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 #include "fabrics.h"
  34
  35 #define NVME_MINORS             (1U << MINORBITS)
  36
  37 unsigned char admin_timeout = 60;
  38 module_param(admin_timeout, byte, 0644);
  39 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  40 EXPORT_SYMBOL_GPL(admin_timeout);
  41
  42 unsigned char nvme_io_timeout = 30;
  43 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
  44 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  45 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  46
  47 unsigned char shutdown_timeout = 5;
  48 module_param(shutdown_timeout, byte, 0644);
  49 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  50
  51 unsigned int nvme_max_retries = 5;
  52 module_param_named(max_retries, nvme_max_retries, uint, 0644);
  53 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  54 EXPORT_SYMBOL_GPL(nvme_max_retries);
  55
  56 static int nvme_char_major;
  57 module_param(nvme_char_major, int, 0);
  58
  59 static LIST_HEAD(nvme_ctrl_list);
  60 static DEFINE_SPINLOCK(dev_list_lock);
  61
  62 static struct class *nvme_class;
  63
  64 void nvme_cancel_request(struct request *req, void *data, bool reserved)
  65 {
  66         int status;
  67
  68         if (!blk_mq_request_started(req))
  69                 return;
  70
  71         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
  72                                 "Cancelling I/O %d", req->tag);
  73
  74         status = NVME_SC_ABORT_REQ;
  75         if (blk_queue_dying(req->q))
  76                 status |= NVME_SC_DNR;
  77         blk_mq_complete_request(req, status);
  78 }
  79 EXPORT_SYMBOL_GPL(nvme_cancel_request);
  80
  81 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
  82                 enum nvme_ctrl_state new_state)
  83 {
  84         enum nvme_ctrl_state old_state;
  85         bool changed = false;
  86
  87         spin_lock_irq(&ctrl->lock);
  88
  89         old_state = ctrl->state;
  90         switch (new_state) {
  91         case NVME_CTRL_LIVE:
  92                 switch (old_state) {
  93                 case NVME_CTRL_NEW:
  94                 case NVME_CTRL_RESETTING:
  95                 case NVME_CTRL_RECONNECTING:
  96                         changed = true;
  97                         /* FALLTHRU */
  98                 default:
  99                         break;
 100                 }
 101                 break;
 102         case NVME_CTRL_RESETTING:
 103                 switch (old_state) {
 104                 case NVME_CTRL_NEW:
 105                 case NVME_CTRL_LIVE:
 106                 case NVME_CTRL_RECONNECTING:
 107                         changed = true;
 108                         /* FALLTHRU */
 109                 default:
 110                         break;
 111                 }
 112                 break;
 113         case NVME_CTRL_RECONNECTING:
 114                 switch (old_state) {
 115                 case NVME_CTRL_LIVE:
 116                         changed = true;
 117                         /* FALLTHRU */
 118                 default:
 119                         break;
 120                 }
 121                 break;
 122         case NVME_CTRL_DELETING:
 123                 switch (old_state) {
 124                 case NVME_CTRL_LIVE:
 125                 case NVME_CTRL_RESETTING:
 126                 case NVME_CTRL_RECONNECTING:
 127                         changed = true;
 128                         /* FALLTHRU */
 129                 default:
 130                         break;
 131                 }
 132                 break;
 133         case NVME_CTRL_DEAD:
 134                 switch (old_state) {
 135                 case NVME_CTRL_DELETING:
 136                         changed = true;
 137                         /* FALLTHRU */
 138                 default:
 139                         break;
 140                 }
 141                 break;
 142         default:
 143                 break;
 144         }
 145
 146         if (changed)
 147                 ctrl->state = new_state;
 148
 149         spin_unlock_irq(&ctrl->lock);
 150
 151         return changed;
 152 }
 153 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 154
 155 static void nvme_free_ns(struct kref *kref)
 156 {
 157         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 158
 159         if (ns->ndev)
 160                 nvme_nvm_unregister(ns);
 161
 162         if (ns->disk) {
 163                 spin_lock(&dev_list_lock);
 164                 ns->disk->private_data = NULL;
 165                 spin_unlock(&dev_list_lock);
 166         }
 167
 168         put_disk(ns->disk);
 169         ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
 170         nvme_put_ctrl(ns->ctrl);
 171         kfree(ns);
 172 }
 173
 174 static void nvme_put_ns(struct nvme_ns *ns)
 175 {
 176         kref_put(&ns->kref, nvme_free_ns);
 177 }
 178
 179 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
 180 {
 181         struct nvme_ns *ns;
 182
 183         spin_lock(&dev_list_lock);
 184         ns = disk->private_data;
 185         if (ns) {
 186                 if (!kref_get_unless_zero(&ns->kref))
 187                         goto fail;
 188                 if (!try_module_get(ns->ctrl->ops->module))
 189                         goto fail_put_ns;
 190         }
 191         spin_unlock(&dev_list_lock);
 192
 193         return ns;
 194
 195 fail_put_ns:
 196         kref_put(&ns->kref, nvme_free_ns);
 197 fail:
 198         spin_unlock(&dev_list_lock);
 199         return NULL;
 200 }
 201
 202 void nvme_requeue_req(struct request *req)
 203 {
 204         unsigned long flags;
 205
 206         blk_mq_requeue_request(req);
 207         spin_lock_irqsave(req->q->queue_lock, flags);
 208         if (!blk_queue_stopped(req->q))
 209                 blk_mq_kick_requeue_list(req->q);
 210         spin_unlock_irqrestore(req->q->queue_lock, flags);
 211 }
 212 EXPORT_SYMBOL_GPL(nvme_requeue_req);
 213
 214 struct request *nvme_alloc_request(struct request_queue *q,
 215                 struct nvme_command *cmd, unsigned int flags, int qid)
 216 {
 217         struct request *req;
 218
 219         if (qid == NVME_QID_ANY) {
 220                 req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
 221         } else {
 222                 req = blk_mq_alloc_request_hctx(q, nvme_is_write(cmd), flags,
 223                                 qid ? qid - 1 : 0);
 224         }
 225         if (IS_ERR(req))
 226                 return req;
 227
 228         req->cmd_type = REQ_TYPE_DRV_PRIV;
 229         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 230         req->cmd = (unsigned char *)cmd;
 231         req->cmd_len = sizeof(struct nvme_command);
 232
 233         return req;
 234 }
 235 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 236
 237 static inline void nvme_setup_flush(struct nvme_ns *ns,
 238                 struct nvme_command *cmnd)
 239 {
 240         memset(cmnd, 0, sizeof(*cmnd));
 241         cmnd->common.opcode = nvme_cmd_flush;
 242         cmnd->common.nsid = cpu_to_le32(ns->ns_id);
 243 }
 244
 245 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 246                 struct nvme_command *cmnd)
 247 {
 248         struct nvme_dsm_range *range;
 249         struct page *page;
 250         int offset;
 251         unsigned int nr_bytes = blk_rq_bytes(req);
 252
 253         range = kmalloc(sizeof(*range), GFP_ATOMIC);
 254         if (!range)
 255                 return BLK_MQ_RQ_QUEUE_BUSY;
 256
 257         range->cattr = cpu_to_le32(0);
 258         range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
 259         range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 260
 261         memset(cmnd, 0, sizeof(*cmnd));
 262         cmnd->dsm.opcode = nvme_cmd_dsm;
 263         cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
 264         cmnd->dsm.nr = 0;
 265         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 266
 267         req->completion_data = range;
 268         page = virt_to_page(range);
 269         offset = offset_in_page(range);
 270         blk_add_request_payload(req, page, offset, sizeof(*range));
 271
 272         /*
 273          * we set __data_len back to the size of the area to be discarded
 274          * on disk. This allows us to report completion on the full amount
 275          * of blocks described by the request.
 276          */
 277         req->__data_len = nr_bytes;
 278
 279         return 0;
 280 }
 281
 282 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
 283                 struct nvme_command *cmnd)
 284 {
 285         u16 control = 0;
 286         u32 dsmgmt = 0;
 287
 288         if (req->cmd_flags & REQ_FUA)
 289                 control |= NVME_RW_FUA;
 290         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 291                 control |= NVME_RW_LR;
 292
 293         if (req->cmd_flags & REQ_RAHEAD)
 294                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 295
 296         memset(cmnd, 0, sizeof(*cmnd));
 297         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 298         cmnd->rw.command_id = req->tag;
 299         cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
 300         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 301         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 302
 303         if (ns->ms) {
 304                 switch (ns->pi_type) {
 305                 case NVME_NS_DPS_PI_TYPE3:
 306                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 307                         break;
 308                 case NVME_NS_DPS_PI_TYPE1:
 309                 case NVME_NS_DPS_PI_TYPE2:
 310                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 311                                         NVME_RW_PRINFO_PRCHK_REF;
 312                         cmnd->rw.reftag = cpu_to_le32(
 313                                         nvme_block_nr(ns, blk_rq_pos(req)));
 314                         break;
 315                 }
 316                 if (!blk_integrity_rq(req))
 317                         control |= NVME_RW_PRINFO_PRACT;
 318         }
 319
 320         cmnd->rw.control = cpu_to_le16(control);
 321         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 322 }
 323
 324 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 325                 struct nvme_command *cmd)
 326 {
 327         int ret = 0;
 328
 329         if (req->cmd_type == REQ_TYPE_DRV_PRIV)
 330                 memcpy(cmd, req->cmd, sizeof(*cmd));
 331         else if (req_op(req) == REQ_OP_FLUSH)
 332                 nvme_setup_flush(ns, cmd);
 333         else if (req_op(req) == REQ_OP_DISCARD)
 334                 ret = nvme_setup_discard(ns, req, cmd);
 335         else
 336                 nvme_setup_rw(ns, req, cmd);
 337
 338         return ret;
 339 }
 340 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 341
 342 /*
 343  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 344  * if the result is positive, it's an NVM Express status code
 345  */
 346 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 347                 struct nvme_completion *cqe, void *buffer, unsigned bufflen,
 348                 unsigned timeout, int qid, int at_head, int flags)
 349 {
 350         struct request *req;
 351         int ret;
 352
 353         req = nvme_alloc_request(q, cmd, flags, qid);
 354         if (IS_ERR(req))
 355                 return PTR_ERR(req);
 356
 357         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 358         req->special = cqe;
 359
 360         if (buffer && bufflen) {
 361                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 362                 if (ret)
 363                         goto out;
 364         }
 365
 366         blk_execute_rq(req->q, NULL, req, at_head);
 367         ret = req->errors;
 368  out:
 369         blk_mq_free_request(req);
 370         return ret;
 371 }
 372 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 373
 374 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 375                 void *buffer, unsigned bufflen)
 376 {
 377         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 378                         NVME_QID_ANY, 0, 0);
 379 }
 380 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 381
 382 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 383                 void __user *ubuffer, unsigned bufflen,
 384                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 385                 u32 *result, unsigned timeout)
 386 {
 387         bool write = nvme_is_write(cmd);
 388         struct nvme_completion cqe;
 389         struct nvme_ns *ns = q->queuedata;
 390         struct gendisk *disk = ns ? ns->disk : NULL;
 391         struct request *req;
 392         struct bio *bio = NULL;
 393         void *meta = NULL;
 394         int ret;
 395
 396         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 397         if (IS_ERR(req))
 398                 return PTR_ERR(req);
 399
 400         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 401         req->special = &cqe;
 402
 403         if (ubuffer && bufflen) {
 404                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 405                                 GFP_KERNEL);
 406                 if (ret)
 407                         goto out;
 408                 bio = req->bio;
 409
 410                 if (!disk)
 411                         goto submit;
 412                 bio->bi_bdev = bdget_disk(disk, 0);
 413                 if (!bio->bi_bdev) {
 414                         ret = -ENODEV;
 415                         goto out_unmap;
 416                 }
 417
 418                 if (meta_buffer && meta_len) {
 419                         struct bio_integrity_payload *bip;
 420
 421                         meta = kmalloc(meta_len, GFP_KERNEL);
 422                         if (!meta) {
 423                                 ret = -ENOMEM;
 424                                 goto out_unmap;
 425                         }
 426
 427                         if (write) {
 428                                 if (copy_from_user(meta, meta_buffer,
 429                                                 meta_len)) {
 430                                         ret = -EFAULT;
 431                                         goto out_free_meta;
 432                                 }
 433                         }
 434
 435                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 436                         if (IS_ERR(bip)) {
 437                                 ret = PTR_ERR(bip);
 438                                 goto out_free_meta;
 439                         }
 440
 441                         bip->bip_iter.bi_size = meta_len;
 442                         bip->bip_iter.bi_sector = meta_seed;
 443
 444                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 445                                         meta_len, offset_in_page(meta));
 446                         if (ret != meta_len) {
 447                                 ret = -ENOMEM;
 448                                 goto out_free_meta;
 449                         }
 450                 }
 451         }
 452  submit:
 453         blk_execute_rq(req->q, disk, req, 0);
 454         ret = req->errors;
 455         if (result)
 456                 *result = le32_to_cpu(cqe.result);
 457         if (meta && !ret && !write) {
 458                 if (copy_to_user(meta_buffer, meta, meta_len))
 459                         ret = -EFAULT;
 460         }
 461  out_free_meta:
 462         kfree(meta);
 463  out_unmap:
 464         if (bio) {
 465                 if (disk && bio->bi_bdev)
 466                         bdput(bio->bi_bdev);
 467                 blk_rq_unmap_user(bio);
 468         }
 469  out:
 470         blk_mq_free_request(req);
 471         return ret;
 472 }
 473
 474 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 475                 void __user *ubuffer, unsigned bufflen, u32 *result,
 476                 unsigned timeout)
 477 {
 478         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 479                         result, timeout);
 480 }
 481
 482 static void nvme_keep_alive_end_io(struct request *rq, int error)
 483 {
 484         struct nvme_ctrl *ctrl = rq->end_io_data;
 485
 486         blk_mq_free_request(rq);
 487
 488         if (error) {
 489                 dev_err(ctrl->device,
 490                         "failed nvme_keep_alive_end_io error=%d\n", error);
 491                 return;
 492         }
 493
 494         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 495 }
 496
 497 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 498 {
 499         struct nvme_command c;
 500         struct request *rq;
 501
 502         memset(&c, 0, sizeof(c));
 503         c.common.opcode = nvme_admin_keep_alive;
 504
 505         rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
 506                         NVME_QID_ANY);
 507         if (IS_ERR(rq))
 508                 return PTR_ERR(rq);
 509
 510         rq->timeout = ctrl->kato * HZ;
 511         rq->end_io_data = ctrl;
 512
 513         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 514
 515         return 0;
 516 }
 517
 518 static void nvme_keep_alive_work(struct work_struct *work)
 519 {
 520         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 521                         struct nvme_ctrl, ka_work);
 522
 523         if (nvme_keep_alive(ctrl)) {
 524                 /* allocation failure, reset the controller */
 525                 dev_err(ctrl->device, "keep-alive failed\n");
 526                 ctrl->ops->reset_ctrl(ctrl);
 527                 return;
 528         }
 529 }
 530
 531 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 532 {
 533         if (unlikely(ctrl->kato == 0))
 534                 return;
 535
 536         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
 537         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 538 }
 539 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
 540
 541 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 542 {
 543         if (unlikely(ctrl->kato == 0))
 544                 return;
 545
 546         cancel_delayed_work_sync(&ctrl->ka_work);
 547 }
 548 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 549
 550 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 551 {
 552         struct nvme_command c = { };
 553         int error;
 554
 555         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 556         c.identify.opcode = nvme_admin_identify;
 557         c.identify.cns = cpu_to_le32(1);
 558
 559         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 560         if (!*id)
 561                 return -ENOMEM;
 562
 563         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 564                         sizeof(struct nvme_id_ctrl));
 565         if (error)
 566                 kfree(*id);
 567         return error;
 568 }
 569
 570 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 571 {
 572         struct nvme_command c = { };
 573
 574         c.identify.opcode = nvme_admin_identify;
 575         c.identify.cns = cpu_to_le32(2);
 576         c.identify.nsid = cpu_to_le32(nsid);
 577         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 578 }
 579
 580 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 581                 struct nvme_id_ns **id)
 582 {
 583         struct nvme_command c = { };
 584         int error;
 585
 586         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 587         c.identify.opcode = nvme_admin_identify,
 588         c.identify.nsid = cpu_to_le32(nsid),
 589
 590         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 591         if (!*id)
 592                 return -ENOMEM;
 593
 594         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 595                         sizeof(struct nvme_id_ns));
 596         if (error)
 597                 kfree(*id);
 598         return error;
 599 }
 600
 601 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 602                       void *buffer, size_t buflen, u32 *result)
 603 {
 604         struct nvme_command c;
 605         struct nvme_completion cqe;
 606         int ret;
 607
 608         memset(&c, 0, sizeof(c));
 609         c.features.opcode = nvme_admin_get_features;
 610         c.features.nsid = cpu_to_le32(nsid);
 611         c.features.fid = cpu_to_le32(fid);
 612
 613         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, buffer, buflen, 0,
 614                         NVME_QID_ANY, 0, 0);
 615         if (ret >= 0 && result)
 616                 *result = le32_to_cpu(cqe.result);
 617         return ret;
 618 }
 619
 620 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 621                       void *buffer, size_t buflen, u32 *result)
 622 {
 623         struct nvme_command c;
 624         struct nvme_completion cqe;
 625         int ret;
 626
 627         memset(&c, 0, sizeof(c));
 628         c.features.opcode = nvme_admin_set_features;
 629         c.features.fid = cpu_to_le32(fid);
 630         c.features.dword11 = cpu_to_le32(dword11);
 631
 632         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe,
 633                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
 634         if (ret >= 0 && result)
 635                 *result = le32_to_cpu(cqe.result);
 636         return ret;
 637 }
 638
 639 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 640 {
 641         struct nvme_command c = { };
 642         int error;
 643
 644         c.common.opcode = nvme_admin_get_log_page,
 645         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 646         c.common.cdw10[0] = cpu_to_le32(
 647                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 648                          NVME_LOG_SMART),
 649
 650         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 651         if (!*log)
 652                 return -ENOMEM;
 653
 654         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 655                         sizeof(struct nvme_smart_log));
 656         if (error)
 657                 kfree(*log);
 658         return error;
 659 }
 660
 661 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 662 {
 663         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 664         u32 result;
 665         int status, nr_io_queues;
 666
 667         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
 668                         &result);
 669         if (status < 0)
 670                 return status;
 671
 672         /*
 673          * Degraded controllers might return an error when setting the queue
 674          * count.  We still want to be able to bring them online and offer
 675          * access to the admin queue, as that might be only way to fix them up.
 676          */
 677         if (status > 0) {
 678                 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
 679                 *count = 0;
 680         } else {
 681                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 682                 *count = min(*count, nr_io_queues);
 683         }
 684
 685         return 0;
 686 }
 687 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
 688
 689 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 690 {
 691         struct nvme_user_io io;
 692         struct nvme_command c;
 693         unsigned length, meta_len;
 694         void __user *metadata;
 695
 696         if (copy_from_user(&io, uio, sizeof(io)))
 697                 return -EFAULT;
 698         if (io.flags)
 699                 return -EINVAL;
 700
 701         switch (io.opcode) {
 702         case nvme_cmd_write:
 703         case nvme_cmd_read:
 704         case nvme_cmd_compare:
 705                 break;
 706         default:
 707                 return -EINVAL;
 708         }
 709
 710         length = (io.nblocks + 1) << ns->lba_shift;
 711         meta_len = (io.nblocks + 1) * ns->ms;
 712         metadata = (void __user *)(uintptr_t)io.metadata;
 713
 714         if (ns->ext) {
 715                 length += meta_len;
 716                 meta_len = 0;
 717         } else if (meta_len) {
 718                 if ((io.metadata & 3) || !io.metadata)
 719                         return -EINVAL;
 720         }
 721
 722         memset(&c, 0, sizeof(c));
 723         c.rw.opcode = io.opcode;
 724         c.rw.flags = io.flags;
 725         c.rw.nsid = cpu_to_le32(ns->ns_id);
 726         c.rw.slba = cpu_to_le64(io.slba);
 727         c.rw.length = cpu_to_le16(io.nblocks);
 728         c.rw.control = cpu_to_le16(io.control);
 729         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 730         c.rw.reftag = cpu_to_le32(io.reftag);
 731         c.rw.apptag = cpu_to_le16(io.apptag);
 732         c.rw.appmask = cpu_to_le16(io.appmask);
 733
 734         return __nvme_submit_user_cmd(ns->queue, &c,
 735                         (void __user *)(uintptr_t)io.addr, length,
 736                         metadata, meta_len, io.slba, NULL, 0);
 737 }
 738
 739 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 740                         struct nvme_passthru_cmd __user *ucmd)
 741 {
 742         struct nvme_passthru_cmd cmd;
 743         struct nvme_command c;
 744         unsigned timeout = 0;
 745         int status;
 746
 747         if (!capable(CAP_SYS_ADMIN))
 748                 return -EACCES;
 749         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 750                 return -EFAULT;
 751         if (cmd.flags)
 752                 return -EINVAL;
 753
 754         memset(&c, 0, sizeof(c));
 755         c.common.opcode = cmd.opcode;
 756         c.common.flags = cmd.flags;
 757         c.common.nsid = cpu_to_le32(cmd.nsid);
 758         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 759         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 760         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 761         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 762         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 763         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 764         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 765         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 766
 767         if (cmd.timeout_ms)
 768                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 769
 770         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 771                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 772                         &cmd.result, timeout);
 773         if (status >= 0) {
 774                 if (put_user(cmd.result, &ucmd->result))
 775                         return -EFAULT;
 776         }
 777
 778         return status;
 779 }
 780
 781 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 782                 unsigned int cmd, unsigned long arg)
 783 {
 784         struct nvme_ns *ns = bdev->bd_disk->private_data;
 785
 786         switch (cmd) {
 787         case NVME_IOCTL_ID:
 788                 force_successful_syscall_return();
 789                 return ns->ns_id;
 790         case NVME_IOCTL_ADMIN_CMD:
 791                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 792         case NVME_IOCTL_IO_CMD:
 793                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 794         case NVME_IOCTL_SUBMIT_IO:
 795                 return nvme_submit_io(ns, (void __user *)arg);
 796 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 797         case SG_GET_VERSION_NUM:
 798                 return nvme_sg_get_version_num((void __user *)arg);
 799         case SG_IO:
 800                 return nvme_sg_io(ns, (void __user *)arg);
 801 #endif
 802         default:
 803                 return -ENOTTY;
 804         }
 805 }
 806
 807 #ifdef CONFIG_COMPAT
 808 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 809                         unsigned int cmd, unsigned long arg)
 810 {
 811         switch (cmd) {
 812         case SG_IO:
 813                 return -ENOIOCTLCMD;
 814         }
 815         return nvme_ioctl(bdev, mode, cmd, arg);
 816 }
 817 #else
 818 #define nvme_compat_ioctl       NULL
 819 #endif
 820
 821 static int nvme_open(struct block_device *bdev, fmode_t mode)
 822 {
 823         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 824 }
 825
 826 static void nvme_release(struct gendisk *disk, fmode_t mode)
 827 {
 828         struct nvme_ns *ns = disk->private_data;
 829
 830         module_put(ns->ctrl->ops->module);
 831         nvme_put_ns(ns);
 832 }
 833
 834 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 835 {
 836         /* some standard values */
 837         geo->heads = 1 << 6;
 838         geo->sectors = 1 << 5;
 839         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 840         return 0;
 841 }
 842
 843 #ifdef CONFIG_BLK_DEV_INTEGRITY
 844 static void nvme_init_integrity(struct nvme_ns *ns)
 845 {
 846         struct blk_integrity integrity;
 847
 848         memset(&integrity, 0, sizeof(integrity));
 849         switch (ns->pi_type) {
 850         case NVME_NS_DPS_PI_TYPE3:
 851                 integrity.profile = &t10_pi_type3_crc;
 852                 integrity.tag_size = sizeof(u16) + sizeof(u32);
 853                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 854                 break;
 855         case NVME_NS_DPS_PI_TYPE1:
 856         case NVME_NS_DPS_PI_TYPE2:
 857                 integrity.profile = &t10_pi_type1_crc;
 858                 integrity.tag_size = sizeof(u16);
 859                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 860                 break;
 861         default:
 862                 integrity.profile = NULL;
 863                 break;
 864         }
 865         integrity.tuple_size = ns->ms;
 866         blk_integrity_register(ns->disk, &integrity);
 867         blk_queue_max_integrity_segments(ns->queue, 1);
 868 }
 869 #else
 870 static void nvme_init_integrity(struct nvme_ns *ns)
 871 {
 872 }
 873 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 874
 875 static void nvme_config_discard(struct nvme_ns *ns)
 876 {
 877         struct nvme_ctrl *ctrl = ns->ctrl;
 878         u32 logical_block_size = queue_logical_block_size(ns->queue);
 879
 880         if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
 881                 ns->queue->limits.discard_zeroes_data = 1;
 882         else
 883                 ns->queue->limits.discard_zeroes_data = 0;
 884
 885         ns->queue->limits.discard_alignment = logical_block_size;
 886         ns->queue->limits.discard_granularity = logical_block_size;
 887         blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
 888         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 889 }
 890
 891 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
 892 {
 893         if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
 894                 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
 895                 return -ENODEV;
 896         }
 897
 898         if ((*id)->ncap == 0) {
 899                 kfree(*id);
 900                 return -ENODEV;
 901         }
 902
 903         if (ns->ctrl->vs >= NVME_VS(1, 1))
 904                 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
 905         if (ns->ctrl->vs >= NVME_VS(1, 2))
 906                 memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
 907
 908         return 0;
 909 }
 910
 911 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
 912 {
 913         struct nvme_ns *ns = disk->private_data;
 914         u8 lbaf, pi_type;
 915         u16 old_ms;
 916         unsigned short bs;
 917
 918         old_ms = ns->ms;
 919         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 920         ns->lba_shift = id->lbaf[lbaf].ds;
 921         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 922         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 923
 924         /*
 925          * If identify namespace failed, use default 512 byte block size so
 926          * block layer can use before failing read/write for 0 capacity.
 927          */
 928         if (ns->lba_shift == 0)
 929                 ns->lba_shift = 9;
 930         bs = 1 << ns->lba_shift;
 931         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 932         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 933                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 934
 935         blk_mq_freeze_queue(disk->queue);
 936         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 937                                 ns->ms != old_ms ||
 938                                 bs != queue_logical_block_size(disk->queue) ||
 939                                 (ns->ms && ns->ext)))
 940                 blk_integrity_unregister(disk);
 941
 942         ns->pi_type = pi_type;
 943         blk_queue_logical_block_size(ns->queue, bs);
 944
 945         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 946                 nvme_init_integrity(ns);
 947         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 948                 set_capacity(disk, 0);
 949         else
 950                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 951
 952         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 953                 nvme_config_discard(ns);
 954         blk_mq_unfreeze_queue(disk->queue);
 955 }
 956
 957 static int nvme_revalidate_disk(struct gendisk *disk)
 958 {
 959         struct nvme_ns *ns = disk->private_data;
 960         struct nvme_id_ns *id = NULL;
 961         int ret;
 962
 963         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
 964                 set_capacity(disk, 0);
 965                 return -ENODEV;
 966         }
 967
 968         ret = nvme_revalidate_ns(ns, &id);
 969         if (ret)
 970                 return ret;
 971
 972         __nvme_revalidate_disk(disk, id);
 973         kfree(id);
 974
 975         return 0;
 976 }
 977
 978 static char nvme_pr_type(enum pr_type type)
 979 {
 980         switch (type) {
 981         case PR_WRITE_EXCLUSIVE:
 982                 return 1;
 983         case PR_EXCLUSIVE_ACCESS:
 984                 return 2;
 985         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 986                 return 3;
 987         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 988                 return 4;
 989         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 990                 return 5;
 991         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 992                 return 6;
 993         default:
 994                 return 0;
 995         }
 996 };
 997
 998 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 999                                 u64 key, u64 sa_key, u8 op)
1000 {
1001         struct nvme_ns *ns = bdev->bd_disk->private_data;
1002         struct nvme_command c;
1003         u8 data[16] = { 0, };
1004
1005         put_unaligned_le64(key, &data[0]);
1006         put_unaligned_le64(sa_key, &data[8]);
1007
1008         memset(&c, 0, sizeof(c));
1009         c.common.opcode = op;
1010         c.common.nsid = cpu_to_le32(ns->ns_id);
1011         c.common.cdw10[0] = cpu_to_le32(cdw10);
1012
1013         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1014 }
1015
1016 static int nvme_pr_register(struct block_device *bdev, u64 old,
1017                 u64 new, unsigned flags)
1018 {
1019         u32 cdw10;
1020
1021         if (flags & ~PR_FL_IGNORE_KEY)
1022                 return -EOPNOTSUPP;
1023
1024         cdw10 = old ? 2 : 0;
1025         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1026         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1027         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1028 }
1029
1030 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1031                 enum pr_type type, unsigned flags)
1032 {
1033         u32 cdw10;
1034
1035         if (flags & ~PR_FL_IGNORE_KEY)
1036                 return -EOPNOTSUPP;
1037
1038         cdw10 = nvme_pr_type(type) << 8;
1039         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1040         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1041 }
1042
1043 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1044                 enum pr_type type, bool abort)
1045 {
1046         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1047         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1048 }
1049
1050 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1051 {
1052         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1053         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1054 }
1055
1056 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1057 {
1058         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1059         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1060 }
1061
1062 static const struct pr_ops nvme_pr_ops = {
1063         .pr_register    = nvme_pr_register,
1064         .pr_reserve     = nvme_pr_reserve,
1065         .pr_release     = nvme_pr_release,
1066         .pr_preempt     = nvme_pr_preempt,
1067         .pr_clear       = nvme_pr_clear,
1068 };
1069
1070 static const struct block_device_operations nvme_fops = {
1071         .owner          = THIS_MODULE,
1072         .ioctl          = nvme_ioctl,
1073         .compat_ioctl   = nvme_compat_ioctl,
1074         .open           = nvme_open,
1075         .release        = nvme_release,
1076         .getgeo         = nvme_getgeo,
1077         .revalidate_disk= nvme_revalidate_disk,
1078         .pr_ops         = &nvme_pr_ops,
1079 };
1080
1081 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1082 {
1083         unsigned long timeout =
1084                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1085         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1086         int ret;
1087
1088         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1089                 if ((csts & NVME_CSTS_RDY) == bit)
1090                         break;
1091
1092                 msleep(100);
1093                 if (fatal_signal_pending(current))
1094                         return -EINTR;
1095                 if (time_after(jiffies, timeout)) {
1096                         dev_err(ctrl->device,
1097                                 "Device not ready; aborting %s\n", enabled ?
1098                                                 "initialisation" : "reset");
1099                         return -ENODEV;
1100                 }
1101         }
1102
1103         return ret;
1104 }
1105
1106 /*
1107  * If the device has been passed off to us in an enabled state, just clear
1108  * the enabled bit.  The spec says we should set the 'shutdown notification
1109  * bits', but doing so may cause the device to complete commands to the
1110  * admin queue ... and we don't know what memory that might be pointing at!
1111  */
1112 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1113 {
1114         int ret;
1115
1116         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1117         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1118
1119         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1120         if (ret)
1121                 return ret;
1122
1123         /* Checking for ctrl->tagset is a trick to avoid sleeping on module
1124          * load, since we only need the quirk on reset_controller. Notice
1125          * that the HGST device needs this delay only in firmware activation
1126          * procedure; unfortunately we have no (easy) way to verify this.
1127          */
1128         if ((ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) && ctrl->tagset)
1129                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1130
1131         return nvme_wait_ready(ctrl, cap, false);
1132 }
1133 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1134
1135 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1136 {
1137         /*
1138          * Default to a 4K page size, with the intention to update this
1139          * path in the future to accomodate architectures with differing
1140          * kernel and IO page sizes.
1141          */
1142         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1143         int ret;
1144
1145         if (page_shift < dev_page_min) {
1146                 dev_err(ctrl->device,
1147                         "Minimum device page size %u too large for host (%u)\n",
1148                         1 << dev_page_min, 1 << page_shift);
1149                 return -ENODEV;
1150         }
1151
1152         ctrl->page_size = 1 << page_shift;
1153
1154         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1155         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1156         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1157         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1158         ctrl->ctrl_config |= NVME_CC_ENABLE;
1159
1160         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1161         if (ret)
1162                 return ret;
1163         return nvme_wait_ready(ctrl, cap, true);
1164 }
1165 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1166
1167 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1168 {
1169         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1170         u32 csts;
1171         int ret;
1172
1173         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1174         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1175
1176         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1177         if (ret)
1178                 return ret;
1179
1180         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1181                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1182                         break;
1183
1184                 msleep(100);
1185                 if (fatal_signal_pending(current))
1186                         return -EINTR;
1187                 if (time_after(jiffies, timeout)) {
1188                         dev_err(ctrl->device,
1189                                 "Device shutdown incomplete; abort shutdown\n");
1190                         return -ENODEV;
1191                 }
1192         }
1193
1194         return ret;
1195 }
1196 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1197
1198 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1199                 struct request_queue *q)
1200 {
1201         bool vwc = false;
1202
1203         if (ctrl->max_hw_sectors) {
1204                 u32 max_segments =
1205                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1206
1207                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1208                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1209         }
1210         if (ctrl->stripe_size)
1211                 blk_queue_chunk_sectors(q, ctrl->stripe_size >> 9);
1212         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1213         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1214                 vwc = true;
1215         blk_queue_write_cache(q, vwc, vwc);
1216 }
1217
1218 /*
1219  * Initialize the cached copies of the Identify data and various controller
1220  * register in our nvme_ctrl structure.  This should be called as soon as
1221  * the admin queue is fully up and running.
1222  */
1223 int nvme_init_identify(struct nvme_ctrl *ctrl)
1224 {
1225         struct nvme_id_ctrl *id;
1226         u64 cap;
1227         int ret, page_shift;
1228         u32 max_hw_sectors;
1229
1230         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1231         if (ret) {
1232                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1233                 return ret;
1234         }
1235
1236         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1237         if (ret) {
1238                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1239                 return ret;
1240         }
1241         page_shift = NVME_CAP_MPSMIN(cap) + 12;
1242
1243         if (ctrl->vs >= NVME_VS(1, 1))
1244                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1245
1246         ret = nvme_identify_ctrl(ctrl, &id);
1247         if (ret) {
1248                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1249                 return -EIO;
1250         }
1251
1252         ctrl->vid = le16_to_cpu(id->vid);
1253         ctrl->oncs = le16_to_cpup(&id->oncs);
1254         atomic_set(&ctrl->abort_limit, id->acl + 1);
1255         ctrl->vwc = id->vwc;
1256         ctrl->cntlid = le16_to_cpup(&id->cntlid);
1257         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1258         memcpy(ctrl->model, id->mn, sizeof(id->mn));
1259         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1260         if (id->mdts)
1261                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1262         else
1263                 max_hw_sectors = UINT_MAX;
1264         ctrl->max_hw_sectors =
1265                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1266
1267         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
1268                 unsigned int max_hw_sectors;
1269
1270                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
1271                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
1272                 if (ctrl->max_hw_sectors) {
1273                         ctrl->max_hw_sectors = min(max_hw_sectors,
1274                                                         ctrl->max_hw_sectors);
1275                 } else {
1276                         ctrl->max_hw_sectors = max_hw_sectors;
1277                 }
1278         }
1279
1280         nvme_set_queue_limits(ctrl, ctrl->admin_q);
1281         ctrl->sgls = le32_to_cpu(id->sgls);
1282         ctrl->kas = le16_to_cpu(id->kas);
1283
1284         if (ctrl->ops->is_fabrics) {
1285                 ctrl->icdoff = le16_to_cpu(id->icdoff);
1286                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1287                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1288                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1289
1290                 /*
1291                  * In fabrics we need to verify the cntlid matches the
1292                  * admin connect
1293                  */
1294                 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1295                         ret = -EINVAL;
1296
1297                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1298                         dev_err(ctrl->dev,
1299                                 "keep-alive support is mandatory for fabrics\n");
1300                         ret = -EINVAL;
1301                 }
1302         } else {
1303                 ctrl->cntlid = le16_to_cpu(id->cntlid);
1304         }
1305
1306         kfree(id);
1307         return ret;
1308 }
1309 EXPORT_SYMBOL_GPL(nvme_init_identify);
1310
1311 static int nvme_dev_open(struct inode *inode, struct file *file)
1312 {
1313         struct nvme_ctrl *ctrl;
1314         int instance = iminor(inode);
1315         int ret = -ENODEV;
1316
1317         spin_lock(&dev_list_lock);
1318         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1319                 if (ctrl->instance != instance)
1320                         continue;
1321
1322                 if (!ctrl->admin_q) {
1323                         ret = -EWOULDBLOCK;
1324                         break;
1325                 }
1326                 if (!kref_get_unless_zero(&ctrl->kref))
1327                         break;
1328                 file->private_data = ctrl;
1329                 ret = 0;
1330                 break;
1331         }
1332         spin_unlock(&dev_list_lock);
1333
1334         return ret;
1335 }
1336
1337 static int nvme_dev_release(struct inode *inode, struct file *file)
1338 {
1339         nvme_put_ctrl(file->private_data);
1340         return 0;
1341 }
1342
1343 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1344 {
1345         struct nvme_ns *ns;
1346         int ret;
1347
1348         mutex_lock(&ctrl->namespaces_mutex);
1349         if (list_empty(&ctrl->namespaces)) {
1350                 ret = -ENOTTY;
1351                 goto out_unlock;
1352         }
1353
1354         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1355         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1356                 dev_warn(ctrl->device,
1357                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1358                 ret = -EINVAL;
1359                 goto out_unlock;
1360         }
1361
1362         dev_warn(ctrl->device,
1363                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1364         kref_get(&ns->kref);
1365         mutex_unlock(&ctrl->namespaces_mutex);
1366
1367         ret = nvme_user_cmd(ctrl, ns, argp);
1368         nvme_put_ns(ns);
1369         return ret;
1370
1371 out_unlock:
1372         mutex_unlock(&ctrl->namespaces_mutex);
1373         return ret;
1374 }
1375
1376 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1377                 unsigned long arg)
1378 {
1379         struct nvme_ctrl *ctrl = file->private_data;
1380         void __user *argp = (void __user *)arg;
1381
1382         switch (cmd) {
1383         case NVME_IOCTL_ADMIN_CMD:
1384                 return nvme_user_cmd(ctrl, NULL, argp);
1385         case NVME_IOCTL_IO_CMD:
1386                 return nvme_dev_user_cmd(ctrl, argp);
1387         case NVME_IOCTL_RESET:
1388                 dev_warn(ctrl->device, "resetting controller\n");
1389                 return ctrl->ops->reset_ctrl(ctrl);
1390         case NVME_IOCTL_SUBSYS_RESET:
1391                 return nvme_reset_subsystem(ctrl);
1392         case NVME_IOCTL_RESCAN:
1393                 nvme_queue_scan(ctrl);
1394                 return 0;
1395         default:
1396                 return -ENOTTY;
1397         }
1398 }
1399
1400 static const struct file_operations nvme_dev_fops = {
1401         .owner          = THIS_MODULE,
1402         .open           = nvme_dev_open,
1403         .release        = nvme_dev_release,
1404         .unlocked_ioctl = nvme_dev_ioctl,
1405         .compat_ioctl   = nvme_dev_ioctl,
1406 };
1407
1408 static ssize_t nvme_sysfs_reset(struct device *dev,
1409                                 struct device_attribute *attr, const char *buf,
1410                                 size_t count)
1411 {
1412         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1413         int ret;
1414
1415         ret = ctrl->ops->reset_ctrl(ctrl);
1416         if (ret < 0)
1417                 return ret;
1418         return count;
1419 }
1420 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1421
1422 static ssize_t nvme_sysfs_rescan(struct device *dev,
1423                                 struct device_attribute *attr, const char *buf,
1424                                 size_t count)
1425 {
1426         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1427
1428         nvme_queue_scan(ctrl);
1429         return count;
1430 }
1431 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1432
1433 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1434                                                                 char *buf)
1435 {
1436         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1437         struct nvme_ctrl *ctrl = ns->ctrl;
1438         int serial_len = sizeof(ctrl->serial);
1439         int model_len = sizeof(ctrl->model);
1440
1441         if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1442                 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1443
1444         if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1445                 return sprintf(buf, "eui.%8phN\n", ns->eui);
1446
1447         while (ctrl->serial[serial_len - 1] == ' ')
1448                 serial_len--;
1449         while (ctrl->model[model_len - 1] == ' ')
1450                 model_len--;
1451
1452         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1453                 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1454 }
1455 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1456
1457 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1458                                                                 char *buf)
1459 {
1460         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1461         return sprintf(buf, "%pU\n", ns->uuid);
1462 }
1463 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1464
1465 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1466                                                                 char *buf)
1467 {
1468         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1469         return sprintf(buf, "%8phd\n", ns->eui);
1470 }
1471 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1472
1473 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1474                                                                 char *buf)
1475 {
1476         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1477         return sprintf(buf, "%d\n", ns->ns_id);
1478 }
1479 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1480
1481 static struct attribute *nvme_ns_attrs[] = {
1482         &dev_attr_wwid.attr,
1483         &dev_attr_uuid.attr,
1484         &dev_attr_eui.attr,
1485         &dev_attr_nsid.attr,
1486         NULL,
1487 };
1488
1489 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1490                 struct attribute *a, int n)
1491 {
1492         struct device *dev = container_of(kobj, struct device, kobj);
1493         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1494
1495         if (a == &dev_attr_uuid.attr) {
1496                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1497                         return 0;
1498         }
1499         if (a == &dev_attr_eui.attr) {
1500                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1501                         return 0;
1502         }
1503         return a->mode;
1504 }
1505
1506 static const struct attribute_group nvme_ns_attr_group = {
1507         .attrs          = nvme_ns_attrs,
1508         .is_visible     = nvme_ns_attrs_are_visible,
1509 };
1510
1511 #define nvme_show_str_function(field)                                           \
1512 static ssize_t  field##_show(struct device *dev,                                \
1513                             struct device_attribute *attr, char *buf)           \
1514 {                                                                               \
1515         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1516         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1517 }                                                                               \
1518 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1519
1520 #define nvme_show_int_function(field)                                           \
1521 static ssize_t  field##_show(struct device *dev,                                \
1522                             struct device_attribute *attr, char *buf)           \
1523 {                                                                               \
1524         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1525         return sprintf(buf, "%d\n", ctrl->field);       \
1526 }                                                                               \
1527 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1528
1529 nvme_show_str_function(model);
1530 nvme_show_str_function(serial);
1531 nvme_show_str_function(firmware_rev);
1532 nvme_show_int_function(cntlid);
1533
1534 static ssize_t nvme_sysfs_delete(struct device *dev,
1535                                 struct device_attribute *attr, const char *buf,
1536                                 size_t count)
1537 {
1538         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1539
1540         if (device_remove_file_self(dev, attr))
1541                 ctrl->ops->delete_ctrl(ctrl);
1542         return count;
1543 }
1544 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1545
1546 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1547                                          struct device_attribute *attr,
1548                                          char *buf)
1549 {
1550         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1551
1552         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1553 }
1554 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1555
1556 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1557                                          struct device_attribute *attr,
1558                                          char *buf)
1559 {
1560         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1561
1562         return snprintf(buf, PAGE_SIZE, "%s\n",
1563                         ctrl->ops->get_subsysnqn(ctrl));
1564 }
1565 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1566
1567 static ssize_t nvme_sysfs_show_address(struct device *dev,
1568                                          struct device_attribute *attr,
1569                                          char *buf)
1570 {
1571         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1572
1573         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1574 }
1575 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1576
1577 static struct attribute *nvme_dev_attrs[] = {
1578         &dev_attr_reset_controller.attr,
1579         &dev_attr_rescan_controller.attr,
1580         &dev_attr_model.attr,
1581         &dev_attr_serial.attr,
1582         &dev_attr_firmware_rev.attr,
1583         &dev_attr_cntlid.attr,
1584         &dev_attr_delete_controller.attr,
1585         &dev_attr_transport.attr,
1586         &dev_attr_subsysnqn.attr,
1587         &dev_attr_address.attr,
1588         NULL
1589 };
1590
1591 #define CHECK_ATTR(ctrl, a, name)               \
1592         if ((a) == &dev_attr_##name.attr &&     \
1593             !(ctrl)->ops->get_##name)           \
1594                 return 0
1595
1596 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1597                 struct attribute *a, int n)
1598 {
1599         struct device *dev = container_of(kobj, struct device, kobj);
1600         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1601
1602         if (a == &dev_attr_delete_controller.attr) {
1603                 if (!ctrl->ops->delete_ctrl)
1604                         return 0;
1605         }
1606
1607         CHECK_ATTR(ctrl, a, subsysnqn);
1608         CHECK_ATTR(ctrl, a, address);
1609
1610         return a->mode;
1611 }
1612
1613 static struct attribute_group nvme_dev_attrs_group = {
1614         .attrs          = nvme_dev_attrs,
1615         .is_visible     = nvme_dev_attrs_are_visible,
1616 };
1617
1618 static const struct attribute_group *nvme_dev_attr_groups[] = {
1619         &nvme_dev_attrs_group,
1620         NULL,
1621 };
1622
1623 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1624 {
1625         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1626         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1627
1628         return nsa->ns_id - nsb->ns_id;
1629 }
1630
1631 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1632 {
1633         struct nvme_ns *ns, *ret = NULL;
1634
1635         mutex_lock(&ctrl->namespaces_mutex);
1636         list_for_each_entry(ns, &ctrl->namespaces, list) {
1637                 if (ns->ns_id == nsid) {
1638                         kref_get(&ns->kref);
1639                         ret = ns;
1640                         break;
1641                 }
1642                 if (ns->ns_id > nsid)
1643                         break;
1644         }
1645         mutex_unlock(&ctrl->namespaces_mutex);
1646         return ret;
1647 }
1648
1649 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1650 {
1651         struct nvme_ns *ns;
1652         struct gendisk *disk;
1653         struct nvme_id_ns *id;
1654         char disk_name[DISK_NAME_LEN];
1655         int node = dev_to_node(ctrl->dev);
1656
1657         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1658         if (!ns)
1659                 return;
1660
1661         ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1662         if (ns->instance < 0)
1663                 goto out_free_ns;
1664
1665         ns->queue = blk_mq_init_queue(ctrl->tagset);
1666         if (IS_ERR(ns->queue))
1667                 goto out_release_instance;
1668         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1669         ns->queue->queuedata = ns;
1670         ns->ctrl = ctrl;
1671
1672         kref_init(&ns->kref);
1673         ns->ns_id = nsid;
1674         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1675
1676         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1677         nvme_set_queue_limits(ctrl, ns->queue);
1678
1679         sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1680
1681         if (nvme_revalidate_ns(ns, &id))
1682                 goto out_free_queue;
1683
1684         if (nvme_nvm_ns_supported(ns, id)) {
1685                 if (nvme_nvm_register(ns, disk_name, node,
1686                                                         &nvme_ns_attr_group)) {
1687                         dev_warn(ctrl->dev, "%s: LightNVM init failure\n",
1688                                                                 __func__);
1689                         goto out_free_id;
1690                 }
1691         } else {
1692                 disk = alloc_disk_node(0, node);
1693                 if (!disk)
1694                         goto out_free_id;
1695
1696                 disk->fops = &nvme_fops;
1697                 disk->private_data = ns;
1698                 disk->queue = ns->queue;
1699                 disk->flags = GENHD_FL_EXT_DEVT;
1700                 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
1701                 ns->disk = disk;
1702
1703                 __nvme_revalidate_disk(disk, id);
1704         }
1705
1706         mutex_lock(&ctrl->namespaces_mutex);
1707         list_add_tail(&ns->list, &ctrl->namespaces);
1708         mutex_unlock(&ctrl->namespaces_mutex);
1709
1710         kref_get(&ctrl->kref);
1711
1712         kfree(id);
1713
1714         if (ns->ndev)
1715                 return;
1716
1717         device_add_disk(ctrl->device, ns->disk);
1718         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1719                                         &nvme_ns_attr_group))
1720                 pr_warn("%s: failed to create sysfs group for identification\n",
1721                         ns->disk->disk_name);
1722         return;
1723  out_free_id:
1724         kfree(id);
1725  out_free_queue:
1726         blk_cleanup_queue(ns->queue);
1727  out_release_instance:
1728         ida_simple_remove(&ctrl->ns_ida, ns->instance);
1729  out_free_ns:
1730         kfree(ns);
1731 }
1732
1733 static void nvme_ns_remove(struct nvme_ns *ns)
1734 {
1735         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1736                 return;
1737
1738         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
1739                 if (blk_get_integrity(ns->disk))
1740                         blk_integrity_unregister(ns->disk);
1741                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1742                                         &nvme_ns_attr_group);
1743                 del_gendisk(ns->disk);
1744                 blk_mq_abort_requeue_list(ns->queue);
1745                 blk_cleanup_queue(ns->queue);
1746         }
1747
1748         mutex_lock(&ns->ctrl->namespaces_mutex);
1749         list_del_init(&ns->list);
1750         mutex_unlock(&ns->ctrl->namespaces_mutex);
1751
1752         nvme_put_ns(ns);
1753 }
1754
1755 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1756 {
1757         struct nvme_ns *ns;
1758
1759         ns = nvme_find_get_ns(ctrl, nsid);
1760         if (ns) {
1761                 if (ns->disk && revalidate_disk(ns->disk))
1762                         nvme_ns_remove(ns);
1763                 nvme_put_ns(ns);
1764         } else
1765                 nvme_alloc_ns(ctrl, nsid);
1766 }
1767
1768 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1769                                         unsigned nsid)
1770 {
1771         struct nvme_ns *ns, *next;
1772
1773         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1774                 if (ns->ns_id > nsid)
1775                         nvme_ns_remove(ns);
1776         }
1777 }
1778
1779 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1780 {
1781         struct nvme_ns *ns;
1782         __le32 *ns_list;
1783         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1784         int ret = 0;
1785
1786         ns_list = kzalloc(0x1000, GFP_KERNEL);
1787         if (!ns_list)
1788                 return -ENOMEM;
1789
1790         for (i = 0; i < num_lists; i++) {
1791                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1792                 if (ret)
1793                         goto free;
1794
1795                 for (j = 0; j < min(nn, 1024U); j++) {
1796                         nsid = le32_to_cpu(ns_list[j]);
1797                         if (!nsid)
1798                                 goto out;
1799
1800                         nvme_validate_ns(ctrl, nsid);
1801
1802                         while (++prev < nsid) {
1803                                 ns = nvme_find_get_ns(ctrl, prev);
1804                                 if (ns) {
1805                                         nvme_ns_remove(ns);
1806                                         nvme_put_ns(ns);
1807                                 }
1808                         }
1809                 }
1810                 nn -= j;
1811         }
1812  out:
1813         nvme_remove_invalid_namespaces(ctrl, prev);
1814  free:
1815         kfree(ns_list);
1816         return ret;
1817 }
1818
1819 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1820 {
1821         unsigned i;
1822
1823         for (i = 1; i <= nn; i++)
1824                 nvme_validate_ns(ctrl, i);
1825
1826         nvme_remove_invalid_namespaces(ctrl, nn);
1827 }
1828
1829 static void nvme_scan_work(struct work_struct *work)
1830 {
1831         struct nvme_ctrl *ctrl =
1832                 container_of(work, struct nvme_ctrl, scan_work);
1833         struct nvme_id_ctrl *id;
1834         unsigned nn;
1835
1836         if (ctrl->state != NVME_CTRL_LIVE)
1837                 return;
1838
1839         if (nvme_identify_ctrl(ctrl, &id))
1840                 return;
1841
1842         nn = le32_to_cpu(id->nn);
1843         if (ctrl->vs >= NVME_VS(1, 1) &&
1844             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1845                 if (!nvme_scan_ns_list(ctrl, nn))
1846                         goto done;
1847         }
1848         nvme_scan_ns_sequential(ctrl, nn);
1849  done:
1850         mutex_lock(&ctrl->namespaces_mutex);
1851         list_sort(NULL, &ctrl->namespaces, ns_cmp);
1852         mutex_unlock(&ctrl->namespaces_mutex);
1853         kfree(id);
1854
1855         if (ctrl->ops->post_scan)
1856                 ctrl->ops->post_scan(ctrl);
1857 }
1858
1859 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1860 {
1861         /*
1862          * Do not queue new scan work when a controller is reset during
1863          * removal.
1864          */
1865         if (ctrl->state == NVME_CTRL_LIVE)
1866                 schedule_work(&ctrl->scan_work);
1867 }
1868 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1869
1870 /*
1871  * This function iterates the namespace list unlocked to allow recovery from
1872  * controller failure. It is up to the caller to ensure the namespace list is
1873  * not modified by scan work while this function is executing.
1874  */
1875 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1876 {
1877         struct nvme_ns *ns, *next;
1878
1879         /*
1880          * The dead states indicates the controller was not gracefully
1881          * disconnected. In that case, we won't be able to flush any data while
1882          * removing the namespaces' disks; fail all the queues now to avoid
1883          * potentially having to clean up the failed sync later.
1884          */
1885         if (ctrl->state == NVME_CTRL_DEAD)
1886                 nvme_kill_queues(ctrl);
1887
1888         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1889                 nvme_ns_remove(ns);
1890 }
1891 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1892
1893 static void nvme_async_event_work(struct work_struct *work)
1894 {
1895         struct nvme_ctrl *ctrl =
1896                 container_of(work, struct nvme_ctrl, async_event_work);
1897
1898         spin_lock_irq(&ctrl->lock);
1899         while (ctrl->event_limit > 0) {
1900                 int aer_idx = --ctrl->event_limit;
1901
1902                 spin_unlock_irq(&ctrl->lock);
1903                 ctrl->ops->submit_async_event(ctrl, aer_idx);
1904                 spin_lock_irq(&ctrl->lock);
1905         }
1906         spin_unlock_irq(&ctrl->lock);
1907 }
1908
1909 void nvme_complete_async_event(struct nvme_ctrl *ctrl,
1910                 struct nvme_completion *cqe)
1911 {
1912         u16 status = le16_to_cpu(cqe->status) >> 1;
1913         u32 result = le32_to_cpu(cqe->result);
1914
1915         if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) {
1916                 ++ctrl->event_limit;
1917                 schedule_work(&ctrl->async_event_work);
1918         }
1919
1920         if (status != NVME_SC_SUCCESS)
1921                 return;
1922
1923         switch (result & 0xff07) {
1924         case NVME_AER_NOTICE_NS_CHANGED:
1925                 dev_info(ctrl->device, "rescanning\n");
1926                 nvme_queue_scan(ctrl);
1927                 break;
1928         default:
1929                 dev_warn(ctrl->device, "async event result %08x\n", result);
1930         }
1931 }
1932 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1933
1934 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1935 {
1936         ctrl->event_limit = NVME_NR_AERS;
1937         schedule_work(&ctrl->async_event_work);
1938 }
1939 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1940
1941 static DEFINE_IDA(nvme_instance_ida);
1942
1943 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1944 {
1945         int instance, error;
1946
1947         do {
1948                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1949                         return -ENODEV;
1950
1951                 spin_lock(&dev_list_lock);
1952                 error = ida_get_new(&nvme_instance_ida, &instance);
1953                 spin_unlock(&dev_list_lock);
1954         } while (error == -EAGAIN);
1955
1956         if (error)
1957                 return -ENODEV;
1958
1959         ctrl->instance = instance;
1960         return 0;
1961 }
1962
1963 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1964 {
1965         spin_lock(&dev_list_lock);
1966         ida_remove(&nvme_instance_ida, ctrl->instance);
1967         spin_unlock(&dev_list_lock);
1968 }
1969
1970 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1971 {
1972         flush_work(&ctrl->async_event_work);
1973         flush_work(&ctrl->scan_work);
1974         nvme_remove_namespaces(ctrl);
1975
1976         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1977
1978         spin_lock(&dev_list_lock);
1979         list_del(&ctrl->node);
1980         spin_unlock(&dev_list_lock);
1981 }
1982 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1983
1984 static void nvme_free_ctrl(struct kref *kref)
1985 {
1986         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1987
1988         put_device(ctrl->device);
1989         nvme_release_instance(ctrl);
1990         ida_destroy(&ctrl->ns_ida);
1991
1992         ctrl->ops->free_ctrl(ctrl);
1993 }
1994
1995 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1996 {
1997         kref_put(&ctrl->kref, nvme_free_ctrl);
1998 }
1999 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2000
2001 /*
2002  * Initialize a NVMe controller structures.  This needs to be called during
2003  * earliest initialization so that we have the initialized structured around
2004  * during probing.
2005  */
2006 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2007                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2008 {
2009         int ret;
2010
2011         ctrl->state = NVME_CTRL_NEW;
2012         spin_lock_init(&ctrl->lock);
2013         INIT_LIST_HEAD(&ctrl->namespaces);
2014         mutex_init(&ctrl->namespaces_mutex);
2015         kref_init(&ctrl->kref);
2016         ctrl->dev = dev;
2017         ctrl->ops = ops;
2018         ctrl->quirks = quirks;
2019         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2020         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2021
2022         ret = nvme_set_instance(ctrl);
2023         if (ret)
2024                 goto out;
2025
2026         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2027                                 MKDEV(nvme_char_major, ctrl->instance),
2028                                 ctrl, nvme_dev_attr_groups,
2029                                 "nvme%d", ctrl->instance);
2030         if (IS_ERR(ctrl->device)) {
2031                 ret = PTR_ERR(ctrl->device);
2032                 goto out_release_instance;
2033         }
2034         get_device(ctrl->device);
2035         ida_init(&ctrl->ns_ida);
2036
2037         spin_lock(&dev_list_lock);
2038         list_add_tail(&ctrl->node, &nvme_ctrl_list);
2039         spin_unlock(&dev_list_lock);
2040
2041         return 0;
2042 out_release_instance:
2043         nvme_release_instance(ctrl);
2044 out:
2045         return ret;
2046 }
2047 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2048
2049 /**
2050  * nvme_kill_queues(): Ends all namespace queues
2051  * @ctrl: the dead controller that needs to end
2052  *
2053  * Call this function when the driver determines it is unable to get the
2054  * controller in a state capable of servicing IO.
2055  */
2056 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2057 {
2058         struct nvme_ns *ns;
2059
2060         mutex_lock(&ctrl->namespaces_mutex);
2061         list_for_each_entry(ns, &ctrl->namespaces, list) {
2062                 /*
2063                  * Revalidating a dead namespace sets capacity to 0. This will
2064                  * end buffered writers dirtying pages that can't be synced.
2065                  */
2066                 if (ns->disk && !test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2067                         revalidate_disk(ns->disk);
2068
2069                 blk_set_queue_dying(ns->queue);
2070                 blk_mq_abort_requeue_list(ns->queue);
2071                 blk_mq_start_stopped_hw_queues(ns->queue, true);
2072         }
2073         mutex_unlock(&ctrl->namespaces_mutex);
2074 }
2075 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2076
2077 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2078 {
2079         struct nvme_ns *ns;
2080
2081         mutex_lock(&ctrl->namespaces_mutex);
2082         list_for_each_entry(ns, &ctrl->namespaces, list) {
2083                 spin_lock_irq(ns->queue->queue_lock);
2084                 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
2085                 spin_unlock_irq(ns->queue->queue_lock);
2086
2087                 blk_mq_cancel_requeue_work(ns->queue);
2088                 blk_mq_stop_hw_queues(ns->queue);
2089         }
2090         mutex_unlock(&ctrl->namespaces_mutex);
2091 }
2092 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2093
2094 void nvme_start_queues(struct nvme_ctrl *ctrl)
2095 {
2096         struct nvme_ns *ns;
2097
2098         mutex_lock(&ctrl->namespaces_mutex);
2099         list_for_each_entry(ns, &ctrl->namespaces, list) {
2100                 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
2101                 blk_mq_start_stopped_hw_queues(ns->queue, true);
2102                 blk_mq_kick_requeue_list(ns->queue);
2103         }
2104         mutex_unlock(&ctrl->namespaces_mutex);
2105 }
2106 EXPORT_SYMBOL_GPL(nvme_start_queues);
2107
2108 int __init nvme_core_init(void)
2109 {
2110         int result;
2111
2112         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2113                                                         &nvme_dev_fops);
2114         if (result < 0)
2115                 return result;
2116         else if (result > 0)
2117                 nvme_char_major = result;
2118
2119         nvme_class = class_create(THIS_MODULE, "nvme");
2120         if (IS_ERR(nvme_class)) {
2121                 result = PTR_ERR(nvme_class);
2122                 goto unregister_chrdev;
2123         }
2124
2125         return 0;
2126
2127  unregister_chrdev:
2128         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2129         return result;
2130 }
2131
2132 void nvme_core_exit(void)
2133 {
2134         class_destroy(nvme_class);
2135         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2136 }
2137
2138 MODULE_LICENSE("GPL");
2139 MODULE_VERSION("1.0");
2140 module_init(nvme_core_init);
2141 module_exit(nvme_core_exit);