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         put_disk(ns->disk);
  73         ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
  74         nvme_put_ctrl(ns->ctrl);
  75         kfree(ns);
  76 }
  77
  78 static void nvme_put_ns(struct nvme_ns *ns)
  79 {
  80         kref_put(&ns->kref, nvme_free_ns);
  81 }
  82
  83 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
  84 {
  85         struct nvme_ns *ns;
  86
  87         spin_lock(&dev_list_lock);
  88         ns = disk->private_data;
  89         if (ns) {
  90                 if (!kref_get_unless_zero(&ns->kref))
  91                         goto fail;
  92                 if (!try_module_get(ns->ctrl->ops->module))
  93                         goto fail_put_ns;
  94         }
  95         spin_unlock(&dev_list_lock);
  96
  97         return ns;
  98
  99 fail_put_ns:
 100         kref_put(&ns->kref, nvme_free_ns);
 101 fail:
 102         spin_unlock(&dev_list_lock);
 103         return NULL;
 104 }
 105
 106 void nvme_requeue_req(struct request *req)
 107 {
 108         unsigned long flags;
 109
 110         blk_mq_requeue_request(req);
 111         spin_lock_irqsave(req->q->queue_lock, flags);
 112         if (!blk_queue_stopped(req->q))
 113                 blk_mq_kick_requeue_list(req->q);
 114         spin_unlock_irqrestore(req->q->queue_lock, flags);
 115 }
 116 EXPORT_SYMBOL_GPL(nvme_requeue_req);
 117
 118 struct request *nvme_alloc_request(struct request_queue *q,
 119                 struct nvme_command *cmd, unsigned int flags)
 120 {
 121         bool write = cmd->common.opcode & 1;
 122         struct request *req;
 123
 124         req = blk_mq_alloc_request(q, write, flags);
 125         if (IS_ERR(req))
 126                 return req;
 127
 128         req->cmd_type = REQ_TYPE_DRV_PRIV;
 129         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 130         req->__data_len = 0;
 131         req->__sector = (sector_t) -1;
 132         req->bio = req->biotail = NULL;
 133
 134         req->cmd = (unsigned char *)cmd;
 135         req->cmd_len = sizeof(struct nvme_command);
 136
 137         return req;
 138 }
 139 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 140
 141 static inline void nvme_setup_flush(struct nvme_ns *ns,
 142                 struct nvme_command *cmnd)
 143 {
 144         memset(cmnd, 0, sizeof(*cmnd));
 145         cmnd->common.opcode = nvme_cmd_flush;
 146         cmnd->common.nsid = cpu_to_le32(ns->ns_id);
 147 }
 148
 149 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 150                 struct nvme_command *cmnd)
 151 {
 152         struct nvme_dsm_range *range;
 153         struct page *page;
 154         int offset;
 155         unsigned int nr_bytes = blk_rq_bytes(req);
 156
 157         range = kmalloc(sizeof(*range), GFP_ATOMIC);
 158         if (!range)
 159                 return BLK_MQ_RQ_QUEUE_BUSY;
 160
 161         range->cattr = cpu_to_le32(0);
 162         range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
 163         range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 164
 165         memset(cmnd, 0, sizeof(*cmnd));
 166         cmnd->dsm.opcode = nvme_cmd_dsm;
 167         cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
 168         cmnd->dsm.nr = 0;
 169         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 170
 171         req->completion_data = range;
 172         page = virt_to_page(range);
 173         offset = offset_in_page(range);
 174         blk_add_request_payload(req, page, offset, sizeof(*range));
 175
 176         /*
 177          * we set __data_len back to the size of the area to be discarded
 178          * on disk. This allows us to report completion on the full amount
 179          * of blocks described by the request.
 180          */
 181         req->__data_len = nr_bytes;
 182
 183         return 0;
 184 }
 185
 186 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
 187                 struct nvme_command *cmnd)
 188 {
 189         u16 control = 0;
 190         u32 dsmgmt = 0;
 191
 192         if (req->cmd_flags & REQ_FUA)
 193                 control |= NVME_RW_FUA;
 194         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 195                 control |= NVME_RW_LR;
 196
 197         if (req->cmd_flags & REQ_RAHEAD)
 198                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 199
 200         memset(cmnd, 0, sizeof(*cmnd));
 201         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 202         cmnd->rw.command_id = req->tag;
 203         cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
 204         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 205         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 206
 207         if (ns->ms) {
 208                 switch (ns->pi_type) {
 209                 case NVME_NS_DPS_PI_TYPE3:
 210                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 211                         break;
 212                 case NVME_NS_DPS_PI_TYPE1:
 213                 case NVME_NS_DPS_PI_TYPE2:
 214                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 215                                         NVME_RW_PRINFO_PRCHK_REF;
 216                         cmnd->rw.reftag = cpu_to_le32(
 217                                         nvme_block_nr(ns, blk_rq_pos(req)));
 218                         break;
 219                 }
 220                 if (!blk_integrity_rq(req))
 221                         control |= NVME_RW_PRINFO_PRACT;
 222         }
 223
 224         cmnd->rw.control = cpu_to_le16(control);
 225         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 226 }
 227
 228 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 229                 struct nvme_command *cmd)
 230 {
 231         int ret = 0;
 232
 233         if (req->cmd_type == REQ_TYPE_DRV_PRIV)
 234                 memcpy(cmd, req->cmd, sizeof(*cmd));
 235         else if (req->cmd_flags & REQ_FLUSH)
 236                 nvme_setup_flush(ns, cmd);
 237         else if (req->cmd_flags & REQ_DISCARD)
 238                 ret = nvme_setup_discard(ns, req, cmd);
 239         else
 240                 nvme_setup_rw(ns, req, cmd);
 241
 242         return ret;
 243 }
 244 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 245
 246 /*
 247  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 248  * if the result is positive, it's an NVM Express status code
 249  */
 250 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 251                 struct nvme_completion *cqe, void *buffer, unsigned bufflen,
 252                 unsigned timeout)
 253 {
 254         struct request *req;
 255         int ret;
 256
 257         req = nvme_alloc_request(q, cmd, 0);
 258         if (IS_ERR(req))
 259                 return PTR_ERR(req);
 260
 261         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 262         req->special = cqe;
 263
 264         if (buffer && bufflen) {
 265                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 266                 if (ret)
 267                         goto out;
 268         }
 269
 270         blk_execute_rq(req->q, NULL, req, 0);
 271         ret = req->errors;
 272  out:
 273         blk_mq_free_request(req);
 274         return ret;
 275 }
 276
 277 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 278                 void *buffer, unsigned bufflen)
 279 {
 280         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0);
 281 }
 282 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 283
 284 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 285                 void __user *ubuffer, unsigned bufflen,
 286                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 287                 u32 *result, unsigned timeout)
 288 {
 289         bool write = cmd->common.opcode & 1;
 290         struct nvme_completion cqe;
 291         struct nvme_ns *ns = q->queuedata;
 292         struct gendisk *disk = ns ? ns->disk : NULL;
 293         struct request *req;
 294         struct bio *bio = NULL;
 295         void *meta = NULL;
 296         int ret;
 297
 298         req = nvme_alloc_request(q, cmd, 0);
 299         if (IS_ERR(req))
 300                 return PTR_ERR(req);
 301
 302         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 303         req->special = &cqe;
 304
 305         if (ubuffer && bufflen) {
 306                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 307                                 GFP_KERNEL);
 308                 if (ret)
 309                         goto out;
 310                 bio = req->bio;
 311
 312                 if (!disk)
 313                         goto submit;
 314                 bio->bi_bdev = bdget_disk(disk, 0);
 315                 if (!bio->bi_bdev) {
 316                         ret = -ENODEV;
 317                         goto out_unmap;
 318                 }
 319
 320                 if (meta_buffer && meta_len) {
 321                         struct bio_integrity_payload *bip;
 322
 323                         meta = kmalloc(meta_len, GFP_KERNEL);
 324                         if (!meta) {
 325                                 ret = -ENOMEM;
 326                                 goto out_unmap;
 327                         }
 328
 329                         if (write) {
 330                                 if (copy_from_user(meta, meta_buffer,
 331                                                 meta_len)) {
 332                                         ret = -EFAULT;
 333                                         goto out_free_meta;
 334                                 }
 335                         }
 336
 337                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 338                         if (IS_ERR(bip)) {
 339                                 ret = PTR_ERR(bip);
 340                                 goto out_free_meta;
 341                         }
 342
 343                         bip->bip_iter.bi_size = meta_len;
 344                         bip->bip_iter.bi_sector = meta_seed;
 345
 346                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 347                                         meta_len, offset_in_page(meta));
 348                         if (ret != meta_len) {
 349                                 ret = -ENOMEM;
 350                                 goto out_free_meta;
 351                         }
 352                 }
 353         }
 354  submit:
 355         blk_execute_rq(req->q, disk, req, 0);
 356         ret = req->errors;
 357         if (result)
 358                 *result = le32_to_cpu(cqe.result);
 359         if (meta && !ret && !write) {
 360                 if (copy_to_user(meta_buffer, meta, meta_len))
 361                         ret = -EFAULT;
 362         }
 363  out_free_meta:
 364         kfree(meta);
 365  out_unmap:
 366         if (bio) {
 367                 if (disk && bio->bi_bdev)
 368                         bdput(bio->bi_bdev);
 369                 blk_rq_unmap_user(bio);
 370         }
 371  out:
 372         blk_mq_free_request(req);
 373         return ret;
 374 }
 375
 376 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 377                 void __user *ubuffer, unsigned bufflen, u32 *result,
 378                 unsigned timeout)
 379 {
 380         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 381                         result, timeout);
 382 }
 383
 384 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 385 {
 386         struct nvme_command c = { };
 387         int error;
 388
 389         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 390         c.identify.opcode = nvme_admin_identify;
 391         c.identify.cns = cpu_to_le32(1);
 392
 393         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 394         if (!*id)
 395                 return -ENOMEM;
 396
 397         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 398                         sizeof(struct nvme_id_ctrl));
 399         if (error)
 400                 kfree(*id);
 401         return error;
 402 }
 403
 404 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 405 {
 406         struct nvme_command c = { };
 407
 408         c.identify.opcode = nvme_admin_identify;
 409         c.identify.cns = cpu_to_le32(2);
 410         c.identify.nsid = cpu_to_le32(nsid);
 411         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 412 }
 413
 414 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 415                 struct nvme_id_ns **id)
 416 {
 417         struct nvme_command c = { };
 418         int error;
 419
 420         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 421         c.identify.opcode = nvme_admin_identify,
 422         c.identify.nsid = cpu_to_le32(nsid),
 423
 424         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 425         if (!*id)
 426                 return -ENOMEM;
 427
 428         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 429                         sizeof(struct nvme_id_ns));
 430         if (error)
 431                 kfree(*id);
 432         return error;
 433 }
 434
 435 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 436                                         dma_addr_t dma_addr, u32 *result)
 437 {
 438         struct nvme_command c;
 439         struct nvme_completion cqe;
 440         int ret;
 441
 442         memset(&c, 0, sizeof(c));
 443         c.features.opcode = nvme_admin_get_features;
 444         c.features.nsid = cpu_to_le32(nsid);
 445         c.features.prp1 = cpu_to_le64(dma_addr);
 446         c.features.fid = cpu_to_le32(fid);
 447
 448         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
 449         if (ret >= 0)
 450                 *result = le32_to_cpu(cqe.result);
 451         return ret;
 452 }
 453
 454 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 455                                         dma_addr_t dma_addr, u32 *result)
 456 {
 457         struct nvme_command c;
 458         struct nvme_completion cqe;
 459         int ret;
 460
 461         memset(&c, 0, sizeof(c));
 462         c.features.opcode = nvme_admin_set_features;
 463         c.features.prp1 = cpu_to_le64(dma_addr);
 464         c.features.fid = cpu_to_le32(fid);
 465         c.features.dword11 = cpu_to_le32(dword11);
 466
 467         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0);
 468         if (ret >= 0)
 469                 *result = le32_to_cpu(cqe.result);
 470         return ret;
 471 }
 472
 473 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 474 {
 475         struct nvme_command c = { };
 476         int error;
 477
 478         c.common.opcode = nvme_admin_get_log_page,
 479         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 480         c.common.cdw10[0] = cpu_to_le32(
 481                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 482                          NVME_LOG_SMART),
 483
 484         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 485         if (!*log)
 486                 return -ENOMEM;
 487
 488         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 489                         sizeof(struct nvme_smart_log));
 490         if (error)
 491                 kfree(*log);
 492         return error;
 493 }
 494
 495 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 496 {
 497         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 498         u32 result;
 499         int status, nr_io_queues;
 500
 501         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
 502                         &result);
 503         if (status)
 504                 return status;
 505
 506         nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 507         *count = min(*count, nr_io_queues);
 508         return 0;
 509 }
 510 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
 511
 512 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 513 {
 514         struct nvme_user_io io;
 515         struct nvme_command c;
 516         unsigned length, meta_len;
 517         void __user *metadata;
 518
 519         if (copy_from_user(&io, uio, sizeof(io)))
 520                 return -EFAULT;
 521         if (io.flags)
 522                 return -EINVAL;
 523
 524         switch (io.opcode) {
 525         case nvme_cmd_write:
 526         case nvme_cmd_read:
 527         case nvme_cmd_compare:
 528                 break;
 529         default:
 530                 return -EINVAL;
 531         }
 532
 533         length = (io.nblocks + 1) << ns->lba_shift;
 534         meta_len = (io.nblocks + 1) * ns->ms;
 535         metadata = (void __user *)(uintptr_t)io.metadata;
 536
 537         if (ns->ext) {
 538                 length += meta_len;
 539                 meta_len = 0;
 540         } else if (meta_len) {
 541                 if ((io.metadata & 3) || !io.metadata)
 542                         return -EINVAL;
 543         }
 544
 545         memset(&c, 0, sizeof(c));
 546         c.rw.opcode = io.opcode;
 547         c.rw.flags = io.flags;
 548         c.rw.nsid = cpu_to_le32(ns->ns_id);
 549         c.rw.slba = cpu_to_le64(io.slba);
 550         c.rw.length = cpu_to_le16(io.nblocks);
 551         c.rw.control = cpu_to_le16(io.control);
 552         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 553         c.rw.reftag = cpu_to_le32(io.reftag);
 554         c.rw.apptag = cpu_to_le16(io.apptag);
 555         c.rw.appmask = cpu_to_le16(io.appmask);
 556
 557         return __nvme_submit_user_cmd(ns->queue, &c,
 558                         (void __user *)(uintptr_t)io.addr, length,
 559                         metadata, meta_len, io.slba, NULL, 0);
 560 }
 561
 562 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 563                         struct nvme_passthru_cmd __user *ucmd)
 564 {
 565         struct nvme_passthru_cmd cmd;
 566         struct nvme_command c;
 567         unsigned timeout = 0;
 568         int status;
 569
 570         if (!capable(CAP_SYS_ADMIN))
 571                 return -EACCES;
 572         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 573                 return -EFAULT;
 574         if (cmd.flags)
 575                 return -EINVAL;
 576
 577         memset(&c, 0, sizeof(c));
 578         c.common.opcode = cmd.opcode;
 579         c.common.flags = cmd.flags;
 580         c.common.nsid = cpu_to_le32(cmd.nsid);
 581         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 582         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 583         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 584         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 585         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 586         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 587         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 588         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 589
 590         if (cmd.timeout_ms)
 591                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 592
 593         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 594                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 595                         &cmd.result, timeout);
 596         if (status >= 0) {
 597                 if (put_user(cmd.result, &ucmd->result))
 598                         return -EFAULT;
 599         }
 600
 601         return status;
 602 }
 603
 604 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 605                 unsigned int cmd, unsigned long arg)
 606 {
 607         struct nvme_ns *ns = bdev->bd_disk->private_data;
 608
 609         switch (cmd) {
 610         case NVME_IOCTL_ID:
 611                 force_successful_syscall_return();
 612                 return ns->ns_id;
 613         case NVME_IOCTL_ADMIN_CMD:
 614                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 615         case NVME_IOCTL_IO_CMD:
 616                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 617         case NVME_IOCTL_SUBMIT_IO:
 618                 return nvme_submit_io(ns, (void __user *)arg);
 619 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 620         case SG_GET_VERSION_NUM:
 621                 return nvme_sg_get_version_num((void __user *)arg);
 622         case SG_IO:
 623                 return nvme_sg_io(ns, (void __user *)arg);
 624 #endif
 625         default:
 626                 return -ENOTTY;
 627         }
 628 }
 629
 630 #ifdef CONFIG_COMPAT
 631 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 632                         unsigned int cmd, unsigned long arg)
 633 {
 634         switch (cmd) {
 635         case SG_IO:
 636                 return -ENOIOCTLCMD;
 637         }
 638         return nvme_ioctl(bdev, mode, cmd, arg);
 639 }
 640 #else
 641 #define nvme_compat_ioctl       NULL
 642 #endif
 643
 644 static int nvme_open(struct block_device *bdev, fmode_t mode)
 645 {
 646         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 647 }
 648
 649 static void nvme_release(struct gendisk *disk, fmode_t mode)
 650 {
 651         struct nvme_ns *ns = disk->private_data;
 652
 653         module_put(ns->ctrl->ops->module);
 654         nvme_put_ns(ns);
 655 }
 656
 657 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 658 {
 659         /* some standard values */
 660         geo->heads = 1 << 6;
 661         geo->sectors = 1 << 5;
 662         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 663         return 0;
 664 }
 665
 666 #ifdef CONFIG_BLK_DEV_INTEGRITY
 667 static void nvme_init_integrity(struct nvme_ns *ns)
 668 {
 669         struct blk_integrity integrity;
 670
 671         switch (ns->pi_type) {
 672         case NVME_NS_DPS_PI_TYPE3:
 673                 integrity.profile = &t10_pi_type3_crc;
 674                 break;
 675         case NVME_NS_DPS_PI_TYPE1:
 676         case NVME_NS_DPS_PI_TYPE2:
 677                 integrity.profile = &t10_pi_type1_crc;
 678                 break;
 679         default:
 680                 integrity.profile = NULL;
 681                 break;
 682         }
 683         integrity.tuple_size = ns->ms;
 684         blk_integrity_register(ns->disk, &integrity);
 685         blk_queue_max_integrity_segments(ns->queue, 1);
 686 }
 687 #else
 688 static void nvme_init_integrity(struct nvme_ns *ns)
 689 {
 690 }
 691 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 692
 693 static void nvme_config_discard(struct nvme_ns *ns)
 694 {
 695         struct nvme_ctrl *ctrl = ns->ctrl;
 696         u32 logical_block_size = queue_logical_block_size(ns->queue);
 697
 698         if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
 699                 ns->queue->limits.discard_zeroes_data = 1;
 700         else
 701                 ns->queue->limits.discard_zeroes_data = 0;
 702
 703         ns->queue->limits.discard_alignment = logical_block_size;
 704         ns->queue->limits.discard_granularity = logical_block_size;
 705         blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
 706         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 707 }
 708
 709 static int nvme_revalidate_disk(struct gendisk *disk)
 710 {
 711         struct nvme_ns *ns = disk->private_data;
 712         struct nvme_id_ns *id;
 713         u8 lbaf, pi_type;
 714         u16 old_ms;
 715         unsigned short bs;
 716
 717         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
 718                 set_capacity(disk, 0);
 719                 return -ENODEV;
 720         }
 721         if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
 722                 dev_warn(disk_to_dev(ns->disk), "%s: Identify failure\n",
 723                                 __func__);
 724                 return -ENODEV;
 725         }
 726         if (id->ncap == 0) {
 727                 kfree(id);
 728                 return -ENODEV;
 729         }
 730
 731         if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
 732                 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
 733                         dev_warn(disk_to_dev(ns->disk),
 734                                 "%s: LightNVM init failure\n", __func__);
 735                         kfree(id);
 736                         return -ENODEV;
 737                 }
 738                 ns->type = NVME_NS_LIGHTNVM;
 739         }
 740
 741         if (ns->ctrl->vs >= NVME_VS(1, 1))
 742                 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
 743         if (ns->ctrl->vs >= NVME_VS(1, 2))
 744                 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
 745
 746         old_ms = ns->ms;
 747         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 748         ns->lba_shift = id->lbaf[lbaf].ds;
 749         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 750         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 751
 752         /*
 753          * If identify namespace failed, use default 512 byte block size so
 754          * block layer can use before failing read/write for 0 capacity.
 755          */
 756         if (ns->lba_shift == 0)
 757                 ns->lba_shift = 9;
 758         bs = 1 << ns->lba_shift;
 759         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 760         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 761                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 762
 763         blk_mq_freeze_queue(disk->queue);
 764         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 765                                 ns->ms != old_ms ||
 766                                 bs != queue_logical_block_size(disk->queue) ||
 767                                 (ns->ms && ns->ext)))
 768                 blk_integrity_unregister(disk);
 769
 770         ns->pi_type = pi_type;
 771         blk_queue_logical_block_size(ns->queue, bs);
 772
 773         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 774                 nvme_init_integrity(ns);
 775         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 776                 set_capacity(disk, 0);
 777         else
 778                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 779
 780         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 781                 nvme_config_discard(ns);
 782         blk_mq_unfreeze_queue(disk->queue);
 783
 784         kfree(id);
 785         return 0;
 786 }
 787
 788 static char nvme_pr_type(enum pr_type type)
 789 {
 790         switch (type) {
 791         case PR_WRITE_EXCLUSIVE:
 792                 return 1;
 793         case PR_EXCLUSIVE_ACCESS:
 794                 return 2;
 795         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 796                 return 3;
 797         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 798                 return 4;
 799         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 800                 return 5;
 801         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 802                 return 6;
 803         default:
 804                 return 0;
 805         }
 806 };
 807
 808 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 809                                 u64 key, u64 sa_key, u8 op)
 810 {
 811         struct nvme_ns *ns = bdev->bd_disk->private_data;
 812         struct nvme_command c;
 813         u8 data[16] = { 0, };
 814
 815         put_unaligned_le64(key, &data[0]);
 816         put_unaligned_le64(sa_key, &data[8]);
 817
 818         memset(&c, 0, sizeof(c));
 819         c.common.opcode = op;
 820         c.common.nsid = cpu_to_le32(ns->ns_id);
 821         c.common.cdw10[0] = cpu_to_le32(cdw10);
 822
 823         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
 824 }
 825
 826 static int nvme_pr_register(struct block_device *bdev, u64 old,
 827                 u64 new, unsigned flags)
 828 {
 829         u32 cdw10;
 830
 831         if (flags & ~PR_FL_IGNORE_KEY)
 832                 return -EOPNOTSUPP;
 833
 834         cdw10 = old ? 2 : 0;
 835         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
 836         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
 837         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
 838 }
 839
 840 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
 841                 enum pr_type type, unsigned flags)
 842 {
 843         u32 cdw10;
 844
 845         if (flags & ~PR_FL_IGNORE_KEY)
 846                 return -EOPNOTSUPP;
 847
 848         cdw10 = nvme_pr_type(type) << 8;
 849         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
 850         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
 851 }
 852
 853 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
 854                 enum pr_type type, bool abort)
 855 {
 856         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
 857         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
 858 }
 859
 860 static int nvme_pr_clear(struct block_device *bdev, u64 key)
 861 {
 862         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
 863         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
 864 }
 865
 866 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
 867 {
 868         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
 869         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
 870 }
 871
 872 static const struct pr_ops nvme_pr_ops = {
 873         .pr_register    = nvme_pr_register,
 874         .pr_reserve     = nvme_pr_reserve,
 875         .pr_release     = nvme_pr_release,
 876         .pr_preempt     = nvme_pr_preempt,
 877         .pr_clear       = nvme_pr_clear,
 878 };
 879
 880 static const struct block_device_operations nvme_fops = {
 881         .owner          = THIS_MODULE,
 882         .ioctl          = nvme_ioctl,
 883         .compat_ioctl   = nvme_compat_ioctl,
 884         .open           = nvme_open,
 885         .release        = nvme_release,
 886         .getgeo         = nvme_getgeo,
 887         .revalidate_disk= nvme_revalidate_disk,
 888         .pr_ops         = &nvme_pr_ops,
 889 };
 890
 891 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
 892 {
 893         unsigned long timeout =
 894                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
 895         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
 896         int ret;
 897
 898         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 899                 if ((csts & NVME_CSTS_RDY) == bit)
 900                         break;
 901
 902                 msleep(100);
 903                 if (fatal_signal_pending(current))
 904                         return -EINTR;
 905                 if (time_after(jiffies, timeout)) {
 906                         dev_err(ctrl->device,
 907                                 "Device not ready; aborting %s\n", enabled ?
 908                                                 "initialisation" : "reset");
 909                         return -ENODEV;
 910                 }
 911         }
 912
 913         return ret;
 914 }
 915
 916 /*
 917  * If the device has been passed off to us in an enabled state, just clear
 918  * the enabled bit.  The spec says we should set the 'shutdown notification
 919  * bits', but doing so may cause the device to complete commands to the
 920  * admin queue ... and we don't know what memory that might be pointing at!
 921  */
 922 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 923 {
 924         int ret;
 925
 926         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 927         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
 928
 929         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 930         if (ret)
 931                 return ret;
 932         return nvme_wait_ready(ctrl, cap, false);
 933 }
 934 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
 935
 936 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 937 {
 938         /*
 939          * Default to a 4K page size, with the intention to update this
 940          * path in the future to accomodate architectures with differing
 941          * kernel and IO page sizes.
 942          */
 943         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
 944         int ret;
 945
 946         if (page_shift < dev_page_min) {
 947                 dev_err(ctrl->device,
 948                         "Minimum device page size %u too large for host (%u)\n",
 949                         1 << dev_page_min, 1 << page_shift);
 950                 return -ENODEV;
 951         }
 952
 953         ctrl->page_size = 1 << page_shift;
 954
 955         ctrl->ctrl_config = NVME_CC_CSS_NVM;
 956         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
 957         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
 958         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
 959         ctrl->ctrl_config |= NVME_CC_ENABLE;
 960
 961         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 962         if (ret)
 963                 return ret;
 964         return nvme_wait_ready(ctrl, cap, true);
 965 }
 966 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
 967
 968 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
 969 {
 970         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
 971         u32 csts;
 972         int ret;
 973
 974         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 975         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
 976
 977         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 978         if (ret)
 979                 return ret;
 980
 981         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 982                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
 983                         break;
 984
 985                 msleep(100);
 986                 if (fatal_signal_pending(current))
 987                         return -EINTR;
 988                 if (time_after(jiffies, timeout)) {
 989                         dev_err(ctrl->device,
 990                                 "Device shutdown incomplete; abort shutdown\n");
 991                         return -ENODEV;
 992                 }
 993         }
 994
 995         return ret;
 996 }
 997 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
 998
 999 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1000                 struct request_queue *q)
1001 {
1002         if (ctrl->max_hw_sectors) {
1003                 u32 max_segments =
1004                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1005
1006                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1007                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1008         }
1009         if (ctrl->stripe_size)
1010                 blk_queue_chunk_sectors(q, ctrl->stripe_size >> 9);
1011         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1012                 blk_queue_flush(q, REQ_FLUSH | REQ_FUA);
1013         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1014 }
1015
1016 /*
1017  * Initialize the cached copies of the Identify data and various controller
1018  * register in our nvme_ctrl structure.  This should be called as soon as
1019  * the admin queue is fully up and running.
1020  */
1021 int nvme_init_identify(struct nvme_ctrl *ctrl)
1022 {
1023         struct nvme_id_ctrl *id;
1024         u64 cap;
1025         int ret, page_shift;
1026
1027         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1028         if (ret) {
1029                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1030                 return ret;
1031         }
1032
1033         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1034         if (ret) {
1035                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1036                 return ret;
1037         }
1038         page_shift = NVME_CAP_MPSMIN(cap) + 12;
1039
1040         if (ctrl->vs >= NVME_VS(1, 1))
1041                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1042
1043         ret = nvme_identify_ctrl(ctrl, &id);
1044         if (ret) {
1045                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1046                 return -EIO;
1047         }
1048
1049         ctrl->vid = le16_to_cpu(id->vid);
1050         ctrl->oncs = le16_to_cpup(&id->oncs);
1051         atomic_set(&ctrl->abort_limit, id->acl + 1);
1052         ctrl->vwc = id->vwc;
1053         ctrl->cntlid = le16_to_cpup(&id->cntlid);
1054         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1055         memcpy(ctrl->model, id->mn, sizeof(id->mn));
1056         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1057         if (id->mdts)
1058                 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1059         else
1060                 ctrl->max_hw_sectors = UINT_MAX;
1061
1062         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
1063                 unsigned int max_hw_sectors;
1064
1065                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
1066                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
1067                 if (ctrl->max_hw_sectors) {
1068                         ctrl->max_hw_sectors = min(max_hw_sectors,
1069                                                         ctrl->max_hw_sectors);
1070                 } else {
1071                         ctrl->max_hw_sectors = max_hw_sectors;
1072                 }
1073         }
1074
1075         nvme_set_queue_limits(ctrl, ctrl->admin_q);
1076
1077         kfree(id);
1078         return 0;
1079 }
1080 EXPORT_SYMBOL_GPL(nvme_init_identify);
1081
1082 static int nvme_dev_open(struct inode *inode, struct file *file)
1083 {
1084         struct nvme_ctrl *ctrl;
1085         int instance = iminor(inode);
1086         int ret = -ENODEV;
1087
1088         spin_lock(&dev_list_lock);
1089         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1090                 if (ctrl->instance != instance)
1091                         continue;
1092
1093                 if (!ctrl->admin_q) {
1094                         ret = -EWOULDBLOCK;
1095                         break;
1096                 }
1097                 if (!kref_get_unless_zero(&ctrl->kref))
1098                         break;
1099                 file->private_data = ctrl;
1100                 ret = 0;
1101                 break;
1102         }
1103         spin_unlock(&dev_list_lock);
1104
1105         return ret;
1106 }
1107
1108 static int nvme_dev_release(struct inode *inode, struct file *file)
1109 {
1110         nvme_put_ctrl(file->private_data);
1111         return 0;
1112 }
1113
1114 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1115 {
1116         struct nvme_ns *ns;
1117         int ret;
1118
1119         mutex_lock(&ctrl->namespaces_mutex);
1120         if (list_empty(&ctrl->namespaces)) {
1121                 ret = -ENOTTY;
1122                 goto out_unlock;
1123         }
1124
1125         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1126         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1127                 dev_warn(ctrl->device,
1128                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1129                 ret = -EINVAL;
1130                 goto out_unlock;
1131         }
1132
1133         dev_warn(ctrl->device,
1134                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1135         kref_get(&ns->kref);
1136         mutex_unlock(&ctrl->namespaces_mutex);
1137
1138         ret = nvme_user_cmd(ctrl, ns, argp);
1139         nvme_put_ns(ns);
1140         return ret;
1141
1142 out_unlock:
1143         mutex_unlock(&ctrl->namespaces_mutex);
1144         return ret;
1145 }
1146
1147 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1148                 unsigned long arg)
1149 {
1150         struct nvme_ctrl *ctrl = file->private_data;
1151         void __user *argp = (void __user *)arg;
1152
1153         switch (cmd) {
1154         case NVME_IOCTL_ADMIN_CMD:
1155                 return nvme_user_cmd(ctrl, NULL, argp);
1156         case NVME_IOCTL_IO_CMD:
1157                 return nvme_dev_user_cmd(ctrl, argp);
1158         case NVME_IOCTL_RESET:
1159                 dev_warn(ctrl->device, "resetting controller\n");
1160                 return ctrl->ops->reset_ctrl(ctrl);
1161         case NVME_IOCTL_SUBSYS_RESET:
1162                 return nvme_reset_subsystem(ctrl);
1163         default:
1164                 return -ENOTTY;
1165         }
1166 }
1167
1168 static const struct file_operations nvme_dev_fops = {
1169         .owner          = THIS_MODULE,
1170         .open           = nvme_dev_open,
1171         .release        = nvme_dev_release,
1172         .unlocked_ioctl = nvme_dev_ioctl,
1173         .compat_ioctl   = nvme_dev_ioctl,
1174 };
1175
1176 static ssize_t nvme_sysfs_reset(struct device *dev,
1177                                 struct device_attribute *attr, const char *buf,
1178                                 size_t count)
1179 {
1180         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1181         int ret;
1182
1183         ret = ctrl->ops->reset_ctrl(ctrl);
1184         if (ret < 0)
1185                 return ret;
1186         return count;
1187 }
1188 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1189
1190 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1191                                                                 char *buf)
1192 {
1193         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1194         struct nvme_ctrl *ctrl = ns->ctrl;
1195         int serial_len = sizeof(ctrl->serial);
1196         int model_len = sizeof(ctrl->model);
1197
1198         if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1199                 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1200
1201         if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1202                 return sprintf(buf, "eui.%8phN\n", ns->eui);
1203
1204         while (ctrl->serial[serial_len - 1] == ' ')
1205                 serial_len--;
1206         while (ctrl->model[model_len - 1] == ' ')
1207                 model_len--;
1208
1209         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1210                 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1211 }
1212 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1213
1214 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1215                                                                 char *buf)
1216 {
1217         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1218         return sprintf(buf, "%pU\n", ns->uuid);
1219 }
1220 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1221
1222 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1223                                                                 char *buf)
1224 {
1225         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1226         return sprintf(buf, "%8phd\n", ns->eui);
1227 }
1228 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1229
1230 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1231                                                                 char *buf)
1232 {
1233         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1234         return sprintf(buf, "%d\n", ns->ns_id);
1235 }
1236 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1237
1238 static struct attribute *nvme_ns_attrs[] = {
1239         &dev_attr_wwid.attr,
1240         &dev_attr_uuid.attr,
1241         &dev_attr_eui.attr,
1242         &dev_attr_nsid.attr,
1243         NULL,
1244 };
1245
1246 static umode_t nvme_attrs_are_visible(struct kobject *kobj,
1247                 struct attribute *a, int n)
1248 {
1249         struct device *dev = container_of(kobj, struct device, kobj);
1250         struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1251
1252         if (a == &dev_attr_uuid.attr) {
1253                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1254                         return 0;
1255         }
1256         if (a == &dev_attr_eui.attr) {
1257                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1258                         return 0;
1259         }
1260         return a->mode;
1261 }
1262
1263 static const struct attribute_group nvme_ns_attr_group = {
1264         .attrs          = nvme_ns_attrs,
1265         .is_visible     = nvme_attrs_are_visible,
1266 };
1267
1268 #define nvme_show_str_function(field)                                           \
1269 static ssize_t  field##_show(struct device *dev,                                \
1270                             struct device_attribute *attr, char *buf)           \
1271 {                                                                               \
1272         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1273         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1274 }                                                                               \
1275 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1276
1277 #define nvme_show_int_function(field)                                           \
1278 static ssize_t  field##_show(struct device *dev,                                \
1279                             struct device_attribute *attr, char *buf)           \
1280 {                                                                               \
1281         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1282         return sprintf(buf, "%d\n", ctrl->field);       \
1283 }                                                                               \
1284 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1285
1286 nvme_show_str_function(model);
1287 nvme_show_str_function(serial);
1288 nvme_show_str_function(firmware_rev);
1289 nvme_show_int_function(cntlid);
1290
1291 static struct attribute *nvme_dev_attrs[] = {
1292         &dev_attr_reset_controller.attr,
1293         &dev_attr_model.attr,
1294         &dev_attr_serial.attr,
1295         &dev_attr_firmware_rev.attr,
1296         &dev_attr_cntlid.attr,
1297         NULL
1298 };
1299
1300 static struct attribute_group nvme_dev_attrs_group = {
1301         .attrs = nvme_dev_attrs,
1302 };
1303
1304 static const struct attribute_group *nvme_dev_attr_groups[] = {
1305         &nvme_dev_attrs_group,
1306         NULL,
1307 };
1308
1309 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1310 {
1311         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1312         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1313
1314         return nsa->ns_id - nsb->ns_id;
1315 }
1316
1317 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1318 {
1319         struct nvme_ns *ns;
1320
1321         lockdep_assert_held(&ctrl->namespaces_mutex);
1322
1323         list_for_each_entry(ns, &ctrl->namespaces, list) {
1324                 if (ns->ns_id == nsid)
1325                         return ns;
1326                 if (ns->ns_id > nsid)
1327                         break;
1328         }
1329         return NULL;
1330 }
1331
1332 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1333 {
1334         struct nvme_ns *ns;
1335         struct gendisk *disk;
1336         int node = dev_to_node(ctrl->dev);
1337
1338         lockdep_assert_held(&ctrl->namespaces_mutex);
1339
1340         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1341         if (!ns)
1342                 return;
1343
1344         ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1345         if (ns->instance < 0)
1346                 goto out_free_ns;
1347
1348         ns->queue = blk_mq_init_queue(ctrl->tagset);
1349         if (IS_ERR(ns->queue))
1350                 goto out_release_instance;
1351         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1352         ns->queue->queuedata = ns;
1353         ns->ctrl = ctrl;
1354
1355         disk = alloc_disk_node(0, node);
1356         if (!disk)
1357                 goto out_free_queue;
1358
1359         kref_init(&ns->kref);
1360         ns->ns_id = nsid;
1361         ns->disk = disk;
1362         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1363
1364
1365         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1366         nvme_set_queue_limits(ctrl, ns->queue);
1367
1368         disk->major = nvme_major;
1369         disk->first_minor = 0;
1370         disk->fops = &nvme_fops;
1371         disk->private_data = ns;
1372         disk->queue = ns->queue;
1373         disk->driverfs_dev = ctrl->device;
1374         disk->flags = GENHD_FL_EXT_DEVT;
1375         sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1376
1377         if (nvme_revalidate_disk(ns->disk))
1378                 goto out_free_disk;
1379
1380         list_add_tail(&ns->list, &ctrl->namespaces);
1381         kref_get(&ctrl->kref);
1382         if (ns->type == NVME_NS_LIGHTNVM)
1383                 return;
1384
1385         add_disk(ns->disk);
1386         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1387                                         &nvme_ns_attr_group))
1388                 pr_warn("%s: failed to create sysfs group for identification\n",
1389                         ns->disk->disk_name);
1390         return;
1391  out_free_disk:
1392         kfree(disk);
1393  out_free_queue:
1394         blk_cleanup_queue(ns->queue);
1395  out_release_instance:
1396         ida_simple_remove(&ctrl->ns_ida, ns->instance);
1397  out_free_ns:
1398         kfree(ns);
1399 }
1400
1401 static void nvme_ns_remove(struct nvme_ns *ns)
1402 {
1403         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1404                 return;
1405
1406         if (ns->disk->flags & GENHD_FL_UP) {
1407                 if (blk_get_integrity(ns->disk))
1408                         blk_integrity_unregister(ns->disk);
1409                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1410                                         &nvme_ns_attr_group);
1411                 del_gendisk(ns->disk);
1412                 blk_mq_abort_requeue_list(ns->queue);
1413                 blk_cleanup_queue(ns->queue);
1414         }
1415         mutex_lock(&ns->ctrl->namespaces_mutex);
1416         list_del_init(&ns->list);
1417         mutex_unlock(&ns->ctrl->namespaces_mutex);
1418         nvme_put_ns(ns);
1419 }
1420
1421 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1422 {
1423         struct nvme_ns *ns;
1424
1425         ns = nvme_find_ns(ctrl, nsid);
1426         if (ns) {
1427                 if (revalidate_disk(ns->disk))
1428                         nvme_ns_remove(ns);
1429         } else
1430                 nvme_alloc_ns(ctrl, nsid);
1431 }
1432
1433 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1434 {
1435         struct nvme_ns *ns;
1436         __le32 *ns_list;
1437         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1438         int ret = 0;
1439
1440         ns_list = kzalloc(0x1000, GFP_KERNEL);
1441         if (!ns_list)
1442                 return -ENOMEM;
1443
1444         for (i = 0; i < num_lists; i++) {
1445                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1446                 if (ret)
1447                         goto out;
1448
1449                 for (j = 0; j < min(nn, 1024U); j++) {
1450                         nsid = le32_to_cpu(ns_list[j]);
1451                         if (!nsid)
1452                                 goto out;
1453
1454                         nvme_validate_ns(ctrl, nsid);
1455
1456                         while (++prev < nsid) {
1457                                 ns = nvme_find_ns(ctrl, prev);
1458                                 if (ns)
1459                                         nvme_ns_remove(ns);
1460                         }
1461                 }
1462                 nn -= j;
1463         }
1464  out:
1465         kfree(ns_list);
1466         return ret;
1467 }
1468
1469 static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn)
1470 {
1471         struct nvme_ns *ns, *next;
1472         unsigned i;
1473
1474         lockdep_assert_held(&ctrl->namespaces_mutex);
1475
1476         for (i = 1; i <= nn; i++)
1477                 nvme_validate_ns(ctrl, i);
1478
1479         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1480                 if (ns->ns_id > nn)
1481                         nvme_ns_remove(ns);
1482         }
1483 }
1484
1485 void nvme_scan_namespaces(struct nvme_ctrl *ctrl)
1486 {
1487         struct nvme_id_ctrl *id;
1488         unsigned nn;
1489
1490         if (nvme_identify_ctrl(ctrl, &id))
1491                 return;
1492
1493         mutex_lock(&ctrl->namespaces_mutex);
1494         nn = le32_to_cpu(id->nn);
1495         if (ctrl->vs >= NVME_VS(1, 1) &&
1496             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1497                 if (!nvme_scan_ns_list(ctrl, nn))
1498                         goto done;
1499         }
1500         __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn));
1501  done:
1502         list_sort(NULL, &ctrl->namespaces, ns_cmp);
1503         mutex_unlock(&ctrl->namespaces_mutex);
1504         kfree(id);
1505 }
1506 EXPORT_SYMBOL_GPL(nvme_scan_namespaces);
1507
1508 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1509 {
1510         struct nvme_ns *ns, *next;
1511
1512         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1513                 nvme_ns_remove(ns);
1514 }
1515 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1516
1517 static DEFINE_IDA(nvme_instance_ida);
1518
1519 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1520 {
1521         int instance, error;
1522
1523         do {
1524                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1525                         return -ENODEV;
1526
1527                 spin_lock(&dev_list_lock);
1528                 error = ida_get_new(&nvme_instance_ida, &instance);
1529                 spin_unlock(&dev_list_lock);
1530         } while (error == -EAGAIN);
1531
1532         if (error)
1533                 return -ENODEV;
1534
1535         ctrl->instance = instance;
1536         return 0;
1537 }
1538
1539 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1540 {
1541         spin_lock(&dev_list_lock);
1542         ida_remove(&nvme_instance_ida, ctrl->instance);
1543         spin_unlock(&dev_list_lock);
1544 }
1545
1546 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1547 {
1548         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1549
1550         spin_lock(&dev_list_lock);
1551         list_del(&ctrl->node);
1552         spin_unlock(&dev_list_lock);
1553 }
1554 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1555
1556 static void nvme_free_ctrl(struct kref *kref)
1557 {
1558         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1559
1560         put_device(ctrl->device);
1561         nvme_release_instance(ctrl);
1562         ida_destroy(&ctrl->ns_ida);
1563
1564         ctrl->ops->free_ctrl(ctrl);
1565 }
1566
1567 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1568 {
1569         kref_put(&ctrl->kref, nvme_free_ctrl);
1570 }
1571 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1572
1573 /*
1574  * Initialize a NVMe controller structures.  This needs to be called during
1575  * earliest initialization so that we have the initialized structured around
1576  * during probing.
1577  */
1578 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1579                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1580 {
1581         int ret;
1582
1583         INIT_LIST_HEAD(&ctrl->namespaces);
1584         mutex_init(&ctrl->namespaces_mutex);
1585         kref_init(&ctrl->kref);
1586         ctrl->dev = dev;
1587         ctrl->ops = ops;
1588         ctrl->quirks = quirks;
1589
1590         ret = nvme_set_instance(ctrl);
1591         if (ret)
1592                 goto out;
1593
1594         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1595                                 MKDEV(nvme_char_major, ctrl->instance),
1596                                 ctrl, nvme_dev_attr_groups,
1597                                 "nvme%d", ctrl->instance);
1598         if (IS_ERR(ctrl->device)) {
1599                 ret = PTR_ERR(ctrl->device);
1600                 goto out_release_instance;
1601         }
1602         get_device(ctrl->device);
1603         ida_init(&ctrl->ns_ida);
1604
1605         spin_lock(&dev_list_lock);
1606         list_add_tail(&ctrl->node, &nvme_ctrl_list);
1607         spin_unlock(&dev_list_lock);
1608
1609         return 0;
1610 out_release_instance:
1611         nvme_release_instance(ctrl);
1612 out:
1613         return ret;
1614 }
1615 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
1616
1617 /**
1618  * nvme_kill_queues(): Ends all namespace queues
1619  * @ctrl: the dead controller that needs to end
1620  *
1621  * Call this function when the driver determines it is unable to get the
1622  * controller in a state capable of servicing IO.
1623  */
1624 void nvme_kill_queues(struct nvme_ctrl *ctrl)
1625 {
1626         struct nvme_ns *ns;
1627
1628         mutex_lock(&ctrl->namespaces_mutex);
1629         list_for_each_entry(ns, &ctrl->namespaces, list) {
1630                 if (!kref_get_unless_zero(&ns->kref))
1631                         continue;
1632
1633                 /*
1634                  * Revalidating a dead namespace sets capacity to 0. This will
1635                  * end buffered writers dirtying pages that can't be synced.
1636                  */
1637                 if (!test_and_set_bit(NVME_NS_DEAD, &ns->flags))
1638                         revalidate_disk(ns->disk);
1639
1640                 blk_set_queue_dying(ns->queue);
1641                 blk_mq_abort_requeue_list(ns->queue);
1642                 blk_mq_start_stopped_hw_queues(ns->queue, true);
1643
1644                 nvme_put_ns(ns);
1645         }
1646         mutex_unlock(&ctrl->namespaces_mutex);
1647 }
1648 EXPORT_SYMBOL_GPL(nvme_kill_queues);
1649
1650 void nvme_stop_queues(struct nvme_ctrl *ctrl)
1651 {
1652         struct nvme_ns *ns;
1653
1654         mutex_lock(&ctrl->namespaces_mutex);
1655         list_for_each_entry(ns, &ctrl->namespaces, list) {
1656                 spin_lock_irq(ns->queue->queue_lock);
1657                 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
1658                 spin_unlock_irq(ns->queue->queue_lock);
1659
1660                 blk_mq_cancel_requeue_work(ns->queue);
1661                 blk_mq_stop_hw_queues(ns->queue);
1662         }
1663         mutex_unlock(&ctrl->namespaces_mutex);
1664 }
1665 EXPORT_SYMBOL_GPL(nvme_stop_queues);
1666
1667 void nvme_start_queues(struct nvme_ctrl *ctrl)
1668 {
1669         struct nvme_ns *ns;
1670
1671         mutex_lock(&ctrl->namespaces_mutex);
1672         list_for_each_entry(ns, &ctrl->namespaces, list) {
1673                 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
1674                 blk_mq_start_stopped_hw_queues(ns->queue, true);
1675                 blk_mq_kick_requeue_list(ns->queue);
1676         }
1677         mutex_unlock(&ctrl->namespaces_mutex);
1678 }
1679 EXPORT_SYMBOL_GPL(nvme_start_queues);
1680
1681 int __init nvme_core_init(void)
1682 {
1683         int result;
1684
1685         result = register_blkdev(nvme_major, "nvme");
1686         if (result < 0)
1687                 return result;
1688         else if (result > 0)
1689                 nvme_major = result;
1690
1691         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1692                                                         &nvme_dev_fops);
1693         if (result < 0)
1694                 goto unregister_blkdev;
1695         else if (result > 0)
1696                 nvme_char_major = result;
1697
1698         nvme_class = class_create(THIS_MODULE, "nvme");
1699         if (IS_ERR(nvme_class)) {
1700                 result = PTR_ERR(nvme_class);
1701                 goto unregister_chrdev;
1702         }
1703
1704         return 0;
1705
1706  unregister_chrdev:
1707         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1708  unregister_blkdev:
1709         unregister_blkdev(nvme_major, "nvme");
1710         return result;
1711 }
1712
1713 void nvme_core_exit(void)
1714 {
1715         unregister_blkdev(nvme_major, "nvme");
1716         class_destroy(nvme_class);
1717         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1718 }
1719
1720 MODULE_LICENSE("GPL");
1721 MODULE_VERSION("1.0");
1722 module_init(nvme_core_init);
1723 module_exit(nvme_core_exit);