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         put_disk(ns->disk);
  59         ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
  60         nvme_put_ctrl(ns->ctrl);
  61         kfree(ns);
  62 }
  63
  64 static void nvme_put_ns(struct nvme_ns *ns)
  65 {
  66         kref_put(&ns->kref, nvme_free_ns);
  67 }
  68
  69 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
  70 {
  71         struct nvme_ns *ns;
  72
  73         spin_lock(&dev_list_lock);
  74         ns = disk->private_data;
  75         if (ns && !kref_get_unless_zero(&ns->kref))
  76                 ns = NULL;
  77         spin_unlock(&dev_list_lock);
  78
  79         return ns;
  80 }
  81
  82 void nvme_requeue_req(struct request *req)
  83 {
  84         unsigned long flags;
  85
  86         blk_mq_requeue_request(req);
  87         spin_lock_irqsave(req->q->queue_lock, flags);
  88         if (!blk_queue_stopped(req->q))
  89                 blk_mq_kick_requeue_list(req->q);
  90         spin_unlock_irqrestore(req->q->queue_lock, flags);
  91 }
  92
  93 struct request *nvme_alloc_request(struct request_queue *q,
  94                 struct nvme_command *cmd, unsigned int flags)
  95 {
  96         bool write = cmd->common.opcode & 1;
  97         struct request *req;
  98
  99         req = blk_mq_alloc_request(q, write, flags);
 100         if (IS_ERR(req))
 101                 return req;
 102
 103         req->cmd_type = REQ_TYPE_DRV_PRIV;
 104         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 105         req->__data_len = 0;
 106         req->__sector = (sector_t) -1;
 107         req->bio = req->biotail = NULL;
 108
 109         req->cmd = (unsigned char *)cmd;
 110         req->cmd_len = sizeof(struct nvme_command);
 111         req->special = (void *)0;
 112
 113         return req;
 114 }
 115
 116 /*
 117  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 118  * if the result is positive, it's an NVM Express status code
 119  */
 120 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 121                 void *buffer, unsigned bufflen, u32 *result, unsigned timeout)
 122 {
 123         struct request *req;
 124         int ret;
 125
 126         req = nvme_alloc_request(q, cmd, 0);
 127         if (IS_ERR(req))
 128                 return PTR_ERR(req);
 129
 130         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 131
 132         if (buffer && bufflen) {
 133                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 134                 if (ret)
 135                         goto out;
 136         }
 137
 138         blk_execute_rq(req->q, NULL, req, 0);
 139         if (result)
 140                 *result = (u32)(uintptr_t)req->special;
 141         ret = req->errors;
 142  out:
 143         blk_mq_free_request(req);
 144         return ret;
 145 }
 146
 147 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 148                 void *buffer, unsigned bufflen)
 149 {
 150         return __nvme_submit_sync_cmd(q, cmd, buffer, bufflen, NULL, 0);
 151 }
 152
 153 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 154                 void __user *ubuffer, unsigned bufflen,
 155                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 156                 u32 *result, unsigned timeout)
 157 {
 158         bool write = cmd->common.opcode & 1;
 159         struct nvme_ns *ns = q->queuedata;
 160         struct gendisk *disk = ns ? ns->disk : NULL;
 161         struct request *req;
 162         struct bio *bio = NULL;
 163         void *meta = NULL;
 164         int ret;
 165
 166         req = nvme_alloc_request(q, cmd, 0);
 167         if (IS_ERR(req))
 168                 return PTR_ERR(req);
 169
 170         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 171
 172         if (ubuffer && bufflen) {
 173                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 174                                 GFP_KERNEL);
 175                 if (ret)
 176                         goto out;
 177                 bio = req->bio;
 178
 179                 if (!disk)
 180                         goto submit;
 181                 bio->bi_bdev = bdget_disk(disk, 0);
 182                 if (!bio->bi_bdev) {
 183                         ret = -ENODEV;
 184                         goto out_unmap;
 185                 }
 186
 187                 if (meta_buffer) {
 188                         struct bio_integrity_payload *bip;
 189
 190                         meta = kmalloc(meta_len, GFP_KERNEL);
 191                         if (!meta) {
 192                                 ret = -ENOMEM;
 193                                 goto out_unmap;
 194                         }
 195
 196                         if (write) {
 197                                 if (copy_from_user(meta, meta_buffer,
 198                                                 meta_len)) {
 199                                         ret = -EFAULT;
 200                                         goto out_free_meta;
 201                                 }
 202                         }
 203
 204                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 205                         if (IS_ERR(bip)) {
 206                                 ret = PTR_ERR(bip);
 207                                 goto out_free_meta;
 208                         }
 209
 210                         bip->bip_iter.bi_size = meta_len;
 211                         bip->bip_iter.bi_sector = meta_seed;
 212
 213                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 214                                         meta_len, offset_in_page(meta));
 215                         if (ret != meta_len) {
 216                                 ret = -ENOMEM;
 217                                 goto out_free_meta;
 218                         }
 219                 }
 220         }
 221  submit:
 222         blk_execute_rq(req->q, disk, req, 0);
 223         ret = req->errors;
 224         if (result)
 225                 *result = (u32)(uintptr_t)req->special;
 226         if (meta && !ret && !write) {
 227                 if (copy_to_user(meta_buffer, meta, meta_len))
 228                         ret = -EFAULT;
 229         }
 230  out_free_meta:
 231         kfree(meta);
 232  out_unmap:
 233         if (bio) {
 234                 if (disk && bio->bi_bdev)
 235                         bdput(bio->bi_bdev);
 236                 blk_rq_unmap_user(bio);
 237         }
 238  out:
 239         blk_mq_free_request(req);
 240         return ret;
 241 }
 242
 243 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 244                 void __user *ubuffer, unsigned bufflen, u32 *result,
 245                 unsigned timeout)
 246 {
 247         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 248                         result, timeout);
 249 }
 250
 251 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 252 {
 253         struct nvme_command c = { };
 254         int error;
 255
 256         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 257         c.identify.opcode = nvme_admin_identify;
 258         c.identify.cns = cpu_to_le32(1);
 259
 260         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 261         if (!*id)
 262                 return -ENOMEM;
 263
 264         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 265                         sizeof(struct nvme_id_ctrl));
 266         if (error)
 267                 kfree(*id);
 268         return error;
 269 }
 270
 271 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 272 {
 273         struct nvme_command c = { };
 274
 275         c.identify.opcode = nvme_admin_identify;
 276         c.identify.cns = cpu_to_le32(2);
 277         c.identify.nsid = cpu_to_le32(nsid);
 278         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 279 }
 280
 281 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 282                 struct nvme_id_ns **id)
 283 {
 284         struct nvme_command c = { };
 285         int error;
 286
 287         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 288         c.identify.opcode = nvme_admin_identify,
 289         c.identify.nsid = cpu_to_le32(nsid),
 290
 291         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 292         if (!*id)
 293                 return -ENOMEM;
 294
 295         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 296                         sizeof(struct nvme_id_ns));
 297         if (error)
 298                 kfree(*id);
 299         return error;
 300 }
 301
 302 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 303                                         dma_addr_t dma_addr, u32 *result)
 304 {
 305         struct nvme_command c;
 306
 307         memset(&c, 0, sizeof(c));
 308         c.features.opcode = nvme_admin_get_features;
 309         c.features.nsid = cpu_to_le32(nsid);
 310         c.features.prp1 = cpu_to_le64(dma_addr);
 311         c.features.fid = cpu_to_le32(fid);
 312
 313         return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
 314 }
 315
 316 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 317                                         dma_addr_t dma_addr, u32 *result)
 318 {
 319         struct nvme_command c;
 320
 321         memset(&c, 0, sizeof(c));
 322         c.features.opcode = nvme_admin_set_features;
 323         c.features.prp1 = cpu_to_le64(dma_addr);
 324         c.features.fid = cpu_to_le32(fid);
 325         c.features.dword11 = cpu_to_le32(dword11);
 326
 327         return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
 328 }
 329
 330 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 331 {
 332         struct nvme_command c = { };
 333         int error;
 334
 335         c.common.opcode = nvme_admin_get_log_page,
 336         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 337         c.common.cdw10[0] = cpu_to_le32(
 338                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 339                          NVME_LOG_SMART),
 340
 341         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 342         if (!*log)
 343                 return -ENOMEM;
 344
 345         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 346                         sizeof(struct nvme_smart_log));
 347         if (error)
 348                 kfree(*log);
 349         return error;
 350 }
 351
 352 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 353 {
 354         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 355         u32 result;
 356         int status, nr_io_queues;
 357
 358         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
 359                         &result);
 360         if (status)
 361                 return status;
 362
 363         nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 364         *count = min(*count, nr_io_queues);
 365         return 0;
 366 }
 367
 368 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 369 {
 370         struct nvme_user_io io;
 371         struct nvme_command c;
 372         unsigned length, meta_len;
 373         void __user *metadata;
 374
 375         if (copy_from_user(&io, uio, sizeof(io)))
 376                 return -EFAULT;
 377         if (io.flags)
 378                 return -EINVAL;
 379
 380         switch (io.opcode) {
 381         case nvme_cmd_write:
 382         case nvme_cmd_read:
 383         case nvme_cmd_compare:
 384                 break;
 385         default:
 386                 return -EINVAL;
 387         }
 388
 389         length = (io.nblocks + 1) << ns->lba_shift;
 390         meta_len = (io.nblocks + 1) * ns->ms;
 391         metadata = (void __user *)(uintptr_t)io.metadata;
 392
 393         if (ns->ext) {
 394                 length += meta_len;
 395                 meta_len = 0;
 396         } else if (meta_len) {
 397                 if ((io.metadata & 3) || !io.metadata)
 398                         return -EINVAL;
 399         }
 400
 401         memset(&c, 0, sizeof(c));
 402         c.rw.opcode = io.opcode;
 403         c.rw.flags = io.flags;
 404         c.rw.nsid = cpu_to_le32(ns->ns_id);
 405         c.rw.slba = cpu_to_le64(io.slba);
 406         c.rw.length = cpu_to_le16(io.nblocks);
 407         c.rw.control = cpu_to_le16(io.control);
 408         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 409         c.rw.reftag = cpu_to_le32(io.reftag);
 410         c.rw.apptag = cpu_to_le16(io.apptag);
 411         c.rw.appmask = cpu_to_le16(io.appmask);
 412
 413         return __nvme_submit_user_cmd(ns->queue, &c,
 414                         (void __user *)(uintptr_t)io.addr, length,
 415                         metadata, meta_len, io.slba, NULL, 0);
 416 }
 417
 418 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 419                         struct nvme_passthru_cmd __user *ucmd)
 420 {
 421         struct nvme_passthru_cmd cmd;
 422         struct nvme_command c;
 423         unsigned timeout = 0;
 424         int status;
 425
 426         if (!capable(CAP_SYS_ADMIN))
 427                 return -EACCES;
 428         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 429                 return -EFAULT;
 430         if (cmd.flags)
 431                 return -EINVAL;
 432
 433         memset(&c, 0, sizeof(c));
 434         c.common.opcode = cmd.opcode;
 435         c.common.flags = cmd.flags;
 436         c.common.nsid = cpu_to_le32(cmd.nsid);
 437         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 438         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 439         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 440         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 441         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 442         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 443         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 444         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 445
 446         if (cmd.timeout_ms)
 447                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 448
 449         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 450                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 451                         &cmd.result, timeout);
 452         if (status >= 0) {
 453                 if (put_user(cmd.result, &ucmd->result))
 454                         return -EFAULT;
 455         }
 456
 457         return status;
 458 }
 459
 460 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 461                 unsigned int cmd, unsigned long arg)
 462 {
 463         struct nvme_ns *ns = bdev->bd_disk->private_data;
 464
 465         switch (cmd) {
 466         case NVME_IOCTL_ID:
 467                 force_successful_syscall_return();
 468                 return ns->ns_id;
 469         case NVME_IOCTL_ADMIN_CMD:
 470                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 471         case NVME_IOCTL_IO_CMD:
 472                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 473         case NVME_IOCTL_SUBMIT_IO:
 474                 return nvme_submit_io(ns, (void __user *)arg);
 475 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 476         case SG_GET_VERSION_NUM:
 477                 return nvme_sg_get_version_num((void __user *)arg);
 478         case SG_IO:
 479                 return nvme_sg_io(ns, (void __user *)arg);
 480 #endif
 481         default:
 482                 return -ENOTTY;
 483         }
 484 }
 485
 486 #ifdef CONFIG_COMPAT
 487 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 488                         unsigned int cmd, unsigned long arg)
 489 {
 490         switch (cmd) {
 491         case SG_IO:
 492                 return -ENOIOCTLCMD;
 493         }
 494         return nvme_ioctl(bdev, mode, cmd, arg);
 495 }
 496 #else
 497 #define nvme_compat_ioctl       NULL
 498 #endif
 499
 500 static int nvme_open(struct block_device *bdev, fmode_t mode)
 501 {
 502         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 503 }
 504
 505 static void nvme_release(struct gendisk *disk, fmode_t mode)
 506 {
 507         nvme_put_ns(disk->private_data);
 508 }
 509
 510 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 511 {
 512         /* some standard values */
 513         geo->heads = 1 << 6;
 514         geo->sectors = 1 << 5;
 515         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 516         return 0;
 517 }
 518
 519 #ifdef CONFIG_BLK_DEV_INTEGRITY
 520 static void nvme_init_integrity(struct nvme_ns *ns)
 521 {
 522         struct blk_integrity integrity;
 523
 524         switch (ns->pi_type) {
 525         case NVME_NS_DPS_PI_TYPE3:
 526                 integrity.profile = &t10_pi_type3_crc;
 527                 break;
 528         case NVME_NS_DPS_PI_TYPE1:
 529         case NVME_NS_DPS_PI_TYPE2:
 530                 integrity.profile = &t10_pi_type1_crc;
 531                 break;
 532         default:
 533                 integrity.profile = NULL;
 534                 break;
 535         }
 536         integrity.tuple_size = ns->ms;
 537         blk_integrity_register(ns->disk, &integrity);
 538         blk_queue_max_integrity_segments(ns->queue, 1);
 539 }
 540 #else
 541 static void nvme_init_integrity(struct nvme_ns *ns)
 542 {
 543 }
 544 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 545
 546 static void nvme_config_discard(struct nvme_ns *ns)
 547 {
 548         u32 logical_block_size = queue_logical_block_size(ns->queue);
 549         ns->queue->limits.discard_zeroes_data = 0;
 550         ns->queue->limits.discard_alignment = logical_block_size;
 551         ns->queue->limits.discard_granularity = logical_block_size;
 552         blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
 553         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 554 }
 555
 556 static int nvme_revalidate_disk(struct gendisk *disk)
 557 {
 558         struct nvme_ns *ns = disk->private_data;
 559         struct nvme_id_ns *id;
 560         u8 lbaf, pi_type;
 561         u16 old_ms;
 562         unsigned short bs;
 563
 564         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
 565                 set_capacity(disk, 0);
 566                 return -ENODEV;
 567         }
 568         if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
 569                 dev_warn(ns->ctrl->dev, "%s: Identify failure nvme%dn%d\n",
 570                                 __func__, ns->ctrl->instance, ns->ns_id);
 571                 return -ENODEV;
 572         }
 573         if (id->ncap == 0) {
 574                 kfree(id);
 575                 return -ENODEV;
 576         }
 577
 578         if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
 579                 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
 580                         dev_warn(ns->ctrl->dev,
 581                                 "%s: LightNVM init failure\n", __func__);
 582                         kfree(id);
 583                         return -ENODEV;
 584                 }
 585                 ns->type = NVME_NS_LIGHTNVM;
 586         }
 587
 588         if (ns->ctrl->vs >= NVME_VS(1, 1))
 589                 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
 590         if (ns->ctrl->vs >= NVME_VS(1, 2))
 591                 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
 592
 593         old_ms = ns->ms;
 594         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 595         ns->lba_shift = id->lbaf[lbaf].ds;
 596         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 597         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 598
 599         /*
 600          * If identify namespace failed, use default 512 byte block size so
 601          * block layer can use before failing read/write for 0 capacity.
 602          */
 603         if (ns->lba_shift == 0)
 604                 ns->lba_shift = 9;
 605         bs = 1 << ns->lba_shift;
 606         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 607         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 608                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 609
 610         blk_mq_freeze_queue(disk->queue);
 611         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 612                                 ns->ms != old_ms ||
 613                                 bs != queue_logical_block_size(disk->queue) ||
 614                                 (ns->ms && ns->ext)))
 615                 blk_integrity_unregister(disk);
 616
 617         ns->pi_type = pi_type;
 618         blk_queue_logical_block_size(ns->queue, bs);
 619
 620         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 621                 nvme_init_integrity(ns);
 622         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 623                 set_capacity(disk, 0);
 624         else
 625                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 626
 627         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 628                 nvme_config_discard(ns);
 629         blk_mq_unfreeze_queue(disk->queue);
 630
 631         kfree(id);
 632         return 0;
 633 }
 634
 635 static char nvme_pr_type(enum pr_type type)
 636 {
 637         switch (type) {
 638         case PR_WRITE_EXCLUSIVE:
 639                 return 1;
 640         case PR_EXCLUSIVE_ACCESS:
 641                 return 2;
 642         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 643                 return 3;
 644         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 645                 return 4;
 646         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 647                 return 5;
 648         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 649                 return 6;
 650         default:
 651                 return 0;
 652         }
 653 };
 654
 655 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 656                                 u64 key, u64 sa_key, u8 op)
 657 {
 658         struct nvme_ns *ns = bdev->bd_disk->private_data;
 659         struct nvme_command c;
 660         u8 data[16] = { 0, };
 661
 662         put_unaligned_le64(key, &data[0]);
 663         put_unaligned_le64(sa_key, &data[8]);
 664
 665         memset(&c, 0, sizeof(c));
 666         c.common.opcode = op;
 667         c.common.nsid = cpu_to_le32(ns->ns_id);
 668         c.common.cdw10[0] = cpu_to_le32(cdw10);
 669
 670         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
 671 }
 672
 673 static int nvme_pr_register(struct block_device *bdev, u64 old,
 674                 u64 new, unsigned flags)
 675 {
 676         u32 cdw10;
 677
 678         if (flags & ~PR_FL_IGNORE_KEY)
 679                 return -EOPNOTSUPP;
 680
 681         cdw10 = old ? 2 : 0;
 682         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
 683         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
 684         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
 685 }
 686
 687 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
 688                 enum pr_type type, unsigned flags)
 689 {
 690         u32 cdw10;
 691
 692         if (flags & ~PR_FL_IGNORE_KEY)
 693                 return -EOPNOTSUPP;
 694
 695         cdw10 = nvme_pr_type(type) << 8;
 696         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
 697         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
 698 }
 699
 700 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
 701                 enum pr_type type, bool abort)
 702 {
 703         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
 704         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
 705 }
 706
 707 static int nvme_pr_clear(struct block_device *bdev, u64 key)
 708 {
 709         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
 710         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
 711 }
 712
 713 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
 714 {
 715         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
 716         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
 717 }
 718
 719 static const struct pr_ops nvme_pr_ops = {
 720         .pr_register    = nvme_pr_register,
 721         .pr_reserve     = nvme_pr_reserve,
 722         .pr_release     = nvme_pr_release,
 723         .pr_preempt     = nvme_pr_preempt,
 724         .pr_clear       = nvme_pr_clear,
 725 };
 726
 727 static const struct block_device_operations nvme_fops = {
 728         .owner          = THIS_MODULE,
 729         .ioctl          = nvme_ioctl,
 730         .compat_ioctl   = nvme_compat_ioctl,
 731         .open           = nvme_open,
 732         .release        = nvme_release,
 733         .getgeo         = nvme_getgeo,
 734         .revalidate_disk= nvme_revalidate_disk,
 735         .pr_ops         = &nvme_pr_ops,
 736 };
 737
 738 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
 739 {
 740         unsigned long timeout =
 741                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
 742         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
 743         int ret;
 744
 745         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 746                 if ((csts & NVME_CSTS_RDY) == bit)
 747                         break;
 748
 749                 msleep(100);
 750                 if (fatal_signal_pending(current))
 751                         return -EINTR;
 752                 if (time_after(jiffies, timeout)) {
 753                         dev_err(ctrl->dev,
 754                                 "Device not ready; aborting %s\n", enabled ?
 755                                                 "initialisation" : "reset");
 756                         return -ENODEV;
 757                 }
 758         }
 759
 760         return ret;
 761 }
 762
 763 /*
 764  * If the device has been passed off to us in an enabled state, just clear
 765  * the enabled bit.  The spec says we should set the 'shutdown notification
 766  * bits', but doing so may cause the device to complete commands to the
 767  * admin queue ... and we don't know what memory that might be pointing at!
 768  */
 769 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 770 {
 771         int ret;
 772
 773         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 774         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
 775
 776         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 777         if (ret)
 778                 return ret;
 779         return nvme_wait_ready(ctrl, cap, false);
 780 }
 781
 782 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 783 {
 784         /*
 785          * Default to a 4K page size, with the intention to update this
 786          * path in the future to accomodate architectures with differing
 787          * kernel and IO page sizes.
 788          */
 789         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
 790         int ret;
 791
 792         if (page_shift < dev_page_min) {
 793                 dev_err(ctrl->dev,
 794                         "Minimum device page size %u too large for host (%u)\n",
 795                         1 << dev_page_min, 1 << page_shift);
 796                 return -ENODEV;
 797         }
 798
 799         ctrl->page_size = 1 << page_shift;
 800
 801         ctrl->ctrl_config = NVME_CC_CSS_NVM;
 802         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
 803         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
 804         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
 805         ctrl->ctrl_config |= NVME_CC_ENABLE;
 806
 807         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 808         if (ret)
 809                 return ret;
 810         return nvme_wait_ready(ctrl, cap, true);
 811 }
 812
 813 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
 814 {
 815         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
 816         u32 csts;
 817         int ret;
 818
 819         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 820         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
 821
 822         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 823         if (ret)
 824                 return ret;
 825
 826         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 827                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
 828                         break;
 829
 830                 msleep(100);
 831                 if (fatal_signal_pending(current))
 832                         return -EINTR;
 833                 if (time_after(jiffies, timeout)) {
 834                         dev_err(ctrl->dev,
 835                                 "Device shutdown incomplete; abort shutdown\n");
 836                         return -ENODEV;
 837                 }
 838         }
 839
 840         return ret;
 841 }
 842
 843 /*
 844  * Initialize the cached copies of the Identify data and various controller
 845  * register in our nvme_ctrl structure.  This should be called as soon as
 846  * the admin queue is fully up and running.
 847  */
 848 int nvme_init_identify(struct nvme_ctrl *ctrl)
 849 {
 850         struct nvme_id_ctrl *id;
 851         u64 cap;
 852         int ret, page_shift;
 853
 854         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
 855         if (ret) {
 856                 dev_err(ctrl->dev, "Reading VS failed (%d)\n", ret);
 857                 return ret;
 858         }
 859
 860         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
 861         if (ret) {
 862                 dev_err(ctrl->dev, "Reading CAP failed (%d)\n", ret);
 863                 return ret;
 864         }
 865         page_shift = NVME_CAP_MPSMIN(cap) + 12;
 866
 867         if (ctrl->vs >= NVME_VS(1, 1))
 868                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
 869
 870         ret = nvme_identify_ctrl(ctrl, &id);
 871         if (ret) {
 872                 dev_err(ctrl->dev, "Identify Controller failed (%d)\n", ret);
 873                 return -EIO;
 874         }
 875
 876         ctrl->oncs = le16_to_cpup(&id->oncs);
 877         atomic_set(&ctrl->abort_limit, id->acl + 1);
 878         ctrl->vwc = id->vwc;
 879         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
 880         memcpy(ctrl->model, id->mn, sizeof(id->mn));
 881         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
 882         if (id->mdts)
 883                 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
 884         else
 885                 ctrl->max_hw_sectors = UINT_MAX;
 886
 887         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
 888                 unsigned int max_hw_sectors;
 889
 890                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
 891                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
 892                 if (ctrl->max_hw_sectors) {
 893                         ctrl->max_hw_sectors = min(max_hw_sectors,
 894                                                         ctrl->max_hw_sectors);
 895                 } else {
 896                         ctrl->max_hw_sectors = max_hw_sectors;
 897                 }
 898         }
 899
 900         kfree(id);
 901         return 0;
 902 }
 903
 904 static int nvme_dev_open(struct inode *inode, struct file *file)
 905 {
 906         struct nvme_ctrl *ctrl;
 907         int instance = iminor(inode);
 908         int ret = -ENODEV;
 909
 910         spin_lock(&dev_list_lock);
 911         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
 912                 if (ctrl->instance != instance)
 913                         continue;
 914
 915                 if (!ctrl->admin_q) {
 916                         ret = -EWOULDBLOCK;
 917                         break;
 918                 }
 919                 if (!kref_get_unless_zero(&ctrl->kref))
 920                         break;
 921                 file->private_data = ctrl;
 922                 ret = 0;
 923                 break;
 924         }
 925         spin_unlock(&dev_list_lock);
 926
 927         return ret;
 928 }
 929
 930 static int nvme_dev_release(struct inode *inode, struct file *file)
 931 {
 932         nvme_put_ctrl(file->private_data);
 933         return 0;
 934 }
 935
 936 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
 937 {
 938         struct nvme_ns *ns;
 939         int ret;
 940
 941         mutex_lock(&ctrl->namespaces_mutex);
 942         if (list_empty(&ctrl->namespaces)) {
 943                 ret = -ENOTTY;
 944                 goto out_unlock;
 945         }
 946
 947         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
 948         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
 949                 dev_warn(ctrl->dev,
 950                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
 951                 ret = -EINVAL;
 952                 goto out_unlock;
 953         }
 954
 955         dev_warn(ctrl->dev,
 956                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
 957         kref_get(&ns->kref);
 958         mutex_unlock(&ctrl->namespaces_mutex);
 959
 960         ret = nvme_user_cmd(ctrl, ns, argp);
 961         nvme_put_ns(ns);
 962         return ret;
 963
 964 out_unlock:
 965         mutex_unlock(&ctrl->namespaces_mutex);
 966         return ret;
 967 }
 968
 969 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
 970                 unsigned long arg)
 971 {
 972         struct nvme_ctrl *ctrl = file->private_data;
 973         void __user *argp = (void __user *)arg;
 974
 975         switch (cmd) {
 976         case NVME_IOCTL_ADMIN_CMD:
 977                 return nvme_user_cmd(ctrl, NULL, argp);
 978         case NVME_IOCTL_IO_CMD:
 979                 return nvme_dev_user_cmd(ctrl, argp);
 980         case NVME_IOCTL_RESET:
 981                 dev_warn(ctrl->dev, "resetting controller\n");
 982                 return ctrl->ops->reset_ctrl(ctrl);
 983         case NVME_IOCTL_SUBSYS_RESET:
 984                 return nvme_reset_subsystem(ctrl);
 985         default:
 986                 return -ENOTTY;
 987         }
 988 }
 989
 990 static const struct file_operations nvme_dev_fops = {
 991         .owner          = THIS_MODULE,
 992         .open           = nvme_dev_open,
 993         .release        = nvme_dev_release,
 994         .unlocked_ioctl = nvme_dev_ioctl,
 995         .compat_ioctl   = nvme_dev_ioctl,
 996 };
 997
 998 static ssize_t nvme_sysfs_reset(struct device *dev,
 999                                 struct device_attribute *attr, const char *buf,
1000                                 size_t count)
1001 {
1002         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1003         int ret;
1004
1005         ret = ctrl->ops->reset_ctrl(ctrl);
1006         if (ret < 0)
1007                 return ret;
1008         return count;
1009 }
1010 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1011
1012 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1013                                                                 char *buf)
1014 {
1015         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1016         return sprintf(buf, "%pU\n", ns->uuid);
1017 }
1018 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1019
1020 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1021                                                                 char *buf)
1022 {
1023         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1024         return sprintf(buf, "%8phd\n", ns->eui);
1025 }
1026 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1027
1028 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1029                                                                 char *buf)
1030 {
1031         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1032         return sprintf(buf, "%d\n", ns->ns_id);
1033 }
1034 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1035
1036 static struct attribute *nvme_ns_attrs[] = {
1037         &dev_attr_uuid.attr,
1038         &dev_attr_eui.attr,
1039         &dev_attr_nsid.attr,
1040         NULL,
1041 };
1042
1043 static umode_t nvme_attrs_are_visible(struct kobject *kobj,
1044                 struct attribute *a, int n)
1045 {
1046         struct device *dev = container_of(kobj, struct device, kobj);
1047         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1048
1049         if (a == &dev_attr_uuid.attr) {
1050                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1051                         return 0;
1052         }
1053         if (a == &dev_attr_eui.attr) {
1054                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1055                         return 0;
1056         }
1057         return a->mode;
1058 }
1059
1060 static const struct attribute_group nvme_ns_attr_group = {
1061         .attrs          = nvme_ns_attrs,
1062         .is_visible     = nvme_attrs_are_visible,
1063 };
1064
1065 #define nvme_show_function(field)                                               \
1066 static ssize_t  field##_show(struct device *dev,                                \
1067                             struct device_attribute *attr, char *buf)           \
1068 {                                                                               \
1069         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1070         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1071 }                                                                               \
1072 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1073
1074 nvme_show_function(model);
1075 nvme_show_function(serial);
1076 nvme_show_function(firmware_rev);
1077
1078 static struct attribute *nvme_dev_attrs[] = {
1079         &dev_attr_reset_controller.attr,
1080         &dev_attr_model.attr,
1081         &dev_attr_serial.attr,
1082         &dev_attr_firmware_rev.attr,
1083         NULL
1084 };
1085
1086 static struct attribute_group nvme_dev_attrs_group = {
1087         .attrs = nvme_dev_attrs,
1088 };
1089
1090 static const struct attribute_group *nvme_dev_attr_groups[] = {
1091         &nvme_dev_attrs_group,
1092         NULL,
1093 };
1094
1095 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1096 {
1097         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1098         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1099
1100         return nsa->ns_id - nsb->ns_id;
1101 }
1102
1103 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1104 {
1105         struct nvme_ns *ns;
1106
1107         lockdep_assert_held(&ctrl->namespaces_mutex);
1108
1109         list_for_each_entry(ns, &ctrl->namespaces, list) {
1110                 if (ns->ns_id == nsid)
1111                         return ns;
1112                 if (ns->ns_id > nsid)
1113                         break;
1114         }
1115         return NULL;
1116 }
1117
1118 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1119 {
1120         struct nvme_ns *ns;
1121         struct gendisk *disk;
1122         int node = dev_to_node(ctrl->dev);
1123
1124         lockdep_assert_held(&ctrl->namespaces_mutex);
1125
1126         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1127         if (!ns)
1128                 return;
1129
1130         ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1131         if (ns->instance < 0)
1132                 goto out_free_ns;
1133
1134         ns->queue = blk_mq_init_queue(ctrl->tagset);
1135         if (IS_ERR(ns->queue))
1136                 goto out_release_instance;
1137         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1138         ns->queue->queuedata = ns;
1139         ns->ctrl = ctrl;
1140
1141         disk = alloc_disk_node(0, node);
1142         if (!disk)
1143                 goto out_free_queue;
1144
1145         kref_init(&ns->kref);
1146         ns->ns_id = nsid;
1147         ns->disk = disk;
1148         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1149
1150         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1151         if (ctrl->max_hw_sectors) {
1152                 blk_queue_max_hw_sectors(ns->queue, ctrl->max_hw_sectors);
1153                 blk_queue_max_segments(ns->queue,
1154                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1);
1155         }
1156         if (ctrl->stripe_size)
1157                 blk_queue_chunk_sectors(ns->queue, ctrl->stripe_size >> 9);
1158         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1159                 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
1160         blk_queue_virt_boundary(ns->queue, ctrl->page_size - 1);
1161
1162         disk->major = nvme_major;
1163         disk->first_minor = 0;
1164         disk->fops = &nvme_fops;
1165         disk->private_data = ns;
1166         disk->queue = ns->queue;
1167         disk->driverfs_dev = ctrl->device;
1168         disk->flags = GENHD_FL_EXT_DEVT;
1169         sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1170
1171         if (nvme_revalidate_disk(ns->disk))
1172                 goto out_free_disk;
1173
1174         list_add_tail(&ns->list, &ctrl->namespaces);
1175         kref_get(&ctrl->kref);
1176         if (ns->type == NVME_NS_LIGHTNVM)
1177                 return;
1178
1179         add_disk(ns->disk);
1180         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1181                                         &nvme_ns_attr_group))
1182                 pr_warn("%s: failed to create sysfs group for identification\n",
1183                         ns->disk->disk_name);
1184         return;
1185  out_free_disk:
1186         kfree(disk);
1187  out_free_queue:
1188         blk_cleanup_queue(ns->queue);
1189  out_release_instance:
1190         ida_simple_remove(&ctrl->ns_ida, ns->instance);
1191  out_free_ns:
1192         kfree(ns);
1193 }
1194
1195 static void nvme_ns_remove(struct nvme_ns *ns)
1196 {
1197         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1198                 return;
1199
1200         if (ns->disk->flags & GENHD_FL_UP) {
1201                 if (blk_get_integrity(ns->disk))
1202                         blk_integrity_unregister(ns->disk);
1203                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1204                                         &nvme_ns_attr_group);
1205                 del_gendisk(ns->disk);
1206                 blk_mq_abort_requeue_list(ns->queue);
1207                 blk_cleanup_queue(ns->queue);
1208         }
1209         mutex_lock(&ns->ctrl->namespaces_mutex);
1210         list_del_init(&ns->list);
1211         mutex_unlock(&ns->ctrl->namespaces_mutex);
1212         nvme_put_ns(ns);
1213 }
1214
1215 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1216 {
1217         struct nvme_ns *ns;
1218
1219         ns = nvme_find_ns(ctrl, nsid);
1220         if (ns) {
1221                 if (revalidate_disk(ns->disk))
1222                         nvme_ns_remove(ns);
1223         } else
1224                 nvme_alloc_ns(ctrl, nsid);
1225 }
1226
1227 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1228 {
1229         struct nvme_ns *ns;
1230         __le32 *ns_list;
1231         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1232         int ret = 0;
1233
1234         ns_list = kzalloc(0x1000, GFP_KERNEL);
1235         if (!ns_list)
1236                 return -ENOMEM;
1237
1238         for (i = 0; i < num_lists; i++) {
1239                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1240                 if (ret)
1241                         goto out;
1242
1243                 for (j = 0; j < min(nn, 1024U); j++) {
1244                         nsid = le32_to_cpu(ns_list[j]);
1245                         if (!nsid)
1246                                 goto out;
1247
1248                         nvme_validate_ns(ctrl, nsid);
1249
1250                         while (++prev < nsid) {
1251                                 ns = nvme_find_ns(ctrl, prev);
1252                                 if (ns)
1253                                         nvme_ns_remove(ns);
1254                         }
1255                 }
1256                 nn -= j;
1257         }
1258  out:
1259         kfree(ns_list);
1260         return ret;
1261 }
1262
1263 static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn)
1264 {
1265         struct nvme_ns *ns, *next;
1266         unsigned i;
1267
1268         lockdep_assert_held(&ctrl->namespaces_mutex);
1269
1270         for (i = 1; i <= nn; i++)
1271                 nvme_validate_ns(ctrl, i);
1272
1273         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1274                 if (ns->ns_id > nn)
1275                         nvme_ns_remove(ns);
1276         }
1277 }
1278
1279 void nvme_scan_namespaces(struct nvme_ctrl *ctrl)
1280 {
1281         struct nvme_id_ctrl *id;
1282         unsigned nn;
1283
1284         if (nvme_identify_ctrl(ctrl, &id))
1285                 return;
1286
1287         mutex_lock(&ctrl->namespaces_mutex);
1288         nn = le32_to_cpu(id->nn);
1289         if (ctrl->vs >= NVME_VS(1, 1) &&
1290             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1291                 if (!nvme_scan_ns_list(ctrl, nn))
1292                         goto done;
1293         }
1294         __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn));
1295  done:
1296         list_sort(NULL, &ctrl->namespaces, ns_cmp);
1297         mutex_unlock(&ctrl->namespaces_mutex);
1298         kfree(id);
1299 }
1300
1301 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1302 {
1303         struct nvme_ns *ns, *next;
1304
1305         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1306                 nvme_ns_remove(ns);
1307 }
1308
1309 static DEFINE_IDA(nvme_instance_ida);
1310
1311 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1312 {
1313         int instance, error;
1314
1315         do {
1316                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1317                         return -ENODEV;
1318
1319                 spin_lock(&dev_list_lock);
1320                 error = ida_get_new(&nvme_instance_ida, &instance);
1321                 spin_unlock(&dev_list_lock);
1322         } while (error == -EAGAIN);
1323
1324         if (error)
1325                 return -ENODEV;
1326
1327         ctrl->instance = instance;
1328         return 0;
1329 }
1330
1331 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1332 {
1333         spin_lock(&dev_list_lock);
1334         ida_remove(&nvme_instance_ida, ctrl->instance);
1335         spin_unlock(&dev_list_lock);
1336 }
1337
1338 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1339  {
1340         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1341
1342         spin_lock(&dev_list_lock);
1343         list_del(&ctrl->node);
1344         spin_unlock(&dev_list_lock);
1345 }
1346
1347 static void nvme_free_ctrl(struct kref *kref)
1348 {
1349         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1350
1351         put_device(ctrl->device);
1352         nvme_release_instance(ctrl);
1353         ida_destroy(&ctrl->ns_ida);
1354
1355         ctrl->ops->free_ctrl(ctrl);
1356 }
1357
1358 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1359 {
1360         kref_put(&ctrl->kref, nvme_free_ctrl);
1361 }
1362
1363 /*
1364  * Initialize a NVMe controller structures.  This needs to be called during
1365  * earliest initialization so that we have the initialized structured around
1366  * during probing.
1367  */
1368 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1369                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1370 {
1371         int ret;
1372
1373         INIT_LIST_HEAD(&ctrl->namespaces);
1374         mutex_init(&ctrl->namespaces_mutex);
1375         kref_init(&ctrl->kref);
1376         ctrl->dev = dev;
1377         ctrl->ops = ops;
1378         ctrl->quirks = quirks;
1379
1380         ret = nvme_set_instance(ctrl);
1381         if (ret)
1382                 goto out;
1383
1384         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1385                                 MKDEV(nvme_char_major, ctrl->instance),
1386                                 dev, nvme_dev_attr_groups,
1387                                 "nvme%d", ctrl->instance);
1388         if (IS_ERR(ctrl->device)) {
1389                 ret = PTR_ERR(ctrl->device);
1390                 goto out_release_instance;
1391         }
1392         get_device(ctrl->device);
1393         dev_set_drvdata(ctrl->device, ctrl);
1394         ida_init(&ctrl->ns_ida);
1395
1396         spin_lock(&dev_list_lock);
1397         list_add_tail(&ctrl->node, &nvme_ctrl_list);
1398         spin_unlock(&dev_list_lock);
1399
1400         return 0;
1401 out_release_instance:
1402         nvme_release_instance(ctrl);
1403 out:
1404         return ret;
1405 }
1406
1407 /**
1408  * nvme_kill_queues(): Ends all namespace queues
1409  * @ctrl: the dead controller that needs to end
1410  *
1411  * Call this function when the driver determines it is unable to get the
1412  * controller in a state capable of servicing IO.
1413  */
1414 void nvme_kill_queues(struct nvme_ctrl *ctrl)
1415 {
1416         struct nvme_ns *ns;
1417
1418         mutex_lock(&ctrl->namespaces_mutex);
1419         list_for_each_entry(ns, &ctrl->namespaces, list) {
1420                 if (!kref_get_unless_zero(&ns->kref))
1421                         continue;
1422
1423                 /*
1424                  * Revalidating a dead namespace sets capacity to 0. This will
1425                  * end buffered writers dirtying pages that can't be synced.
1426                  */
1427                 if (!test_and_set_bit(NVME_NS_DEAD, &ns->flags))
1428                         revalidate_disk(ns->disk);
1429
1430                 blk_set_queue_dying(ns->queue);
1431                 blk_mq_abort_requeue_list(ns->queue);
1432                 blk_mq_start_stopped_hw_queues(ns->queue, true);
1433
1434                 nvme_put_ns(ns);
1435         }
1436         mutex_unlock(&ctrl->namespaces_mutex);
1437 }
1438
1439 void nvme_stop_queues(struct nvme_ctrl *ctrl)
1440 {
1441         struct nvme_ns *ns;
1442
1443         mutex_lock(&ctrl->namespaces_mutex);
1444         list_for_each_entry(ns, &ctrl->namespaces, list) {
1445                 spin_lock_irq(ns->queue->queue_lock);
1446                 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1447                 spin_unlock_irq(ns->queue->queue_lock);
1448
1449                 blk_mq_cancel_requeue_work(ns->queue);
1450                 blk_mq_stop_hw_queues(ns->queue);
1451         }
1452         mutex_unlock(&ctrl->namespaces_mutex);
1453 }
1454
1455 void nvme_start_queues(struct nvme_ctrl *ctrl)
1456 {
1457         struct nvme_ns *ns;
1458
1459         mutex_lock(&ctrl->namespaces_mutex);
1460         list_for_each_entry(ns, &ctrl->namespaces, list) {
1461                 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1462                 blk_mq_start_stopped_hw_queues(ns->queue, true);
1463                 blk_mq_kick_requeue_list(ns->queue);
1464         }
1465         mutex_unlock(&ctrl->namespaces_mutex);
1466 }
1467
1468 int __init nvme_core_init(void)
1469 {
1470         int result;
1471
1472         result = register_blkdev(nvme_major, "nvme");
1473         if (result < 0)
1474                 return result;
1475         else if (result > 0)
1476                 nvme_major = result;
1477
1478         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1479                                                         &nvme_dev_fops);
1480         if (result < 0)
1481                 goto unregister_blkdev;
1482         else if (result > 0)
1483                 nvme_char_major = result;
1484
1485         nvme_class = class_create(THIS_MODULE, "nvme");
1486         if (IS_ERR(nvme_class)) {
1487                 result = PTR_ERR(nvme_class);
1488                 goto unregister_chrdev;
1489         }
1490
1491         return 0;
1492
1493  unregister_chrdev:
1494         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1495  unregister_blkdev:
1496         unregister_blkdev(nvme_major, "nvme");
1497         return result;
1498 }
1499
1500 void nvme_core_exit(void)
1501 {
1502         unregister_blkdev(nvme_major, "nvme");
1503         class_destroy(nvme_class);
1504         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1505 }