drivers/nvme/host/core.c

   1 /*
   2  * NVM Express device driver
   3  * Copyright (c) 2011-2014, Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14
  15 #include <linux/blkdev.h>
  16 #include <linux/blk-mq.h>
  17 #include <linux/delay.h>
  18 #include <linux/errno.h>
  19 #include <linux/hdreg.h>
  20 #include <linux/kernel.h>
  21 #include <linux/module.h>
  22 #include <linux/list_sort.h>
  23 #include <linux/slab.h>
  24 #include <linux/types.h>
  25 #include <linux/pr.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/nvme_ioctl.h>
  28 #include <linux/t10-pi.h>
  29 #include <scsi/sg.h>
  30 #include <asm/unaligned.h>
  31
  32 #include "nvme.h"
  33
  34 #define NVME_MINORS             (1U << MINORBITS)
  35
  36 unsigned char admin_timeout = 60;
  37 module_param(admin_timeout, byte, 0644);
  38 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  39 EXPORT_SYMBOL_GPL(admin_timeout);
  40
  41 unsigned char nvme_io_timeout = 30;
  42 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
  43 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  44 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  45
  46 unsigned char shutdown_timeout = 5;
  47 module_param(shutdown_timeout, byte, 0644);
  48 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  49
  50 static int nvme_major;
  51 module_param(nvme_major, int, 0);
  52
  53 static int nvme_char_major;
  54 module_param(nvme_char_major, int, 0);
  55
  56 static LIST_HEAD(nvme_ctrl_list);
  57 static DEFINE_SPINLOCK(dev_list_lock);
  58
  59 static struct class *nvme_class;
  60
  61 static void nvme_free_ns(struct kref *kref)
  62 {
  63         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
  64
  65         if (ns->type == NVME_NS_LIGHTNVM)
  66                 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
  67
  68         spin_lock(&dev_list_lock);
  69         ns->disk->private_data = NULL;
  70         spin_unlock(&dev_list_lock);
  71
  72         nvme_put_ctrl(ns->ctrl);
  73         put_disk(ns->disk);
  74         kfree(ns);
  75 }
  76
  77 static void nvme_put_ns(struct nvme_ns *ns)
  78 {
  79         kref_put(&ns->kref, nvme_free_ns);
  80 }
  81
  82 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
  83 {
  84         struct nvme_ns *ns;
  85
  86         spin_lock(&dev_list_lock);
  87         ns = disk->private_data;
  88         if (ns) {
  89                 if (!kref_get_unless_zero(&ns->kref))
  90                         goto fail;
  91                 if (!try_module_get(ns->ctrl->ops->module))
  92                         goto fail_put_ns;
  93         }
  94         spin_unlock(&dev_list_lock);
  95
  96         return ns;
  97
  98 fail_put_ns:
  99         kref_put(&ns->kref, nvme_free_ns);
 100 fail:
 101         spin_unlock(&dev_list_lock);
 102         return NULL;
 103 }
 104
 105 void nvme_requeue_req(struct request *req)
 106 {
 107         unsigned long flags;
 108
 109         blk_mq_requeue_request(req);
 110         spin_lock_irqsave(req->q->queue_lock, flags);
 111         if (!blk_queue_stopped(req->q))
 112                 blk_mq_kick_requeue_list(req->q);
 113         spin_unlock_irqrestore(req->q->queue_lock, flags);
 114 }
 115 EXPORT_SYMBOL_GPL(nvme_requeue_req);
 116
 117 struct request *nvme_alloc_request(struct request_queue *q,
 118                 struct nvme_command *cmd, unsigned int flags)
 119 {
 120         bool write = cmd->common.opcode & 1;
 121         struct request *req;
 122
 123         req = blk_mq_alloc_request(q, write, flags);
 124         if (IS_ERR(req))
 125                 return req;
 126
 127         req->cmd_type = REQ_TYPE_DRV_PRIV;
 128         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 129         req->__data_len = 0;
 130         req->__sector = (sector_t) -1;
 131         req->bio = req->biotail = NULL;
 132
 133         req->cmd = (unsigned char *)cmd;
 134         req->cmd_len = sizeof(struct nvme_command);
 135
 136         return req;
 137 }
 138 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 139
 140 /*
 141  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 142  * if the result is positive, it's an NVM Express status code
 143  */
 144 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 145                 struct nvme_completion *cqe, void *buffer, unsigned bufflen,
 146                 unsigned timeout)
 147 {
 148         struct request *req;
 149         int ret;
 150
 151         req = nvme_alloc_request(q, cmd, 0);
 152         if (IS_ERR(req))
 153                 return PTR_ERR(req);
 154
 155         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 156         req->special = cqe;
 157
 158         if (buffer && bufflen) {
 159                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 160                 if (ret)
 161                         goto out;
 162         }
 163
 164         blk_execute_rq(req->q, NULL, req, 0);
 165         ret = req->errors;
 166  out:
 167         blk_mq_free_request(req);
 168         return ret;
 169 }
 170
 171 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 172                 void *buffer, unsigned bufflen)
 173 {
 174         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0);
 175 }
 176 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 177
 178 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 179                 void __user *ubuffer, unsigned bufflen,
 180                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 181                 u32 *result, unsigned timeout)
 182 {
 183         bool write = cmd->common.opcode & 1;
 184         struct nvme_completion cqe;
 185         struct nvme_ns *ns = q->queuedata;
 186         struct gendisk *disk = ns ? ns->disk : NULL;
 187         struct request *req;
 188         struct bio *bio = NULL;
 189         void *meta = NULL;
 190         int ret;
 191
 192         req = nvme_alloc_request(q, cmd, 0);
 193         if (IS_ERR(req))
 194                 return PTR_ERR(req);
 195
 196         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 197         req->special = &cqe;
 198
 199         if (ubuffer && bufflen) {
 200                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 201                                 GFP_KERNEL);
 202                 if (ret)
 203                         goto out;
 204                 bio = req->bio;
 205
 206                 if (!disk)
 207                         goto submit;
 208                 bio->bi_bdev = bdget_disk(disk, 0);
 209                 if (!bio->bi_bdev) {
 210                         ret = -ENODEV;
 211                         goto out_unmap;
 212                 }
 213
 214                 if (meta_buffer) {
 215                         struct bio_integrity_payload *bip;
 216
 217                         meta = kmalloc(meta_len, GFP_KERNEL);
 218                         if (!meta) {
 219                                 ret = -ENOMEM;
 220                                 goto out_unmap;
 221                         }
 222
 223                         if (write) {
 224                                 if (copy_from_user(meta, meta_buffer,
 225                                                 meta_len)) {
 226                                         ret = -EFAULT;
 227                                         goto out_free_meta;
 228                                 }
 229                         }
 230
 231                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 232                         if (IS_ERR(bip)) {
 233                                 ret = PTR_ERR(bip);
 234                                 goto out_free_meta;
 235                         }
 236
 237                         bip->bip_iter.bi_size = meta_len;
 238                         bip->bip_iter.bi_sector = meta_seed;
 239
 240                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 241                                         meta_len, offset_in_page(meta));
 242                         if (ret != meta_len) {
 243                                 ret = -ENOMEM;
 244                                 goto out_free_meta;
 245                         }
 246                 }
 247         }
 248  submit:
 249         blk_execute_rq(req->q, disk, req, 0);
 250         ret = req->errors;
 251         if (result)
 252                 *result = le32_to_cpu(cqe.result);
 253         if (meta && !ret && !write) {
 254                 if (copy_to_user(meta_buffer, meta, meta_len))
 255                         ret = -EFAULT;
 256         }
 257  out_free_meta:
 258         kfree(meta);
 259  out_unmap:
 260         if (bio) {
 261                 if (disk && bio->bi_bdev)
 262                         bdput(bio->bi_bdev);
 263                 blk_rq_unmap_user(bio);
 264         }
 265  out:
 266         blk_mq_free_request(req);
 267         return ret;
 268 }
 269
 270 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 271                 void __user *ubuffer, unsigned bufflen, u32 *result,
 272                 unsigned timeout)
 273 {
 274         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 275                         result, timeout);
 276 }
 277
 278 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 279 {
 280         struct nvme_command c = { };
 281         int error;
 282
 283         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 284         c.identify.opcode = nvme_admin_identify;
 285         c.identify.cns = cpu_to_le32(1);
 286
 287         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 288         if (!*id)
 289                 return -ENOMEM;
 290
 291         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 292                         sizeof(struct nvme_id_ctrl));
 293         if (error)
 294                 kfree(*id);
 295         return error;
 296 }
 297
 298 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 299 {
 300         struct nvme_command c = { };
 301
 302         c.identify.opcode = nvme_admin_identify;
 303         c.identify.cns = cpu_to_le32(2);
 304         c.identify.nsid = cpu_to_le32(nsid);
 305         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 306 }
 307
 308 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 309                 struct nvme_id_ns **id)
 310 {
 311         struct nvme_command c = { };
 312         int error;
 313
 314         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 315         c.identify.opcode = nvme_admin_identify,
 316         c.identify.nsid = cpu_to_le32(nsid),
 317
 318         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 319         if (!*id)
 320                 return -ENOMEM;
 321
 322         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 323                         sizeof(struct nvme_id_ns));
 324         if (error)
 325                 kfree(*id);
 326         return error;
 327 }
 328
 329 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 330                                         dma_addr_t dma_addr, u32 *result)
 331 {
 332         struct nvme_command c;
 333         struct nvme_completion cqe;
 334         int ret;
 335
 336         memset(&c, 0, sizeof(c));
 337         c.features.opcode = nvme_admin_get_features;
 338         c.features.nsid = cpu_to_le32(nsid);
 339         c.features.prp1 = cpu_to_le64(dma_addr);
 340         c.features.fid = cpu_to_le32(fid);
 341
 342         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
 343         if (ret >= 0)
 344                 *result = le32_to_cpu(cqe.result);
 345         return ret;
 346 }
 347
 348 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 349                                         dma_addr_t dma_addr, u32 *result)
 350 {
 351         struct nvme_command c;
 352         struct nvme_completion cqe;
 353         int ret;
 354
 355         memset(&c, 0, sizeof(c));
 356         c.features.opcode = nvme_admin_set_features;
 357         c.features.prp1 = cpu_to_le64(dma_addr);
 358         c.features.fid = cpu_to_le32(fid);
 359         c.features.dword11 = cpu_to_le32(dword11);
 360
 361         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
 362         if (ret >= 0)
 363                 *result = le32_to_cpu(cqe.result);
 364         return ret;
 365 }
 366
 367 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 368 {
 369         struct nvme_command c = { };
 370         int error;
 371
 372         c.common.opcode = nvme_admin_get_log_page,
 373         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 374         c.common.cdw10[0] = cpu_to_le32(
 375                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 376                          NVME_LOG_SMART),
 377
 378         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 379         if (!*log)
 380                 return -ENOMEM;
 381
 382         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 383                         sizeof(struct nvme_smart_log));
 384         if (error)
 385                 kfree(*log);
 386         return error;
 387 }
 388
 389 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 390 {
 391         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 392         u32 result;
 393         int status, nr_io_queues;
 394
 395         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
 396                         &result);
 397         if (status)
 398                 return status;
 399
 400         nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 401         *count = min(*count, nr_io_queues);
 402         return 0;
 403 }
 404 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
 405
 406 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 407 {
 408         struct nvme_user_io io;
 409         struct nvme_command c;
 410         unsigned length, meta_len;
 411         void __user *metadata;
 412
 413         if (copy_from_user(&io, uio, sizeof(io)))
 414                 return -EFAULT;
 415
 416         switch (io.opcode) {
 417         case nvme_cmd_write:
 418         case nvme_cmd_read:
 419         case nvme_cmd_compare:
 420                 break;
 421         default:
 422                 return -EINVAL;
 423         }
 424
 425         length = (io.nblocks + 1) << ns->lba_shift;
 426         meta_len = (io.nblocks + 1) * ns->ms;
 427         metadata = (void __user *)(uintptr_t)io.metadata;
 428
 429         if (ns->ext) {
 430                 length += meta_len;
 431                 meta_len = 0;
 432         } else if (meta_len) {
 433                 if ((io.metadata & 3) || !io.metadata)
 434                         return -EINVAL;
 435         }
 436
 437         memset(&c, 0, sizeof(c));
 438         c.rw.opcode = io.opcode;
 439         c.rw.flags = io.flags;
 440         c.rw.nsid = cpu_to_le32(ns->ns_id);
 441         c.rw.slba = cpu_to_le64(io.slba);
 442         c.rw.length = cpu_to_le16(io.nblocks);
 443         c.rw.control = cpu_to_le16(io.control);
 444         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 445         c.rw.reftag = cpu_to_le32(io.reftag);
 446         c.rw.apptag = cpu_to_le16(io.apptag);
 447         c.rw.appmask = cpu_to_le16(io.appmask);
 448
 449         return __nvme_submit_user_cmd(ns->queue, &c,
 450                         (void __user *)(uintptr_t)io.addr, length,
 451                         metadata, meta_len, io.slba, NULL, 0);
 452 }
 453
 454 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 455                         struct nvme_passthru_cmd __user *ucmd)
 456 {
 457         struct nvme_passthru_cmd cmd;
 458         struct nvme_command c;
 459         unsigned timeout = 0;
 460         int status;
 461
 462         if (!capable(CAP_SYS_ADMIN))
 463                 return -EACCES;
 464         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 465                 return -EFAULT;
 466
 467         memset(&c, 0, sizeof(c));
 468         c.common.opcode = cmd.opcode;
 469         c.common.flags = cmd.flags;
 470         c.common.nsid = cpu_to_le32(cmd.nsid);
 471         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 472         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 473         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 474         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 475         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 476         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 477         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 478         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 479
 480         if (cmd.timeout_ms)
 481                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 482
 483         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 484                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 485                         &cmd.result, timeout);
 486         if (status >= 0) {
 487                 if (put_user(cmd.result, &ucmd->result))
 488                         return -EFAULT;
 489         }
 490
 491         return status;
 492 }
 493
 494 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 495                 unsigned int cmd, unsigned long arg)
 496 {
 497         struct nvme_ns *ns = bdev->bd_disk->private_data;
 498
 499         switch (cmd) {
 500         case NVME_IOCTL_ID:
 501                 force_successful_syscall_return();
 502                 return ns->ns_id;
 503         case NVME_IOCTL_ADMIN_CMD:
 504                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 505         case NVME_IOCTL_IO_CMD:
 506                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 507         case NVME_IOCTL_SUBMIT_IO:
 508                 return nvme_submit_io(ns, (void __user *)arg);
 509 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 510         case SG_GET_VERSION_NUM:
 511                 return nvme_sg_get_version_num((void __user *)arg);
 512         case SG_IO:
 513                 return nvme_sg_io(ns, (void __user *)arg);
 514 #endif
 515         default:
 516                 return -ENOTTY;
 517         }
 518 }
 519
 520 #ifdef CONFIG_COMPAT
 521 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 522                         unsigned int cmd, unsigned long arg)
 523 {
 524         switch (cmd) {
 525         case SG_IO:
 526                 return -ENOIOCTLCMD;
 527         }
 528         return nvme_ioctl(bdev, mode, cmd, arg);
 529 }
 530 #else
 531 #define nvme_compat_ioctl       NULL
 532 #endif
 533
 534 static int nvme_open(struct block_device *bdev, fmode_t mode)
 535 {
 536         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 537 }
 538
 539 static void nvme_release(struct gendisk *disk, fmode_t mode)
 540 {
 541         struct nvme_ns *ns = disk->private_data;
 542
 543         module_put(ns->ctrl->ops->module);
 544         nvme_put_ns(ns);
 545 }
 546
 547 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 548 {
 549         /* some standard values */
 550         geo->heads = 1 << 6;
 551         geo->sectors = 1 << 5;
 552         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 553         return 0;
 554 }
 555
 556 #ifdef CONFIG_BLK_DEV_INTEGRITY
 557 static void nvme_init_integrity(struct nvme_ns *ns)
 558 {
 559         struct blk_integrity integrity;
 560
 561         switch (ns->pi_type) {
 562         case NVME_NS_DPS_PI_TYPE3:
 563                 integrity.profile = &t10_pi_type3_crc;
 564                 break;
 565         case NVME_NS_DPS_PI_TYPE1:
 566         case NVME_NS_DPS_PI_TYPE2:
 567                 integrity.profile = &t10_pi_type1_crc;
 568                 break;
 569         default:
 570                 integrity.profile = NULL;
 571                 break;
 572         }
 573         integrity.tuple_size = ns->ms;
 574         blk_integrity_register(ns->disk, &integrity);
 575         blk_queue_max_integrity_segments(ns->queue, 1);
 576 }
 577 #else
 578 static void nvme_init_integrity(struct nvme_ns *ns)
 579 {
 580 }
 581 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 582
 583 static void nvme_config_discard(struct nvme_ns *ns)
 584 {
 585         struct nvme_ctrl *ctrl = ns->ctrl;
 586         u32 logical_block_size = queue_logical_block_size(ns->queue);
 587
 588         if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
 589                 ns->queue->limits.discard_zeroes_data = 1;
 590         else
 591                 ns->queue->limits.discard_zeroes_data = 0;
 592
 593         ns->queue->limits.discard_alignment = logical_block_size;
 594         ns->queue->limits.discard_granularity = logical_block_size;
 595         blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
 596         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 597 }
 598
 599 static int nvme_revalidate_disk(struct gendisk *disk)
 600 {
 601         struct nvme_ns *ns = disk->private_data;
 602         struct nvme_id_ns *id;
 603         u8 lbaf, pi_type;
 604         u16 old_ms;
 605         unsigned short bs;
 606
 607         if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
 608                 dev_warn(disk_to_dev(ns->disk), "%s: Identify failure\n",
 609                                 __func__);
 610                 return -ENODEV;
 611         }
 612         if (id->ncap == 0) {
 613                 kfree(id);
 614                 return -ENODEV;
 615         }
 616
 617         if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
 618                 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
 619                         dev_warn(disk_to_dev(ns->disk),
 620                                 "%s: LightNVM init failure\n", __func__);
 621                         kfree(id);
 622                         return -ENODEV;
 623                 }
 624                 ns->type = NVME_NS_LIGHTNVM;
 625         }
 626
 627         if (ns->ctrl->vs >= NVME_VS(1, 1))
 628                 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
 629         if (ns->ctrl->vs >= NVME_VS(1, 2))
 630                 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
 631
 632         old_ms = ns->ms;
 633         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 634         ns->lba_shift = id->lbaf[lbaf].ds;
 635         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 636         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 637
 638         /*
 639          * If identify namespace failed, use default 512 byte block size so
 640          * block layer can use before failing read/write for 0 capacity.
 641          */
 642         if (ns->lba_shift == 0)
 643                 ns->lba_shift = 9;
 644         bs = 1 << ns->lba_shift;
 645         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 646         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 647                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 648
 649         blk_mq_freeze_queue(disk->queue);
 650         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 651                                 ns->ms != old_ms ||
 652                                 bs != queue_logical_block_size(disk->queue) ||
 653                                 (ns->ms && ns->ext)))
 654                 blk_integrity_unregister(disk);
 655
 656         ns->pi_type = pi_type;
 657         blk_queue_logical_block_size(ns->queue, bs);
 658
 659         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 660                 nvme_init_integrity(ns);
 661         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 662                 set_capacity(disk, 0);
 663         else
 664                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 665
 666         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 667                 nvme_config_discard(ns);
 668         blk_mq_unfreeze_queue(disk->queue);
 669
 670         kfree(id);
 671         return 0;
 672 }
 673
 674 static char nvme_pr_type(enum pr_type type)
 675 {
 676         switch (type) {
 677         case PR_WRITE_EXCLUSIVE:
 678                 return 1;
 679         case PR_EXCLUSIVE_ACCESS:
 680                 return 2;
 681         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 682                 return 3;
 683         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 684                 return 4;
 685         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 686                 return 5;
 687         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 688                 return 6;
 689         default:
 690                 return 0;
 691         }
 692 };
 693
 694 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 695                                 u64 key, u64 sa_key, u8 op)
 696 {
 697         struct nvme_ns *ns = bdev->bd_disk->private_data;
 698         struct nvme_command c;
 699         u8 data[16] = { 0, };
 700
 701         put_unaligned_le64(key, &data[0]);
 702         put_unaligned_le64(sa_key, &data[8]);
 703
 704         memset(&c, 0, sizeof(c));
 705         c.common.opcode = op;
 706         c.common.nsid = cpu_to_le32(ns->ns_id);
 707         c.common.cdw10[0] = cpu_to_le32(cdw10);
 708
 709         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
 710 }
 711
 712 static int nvme_pr_register(struct block_device *bdev, u64 old,
 713                 u64 new, unsigned flags)
 714 {
 715         u32 cdw10;
 716
 717         if (flags & ~PR_FL_IGNORE_KEY)
 718                 return -EOPNOTSUPP;
 719
 720         cdw10 = old ? 2 : 0;
 721         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
 722         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
 723         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
 724 }
 725
 726 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
 727                 enum pr_type type, unsigned flags)
 728 {
 729         u32 cdw10;
 730
 731         if (flags & ~PR_FL_IGNORE_KEY)
 732                 return -EOPNOTSUPP;
 733
 734         cdw10 = nvme_pr_type(type) << 8;
 735         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
 736         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
 737 }
 738
 739 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
 740                 enum pr_type type, bool abort)
 741 {
 742         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
 743         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
 744 }
 745
 746 static int nvme_pr_clear(struct block_device *bdev, u64 key)
 747 {
 748         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
 749         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
 750 }
 751
 752 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
 753 {
 754         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
 755         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
 756 }
 757
 758 static const struct pr_ops nvme_pr_ops = {
 759         .pr_register    = nvme_pr_register,
 760         .pr_reserve     = nvme_pr_reserve,
 761         .pr_release     = nvme_pr_release,
 762         .pr_preempt     = nvme_pr_preempt,
 763         .pr_clear       = nvme_pr_clear,
 764 };
 765
 766 static const struct block_device_operations nvme_fops = {
 767         .owner          = THIS_MODULE,
 768         .ioctl          = nvme_ioctl,
 769         .compat_ioctl   = nvme_compat_ioctl,
 770         .open           = nvme_open,
 771         .release        = nvme_release,
 772         .getgeo         = nvme_getgeo,
 773         .revalidate_disk= nvme_revalidate_disk,
 774         .pr_ops         = &nvme_pr_ops,
 775 };
 776
 777 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
 778 {
 779         unsigned long timeout =
 780                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
 781         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
 782         int ret;
 783
 784         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 785                 if ((csts & NVME_CSTS_RDY) == bit)
 786                         break;
 787
 788                 msleep(100);
 789                 if (fatal_signal_pending(current))
 790                         return -EINTR;
 791                 if (time_after(jiffies, timeout)) {
 792                         dev_err(ctrl->device,
 793                                 "Device not ready; aborting %s\n", enabled ?
 794                                                 "initialisation" : "reset");
 795                         return -ENODEV;
 796                 }
 797         }
 798
 799         return ret;
 800 }
 801
 802 /*
 803  * If the device has been passed off to us in an enabled state, just clear
 804  * the enabled bit.  The spec says we should set the 'shutdown notification
 805  * bits', but doing so may cause the device to complete commands to the
 806  * admin queue ... and we don't know what memory that might be pointing at!
 807  */
 808 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 809 {
 810         int ret;
 811
 812         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 813         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
 814
 815         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 816         if (ret)
 817                 return ret;
 818         return nvme_wait_ready(ctrl, cap, false);
 819 }
 820 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
 821
 822 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 823 {
 824         /*
 825          * Default to a 4K page size, with the intention to update this
 826          * path in the future to accomodate architectures with differing
 827          * kernel and IO page sizes.
 828          */
 829         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
 830         int ret;
 831
 832         if (page_shift < dev_page_min) {
 833                 dev_err(ctrl->device,
 834                         "Minimum device page size %u too large for host (%u)\n",
 835                         1 << dev_page_min, 1 << page_shift);
 836                 return -ENODEV;
 837         }
 838
 839         ctrl->page_size = 1 << page_shift;
 840
 841         ctrl->ctrl_config = NVME_CC_CSS_NVM;
 842         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
 843         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
 844         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
 845         ctrl->ctrl_config |= NVME_CC_ENABLE;
 846
 847         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 848         if (ret)
 849                 return ret;
 850         return nvme_wait_ready(ctrl, cap, true);
 851 }
 852 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
 853
 854 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
 855 {
 856         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
 857         u32 csts;
 858         int ret;
 859
 860         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 861         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
 862
 863         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 864         if (ret)
 865                 return ret;
 866
 867         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 868                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
 869                         break;
 870
 871                 msleep(100);
 872                 if (fatal_signal_pending(current))
 873                         return -EINTR;
 874                 if (time_after(jiffies, timeout)) {
 875                         dev_err(ctrl->device,
 876                                 "Device shutdown incomplete; abort shutdown\n");
 877                         return -ENODEV;
 878                 }
 879         }
 880
 881         return ret;
 882 }
 883 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
 884
 885 /*
 886  * Initialize the cached copies of the Identify data and various controller
 887  * register in our nvme_ctrl structure.  This should be called as soon as
 888  * the admin queue is fully up and running.
 889  */
 890 int nvme_init_identify(struct nvme_ctrl *ctrl)
 891 {
 892         struct nvme_id_ctrl *id;
 893         u64 cap;
 894         int ret, page_shift;
 895
 896         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
 897         if (ret) {
 898                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
 899                 return ret;
 900         }
 901
 902         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
 903         if (ret) {
 904                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
 905                 return ret;
 906         }
 907         page_shift = NVME_CAP_MPSMIN(cap) + 12;
 908
 909         if (ctrl->vs >= NVME_VS(1, 1))
 910                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
 911
 912         ret = nvme_identify_ctrl(ctrl, &id);
 913         if (ret) {
 914                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
 915                 return -EIO;
 916         }
 917
 918         ctrl->vid = le16_to_cpu(id->vid);
 919         ctrl->oncs = le16_to_cpup(&id->oncs);
 920         atomic_set(&ctrl->abort_limit, id->acl + 1);
 921         ctrl->vwc = id->vwc;
 922         ctrl->cntlid = le16_to_cpup(&id->cntlid);
 923         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
 924         memcpy(ctrl->model, id->mn, sizeof(id->mn));
 925         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
 926         if (id->mdts)
 927                 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
 928         else
 929                 ctrl->max_hw_sectors = UINT_MAX;
 930
 931         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
 932                 unsigned int max_hw_sectors;
 933
 934                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
 935                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
 936                 if (ctrl->max_hw_sectors) {
 937                         ctrl->max_hw_sectors = min(max_hw_sectors,
 938                                                         ctrl->max_hw_sectors);
 939                 } else {
 940                         ctrl->max_hw_sectors = max_hw_sectors;
 941                 }
 942         }
 943
 944         kfree(id);
 945         return 0;
 946 }
 947 EXPORT_SYMBOL_GPL(nvme_init_identify);
 948
 949 static int nvme_dev_open(struct inode *inode, struct file *file)
 950 {
 951         struct nvme_ctrl *ctrl;
 952         int instance = iminor(inode);
 953         int ret = -ENODEV;
 954
 955         spin_lock(&dev_list_lock);
 956         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
 957                 if (ctrl->instance != instance)
 958                         continue;
 959
 960                 if (!ctrl->admin_q) {
 961                         ret = -EWOULDBLOCK;
 962                         break;
 963                 }
 964                 if (!kref_get_unless_zero(&ctrl->kref))
 965                         break;
 966                 file->private_data = ctrl;
 967                 ret = 0;
 968                 break;
 969         }
 970         spin_unlock(&dev_list_lock);
 971
 972         return ret;
 973 }
 974
 975 static int nvme_dev_release(struct inode *inode, struct file *file)
 976 {
 977         nvme_put_ctrl(file->private_data);
 978         return 0;
 979 }
 980
 981 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
 982 {
 983         struct nvme_ns *ns;
 984         int ret;
 985
 986         mutex_lock(&ctrl->namespaces_mutex);
 987         if (list_empty(&ctrl->namespaces)) {
 988                 ret = -ENOTTY;
 989                 goto out_unlock;
 990         }
 991
 992         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
 993         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
 994                 dev_warn(ctrl->device,
 995                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
 996                 ret = -EINVAL;
 997                 goto out_unlock;
 998         }
 999
1000         dev_warn(ctrl->device,
1001                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1002         kref_get(&ns->kref);
1003         mutex_unlock(&ctrl->namespaces_mutex);
1004
1005         ret = nvme_user_cmd(ctrl, ns, argp);
1006         nvme_put_ns(ns);
1007         return ret;
1008
1009 out_unlock:
1010         mutex_unlock(&ctrl->namespaces_mutex);
1011         return ret;
1012 }
1013
1014 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1015                 unsigned long arg)
1016 {
1017         struct nvme_ctrl *ctrl = file->private_data;
1018         void __user *argp = (void __user *)arg;
1019
1020         switch (cmd) {
1021         case NVME_IOCTL_ADMIN_CMD:
1022                 return nvme_user_cmd(ctrl, NULL, argp);
1023         case NVME_IOCTL_IO_CMD:
1024                 return nvme_dev_user_cmd(ctrl, argp);
1025         case NVME_IOCTL_RESET:
1026                 dev_warn(ctrl->device, "resetting controller\n");
1027                 return ctrl->ops->reset_ctrl(ctrl);
1028         case NVME_IOCTL_SUBSYS_RESET:
1029                 return nvme_reset_subsystem(ctrl);
1030         default:
1031                 return -ENOTTY;
1032         }
1033 }
1034
1035 static const struct file_operations nvme_dev_fops = {
1036         .owner          = THIS_MODULE,
1037         .open           = nvme_dev_open,
1038         .release        = nvme_dev_release,
1039         .unlocked_ioctl = nvme_dev_ioctl,
1040         .compat_ioctl   = nvme_dev_ioctl,
1041 };
1042
1043 static ssize_t nvme_sysfs_reset(struct device *dev,
1044                                 struct device_attribute *attr, const char *buf,
1045                                 size_t count)
1046 {
1047         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1048         int ret;
1049
1050         ret = ctrl->ops->reset_ctrl(ctrl);
1051         if (ret < 0)
1052                 return ret;
1053         return count;
1054 }
1055 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1056
1057 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1058                                                                 char *buf)
1059 {
1060         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1061         struct nvme_ctrl *ctrl = ns->ctrl;
1062         int serial_len = sizeof(ctrl->serial);
1063         int model_len = sizeof(ctrl->model);
1064
1065         if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1066                 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1067
1068         if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1069                 return sprintf(buf, "eui.%8phN\n", ns->eui);
1070
1071         while (ctrl->serial[serial_len - 1] == ' ')
1072                 serial_len--;
1073         while (ctrl->model[model_len - 1] == ' ')
1074                 model_len--;
1075
1076         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1077                 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1078 }
1079 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1080
1081 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1082                                                                 char *buf)
1083 {
1084         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1085         return sprintf(buf, "%pU\n", ns->uuid);
1086 }
1087 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1088
1089 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1090                                                                 char *buf)
1091 {
1092         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1093         return sprintf(buf, "%8phd\n", ns->eui);
1094 }
1095 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1096
1097 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1098                                                                 char *buf)
1099 {
1100         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1101         return sprintf(buf, "%d\n", ns->ns_id);
1102 }
1103 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1104
1105 static struct attribute *nvme_ns_attrs[] = {
1106         &dev_attr_wwid.attr,
1107         &dev_attr_uuid.attr,
1108         &dev_attr_eui.attr,
1109         &dev_attr_nsid.attr,
1110         NULL,
1111 };
1112
1113 static umode_t nvme_attrs_are_visible(struct kobject *kobj,
1114                 struct attribute *a, int n)
1115 {
1116         struct device *dev = container_of(kobj, struct device, kobj);
1117         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1118
1119         if (a == &dev_attr_uuid.attr) {
1120                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1121                         return 0;
1122         }
1123         if (a == &dev_attr_eui.attr) {
1124                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1125                         return 0;
1126         }
1127         return a->mode;
1128 }
1129
1130 static const struct attribute_group nvme_ns_attr_group = {
1131         .attrs          = nvme_ns_attrs,
1132         .is_visible     = nvme_attrs_are_visible,
1133 };
1134
1135 #define nvme_show_str_function(field)                                           \
1136 static ssize_t  field##_show(struct device *dev,                                \
1137                             struct device_attribute *attr, char *buf)           \
1138 {                                                                               \
1139         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1140         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1141 }                                                                               \
1142 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1143
1144 #define nvme_show_int_function(field)                                           \
1145 static ssize_t  field##_show(struct device *dev,                                \
1146                             struct device_attribute *attr, char *buf)           \
1147 {                                                                               \
1148         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1149         return sprintf(buf, "%d\n", ctrl->field);       \
1150 }                                                                               \
1151 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1152
1153 nvme_show_str_function(model);
1154 nvme_show_str_function(serial);
1155 nvme_show_str_function(firmware_rev);
1156 nvme_show_int_function(cntlid);
1157
1158 static struct attribute *nvme_dev_attrs[] = {
1159         &dev_attr_reset_controller.attr,
1160         &dev_attr_model.attr,
1161         &dev_attr_serial.attr,
1162         &dev_attr_firmware_rev.attr,
1163         &dev_attr_cntlid.attr,
1164         NULL
1165 };
1166
1167 static struct attribute_group nvme_dev_attrs_group = {
1168         .attrs = nvme_dev_attrs,
1169 };
1170
1171 static const struct attribute_group *nvme_dev_attr_groups[] = {
1172         &nvme_dev_attrs_group,
1173         NULL,
1174 };
1175
1176 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1177 {
1178         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1179         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1180
1181         return nsa->ns_id - nsb->ns_id;
1182 }
1183
1184 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1185 {
1186         struct nvme_ns *ns;
1187
1188         lockdep_assert_held(&ctrl->namespaces_mutex);
1189
1190         list_for_each_entry(ns, &ctrl->namespaces, list) {
1191                 if (ns->ns_id == nsid)
1192                         return ns;
1193                 if (ns->ns_id > nsid)
1194                         break;
1195         }
1196         return NULL;
1197 }
1198
1199 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1200 {
1201         struct nvme_ns *ns;
1202         struct gendisk *disk;
1203         int node = dev_to_node(ctrl->dev);
1204
1205         lockdep_assert_held(&ctrl->namespaces_mutex);
1206
1207         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1208         if (!ns)
1209                 return;
1210
1211         ns->queue = blk_mq_init_queue(ctrl->tagset);
1212         if (IS_ERR(ns->queue))
1213                 goto out_free_ns;
1214         queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
1215         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1216         ns->queue->queuedata = ns;
1217         ns->ctrl = ctrl;
1218
1219         disk = alloc_disk_node(0, node);
1220         if (!disk)
1221                 goto out_free_queue;
1222
1223         kref_init(&ns->kref);
1224         ns->ns_id = nsid;
1225         ns->disk = disk;
1226         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1227
1228         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1229         if (ctrl->max_hw_sectors) {
1230                 blk_queue_max_hw_sectors(ns->queue, ctrl->max_hw_sectors);
1231                 blk_queue_max_segments(ns->queue,
1232                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1);
1233         }
1234         if (ctrl->stripe_size)
1235                 blk_queue_chunk_sectors(ns->queue, ctrl->stripe_size >> 9);
1236         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1237                 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
1238         blk_queue_virt_boundary(ns->queue, ctrl->page_size - 1);
1239
1240         disk->major = nvme_major;
1241         disk->first_minor = 0;
1242         disk->fops = &nvme_fops;
1243         disk->private_data = ns;
1244         disk->queue = ns->queue;
1245         disk->driverfs_dev = ctrl->device;
1246         disk->flags = GENHD_FL_EXT_DEVT;
1247         sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, nsid);
1248
1249         if (nvme_revalidate_disk(ns->disk))
1250                 goto out_free_disk;
1251
1252         list_add_tail(&ns->list, &ctrl->namespaces);
1253         kref_get(&ctrl->kref);
1254         if (ns->type == NVME_NS_LIGHTNVM)
1255                 return;
1256
1257         add_disk(ns->disk);
1258         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1259                                         &nvme_ns_attr_group))
1260                 pr_warn("%s: failed to create sysfs group for identification\n",
1261                         ns->disk->disk_name);
1262         return;
1263  out_free_disk:
1264         kfree(disk);
1265  out_free_queue:
1266         blk_cleanup_queue(ns->queue);
1267  out_free_ns:
1268         kfree(ns);
1269 }
1270
1271 static void nvme_ns_remove(struct nvme_ns *ns)
1272 {
1273         bool kill = nvme_io_incapable(ns->ctrl) &&
1274                         !blk_queue_dying(ns->queue);
1275
1276         lockdep_assert_held(&ns->ctrl->namespaces_mutex);
1277
1278         if (kill) {
1279                 blk_set_queue_dying(ns->queue);
1280
1281                 /*
1282                  * The controller was shutdown first if we got here through
1283                  * device removal. The shutdown may requeue outstanding
1284                  * requests. These need to be aborted immediately so
1285                  * del_gendisk doesn't block indefinitely for their completion.
1286                  */
1287                 blk_mq_abort_requeue_list(ns->queue);
1288         }
1289         if (ns->disk->flags & GENHD_FL_UP) {
1290                 if (blk_get_integrity(ns->disk))
1291                         blk_integrity_unregister(ns->disk);
1292                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1293                                         &nvme_ns_attr_group);
1294                 del_gendisk(ns->disk);
1295         }
1296         if (kill || !blk_queue_dying(ns->queue)) {
1297                 blk_mq_abort_requeue_list(ns->queue);
1298                 blk_cleanup_queue(ns->queue);
1299         }
1300         list_del_init(&ns->list);
1301         nvme_put_ns(ns);
1302 }
1303
1304 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1305 {
1306         struct nvme_ns *ns;
1307
1308         ns = nvme_find_ns(ctrl, nsid);
1309         if (ns) {
1310                 if (revalidate_disk(ns->disk))
1311                         nvme_ns_remove(ns);
1312         } else
1313                 nvme_alloc_ns(ctrl, nsid);
1314 }
1315
1316 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1317 {
1318         struct nvme_ns *ns;
1319         __le32 *ns_list;
1320         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1321         int ret = 0;
1322
1323         ns_list = kzalloc(0x1000, GFP_KERNEL);
1324         if (!ns_list)
1325                 return -ENOMEM;
1326
1327         for (i = 0; i < num_lists; i++) {
1328                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1329                 if (ret)
1330                         goto out;
1331
1332                 for (j = 0; j < min(nn, 1024U); j++) {
1333                         nsid = le32_to_cpu(ns_list[j]);
1334                         if (!nsid)
1335                                 goto out;
1336
1337                         nvme_validate_ns(ctrl, nsid);
1338
1339                         while (++prev < nsid) {
1340                                 ns = nvme_find_ns(ctrl, prev);
1341                                 if (ns)
1342                                         nvme_ns_remove(ns);
1343                         }
1344                 }
1345                 nn -= j;
1346         }
1347  out:
1348         kfree(ns_list);
1349         return ret;
1350 }
1351
1352 static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn)
1353 {
1354         struct nvme_ns *ns, *next;
1355         unsigned i;
1356
1357         lockdep_assert_held(&ctrl->namespaces_mutex);
1358
1359         for (i = 1; i <= nn; i++)
1360                 nvme_validate_ns(ctrl, i);
1361
1362         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1363                 if (ns->ns_id > nn)
1364                         nvme_ns_remove(ns);
1365         }
1366 }
1367
1368 void nvme_scan_namespaces(struct nvme_ctrl *ctrl)
1369 {
1370         struct nvme_id_ctrl *id;
1371         unsigned nn;
1372
1373         if (nvme_identify_ctrl(ctrl, &id))
1374                 return;
1375
1376         mutex_lock(&ctrl->namespaces_mutex);
1377         nn = le32_to_cpu(id->nn);
1378         if (ctrl->vs >= NVME_VS(1, 1) &&
1379             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1380                 if (!nvme_scan_ns_list(ctrl, nn))
1381                         goto done;
1382         }
1383         __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn));
1384  done:
1385         list_sort(NULL, &ctrl->namespaces, ns_cmp);
1386         mutex_unlock(&ctrl->namespaces_mutex);
1387         kfree(id);
1388 }
1389 EXPORT_SYMBOL_GPL(nvme_scan_namespaces);
1390
1391 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1392 {
1393         struct nvme_ns *ns, *next;
1394
1395         mutex_lock(&ctrl->namespaces_mutex);
1396         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1397                 nvme_ns_remove(ns);
1398         mutex_unlock(&ctrl->namespaces_mutex);
1399 }
1400 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1401
1402 static DEFINE_IDA(nvme_instance_ida);
1403
1404 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1405 {
1406         int instance, error;
1407
1408         do {
1409                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1410                         return -ENODEV;
1411
1412                 spin_lock(&dev_list_lock);
1413                 error = ida_get_new(&nvme_instance_ida, &instance);
1414                 spin_unlock(&dev_list_lock);
1415         } while (error == -EAGAIN);
1416
1417         if (error)
1418                 return -ENODEV;
1419
1420         ctrl->instance = instance;
1421         return 0;
1422 }
1423
1424 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1425 {
1426         spin_lock(&dev_list_lock);
1427         ida_remove(&nvme_instance_ida, ctrl->instance);
1428         spin_unlock(&dev_list_lock);
1429 }
1430
1431 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1432 {
1433         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1434
1435         spin_lock(&dev_list_lock);
1436         list_del(&ctrl->node);
1437         spin_unlock(&dev_list_lock);
1438 }
1439 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1440
1441 static void nvme_free_ctrl(struct kref *kref)
1442 {
1443         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1444
1445         put_device(ctrl->device);
1446         nvme_release_instance(ctrl);
1447
1448         ctrl->ops->free_ctrl(ctrl);
1449 }
1450
1451 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1452 {
1453         kref_put(&ctrl->kref, nvme_free_ctrl);
1454 }
1455 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1456
1457 /*
1458  * Initialize a NVMe controller structures.  This needs to be called during
1459  * earliest initialization so that we have the initialized structured around
1460  * during probing.
1461  */
1462 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1463                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1464 {
1465         int ret;
1466
1467         INIT_LIST_HEAD(&ctrl->namespaces);
1468         mutex_init(&ctrl->namespaces_mutex);
1469         kref_init(&ctrl->kref);
1470         ctrl->dev = dev;
1471         ctrl->ops = ops;
1472         ctrl->quirks = quirks;
1473
1474         ret = nvme_set_instance(ctrl);
1475         if (ret)
1476                 goto out;
1477
1478         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1479                                 MKDEV(nvme_char_major, ctrl->instance),
1480                                 ctrl, nvme_dev_attr_groups,
1481                                 "nvme%d", ctrl->instance);
1482         if (IS_ERR(ctrl->device)) {
1483                 ret = PTR_ERR(ctrl->device);
1484                 goto out_release_instance;
1485         }
1486         get_device(ctrl->device);
1487
1488         spin_lock(&dev_list_lock);
1489         list_add_tail(&ctrl->node, &nvme_ctrl_list);
1490         spin_unlock(&dev_list_lock);
1491
1492         return 0;
1493 out_release_instance:
1494         nvme_release_instance(ctrl);
1495 out:
1496         return ret;
1497 }
1498 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
1499
1500 void nvme_stop_queues(struct nvme_ctrl *ctrl)
1501 {
1502         struct nvme_ns *ns;
1503
1504         mutex_lock(&ctrl->namespaces_mutex);
1505         list_for_each_entry(ns, &ctrl->namespaces, list) {
1506                 spin_lock_irq(ns->queue->queue_lock);
1507                 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1508                 spin_unlock_irq(ns->queue->queue_lock);
1509
1510                 blk_mq_cancel_requeue_work(ns->queue);
1511                 blk_mq_stop_hw_queues(ns->queue);
1512         }
1513         mutex_unlock(&ctrl->namespaces_mutex);
1514 }
1515 EXPORT_SYMBOL_GPL(nvme_stop_queues);
1516
1517 void nvme_start_queues(struct nvme_ctrl *ctrl)
1518 {
1519         struct nvme_ns *ns;
1520
1521         mutex_lock(&ctrl->namespaces_mutex);
1522         list_for_each_entry(ns, &ctrl->namespaces, list) {
1523                 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1524                 blk_mq_start_stopped_hw_queues(ns->queue, true);
1525                 blk_mq_kick_requeue_list(ns->queue);
1526         }
1527         mutex_unlock(&ctrl->namespaces_mutex);
1528 }
1529 EXPORT_SYMBOL_GPL(nvme_start_queues);
1530
1531 int __init nvme_core_init(void)
1532 {
1533         int result;
1534
1535         result = register_blkdev(nvme_major, "nvme");
1536         if (result < 0)
1537                 return result;
1538         else if (result > 0)
1539                 nvme_major = result;
1540
1541         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1542                                                         &nvme_dev_fops);
1543         if (result < 0)
1544                 goto unregister_blkdev;
1545         else if (result > 0)
1546                 nvme_char_major = result;
1547
1548         nvme_class = class_create(THIS_MODULE, "nvme");
1549         if (IS_ERR(nvme_class)) {
1550                 result = PTR_ERR(nvme_class);
1551                 goto unregister_chrdev;
1552         }
1553
1554         return 0;
1555
1556  unregister_chrdev:
1557         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1558  unregister_blkdev:
1559         unregister_blkdev(nvme_major, "nvme");
1560         return result;
1561 }
1562
1563 void nvme_core_exit(void)
1564 {
1565         unregister_blkdev(nvme_major, "nvme");
1566         class_destroy(nvme_class);
1567         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1568 }
1569
1570 MODULE_LICENSE("GPL");
1571 MODULE_VERSION("1.0");
1572 module_init(nvme_core_init);
1573 module_exit(nvme_core_exit);