_ns;
1124         queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
1125         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1126         ns->queue->queuedata = ns;
1127         ns->ctrl = ctrl;
1128
1129         disk = alloc_disk_node(0, node);
1130         if (!disk)
1131                 goto out_free_queue;
1132
1133         kref_init(&ns->kref);
1134         ns->ns_id = nsid;
1135         ns->disk = disk;
1136         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1137
1138         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1139         if (ctrl->max_hw_sectors) {
1140                 blk_queue_max_hw_sectors(ns->queue, ctrl->max_hw_sectors);
1141                 blk_queue_max_segments(ns->queue,
1142                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1);
1143         }
1144         if (ctrl->stripe_size)
1145                 blk_queue_chunk_sectors(ns->queue, ctrl->stripe_size >> 9);
1146         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1147                 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
1148         blk_queue_virt_boundary(ns->queue, ctrl->page_size - 1);
1149
1150         disk->major = nvme_major;
1151         disk->first_minor = 0;
1152         disk->fops = &nvme_fops;
1153         disk->private_data = ns;
1154         disk->queue = ns->queue;
1155         disk->driverfs_dev = ctrl->device;
1156         disk->flags = GENHD_FL_EXT_DEVT;
1157         sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, nsid);
1158
1159         if (nvme_revalidate_disk(ns->disk))
1160                 goto out_free_disk;
1161
1162         list_add_tail(&ns->list, &ctrl->namespaces);
1163         kref_get(&ctrl->kref);
1164         if (ns->type == NVME_NS_LIGHTNVM)
1165                 return;
1166
1167         add_disk(ns->disk);
1168         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1169                                         &nvme_ns_attr_group))
1170                 pr_warn("%s: failed to create sysfs group for identification\n",
1171                         ns->disk->disk_name);
1172         return;
1173  out_free_disk:
1174         kfree(disk);
1175  out_free_queue:
1176         blk_cleanup_queue(ns->queue);
1177  out_free_ns:
1178         kfree(ns);
1179 }
1180
1181 static void nvme_ns_remove(struct nvme_ns *ns)
1182 {
1183         bool kill = nvme_io_incapable(ns->ctrl) &&
1184                         !blk_queue_dying(ns->queue);
1185
1186         lockdep_assert_held(&ns->ctrl->namespaces_mutex);
1187
1188         if (kill)
1189                 blk_set_queue_dying(ns->queue);
1190         if (ns->disk->flags & GENHD_FL_UP) {
1191                 if (blk_get_integrity(ns->disk))
1192                         blk_integrity_unregister(ns->disk);
1193                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1194                                         &nvme_ns_attr_group);
1195                 del_gendisk(ns->disk);
1196         }
1197         if (kill || !blk_queue_dying(ns->queue)) {
1198                 blk_mq_abort_requeue_list(ns->queue);
1199                 blk_cleanup_queue(ns->queue);
1200         }
1201         list_del_init(&ns->list);
1202         nvme_put_ns(ns);
1203 }
1204
1205 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1206 {
1207         struct nvme_ns *ns;
1208
1209         ns = nvme_find_ns(ctrl, nsid);
1210         if (ns) {
1211                 if (revalidate_disk(ns->disk))
1212                         nvme_ns_remove(ns);
1213         } else
1214                 nvme_alloc_ns(ctrl, nsid);
1215 }
1216
1217 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1218 {
1219         struct nvme_ns *ns;
1220         __le32 *ns_list;
1221         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1222         int ret = 0;
1223
1224         ns_list = kzalloc(0x1000, GFP_KERNEL);
1225         if (!ns_list)
1226                 return -ENOMEM;
1227
1228         for (i = 0; i < num_lists; i++) {
1229                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1230                 if (ret)
1231                         goto out;
1232
1233                 for (j = 0; j < min(nn, 1024U); j++) {
1234                         nsid = le32_to_cpu(ns_list[j]);
1235                         if (!nsid)
1236                                 goto out;
1237
1238                         nvme_validate_ns(ctrl, nsid);
1239
1240                         while (++prev < nsid) {
1241                                 ns = nvme_find_ns(ctrl, prev);
1242                                 if (ns)
1243                                         nvme_ns_remove(ns);
1244                         }
1245                 }
1246                 nn -= j;
1247         }
1248  out:
1249         kfree(ns_list);
1250         return ret;
1251 }
1252
1253 static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn)
1254 {
1255         struct nvme_ns *ns, *next;
1256         unsigned i;
1257
1258         lockdep_assert_held(&ctrl->namespaces_mutex);
1259
1260         for (i = 1; i <= nn; i++)
1261                 nvme_validate_ns(ctrl, i);
1262
1263         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1264                 if (ns->ns_id > nn)
1265                         nvme_ns_remove(ns);
1266         }
1267 }
1268
1269 void nvme_scan_namespaces(struct nvme_ctrl *ctrl)
1270 {
1271         struct nvme_id_ctrl *id;
1272         unsigned nn;
1273
1274         if (nvme_identify_ctrl(ctrl, &id))
1275                 return;
1276
1277         mutex_lock(&ctrl->namespaces_mutex);
1278         nn = le32_to_cpu(id->nn);
1279         if (ctrl->vs >= NVME_VS(1, 1) &&
1280             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1281                 if (!nvme_scan_ns_list(ctrl, nn))
1282                         goto done;
1283         }
1284         __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn));
1285  done:
1286         list_sort(NULL, &ctrl->namespaces, ns_cmp);
1287         mutex_unlock(&ctrl->namespaces_mutex);
1288         kfree(id);
1289 }
1290
1291 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1292 {
1293         struct nvme_ns *ns, *next;
1294
1295         mutex_lock(&ctrl->namespaces_mutex);
1296         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1297                 nvme_ns_remove(ns);
1298         mutex_unlock(&ctrl->namespaces_mutex);
1299 }
1300
1301 static DEFINE_IDA(nvme_instance_ida);
1302
1303 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1304 {
1305         int instance, error;
1306
1307         do {
1308                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1309                         return -ENODEV;
1310
1311                 spin_lock(&dev_list_lock);
1312                 error = ida_get_new(&nvme_instance_ida, &instance);
1313                 spin_unlock(&dev_list_lock);
1314         } while (error == -EAGAIN);
1315
1316         if (error)
1317                 return -ENODEV;
1318
1319         ctrl->instance = instance;
1320         return 0;
1321 }
1322
1323 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1324 {
1325         spin_lock(&dev_list_lock);
1326         ida_remove(&nvme_instance_ida, ctrl->instance);
1327         spin_unlock(&dev_list_lock);
1328 }
1329
1330 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1331  {
1332         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1333
1334         spin_lock(&dev_list_lock);
1335         list_del(&ctrl->node);
1336         spin_unlock(&dev_list_lock);
1337 }
1338
1339 static void nvme_free_ctrl(struct kref *kref)
1340 {
1341         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1342
1343         put_device(ctrl->device);
1344         nvme_release_instance(ctrl);
1345
1346         ctrl->ops->free_ctrl(ctrl);
1347 }
1348
1349 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1350 {
1351         kref_put(&ctrl->kref, nvme_free_ctrl);
1352 }
1353
1354 /*
1355  * Initialize a NVMe controller structures.  This needs to be called during
1356  * earliest initialization so that we have the initialized structured around
1357  * during probing.
1358  */
1359 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1360                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1361 {
1362         int ret;
1363
1364         INIT_LIST_HEAD(&ctrl->namespaces);
1365         mutex_init(&ctrl->namespaces_mutex);
1366         kref_init(&ctrl->kref);
1367         ctrl->dev = dev;
1368         ctrl->ops = ops;
1369         ctrl->quirks = quirks;
1370
1371         ret = nvme_set_instance(ctrl);
1372         if (ret)
1373                 goto out;
1374
1375         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1376                                 MKDEV(nvme_char_major, ctrl->instance),
1377                                 dev, nvme_dev_attr_groups,
1378                                 "nvme%d", ctrl->instance);
1379         if (IS_ERR(ctrl->device)) {
1380                 ret = PTR_ERR(ctrl->device);
1381                 goto out_release_instance;
1382         }
1383         get_device(ctrl->device);
1384         dev_set_drvdata(ctrl->device, ctrl);
1385
1386         spin_lock(&dev_list_lock);
1387         list_add_tail(&ctrl->node, &nvme_ctrl_list);
1388         spin_unlock(&dev_list_lock);
1389
1390         return 0;
1391 out_release_instance:
1392         nvme_release_instance(ctrl);
1393 out:
1394         return ret;
1395 }
1396
1397 void nvme_stop_queues(struct nvme_ctrl *ctrl)
1398 {
1399         struct nvme_ns *ns;
1400
1401         mutex_lock(&ctrl->namespaces_mutex);
1402         list_for_each_entry(ns, &ctrl->namespaces, list) {
1403                 spin_lock_irq(ns->queue->queue_lock);
1404                 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1405                 spin_unlock_irq(ns->queue->queue_lock);
1406
1407                 blk_mq_cancel_requeue_work(ns->queue);
1408                 blk_mq_stop_hw_queues(ns->queue);
1409         }
1410         mutex_unlock(&ctrl->namespaces_mutex);
1411 }
1412
1413 void nvme_start_queues(struct nvme_ctrl *ctrl)
1414 {
1415         struct nvme_ns *ns;
1416
1417         mutex_lock(&ctrl->namespaces_mutex);
1418         list_for_each_entry(ns, &ctrl->namespaces, list) {
1419                 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1420                 blk_mq_start_stopped_hw_queues(ns->queue, true);
1421                 blk_mq_kick_requeue_list(ns->queue);
1422         }
1423         mutex_unlock(&ctrl->namespaces_mutex);
1424 }
1425
1426 int __init nvme_core_init(void)
1427 {
1428         int result;
1429
1430         result = register_blkdev(nvme_major, "nvme");
1431         if (result < 0)
1432                 return result;
1433         else if (result > 0)
1434                 nvme_major = result;
1435
1436         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1437                                                         &nvme_dev_fops);
1438         if (result < 0)
1439                 goto unregister_blkdev;
1440         else if (result > 0)
1441                 nvme_char_major = result;
1442
1443         nvme_class = class_create(THIS_MODULE, "nvme");
1444         if (IS_ERR(nvme_class)) {
1445                 result = PTR_ERR(nvme_class);
1446                 goto unregister_chrdev;
1447         }
1448
1449         return 0;
1450
1451  unregister_chrdev:
1452         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1453  unregister_blkdev:
1454         unregister_blkdev(nvme_major, "nvme");
1455         return result;
1456 }
1457
1458 void nvme_core_exit(void)
1459 {
1460         unregister_blkdev(nvme_major, "nvme");
1461         class_destroy(nvme_class);
1462         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1463 }