1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * NVM Express device driver 4 * Copyright (c) 2011-2014, Intel Corporation. 5 */ 6 7 #include <linux/blkdev.h> 8 #include <linux/blk-mq.h> 9 #include <linux/delay.h> 10 #include <linux/errno.h> 11 #include <linux/hdreg.h> 12 #include <linux/kernel.h> 13 #include <linux/module.h> 14 #include <linux/list_sort.h> 15 #include <linux/slab.h> 16 #include <linux/types.h> 17 #include <linux/pr.h> 18 #include <linux/ptrace.h> 19 #include <linux/nvme_ioctl.h> 20 #include <linux/t10-pi.h> 21 #include <linux/pm_qos.h> 22 #include <asm/unaligned.h> 23 24 #define CREATE_TRACE_POINTS 25 #include "trace.h" 26 27 #include "nvme.h" 28 #include "fabrics.h" 29 30 #define NVME_MINORS (1U << MINORBITS) 31 32 unsigned int admin_timeout = 60; 33 module_param(admin_timeout, uint, 0644); 34 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands"); 35 EXPORT_SYMBOL_GPL(admin_timeout); 36 37 unsigned int nvme_io_timeout = 30; 38 module_param_named(io_timeout, nvme_io_timeout, uint, 0644); 39 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O"); 40 EXPORT_SYMBOL_GPL(nvme_io_timeout); 41 42 static unsigned char shutdown_timeout = 5; 43 module_param(shutdown_timeout, byte, 0644); 44 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown"); 45 46 static u8 nvme_max_retries = 5; 47 module_param_named(max_retries, nvme_max_retries, byte, 0644); 48 MODULE_PARM_DESC(max_retries, "max number of retries a command may have"); 49 50 static unsigned long default_ps_max_latency_us = 100000; 51 module_param(default_ps_max_latency_us, ulong, 0644); 52 MODULE_PARM_DESC(default_ps_max_latency_us, 53 "max power saving latency for new devices; use PM QOS to change per device"); 54 55 static bool force_apst; 56 module_param(force_apst, bool, 0644); 57 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off"); 58 59 static bool streams; 60 module_param(streams, bool, 0644); 61 MODULE_PARM_DESC(streams, "turn on support for Streams write directives"); 62 63 /* 64 * nvme_wq - hosts nvme related works that are not reset or delete 65 * nvme_reset_wq - hosts nvme reset works 66 * nvme_delete_wq - hosts nvme delete works 67 * 68 * nvme_wq will host works such are scan, aen handling, fw activation, 69 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq 70 * runs reset works which also flush works hosted on nvme_wq for 71 * serialization purposes. nvme_delete_wq host controller deletion 72 * works which flush reset works for serialization. 73 */ 74 struct workqueue_struct *nvme_wq; 75 EXPORT_SYMBOL_GPL(nvme_wq); 76 77 struct workqueue_struct *nvme_reset_wq; 78 EXPORT_SYMBOL_GPL(nvme_reset_wq); 79 80 struct workqueue_struct *nvme_delete_wq; 81 EXPORT_SYMBOL_GPL(nvme_delete_wq); 82 83 static DEFINE_IDA(nvme_subsystems_ida); 84 static LIST_HEAD(nvme_subsystems); 85 static DEFINE_MUTEX(nvme_subsystems_lock); 86 87 static DEFINE_IDA(nvme_instance_ida); 88 static dev_t nvme_chr_devt; 89 static struct class *nvme_class; 90 static struct class *nvme_subsys_class; 91 92 static int nvme_revalidate_disk(struct gendisk *disk); 93 static void nvme_put_subsystem(struct nvme_subsystem *subsys); 94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl, 95 unsigned nsid); 96 97 static void nvme_set_queue_dying(struct nvme_ns *ns) 98 { 99 /* 100 * Revalidating a dead namespace sets capacity to 0. This will end 101 * buffered writers dirtying pages that can't be synced. 102 */ 103 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags)) 104 return; 105 revalidate_disk(ns->disk); 106 blk_set_queue_dying(ns->queue); 107 /* Forcibly unquiesce queues to avoid blocking dispatch */ 108 blk_mq_unquiesce_queue(ns->queue); 109 } 110 111 static void nvme_queue_scan(struct nvme_ctrl *ctrl) 112 { 113 /* 114 * Only new queue scan work when admin and IO queues are both alive 115 */ 116 if (ctrl->state == NVME_CTRL_LIVE) 117 queue_work(nvme_wq, &ctrl->scan_work); 118 } 119 120 int nvme_reset_ctrl(struct nvme_ctrl *ctrl) 121 { 122 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) 123 return -EBUSY; 124 if (!queue_work(nvme_reset_wq, &ctrl->reset_work)) 125 return -EBUSY; 126 return 0; 127 } 128 EXPORT_SYMBOL_GPL(nvme_reset_ctrl); 129 130 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl) 131 { 132 int ret; 133 134 ret = nvme_reset_ctrl(ctrl); 135 if (!ret) { 136 flush_work(&ctrl->reset_work); 137 if (ctrl->state != NVME_CTRL_LIVE && 138 ctrl->state != NVME_CTRL_ADMIN_ONLY) 139 ret = -ENETRESET; 140 } 141 142 return ret; 143 } 144 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync); 145 146 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl) 147 { 148 dev_info(ctrl->device, 149 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn); 150 151 flush_work(&ctrl->reset_work); 152 nvme_stop_ctrl(ctrl); 153 nvme_remove_namespaces(ctrl); 154 ctrl->ops->delete_ctrl(ctrl); 155 nvme_uninit_ctrl(ctrl); 156 nvme_put_ctrl(ctrl); 157 } 158 159 static void nvme_delete_ctrl_work(struct work_struct *work) 160 { 161 struct nvme_ctrl *ctrl = 162 container_of(work, struct nvme_ctrl, delete_work); 163 164 nvme_do_delete_ctrl(ctrl); 165 } 166 167 int nvme_delete_ctrl(struct nvme_ctrl *ctrl) 168 { 169 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING)) 170 return -EBUSY; 171 if (!queue_work(nvme_delete_wq, &ctrl->delete_work)) 172 return -EBUSY; 173 return 0; 174 } 175 EXPORT_SYMBOL_GPL(nvme_delete_ctrl); 176 177 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl) 178 { 179 int ret = 0; 180 181 /* 182 * Keep a reference until nvme_do_delete_ctrl() complete, 183 * since ->delete_ctrl can free the controller. 184 */ 185 nvme_get_ctrl(ctrl); 186 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING)) 187 ret = -EBUSY; 188 if (!ret) 189 nvme_do_delete_ctrl(ctrl); 190 nvme_put_ctrl(ctrl); 191 return ret; 192 } 193 194 static inline bool nvme_ns_has_pi(struct nvme_ns *ns) 195 { 196 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple); 197 } 198 199 static blk_status_t nvme_error_status(struct request *req) 200 { 201 switch (nvme_req(req)->status & 0x7ff) { 202 case NVME_SC_SUCCESS: 203 return BLK_STS_OK; 204 case NVME_SC_CAP_EXCEEDED: 205 return BLK_STS_NOSPC; 206 case NVME_SC_LBA_RANGE: 207 return BLK_STS_TARGET; 208 case NVME_SC_BAD_ATTRIBUTES: 209 case NVME_SC_ONCS_NOT_SUPPORTED: 210 case NVME_SC_INVALID_OPCODE: 211 case NVME_SC_INVALID_FIELD: 212 case NVME_SC_INVALID_NS: 213 return BLK_STS_NOTSUPP; 214 case NVME_SC_WRITE_FAULT: 215 case NVME_SC_READ_ERROR: 216 case NVME_SC_UNWRITTEN_BLOCK: 217 case NVME_SC_ACCESS_DENIED: 218 case NVME_SC_READ_ONLY: 219 case NVME_SC_COMPARE_FAILED: 220 return BLK_STS_MEDIUM; 221 case NVME_SC_GUARD_CHECK: 222 case NVME_SC_APPTAG_CHECK: 223 case NVME_SC_REFTAG_CHECK: 224 case NVME_SC_INVALID_PI: 225 return BLK_STS_PROTECTION; 226 case NVME_SC_RESERVATION_CONFLICT: 227 return BLK_STS_NEXUS; 228 default: 229 return BLK_STS_IOERR; 230 } 231 } 232 233 static inline bool nvme_req_needs_retry(struct request *req) 234 { 235 if (blk_noretry_request(req)) 236 return false; 237 if (nvme_req(req)->status & NVME_SC_DNR) 238 return false; 239 if (nvme_req(req)->retries >= nvme_max_retries) 240 return false; 241 return true; 242 } 243 244 static void nvme_retry_req(struct request *req) 245 { 246 struct nvme_ns *ns = req->q->queuedata; 247 unsigned long delay = 0; 248 u16 crd; 249 250 /* The mask and shift result must be <= 3 */ 251 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11; 252 if (ns && crd) 253 delay = ns->ctrl->crdt[crd - 1] * 100; 254 255 nvme_req(req)->retries++; 256 blk_mq_requeue_request(req, false); 257 blk_mq_delay_kick_requeue_list(req->q, delay); 258 } 259 260 void nvme_complete_rq(struct request *req) 261 { 262 blk_status_t status = nvme_error_status(req); 263 264 trace_nvme_complete_rq(req); 265 266 if (nvme_req(req)->ctrl->kas) 267 nvme_req(req)->ctrl->comp_seen = true; 268 269 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) { 270 if ((req->cmd_flags & REQ_NVME_MPATH) && 271 blk_path_error(status)) { 272 nvme_failover_req(req); 273 return; 274 } 275 276 if (!blk_queue_dying(req->q)) { 277 nvme_retry_req(req); 278 return; 279 } 280 } 281 blk_mq_end_request(req, status); 282 } 283 EXPORT_SYMBOL_GPL(nvme_complete_rq); 284 285 bool nvme_cancel_request(struct request *req, void *data, bool reserved) 286 { 287 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device, 288 "Cancelling I/O %d", req->tag); 289 290 nvme_req(req)->status = NVME_SC_ABORT_REQ; 291 blk_mq_complete_request_sync(req); 292 return true; 293 } 294 EXPORT_SYMBOL_GPL(nvme_cancel_request); 295 296 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl, 297 enum nvme_ctrl_state new_state) 298 { 299 enum nvme_ctrl_state old_state; 300 unsigned long flags; 301 bool changed = false; 302 303 spin_lock_irqsave(&ctrl->lock, flags); 304 305 old_state = ctrl->state; 306 switch (new_state) { 307 case NVME_CTRL_ADMIN_ONLY: 308 switch (old_state) { 309 case NVME_CTRL_CONNECTING: 310 changed = true; 311 /* FALLTHRU */ 312 default: 313 break; 314 } 315 break; 316 case NVME_CTRL_LIVE: 317 switch (old_state) { 318 case NVME_CTRL_NEW: 319 case NVME_CTRL_RESETTING: 320 case NVME_CTRL_CONNECTING: 321 changed = true; 322 /* FALLTHRU */ 323 default: 324 break; 325 } 326 break; 327 case NVME_CTRL_RESETTING: 328 switch (old_state) { 329 case NVME_CTRL_NEW: 330 case NVME_CTRL_LIVE: 331 case NVME_CTRL_ADMIN_ONLY: 332 changed = true; 333 /* FALLTHRU */ 334 default: 335 break; 336 } 337 break; 338 case NVME_CTRL_CONNECTING: 339 switch (old_state) { 340 case NVME_CTRL_NEW: 341 case NVME_CTRL_RESETTING: 342 changed = true; 343 /* FALLTHRU */ 344 default: 345 break; 346 } 347 break; 348 case NVME_CTRL_DELETING: 349 switch (old_state) { 350 case NVME_CTRL_LIVE: 351 case NVME_CTRL_ADMIN_ONLY: 352 case NVME_CTRL_RESETTING: 353 case NVME_CTRL_CONNECTING: 354 changed = true; 355 /* FALLTHRU */ 356 default: 357 break; 358 } 359 break; 360 case NVME_CTRL_DEAD: 361 switch (old_state) { 362 case NVME_CTRL_DELETING: 363 changed = true; 364 /* FALLTHRU */ 365 default: 366 break; 367 } 368 break; 369 default: 370 break; 371 } 372 373 if (changed) 374 ctrl->state = new_state; 375 376 spin_unlock_irqrestore(&ctrl->lock, flags); 377 if (changed && ctrl->state == NVME_CTRL_LIVE) 378 nvme_kick_requeue_lists(ctrl); 379 return changed; 380 } 381 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state); 382 383 static void nvme_free_ns_head(struct kref *ref) 384 { 385 struct nvme_ns_head *head = 386 container_of(ref, struct nvme_ns_head, ref); 387 388 nvme_mpath_remove_disk(head); 389 ida_simple_remove(&head->subsys->ns_ida, head->instance); 390 list_del_init(&head->entry); 391 cleanup_srcu_struct(&head->srcu); 392 nvme_put_subsystem(head->subsys); 393 kfree(head); 394 } 395 396 static void nvme_put_ns_head(struct nvme_ns_head *head) 397 { 398 kref_put(&head->ref, nvme_free_ns_head); 399 } 400 401 static void nvme_free_ns(struct kref *kref) 402 { 403 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref); 404 405 if (ns->ndev) 406 nvme_nvm_unregister(ns); 407 408 put_disk(ns->disk); 409 nvme_put_ns_head(ns->head); 410 nvme_put_ctrl(ns->ctrl); 411 kfree(ns); 412 } 413 414 static void nvme_put_ns(struct nvme_ns *ns) 415 { 416 kref_put(&ns->kref, nvme_free_ns); 417 } 418 419 static inline void nvme_clear_nvme_request(struct request *req) 420 { 421 if (!(req->rq_flags & RQF_DONTPREP)) { 422 nvme_req(req)->retries = 0; 423 nvme_req(req)->flags = 0; 424 req->rq_flags |= RQF_DONTPREP; 425 } 426 } 427 428 struct request *nvme_alloc_request(struct request_queue *q, 429 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid) 430 { 431 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN; 432 struct request *req; 433 434 if (qid == NVME_QID_ANY) { 435 req = blk_mq_alloc_request(q, op, flags); 436 } else { 437 req = blk_mq_alloc_request_hctx(q, op, flags, 438 qid ? qid - 1 : 0); 439 } 440 if (IS_ERR(req)) 441 return req; 442 443 req->cmd_flags |= REQ_FAILFAST_DRIVER; 444 nvme_clear_nvme_request(req); 445 nvme_req(req)->cmd = cmd; 446 447 return req; 448 } 449 EXPORT_SYMBOL_GPL(nvme_alloc_request); 450 451 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable) 452 { 453 struct nvme_command c; 454 455 memset(&c, 0, sizeof(c)); 456 457 c.directive.opcode = nvme_admin_directive_send; 458 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL); 459 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE; 460 c.directive.dtype = NVME_DIR_IDENTIFY; 461 c.directive.tdtype = NVME_DIR_STREAMS; 462 c.directive.endir = enable ? NVME_DIR_ENDIR : 0; 463 464 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0); 465 } 466 467 static int nvme_disable_streams(struct nvme_ctrl *ctrl) 468 { 469 return nvme_toggle_streams(ctrl, false); 470 } 471 472 static int nvme_enable_streams(struct nvme_ctrl *ctrl) 473 { 474 return nvme_toggle_streams(ctrl, true); 475 } 476 477 static int nvme_get_stream_params(struct nvme_ctrl *ctrl, 478 struct streams_directive_params *s, u32 nsid) 479 { 480 struct nvme_command c; 481 482 memset(&c, 0, sizeof(c)); 483 memset(s, 0, sizeof(*s)); 484 485 c.directive.opcode = nvme_admin_directive_recv; 486 c.directive.nsid = cpu_to_le32(nsid); 487 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1); 488 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM; 489 c.directive.dtype = NVME_DIR_STREAMS; 490 491 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s)); 492 } 493 494 static int nvme_configure_directives(struct nvme_ctrl *ctrl) 495 { 496 struct streams_directive_params s; 497 int ret; 498 499 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES)) 500 return 0; 501 if (!streams) 502 return 0; 503 504 ret = nvme_enable_streams(ctrl); 505 if (ret) 506 return ret; 507 508 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL); 509 if (ret) 510 return ret; 511 512 ctrl->nssa = le16_to_cpu(s.nssa); 513 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) { 514 dev_info(ctrl->device, "too few streams (%u) available\n", 515 ctrl->nssa); 516 nvme_disable_streams(ctrl); 517 return 0; 518 } 519 520 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1); 521 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams); 522 return 0; 523 } 524 525 /* 526 * Check if 'req' has a write hint associated with it. If it does, assign 527 * a valid namespace stream to the write. 528 */ 529 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl, 530 struct request *req, u16 *control, 531 u32 *dsmgmt) 532 { 533 enum rw_hint streamid = req->write_hint; 534 535 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE) 536 streamid = 0; 537 else { 538 streamid--; 539 if (WARN_ON_ONCE(streamid > ctrl->nr_streams)) 540 return; 541 542 *control |= NVME_RW_DTYPE_STREAMS; 543 *dsmgmt |= streamid << 16; 544 } 545 546 if (streamid < ARRAY_SIZE(req->q->write_hints)) 547 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9; 548 } 549 550 static inline void nvme_setup_flush(struct nvme_ns *ns, 551 struct nvme_command *cmnd) 552 { 553 cmnd->common.opcode = nvme_cmd_flush; 554 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id); 555 } 556 557 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req, 558 struct nvme_command *cmnd) 559 { 560 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0; 561 struct nvme_dsm_range *range; 562 struct bio *bio; 563 564 range = kmalloc_array(segments, sizeof(*range), 565 GFP_ATOMIC | __GFP_NOWARN); 566 if (!range) { 567 /* 568 * If we fail allocation our range, fallback to the controller 569 * discard page. If that's also busy, it's safe to return 570 * busy, as we know we can make progress once that's freed. 571 */ 572 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy)) 573 return BLK_STS_RESOURCE; 574 575 range = page_address(ns->ctrl->discard_page); 576 } 577 578 __rq_for_each_bio(bio, req) { 579 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector); 580 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift; 581 582 if (n < segments) { 583 range[n].cattr = cpu_to_le32(0); 584 range[n].nlb = cpu_to_le32(nlb); 585 range[n].slba = cpu_to_le64(slba); 586 } 587 n++; 588 } 589 590 if (WARN_ON_ONCE(n != segments)) { 591 if (virt_to_page(range) == ns->ctrl->discard_page) 592 clear_bit_unlock(0, &ns->ctrl->discard_page_busy); 593 else 594 kfree(range); 595 return BLK_STS_IOERR; 596 } 597 598 cmnd->dsm.opcode = nvme_cmd_dsm; 599 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id); 600 cmnd->dsm.nr = cpu_to_le32(segments - 1); 601 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD); 602 603 req->special_vec.bv_page = virt_to_page(range); 604 req->special_vec.bv_offset = offset_in_page(range); 605 req->special_vec.bv_len = sizeof(*range) * segments; 606 req->rq_flags |= RQF_SPECIAL_PAYLOAD; 607 608 return BLK_STS_OK; 609 } 610 611 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns, 612 struct request *req, struct nvme_command *cmnd) 613 { 614 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES) 615 return nvme_setup_discard(ns, req, cmnd); 616 617 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes; 618 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id); 619 cmnd->write_zeroes.slba = 620 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); 621 cmnd->write_zeroes.length = 622 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1); 623 cmnd->write_zeroes.control = 0; 624 return BLK_STS_OK; 625 } 626 627 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns, 628 struct request *req, struct nvme_command *cmnd) 629 { 630 struct nvme_ctrl *ctrl = ns->ctrl; 631 u16 control = 0; 632 u32 dsmgmt = 0; 633 634 if (req->cmd_flags & REQ_FUA) 635 control |= NVME_RW_FUA; 636 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD)) 637 control |= NVME_RW_LR; 638 639 if (req->cmd_flags & REQ_RAHEAD) 640 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH; 641 642 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read); 643 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id); 644 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req))); 645 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1); 646 647 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams) 648 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt); 649 650 if (ns->ms) { 651 /* 652 * If formated with metadata, the block layer always provides a 653 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else 654 * we enable the PRACT bit for protection information or set the 655 * namespace capacity to zero to prevent any I/O. 656 */ 657 if (!blk_integrity_rq(req)) { 658 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns))) 659 return BLK_STS_NOTSUPP; 660 control |= NVME_RW_PRINFO_PRACT; 661 } else if (req_op(req) == REQ_OP_WRITE) { 662 t10_pi_prepare(req, ns->pi_type); 663 } 664 665 switch (ns->pi_type) { 666 case NVME_NS_DPS_PI_TYPE3: 667 control |= NVME_RW_PRINFO_PRCHK_GUARD; 668 break; 669 case NVME_NS_DPS_PI_TYPE1: 670 case NVME_NS_DPS_PI_TYPE2: 671 control |= NVME_RW_PRINFO_PRCHK_GUARD | 672 NVME_RW_PRINFO_PRCHK_REF; 673 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req)); 674 break; 675 } 676 } 677 678 cmnd->rw.control = cpu_to_le16(control); 679 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt); 680 return 0; 681 } 682 683 void nvme_cleanup_cmd(struct request *req) 684 { 685 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ && 686 nvme_req(req)->status == 0) { 687 struct nvme_ns *ns = req->rq_disk->private_data; 688 689 t10_pi_complete(req, ns->pi_type, 690 blk_rq_bytes(req) >> ns->lba_shift); 691 } 692 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) { 693 struct nvme_ns *ns = req->rq_disk->private_data; 694 struct page *page = req->special_vec.bv_page; 695 696 if (page == ns->ctrl->discard_page) 697 clear_bit_unlock(0, &ns->ctrl->discard_page_busy); 698 else 699 kfree(page_address(page) + req->special_vec.bv_offset); 700 } 701 } 702 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd); 703 704 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req, 705 struct nvme_command *cmd) 706 { 707 blk_status_t ret = BLK_STS_OK; 708 709 nvme_clear_nvme_request(req); 710 711 memset(cmd, 0, sizeof(*cmd)); 712 switch (req_op(req)) { 713 case REQ_OP_DRV_IN: 714 case REQ_OP_DRV_OUT: 715 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd)); 716 break; 717 case REQ_OP_FLUSH: 718 nvme_setup_flush(ns, cmd); 719 break; 720 case REQ_OP_WRITE_ZEROES: 721 ret = nvme_setup_write_zeroes(ns, req, cmd); 722 break; 723 case REQ_OP_DISCARD: 724 ret = nvme_setup_discard(ns, req, cmd); 725 break; 726 case REQ_OP_READ: 727 case REQ_OP_WRITE: 728 ret = nvme_setup_rw(ns, req, cmd); 729 break; 730 default: 731 WARN_ON_ONCE(1); 732 return BLK_STS_IOERR; 733 } 734 735 cmd->common.command_id = req->tag; 736 trace_nvme_setup_cmd(req, cmd); 737 return ret; 738 } 739 EXPORT_SYMBOL_GPL(nvme_setup_cmd); 740 741 static void nvme_end_sync_rq(struct request *rq, blk_status_t error) 742 { 743 struct completion *waiting = rq->end_io_data; 744 745 rq->end_io_data = NULL; 746 complete(waiting); 747 } 748 749 static void nvme_execute_rq_polled(struct request_queue *q, 750 struct gendisk *bd_disk, struct request *rq, int at_head) 751 { 752 DECLARE_COMPLETION_ONSTACK(wait); 753 754 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)); 755 756 rq->cmd_flags |= REQ_HIPRI; 757 rq->end_io_data = &wait; 758 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq); 759 760 while (!completion_done(&wait)) { 761 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true); 762 cond_resched(); 763 } 764 } 765 766 /* 767 * Returns 0 on success. If the result is negative, it's a Linux error code; 768 * if the result is positive, it's an NVM Express status code 769 */ 770 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 771 union nvme_result *result, void *buffer, unsigned bufflen, 772 unsigned timeout, int qid, int at_head, 773 blk_mq_req_flags_t flags, bool poll) 774 { 775 struct request *req; 776 int ret; 777 778 req = nvme_alloc_request(q, cmd, flags, qid); 779 if (IS_ERR(req)) 780 return PTR_ERR(req); 781 782 req->timeout = timeout ? timeout : ADMIN_TIMEOUT; 783 784 if (buffer && bufflen) { 785 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL); 786 if (ret) 787 goto out; 788 } 789 790 if (poll) 791 nvme_execute_rq_polled(req->q, NULL, req, at_head); 792 else 793 blk_execute_rq(req->q, NULL, req, at_head); 794 if (result) 795 *result = nvme_req(req)->result; 796 if (nvme_req(req)->flags & NVME_REQ_CANCELLED) 797 ret = -EINTR; 798 else 799 ret = nvme_req(req)->status; 800 out: 801 blk_mq_free_request(req); 802 return ret; 803 } 804 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd); 805 806 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 807 void *buffer, unsigned bufflen) 808 { 809 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0, 810 NVME_QID_ANY, 0, 0, false); 811 } 812 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd); 813 814 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf, 815 unsigned len, u32 seed, bool write) 816 { 817 struct bio_integrity_payload *bip; 818 int ret = -ENOMEM; 819 void *buf; 820 821 buf = kmalloc(len, GFP_KERNEL); 822 if (!buf) 823 goto out; 824 825 ret = -EFAULT; 826 if (write && copy_from_user(buf, ubuf, len)) 827 goto out_free_meta; 828 829 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1); 830 if (IS_ERR(bip)) { 831 ret = PTR_ERR(bip); 832 goto out_free_meta; 833 } 834 835 bip->bip_iter.bi_size = len; 836 bip->bip_iter.bi_sector = seed; 837 ret = bio_integrity_add_page(bio, virt_to_page(buf), len, 838 offset_in_page(buf)); 839 if (ret == len) 840 return buf; 841 ret = -ENOMEM; 842 out_free_meta: 843 kfree(buf); 844 out: 845 return ERR_PTR(ret); 846 } 847 848 static int nvme_submit_user_cmd(struct request_queue *q, 849 struct nvme_command *cmd, void __user *ubuffer, 850 unsigned bufflen, void __user *meta_buffer, unsigned meta_len, 851 u32 meta_seed, u32 *result, unsigned timeout) 852 { 853 bool write = nvme_is_write(cmd); 854 struct nvme_ns *ns = q->queuedata; 855 struct gendisk *disk = ns ? ns->disk : NULL; 856 struct request *req; 857 struct bio *bio = NULL; 858 void *meta = NULL; 859 int ret; 860 861 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY); 862 if (IS_ERR(req)) 863 return PTR_ERR(req); 864 865 req->timeout = timeout ? timeout : ADMIN_TIMEOUT; 866 nvme_req(req)->flags |= NVME_REQ_USERCMD; 867 868 if (ubuffer && bufflen) { 869 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, 870 GFP_KERNEL); 871 if (ret) 872 goto out; 873 bio = req->bio; 874 bio->bi_disk = disk; 875 if (disk && meta_buffer && meta_len) { 876 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len, 877 meta_seed, write); 878 if (IS_ERR(meta)) { 879 ret = PTR_ERR(meta); 880 goto out_unmap; 881 } 882 req->cmd_flags |= REQ_INTEGRITY; 883 } 884 } 885 886 blk_execute_rq(req->q, disk, req, 0); 887 if (nvme_req(req)->flags & NVME_REQ_CANCELLED) 888 ret = -EINTR; 889 else 890 ret = nvme_req(req)->status; 891 if (result) 892 *result = le32_to_cpu(nvme_req(req)->result.u32); 893 if (meta && !ret && !write) { 894 if (copy_to_user(meta_buffer, meta, meta_len)) 895 ret = -EFAULT; 896 } 897 kfree(meta); 898 out_unmap: 899 if (bio) 900 blk_rq_unmap_user(bio); 901 out: 902 blk_mq_free_request(req); 903 return ret; 904 } 905 906 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status) 907 { 908 struct nvme_ctrl *ctrl = rq->end_io_data; 909 unsigned long flags; 910 bool startka = false; 911 912 blk_mq_free_request(rq); 913 914 if (status) { 915 dev_err(ctrl->device, 916 "failed nvme_keep_alive_end_io error=%d\n", 917 status); 918 return; 919 } 920 921 ctrl->comp_seen = false; 922 spin_lock_irqsave(&ctrl->lock, flags); 923 if (ctrl->state == NVME_CTRL_LIVE || 924 ctrl->state == NVME_CTRL_CONNECTING) 925 startka = true; 926 spin_unlock_irqrestore(&ctrl->lock, flags); 927 if (startka) 928 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); 929 } 930 931 static int nvme_keep_alive(struct nvme_ctrl *ctrl) 932 { 933 struct request *rq; 934 935 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED, 936 NVME_QID_ANY); 937 if (IS_ERR(rq)) 938 return PTR_ERR(rq); 939 940 rq->timeout = ctrl->kato * HZ; 941 rq->end_io_data = ctrl; 942 943 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io); 944 945 return 0; 946 } 947 948 static void nvme_keep_alive_work(struct work_struct *work) 949 { 950 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work), 951 struct nvme_ctrl, ka_work); 952 bool comp_seen = ctrl->comp_seen; 953 954 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) { 955 dev_dbg(ctrl->device, 956 "reschedule traffic based keep-alive timer\n"); 957 ctrl->comp_seen = false; 958 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); 959 return; 960 } 961 962 if (nvme_keep_alive(ctrl)) { 963 /* allocation failure, reset the controller */ 964 dev_err(ctrl->device, "keep-alive failed\n"); 965 nvme_reset_ctrl(ctrl); 966 return; 967 } 968 } 969 970 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl) 971 { 972 if (unlikely(ctrl->kato == 0)) 973 return; 974 975 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); 976 } 977 978 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl) 979 { 980 if (unlikely(ctrl->kato == 0)) 981 return; 982 983 cancel_delayed_work_sync(&ctrl->ka_work); 984 } 985 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive); 986 987 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id) 988 { 989 struct nvme_command c = { }; 990 int error; 991 992 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ 993 c.identify.opcode = nvme_admin_identify; 994 c.identify.cns = NVME_ID_CNS_CTRL; 995 996 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL); 997 if (!*id) 998 return -ENOMEM; 999 1000 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, 1001 sizeof(struct nvme_id_ctrl)); 1002 if (error) 1003 kfree(*id); 1004 return error; 1005 } 1006 1007 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid, 1008 struct nvme_ns_ids *ids) 1009 { 1010 struct nvme_command c = { }; 1011 int status; 1012 void *data; 1013 int pos; 1014 int len; 1015 1016 c.identify.opcode = nvme_admin_identify; 1017 c.identify.nsid = cpu_to_le32(nsid); 1018 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST; 1019 1020 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL); 1021 if (!data) 1022 return -ENOMEM; 1023 1024 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data, 1025 NVME_IDENTIFY_DATA_SIZE); 1026 if (status) 1027 goto free_data; 1028 1029 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) { 1030 struct nvme_ns_id_desc *cur = data + pos; 1031 1032 if (cur->nidl == 0) 1033 break; 1034 1035 switch (cur->nidt) { 1036 case NVME_NIDT_EUI64: 1037 if (cur->nidl != NVME_NIDT_EUI64_LEN) { 1038 dev_warn(ctrl->device, 1039 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n", 1040 cur->nidl); 1041 goto free_data; 1042 } 1043 len = NVME_NIDT_EUI64_LEN; 1044 memcpy(ids->eui64, data + pos + sizeof(*cur), len); 1045 break; 1046 case NVME_NIDT_NGUID: 1047 if (cur->nidl != NVME_NIDT_NGUID_LEN) { 1048 dev_warn(ctrl->device, 1049 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n", 1050 cur->nidl); 1051 goto free_data; 1052 } 1053 len = NVME_NIDT_NGUID_LEN; 1054 memcpy(ids->nguid, data + pos + sizeof(*cur), len); 1055 break; 1056 case NVME_NIDT_UUID: 1057 if (cur->nidl != NVME_NIDT_UUID_LEN) { 1058 dev_warn(ctrl->device, 1059 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n", 1060 cur->nidl); 1061 goto free_data; 1062 } 1063 len = NVME_NIDT_UUID_LEN; 1064 uuid_copy(&ids->uuid, data + pos + sizeof(*cur)); 1065 break; 1066 default: 1067 /* Skip unknown types */ 1068 len = cur->nidl; 1069 break; 1070 } 1071 1072 len += sizeof(*cur); 1073 } 1074 free_data: 1075 kfree(data); 1076 return status; 1077 } 1078 1079 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list) 1080 { 1081 struct nvme_command c = { }; 1082 1083 c.identify.opcode = nvme_admin_identify; 1084 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST; 1085 c.identify.nsid = cpu_to_le32(nsid); 1086 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 1087 NVME_IDENTIFY_DATA_SIZE); 1088 } 1089 1090 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl, 1091 unsigned nsid) 1092 { 1093 struct nvme_id_ns *id; 1094 struct nvme_command c = { }; 1095 int error; 1096 1097 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ 1098 c.identify.opcode = nvme_admin_identify; 1099 c.identify.nsid = cpu_to_le32(nsid); 1100 c.identify.cns = NVME_ID_CNS_NS; 1101 1102 id = kmalloc(sizeof(*id), GFP_KERNEL); 1103 if (!id) 1104 return NULL; 1105 1106 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id)); 1107 if (error) { 1108 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error); 1109 kfree(id); 1110 return NULL; 1111 } 1112 1113 return id; 1114 } 1115 1116 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid, 1117 unsigned int dword11, void *buffer, size_t buflen, u32 *result) 1118 { 1119 struct nvme_command c; 1120 union nvme_result res; 1121 int ret; 1122 1123 memset(&c, 0, sizeof(c)); 1124 c.features.opcode = op; 1125 c.features.fid = cpu_to_le32(fid); 1126 c.features.dword11 = cpu_to_le32(dword11); 1127 1128 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, 1129 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false); 1130 if (ret >= 0 && result) 1131 *result = le32_to_cpu(res.u32); 1132 return ret; 1133 } 1134 1135 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid, 1136 unsigned int dword11, void *buffer, size_t buflen, 1137 u32 *result) 1138 { 1139 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer, 1140 buflen, result); 1141 } 1142 EXPORT_SYMBOL_GPL(nvme_set_features); 1143 1144 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid, 1145 unsigned int dword11, void *buffer, size_t buflen, 1146 u32 *result) 1147 { 1148 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer, 1149 buflen, result); 1150 } 1151 EXPORT_SYMBOL_GPL(nvme_get_features); 1152 1153 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count) 1154 { 1155 u32 q_count = (*count - 1) | ((*count - 1) << 16); 1156 u32 result; 1157 int status, nr_io_queues; 1158 1159 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0, 1160 &result); 1161 if (status < 0) 1162 return status; 1163 1164 /* 1165 * Degraded controllers might return an error when setting the queue 1166 * count. We still want to be able to bring them online and offer 1167 * access to the admin queue, as that might be only way to fix them up. 1168 */ 1169 if (status > 0) { 1170 dev_err(ctrl->device, "Could not set queue count (%d)\n", status); 1171 *count = 0; 1172 } else { 1173 nr_io_queues = min(result & 0xffff, result >> 16) + 1; 1174 *count = min(*count, nr_io_queues); 1175 } 1176 1177 return 0; 1178 } 1179 EXPORT_SYMBOL_GPL(nvme_set_queue_count); 1180 1181 #define NVME_AEN_SUPPORTED \ 1182 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE) 1183 1184 static void nvme_enable_aen(struct nvme_ctrl *ctrl) 1185 { 1186 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED; 1187 int status; 1188 1189 if (!supported_aens) 1190 return; 1191 1192 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens, 1193 NULL, 0, &result); 1194 if (status) 1195 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n", 1196 supported_aens); 1197 } 1198 1199 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) 1200 { 1201 struct nvme_user_io io; 1202 struct nvme_command c; 1203 unsigned length, meta_len; 1204 void __user *metadata; 1205 1206 if (copy_from_user(&io, uio, sizeof(io))) 1207 return -EFAULT; 1208 if (io.flags) 1209 return -EINVAL; 1210 1211 switch (io.opcode) { 1212 case nvme_cmd_write: 1213 case nvme_cmd_read: 1214 case nvme_cmd_compare: 1215 break; 1216 default: 1217 return -EINVAL; 1218 } 1219 1220 length = (io.nblocks + 1) << ns->lba_shift; 1221 meta_len = (io.nblocks + 1) * ns->ms; 1222 metadata = (void __user *)(uintptr_t)io.metadata; 1223 1224 if (ns->ext) { 1225 length += meta_len; 1226 meta_len = 0; 1227 } else if (meta_len) { 1228 if ((io.metadata & 3) || !io.metadata) 1229 return -EINVAL; 1230 } 1231 1232 memset(&c, 0, sizeof(c)); 1233 c.rw.opcode = io.opcode; 1234 c.rw.flags = io.flags; 1235 c.rw.nsid = cpu_to_le32(ns->head->ns_id); 1236 c.rw.slba = cpu_to_le64(io.slba); 1237 c.rw.length = cpu_to_le16(io.nblocks); 1238 c.rw.control = cpu_to_le16(io.control); 1239 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt); 1240 c.rw.reftag = cpu_to_le32(io.reftag); 1241 c.rw.apptag = cpu_to_le16(io.apptag); 1242 c.rw.appmask = cpu_to_le16(io.appmask); 1243 1244 return nvme_submit_user_cmd(ns->queue, &c, 1245 (void __user *)(uintptr_t)io.addr, length, 1246 metadata, meta_len, lower_32_bits(io.slba), NULL, 0); 1247 } 1248 1249 static u32 nvme_known_admin_effects(u8 opcode) 1250 { 1251 switch (opcode) { 1252 case nvme_admin_format_nvm: 1253 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC | 1254 NVME_CMD_EFFECTS_CSE_MASK; 1255 case nvme_admin_sanitize_nvm: 1256 return NVME_CMD_EFFECTS_CSE_MASK; 1257 default: 1258 break; 1259 } 1260 return 0; 1261 } 1262 1263 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, 1264 u8 opcode) 1265 { 1266 u32 effects = 0; 1267 1268 if (ns) { 1269 if (ctrl->effects) 1270 effects = le32_to_cpu(ctrl->effects->iocs[opcode]); 1271 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC)) 1272 dev_warn(ctrl->device, 1273 "IO command:%02x has unhandled effects:%08x\n", 1274 opcode, effects); 1275 return 0; 1276 } 1277 1278 if (ctrl->effects) 1279 effects = le32_to_cpu(ctrl->effects->acs[opcode]); 1280 effects |= nvme_known_admin_effects(opcode); 1281 1282 /* 1283 * For simplicity, IO to all namespaces is quiesced even if the command 1284 * effects say only one namespace is affected. 1285 */ 1286 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) { 1287 mutex_lock(&ctrl->scan_lock); 1288 nvme_start_freeze(ctrl); 1289 nvme_wait_freeze(ctrl); 1290 } 1291 return effects; 1292 } 1293 1294 static void nvme_update_formats(struct nvme_ctrl *ctrl) 1295 { 1296 struct nvme_ns *ns; 1297 1298 down_read(&ctrl->namespaces_rwsem); 1299 list_for_each_entry(ns, &ctrl->namespaces, list) 1300 if (ns->disk && nvme_revalidate_disk(ns->disk)) 1301 nvme_set_queue_dying(ns); 1302 up_read(&ctrl->namespaces_rwsem); 1303 1304 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL); 1305 } 1306 1307 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects) 1308 { 1309 /* 1310 * Revalidate LBA changes prior to unfreezing. This is necessary to 1311 * prevent memory corruption if a logical block size was changed by 1312 * this command. 1313 */ 1314 if (effects & NVME_CMD_EFFECTS_LBCC) 1315 nvme_update_formats(ctrl); 1316 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) { 1317 nvme_unfreeze(ctrl); 1318 mutex_unlock(&ctrl->scan_lock); 1319 } 1320 if (effects & NVME_CMD_EFFECTS_CCC) 1321 nvme_init_identify(ctrl); 1322 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) 1323 nvme_queue_scan(ctrl); 1324 } 1325 1326 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns, 1327 struct nvme_passthru_cmd __user *ucmd) 1328 { 1329 struct nvme_passthru_cmd cmd; 1330 struct nvme_command c; 1331 unsigned timeout = 0; 1332 u32 effects; 1333 int status; 1334 1335 if (!capable(CAP_SYS_ADMIN)) 1336 return -EACCES; 1337 if (copy_from_user(&cmd, ucmd, sizeof(cmd))) 1338 return -EFAULT; 1339 if (cmd.flags) 1340 return -EINVAL; 1341 1342 memset(&c, 0, sizeof(c)); 1343 c.common.opcode = cmd.opcode; 1344 c.common.flags = cmd.flags; 1345 c.common.nsid = cpu_to_le32(cmd.nsid); 1346 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); 1347 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); 1348 c.common.cdw10 = cpu_to_le32(cmd.cdw10); 1349 c.common.cdw11 = cpu_to_le32(cmd.cdw11); 1350 c.common.cdw12 = cpu_to_le32(cmd.cdw12); 1351 c.common.cdw13 = cpu_to_le32(cmd.cdw13); 1352 c.common.cdw14 = cpu_to_le32(cmd.cdw14); 1353 c.common.cdw15 = cpu_to_le32(cmd.cdw15); 1354 1355 if (cmd.timeout_ms) 1356 timeout = msecs_to_jiffies(cmd.timeout_ms); 1357 1358 effects = nvme_passthru_start(ctrl, ns, cmd.opcode); 1359 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c, 1360 (void __user *)(uintptr_t)cmd.addr, cmd.data_len, 1361 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len, 1362 0, &cmd.result, timeout); 1363 nvme_passthru_end(ctrl, effects); 1364 1365 if (status >= 0) { 1366 if (put_user(cmd.result, &ucmd->result)) 1367 return -EFAULT; 1368 } 1369 1370 return status; 1371 } 1372 1373 /* 1374 * Issue ioctl requests on the first available path. Note that unlike normal 1375 * block layer requests we will not retry failed request on another controller. 1376 */ 1377 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk, 1378 struct nvme_ns_head **head, int *srcu_idx) 1379 { 1380 #ifdef CONFIG_NVME_MULTIPATH 1381 if (disk->fops == &nvme_ns_head_ops) { 1382 struct nvme_ns *ns; 1383 1384 *head = disk->private_data; 1385 *srcu_idx = srcu_read_lock(&(*head)->srcu); 1386 ns = nvme_find_path(*head); 1387 if (!ns) 1388 srcu_read_unlock(&(*head)->srcu, *srcu_idx); 1389 return ns; 1390 } 1391 #endif 1392 *head = NULL; 1393 *srcu_idx = -1; 1394 return disk->private_data; 1395 } 1396 1397 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx) 1398 { 1399 if (head) 1400 srcu_read_unlock(&head->srcu, idx); 1401 } 1402 1403 static int nvme_ioctl(struct block_device *bdev, fmode_t mode, 1404 unsigned int cmd, unsigned long arg) 1405 { 1406 struct nvme_ns_head *head = NULL; 1407 void __user *argp = (void __user *)arg; 1408 struct nvme_ns *ns; 1409 int srcu_idx, ret; 1410 1411 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx); 1412 if (unlikely(!ns)) 1413 return -EWOULDBLOCK; 1414 1415 /* 1416 * Handle ioctls that apply to the controller instead of the namespace 1417 * seperately and drop the ns SRCU reference early. This avoids a 1418 * deadlock when deleting namespaces using the passthrough interface. 1419 */ 1420 if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) { 1421 struct nvme_ctrl *ctrl = ns->ctrl; 1422 1423 nvme_get_ctrl(ns->ctrl); 1424 nvme_put_ns_from_disk(head, srcu_idx); 1425 1426 if (cmd == NVME_IOCTL_ADMIN_CMD) 1427 ret = nvme_user_cmd(ctrl, NULL, argp); 1428 else 1429 ret = sed_ioctl(ctrl->opal_dev, cmd, argp); 1430 1431 nvme_put_ctrl(ctrl); 1432 return ret; 1433 } 1434 1435 switch (cmd) { 1436 case NVME_IOCTL_ID: 1437 force_successful_syscall_return(); 1438 ret = ns->head->ns_id; 1439 break; 1440 case NVME_IOCTL_IO_CMD: 1441 ret = nvme_user_cmd(ns->ctrl, ns, argp); 1442 break; 1443 case NVME_IOCTL_SUBMIT_IO: 1444 ret = nvme_submit_io(ns, argp); 1445 break; 1446 default: 1447 if (ns->ndev) 1448 ret = nvme_nvm_ioctl(ns, cmd, arg); 1449 else 1450 ret = -ENOTTY; 1451 } 1452 1453 nvme_put_ns_from_disk(head, srcu_idx); 1454 return ret; 1455 } 1456 1457 static int nvme_open(struct block_device *bdev, fmode_t mode) 1458 { 1459 struct nvme_ns *ns = bdev->bd_disk->private_data; 1460 1461 #ifdef CONFIG_NVME_MULTIPATH 1462 /* should never be called due to GENHD_FL_HIDDEN */ 1463 if (WARN_ON_ONCE(ns->head->disk)) 1464 goto fail; 1465 #endif 1466 if (!kref_get_unless_zero(&ns->kref)) 1467 goto fail; 1468 if (!try_module_get(ns->ctrl->ops->module)) 1469 goto fail_put_ns; 1470 1471 return 0; 1472 1473 fail_put_ns: 1474 nvme_put_ns(ns); 1475 fail: 1476 return -ENXIO; 1477 } 1478 1479 static void nvme_release(struct gendisk *disk, fmode_t mode) 1480 { 1481 struct nvme_ns *ns = disk->private_data; 1482 1483 module_put(ns->ctrl->ops->module); 1484 nvme_put_ns(ns); 1485 } 1486 1487 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo) 1488 { 1489 /* some standard values */ 1490 geo->heads = 1 << 6; 1491 geo->sectors = 1 << 5; 1492 geo->cylinders = get_capacity(bdev->bd_disk) >> 11; 1493 return 0; 1494 } 1495 1496 #ifdef CONFIG_BLK_DEV_INTEGRITY 1497 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type) 1498 { 1499 struct blk_integrity integrity; 1500 1501 memset(&integrity, 0, sizeof(integrity)); 1502 switch (pi_type) { 1503 case NVME_NS_DPS_PI_TYPE3: 1504 integrity.profile = &t10_pi_type3_crc; 1505 integrity.tag_size = sizeof(u16) + sizeof(u32); 1506 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE; 1507 break; 1508 case NVME_NS_DPS_PI_TYPE1: 1509 case NVME_NS_DPS_PI_TYPE2: 1510 integrity.profile = &t10_pi_type1_crc; 1511 integrity.tag_size = sizeof(u16); 1512 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE; 1513 break; 1514 default: 1515 integrity.profile = NULL; 1516 break; 1517 } 1518 integrity.tuple_size = ms; 1519 blk_integrity_register(disk, &integrity); 1520 blk_queue_max_integrity_segments(disk->queue, 1); 1521 } 1522 #else 1523 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type) 1524 { 1525 } 1526 #endif /* CONFIG_BLK_DEV_INTEGRITY */ 1527 1528 static void nvme_set_chunk_size(struct nvme_ns *ns) 1529 { 1530 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9)); 1531 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size)); 1532 } 1533 1534 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns) 1535 { 1536 struct nvme_ctrl *ctrl = ns->ctrl; 1537 struct request_queue *queue = disk->queue; 1538 u32 size = queue_logical_block_size(queue); 1539 1540 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) { 1541 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue); 1542 return; 1543 } 1544 1545 if (ctrl->nr_streams && ns->sws && ns->sgs) 1546 size *= ns->sws * ns->sgs; 1547 1548 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) < 1549 NVME_DSM_MAX_RANGES); 1550 1551 queue->limits.discard_alignment = 0; 1552 queue->limits.discard_granularity = size; 1553 1554 /* If discard is already enabled, don't reset queue limits */ 1555 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue)) 1556 return; 1557 1558 blk_queue_max_discard_sectors(queue, UINT_MAX); 1559 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES); 1560 1561 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES) 1562 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX); 1563 } 1564 1565 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns) 1566 { 1567 u32 max_sectors; 1568 unsigned short bs = 1 << ns->lba_shift; 1569 1570 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) || 1571 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES)) 1572 return; 1573 /* 1574 * Even though NVMe spec explicitly states that MDTS is not 1575 * applicable to the write-zeroes:- "The restriction does not apply to 1576 * commands that do not transfer data between the host and the 1577 * controller (e.g., Write Uncorrectable ro Write Zeroes command).". 1578 * In order to be more cautious use controller's max_hw_sectors value 1579 * to configure the maximum sectors for the write-zeroes which is 1580 * configured based on the controller's MDTS field in the 1581 * nvme_init_identify() if available. 1582 */ 1583 if (ns->ctrl->max_hw_sectors == UINT_MAX) 1584 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9; 1585 else 1586 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9; 1587 1588 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors); 1589 } 1590 1591 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid, 1592 struct nvme_id_ns *id, struct nvme_ns_ids *ids) 1593 { 1594 memset(ids, 0, sizeof(*ids)); 1595 1596 if (ctrl->vs >= NVME_VS(1, 1, 0)) 1597 memcpy(ids->eui64, id->eui64, sizeof(id->eui64)); 1598 if (ctrl->vs >= NVME_VS(1, 2, 0)) 1599 memcpy(ids->nguid, id->nguid, sizeof(id->nguid)); 1600 if (ctrl->vs >= NVME_VS(1, 3, 0)) { 1601 /* Don't treat error as fatal we potentially 1602 * already have a NGUID or EUI-64 1603 */ 1604 if (nvme_identify_ns_descs(ctrl, nsid, ids)) 1605 dev_warn(ctrl->device, 1606 "%s: Identify Descriptors failed\n", __func__); 1607 } 1608 } 1609 1610 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids) 1611 { 1612 return !uuid_is_null(&ids->uuid) || 1613 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) || 1614 memchr_inv(ids->eui64, 0, sizeof(ids->eui64)); 1615 } 1616 1617 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b) 1618 { 1619 return uuid_equal(&a->uuid, &b->uuid) && 1620 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 && 1621 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0; 1622 } 1623 1624 static void nvme_update_disk_info(struct gendisk *disk, 1625 struct nvme_ns *ns, struct nvme_id_ns *id) 1626 { 1627 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9); 1628 unsigned short bs = 1 << ns->lba_shift; 1629 1630 if (ns->lba_shift > PAGE_SHIFT) { 1631 /* unsupported block size, set capacity to 0 later */ 1632 bs = (1 << 9); 1633 } 1634 blk_mq_freeze_queue(disk->queue); 1635 blk_integrity_unregister(disk); 1636 1637 blk_queue_logical_block_size(disk->queue, bs); 1638 blk_queue_physical_block_size(disk->queue, bs); 1639 blk_queue_io_min(disk->queue, bs); 1640 1641 if (ns->ms && !ns->ext && 1642 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)) 1643 nvme_init_integrity(disk, ns->ms, ns->pi_type); 1644 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) || 1645 ns->lba_shift > PAGE_SHIFT) 1646 capacity = 0; 1647 1648 set_capacity(disk, capacity); 1649 1650 nvme_config_discard(disk, ns); 1651 nvme_config_write_zeroes(disk, ns); 1652 1653 if (id->nsattr & (1 << 0)) 1654 set_disk_ro(disk, true); 1655 else 1656 set_disk_ro(disk, false); 1657 1658 blk_mq_unfreeze_queue(disk->queue); 1659 } 1660 1661 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id) 1662 { 1663 struct nvme_ns *ns = disk->private_data; 1664 1665 /* 1666 * If identify namespace failed, use default 512 byte block size so 1667 * block layer can use before failing read/write for 0 capacity. 1668 */ 1669 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds; 1670 if (ns->lba_shift == 0) 1671 ns->lba_shift = 9; 1672 ns->noiob = le16_to_cpu(id->noiob); 1673 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms); 1674 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT); 1675 /* the PI implementation requires metadata equal t10 pi tuple size */ 1676 if (ns->ms == sizeof(struct t10_pi_tuple)) 1677 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK; 1678 else 1679 ns->pi_type = 0; 1680 1681 if (ns->noiob) 1682 nvme_set_chunk_size(ns); 1683 nvme_update_disk_info(disk, ns, id); 1684 #ifdef CONFIG_NVME_MULTIPATH 1685 if (ns->head->disk) { 1686 nvme_update_disk_info(ns->head->disk, ns, id); 1687 blk_queue_stack_limits(ns->head->disk->queue, ns->queue); 1688 } 1689 #endif 1690 } 1691 1692 static int nvme_revalidate_disk(struct gendisk *disk) 1693 { 1694 struct nvme_ns *ns = disk->private_data; 1695 struct nvme_ctrl *ctrl = ns->ctrl; 1696 struct nvme_id_ns *id; 1697 struct nvme_ns_ids ids; 1698 int ret = 0; 1699 1700 if (test_bit(NVME_NS_DEAD, &ns->flags)) { 1701 set_capacity(disk, 0); 1702 return -ENODEV; 1703 } 1704 1705 id = nvme_identify_ns(ctrl, ns->head->ns_id); 1706 if (!id) 1707 return -ENODEV; 1708 1709 if (id->ncap == 0) { 1710 ret = -ENODEV; 1711 goto out; 1712 } 1713 1714 __nvme_revalidate_disk(disk, id); 1715 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids); 1716 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) { 1717 dev_err(ctrl->device, 1718 "identifiers changed for nsid %d\n", ns->head->ns_id); 1719 ret = -ENODEV; 1720 } 1721 1722 out: 1723 kfree(id); 1724 return ret; 1725 } 1726 1727 static char nvme_pr_type(enum pr_type type) 1728 { 1729 switch (type) { 1730 case PR_WRITE_EXCLUSIVE: 1731 return 1; 1732 case PR_EXCLUSIVE_ACCESS: 1733 return 2; 1734 case PR_WRITE_EXCLUSIVE_REG_ONLY: 1735 return 3; 1736 case PR_EXCLUSIVE_ACCESS_REG_ONLY: 1737 return 4; 1738 case PR_WRITE_EXCLUSIVE_ALL_REGS: 1739 return 5; 1740 case PR_EXCLUSIVE_ACCESS_ALL_REGS: 1741 return 6; 1742 default: 1743 return 0; 1744 } 1745 }; 1746 1747 static int nvme_pr_command(struct block_device *bdev, u32 cdw10, 1748 u64 key, u64 sa_key, u8 op) 1749 { 1750 struct nvme_ns_head *head = NULL; 1751 struct nvme_ns *ns; 1752 struct nvme_command c; 1753 int srcu_idx, ret; 1754 u8 data[16] = { 0, }; 1755 1756 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx); 1757 if (unlikely(!ns)) 1758 return -EWOULDBLOCK; 1759 1760 put_unaligned_le64(key, &data[0]); 1761 put_unaligned_le64(sa_key, &data[8]); 1762 1763 memset(&c, 0, sizeof(c)); 1764 c.common.opcode = op; 1765 c.common.nsid = cpu_to_le32(ns->head->ns_id); 1766 c.common.cdw10 = cpu_to_le32(cdw10); 1767 1768 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16); 1769 nvme_put_ns_from_disk(head, srcu_idx); 1770 return ret; 1771 } 1772 1773 static int nvme_pr_register(struct block_device *bdev, u64 old, 1774 u64 new, unsigned flags) 1775 { 1776 u32 cdw10; 1777 1778 if (flags & ~PR_FL_IGNORE_KEY) 1779 return -EOPNOTSUPP; 1780 1781 cdw10 = old ? 2 : 0; 1782 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0; 1783 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */ 1784 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register); 1785 } 1786 1787 static int nvme_pr_reserve(struct block_device *bdev, u64 key, 1788 enum pr_type type, unsigned flags) 1789 { 1790 u32 cdw10; 1791 1792 if (flags & ~PR_FL_IGNORE_KEY) 1793 return -EOPNOTSUPP; 1794 1795 cdw10 = nvme_pr_type(type) << 8; 1796 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0); 1797 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire); 1798 } 1799 1800 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new, 1801 enum pr_type type, bool abort) 1802 { 1803 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1); 1804 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire); 1805 } 1806 1807 static int nvme_pr_clear(struct block_device *bdev, u64 key) 1808 { 1809 u32 cdw10 = 1 | (key ? 1 << 3 : 0); 1810 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register); 1811 } 1812 1813 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type) 1814 { 1815 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0); 1816 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release); 1817 } 1818 1819 static const struct pr_ops nvme_pr_ops = { 1820 .pr_register = nvme_pr_register, 1821 .pr_reserve = nvme_pr_reserve, 1822 .pr_release = nvme_pr_release, 1823 .pr_preempt = nvme_pr_preempt, 1824 .pr_clear = nvme_pr_clear, 1825 }; 1826 1827 #ifdef CONFIG_BLK_SED_OPAL 1828 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len, 1829 bool send) 1830 { 1831 struct nvme_ctrl *ctrl = data; 1832 struct nvme_command cmd; 1833 1834 memset(&cmd, 0, sizeof(cmd)); 1835 if (send) 1836 cmd.common.opcode = nvme_admin_security_send; 1837 else 1838 cmd.common.opcode = nvme_admin_security_recv; 1839 cmd.common.nsid = 0; 1840 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8); 1841 cmd.common.cdw11 = cpu_to_le32(len); 1842 1843 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 1844 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false); 1845 } 1846 EXPORT_SYMBOL_GPL(nvme_sec_submit); 1847 #endif /* CONFIG_BLK_SED_OPAL */ 1848 1849 static const struct block_device_operations nvme_fops = { 1850 .owner = THIS_MODULE, 1851 .ioctl = nvme_ioctl, 1852 .compat_ioctl = nvme_ioctl, 1853 .open = nvme_open, 1854 .release = nvme_release, 1855 .getgeo = nvme_getgeo, 1856 .revalidate_disk= nvme_revalidate_disk, 1857 .pr_ops = &nvme_pr_ops, 1858 }; 1859 1860 #ifdef CONFIG_NVME_MULTIPATH 1861 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode) 1862 { 1863 struct nvme_ns_head *head = bdev->bd_disk->private_data; 1864 1865 if (!kref_get_unless_zero(&head->ref)) 1866 return -ENXIO; 1867 return 0; 1868 } 1869 1870 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode) 1871 { 1872 nvme_put_ns_head(disk->private_data); 1873 } 1874 1875 const struct block_device_operations nvme_ns_head_ops = { 1876 .owner = THIS_MODULE, 1877 .open = nvme_ns_head_open, 1878 .release = nvme_ns_head_release, 1879 .ioctl = nvme_ioctl, 1880 .compat_ioctl = nvme_ioctl, 1881 .getgeo = nvme_getgeo, 1882 .pr_ops = &nvme_pr_ops, 1883 }; 1884 #endif /* CONFIG_NVME_MULTIPATH */ 1885 1886 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled) 1887 { 1888 unsigned long timeout = 1889 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; 1890 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0; 1891 int ret; 1892 1893 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { 1894 if (csts == ~0) 1895 return -ENODEV; 1896 if ((csts & NVME_CSTS_RDY) == bit) 1897 break; 1898 1899 msleep(100); 1900 if (fatal_signal_pending(current)) 1901 return -EINTR; 1902 if (time_after(jiffies, timeout)) { 1903 dev_err(ctrl->device, 1904 "Device not ready; aborting %s\n", enabled ? 1905 "initialisation" : "reset"); 1906 return -ENODEV; 1907 } 1908 } 1909 1910 return ret; 1911 } 1912 1913 /* 1914 * If the device has been passed off to us in an enabled state, just clear 1915 * the enabled bit. The spec says we should set the 'shutdown notification 1916 * bits', but doing so may cause the device to complete commands to the 1917 * admin queue ... and we don't know what memory that might be pointing at! 1918 */ 1919 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap) 1920 { 1921 int ret; 1922 1923 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; 1924 ctrl->ctrl_config &= ~NVME_CC_ENABLE; 1925 1926 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); 1927 if (ret) 1928 return ret; 1929 1930 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) 1931 msleep(NVME_QUIRK_DELAY_AMOUNT); 1932 1933 return nvme_wait_ready(ctrl, cap, false); 1934 } 1935 EXPORT_SYMBOL_GPL(nvme_disable_ctrl); 1936 1937 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap) 1938 { 1939 /* 1940 * Default to a 4K page size, with the intention to update this 1941 * path in the future to accomodate architectures with differing 1942 * kernel and IO page sizes. 1943 */ 1944 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12; 1945 int ret; 1946 1947 if (page_shift < dev_page_min) { 1948 dev_err(ctrl->device, 1949 "Minimum device page size %u too large for host (%u)\n", 1950 1 << dev_page_min, 1 << page_shift); 1951 return -ENODEV; 1952 } 1953 1954 ctrl->page_size = 1 << page_shift; 1955 1956 ctrl->ctrl_config = NVME_CC_CSS_NVM; 1957 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT; 1958 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE; 1959 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; 1960 ctrl->ctrl_config |= NVME_CC_ENABLE; 1961 1962 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); 1963 if (ret) 1964 return ret; 1965 return nvme_wait_ready(ctrl, cap, true); 1966 } 1967 EXPORT_SYMBOL_GPL(nvme_enable_ctrl); 1968 1969 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl) 1970 { 1971 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ); 1972 u32 csts; 1973 int ret; 1974 1975 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; 1976 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL; 1977 1978 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); 1979 if (ret) 1980 return ret; 1981 1982 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { 1983 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT) 1984 break; 1985 1986 msleep(100); 1987 if (fatal_signal_pending(current)) 1988 return -EINTR; 1989 if (time_after(jiffies, timeout)) { 1990 dev_err(ctrl->device, 1991 "Device shutdown incomplete; abort shutdown\n"); 1992 return -ENODEV; 1993 } 1994 } 1995 1996 return ret; 1997 } 1998 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl); 1999 2000 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl, 2001 struct request_queue *q) 2002 { 2003 bool vwc = false; 2004 2005 if (ctrl->max_hw_sectors) { 2006 u32 max_segments = 2007 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1; 2008 2009 max_segments = min_not_zero(max_segments, ctrl->max_segments); 2010 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors); 2011 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX)); 2012 } 2013 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && 2014 is_power_of_2(ctrl->max_hw_sectors)) 2015 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors); 2016 blk_queue_virt_boundary(q, ctrl->page_size - 1); 2017 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT) 2018 vwc = true; 2019 blk_queue_write_cache(q, vwc, vwc); 2020 } 2021 2022 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl) 2023 { 2024 __le64 ts; 2025 int ret; 2026 2027 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP)) 2028 return 0; 2029 2030 ts = cpu_to_le64(ktime_to_ms(ktime_get_real())); 2031 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts), 2032 NULL); 2033 if (ret) 2034 dev_warn_once(ctrl->device, 2035 "could not set timestamp (%d)\n", ret); 2036 return ret; 2037 } 2038 2039 static int nvme_configure_acre(struct nvme_ctrl *ctrl) 2040 { 2041 struct nvme_feat_host_behavior *host; 2042 int ret; 2043 2044 /* Don't bother enabling the feature if retry delay is not reported */ 2045 if (!ctrl->crdt[0]) 2046 return 0; 2047 2048 host = kzalloc(sizeof(*host), GFP_KERNEL); 2049 if (!host) 2050 return 0; 2051 2052 host->acre = NVME_ENABLE_ACRE; 2053 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0, 2054 host, sizeof(*host), NULL); 2055 kfree(host); 2056 return ret; 2057 } 2058 2059 static int nvme_configure_apst(struct nvme_ctrl *ctrl) 2060 { 2061 /* 2062 * APST (Autonomous Power State Transition) lets us program a 2063 * table of power state transitions that the controller will 2064 * perform automatically. We configure it with a simple 2065 * heuristic: we are willing to spend at most 2% of the time 2066 * transitioning between power states. Therefore, when running 2067 * in any given state, we will enter the next lower-power 2068 * non-operational state after waiting 50 * (enlat + exlat) 2069 * microseconds, as long as that state's exit latency is under 2070 * the requested maximum latency. 2071 * 2072 * We will not autonomously enter any non-operational state for 2073 * which the total latency exceeds ps_max_latency_us. Users 2074 * can set ps_max_latency_us to zero to turn off APST. 2075 */ 2076 2077 unsigned apste; 2078 struct nvme_feat_auto_pst *table; 2079 u64 max_lat_us = 0; 2080 int max_ps = -1; 2081 int ret; 2082 2083 /* 2084 * If APST isn't supported or if we haven't been initialized yet, 2085 * then don't do anything. 2086 */ 2087 if (!ctrl->apsta) 2088 return 0; 2089 2090 if (ctrl->npss > 31) { 2091 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n"); 2092 return 0; 2093 } 2094 2095 table = kzalloc(sizeof(*table), GFP_KERNEL); 2096 if (!table) 2097 return 0; 2098 2099 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) { 2100 /* Turn off APST. */ 2101 apste = 0; 2102 dev_dbg(ctrl->device, "APST disabled\n"); 2103 } else { 2104 __le64 target = cpu_to_le64(0); 2105 int state; 2106 2107 /* 2108 * Walk through all states from lowest- to highest-power. 2109 * According to the spec, lower-numbered states use more 2110 * power. NPSS, despite the name, is the index of the 2111 * lowest-power state, not the number of states. 2112 */ 2113 for (state = (int)ctrl->npss; state >= 0; state--) { 2114 u64 total_latency_us, exit_latency_us, transition_ms; 2115 2116 if (target) 2117 table->entries[state] = target; 2118 2119 /* 2120 * Don't allow transitions to the deepest state 2121 * if it's quirked off. 2122 */ 2123 if (state == ctrl->npss && 2124 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) 2125 continue; 2126 2127 /* 2128 * Is this state a useful non-operational state for 2129 * higher-power states to autonomously transition to? 2130 */ 2131 if (!(ctrl->psd[state].flags & 2132 NVME_PS_FLAGS_NON_OP_STATE)) 2133 continue; 2134 2135 exit_latency_us = 2136 (u64)le32_to_cpu(ctrl->psd[state].exit_lat); 2137 if (exit_latency_us > ctrl->ps_max_latency_us) 2138 continue; 2139 2140 total_latency_us = 2141 exit_latency_us + 2142 le32_to_cpu(ctrl->psd[state].entry_lat); 2143 2144 /* 2145 * This state is good. Use it as the APST idle 2146 * target for higher power states. 2147 */ 2148 transition_ms = total_latency_us + 19; 2149 do_div(transition_ms, 20); 2150 if (transition_ms > (1 << 24) - 1) 2151 transition_ms = (1 << 24) - 1; 2152 2153 target = cpu_to_le64((state << 3) | 2154 (transition_ms << 8)); 2155 2156 if (max_ps == -1) 2157 max_ps = state; 2158 2159 if (total_latency_us > max_lat_us) 2160 max_lat_us = total_latency_us; 2161 } 2162 2163 apste = 1; 2164 2165 if (max_ps == -1) { 2166 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n"); 2167 } else { 2168 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n", 2169 max_ps, max_lat_us, (int)sizeof(*table), table); 2170 } 2171 } 2172 2173 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste, 2174 table, sizeof(*table), NULL); 2175 if (ret) 2176 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret); 2177 2178 kfree(table); 2179 return ret; 2180 } 2181 2182 static void nvme_set_latency_tolerance(struct device *dev, s32 val) 2183 { 2184 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); 2185 u64 latency; 2186 2187 switch (val) { 2188 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT: 2189 case PM_QOS_LATENCY_ANY: 2190 latency = U64_MAX; 2191 break; 2192 2193 default: 2194 latency = val; 2195 } 2196 2197 if (ctrl->ps_max_latency_us != latency) { 2198 ctrl->ps_max_latency_us = latency; 2199 nvme_configure_apst(ctrl); 2200 } 2201 } 2202 2203 struct nvme_core_quirk_entry { 2204 /* 2205 * NVMe model and firmware strings are padded with spaces. For 2206 * simplicity, strings in the quirk table are padded with NULLs 2207 * instead. 2208 */ 2209 u16 vid; 2210 const char *mn; 2211 const char *fr; 2212 unsigned long quirks; 2213 }; 2214 2215 static const struct nvme_core_quirk_entry core_quirks[] = { 2216 { 2217 /* 2218 * This Toshiba device seems to die using any APST states. See: 2219 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11 2220 */ 2221 .vid = 0x1179, 2222 .mn = "THNSF5256GPUK TOSHIBA", 2223 .quirks = NVME_QUIRK_NO_APST, 2224 } 2225 }; 2226 2227 /* match is null-terminated but idstr is space-padded. */ 2228 static bool string_matches(const char *idstr, const char *match, size_t len) 2229 { 2230 size_t matchlen; 2231 2232 if (!match) 2233 return true; 2234 2235 matchlen = strlen(match); 2236 WARN_ON_ONCE(matchlen > len); 2237 2238 if (memcmp(idstr, match, matchlen)) 2239 return false; 2240 2241 for (; matchlen < len; matchlen++) 2242 if (idstr[matchlen] != ' ') 2243 return false; 2244 2245 return true; 2246 } 2247 2248 static bool quirk_matches(const struct nvme_id_ctrl *id, 2249 const struct nvme_core_quirk_entry *q) 2250 { 2251 return q->vid == le16_to_cpu(id->vid) && 2252 string_matches(id->mn, q->mn, sizeof(id->mn)) && 2253 string_matches(id->fr, q->fr, sizeof(id->fr)); 2254 } 2255 2256 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl, 2257 struct nvme_id_ctrl *id) 2258 { 2259 size_t nqnlen; 2260 int off; 2261 2262 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) { 2263 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE); 2264 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) { 2265 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE); 2266 return; 2267 } 2268 2269 if (ctrl->vs >= NVME_VS(1, 2, 1)) 2270 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n"); 2271 } 2272 2273 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */ 2274 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE, 2275 "nqn.2014.08.org.nvmexpress:%04x%04x", 2276 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid)); 2277 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn)); 2278 off += sizeof(id->sn); 2279 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn)); 2280 off += sizeof(id->mn); 2281 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off); 2282 } 2283 2284 static void __nvme_release_subsystem(struct nvme_subsystem *subsys) 2285 { 2286 ida_simple_remove(&nvme_subsystems_ida, subsys->instance); 2287 kfree(subsys); 2288 } 2289 2290 static void nvme_release_subsystem(struct device *dev) 2291 { 2292 __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev)); 2293 } 2294 2295 static void nvme_destroy_subsystem(struct kref *ref) 2296 { 2297 struct nvme_subsystem *subsys = 2298 container_of(ref, struct nvme_subsystem, ref); 2299 2300 mutex_lock(&nvme_subsystems_lock); 2301 list_del(&subsys->entry); 2302 mutex_unlock(&nvme_subsystems_lock); 2303 2304 ida_destroy(&subsys->ns_ida); 2305 device_del(&subsys->dev); 2306 put_device(&subsys->dev); 2307 } 2308 2309 static void nvme_put_subsystem(struct nvme_subsystem *subsys) 2310 { 2311 kref_put(&subsys->ref, nvme_destroy_subsystem); 2312 } 2313 2314 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn) 2315 { 2316 struct nvme_subsystem *subsys; 2317 2318 lockdep_assert_held(&nvme_subsystems_lock); 2319 2320 list_for_each_entry(subsys, &nvme_subsystems, entry) { 2321 if (strcmp(subsys->subnqn, subsysnqn)) 2322 continue; 2323 if (!kref_get_unless_zero(&subsys->ref)) 2324 continue; 2325 return subsys; 2326 } 2327 2328 return NULL; 2329 } 2330 2331 #define SUBSYS_ATTR_RO(_name, _mode, _show) \ 2332 struct device_attribute subsys_attr_##_name = \ 2333 __ATTR(_name, _mode, _show, NULL) 2334 2335 static ssize_t nvme_subsys_show_nqn(struct device *dev, 2336 struct device_attribute *attr, 2337 char *buf) 2338 { 2339 struct nvme_subsystem *subsys = 2340 container_of(dev, struct nvme_subsystem, dev); 2341 2342 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn); 2343 } 2344 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn); 2345 2346 #define nvme_subsys_show_str_function(field) \ 2347 static ssize_t subsys_##field##_show(struct device *dev, \ 2348 struct device_attribute *attr, char *buf) \ 2349 { \ 2350 struct nvme_subsystem *subsys = \ 2351 container_of(dev, struct nvme_subsystem, dev); \ 2352 return sprintf(buf, "%.*s\n", \ 2353 (int)sizeof(subsys->field), subsys->field); \ 2354 } \ 2355 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show); 2356 2357 nvme_subsys_show_str_function(model); 2358 nvme_subsys_show_str_function(serial); 2359 nvme_subsys_show_str_function(firmware_rev); 2360 2361 static struct attribute *nvme_subsys_attrs[] = { 2362 &subsys_attr_model.attr, 2363 &subsys_attr_serial.attr, 2364 &subsys_attr_firmware_rev.attr, 2365 &subsys_attr_subsysnqn.attr, 2366 #ifdef CONFIG_NVME_MULTIPATH 2367 &subsys_attr_iopolicy.attr, 2368 #endif 2369 NULL, 2370 }; 2371 2372 static struct attribute_group nvme_subsys_attrs_group = { 2373 .attrs = nvme_subsys_attrs, 2374 }; 2375 2376 static const struct attribute_group *nvme_subsys_attrs_groups[] = { 2377 &nvme_subsys_attrs_group, 2378 NULL, 2379 }; 2380 2381 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys, 2382 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id) 2383 { 2384 struct nvme_ctrl *tmp; 2385 2386 lockdep_assert_held(&nvme_subsystems_lock); 2387 2388 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) { 2389 if (ctrl->state == NVME_CTRL_DELETING || 2390 ctrl->state == NVME_CTRL_DEAD) 2391 continue; 2392 2393 if (tmp->cntlid == ctrl->cntlid) { 2394 dev_err(ctrl->device, 2395 "Duplicate cntlid %u with %s, rejecting\n", 2396 ctrl->cntlid, dev_name(tmp->device)); 2397 return false; 2398 } 2399 2400 if ((id->cmic & (1 << 1)) || 2401 (ctrl->opts && ctrl->opts->discovery_nqn)) 2402 continue; 2403 2404 dev_err(ctrl->device, 2405 "Subsystem does not support multiple controllers\n"); 2406 return false; 2407 } 2408 2409 return true; 2410 } 2411 2412 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id) 2413 { 2414 struct nvme_subsystem *subsys, *found; 2415 int ret; 2416 2417 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL); 2418 if (!subsys) 2419 return -ENOMEM; 2420 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL); 2421 if (ret < 0) { 2422 kfree(subsys); 2423 return ret; 2424 } 2425 subsys->instance = ret; 2426 mutex_init(&subsys->lock); 2427 kref_init(&subsys->ref); 2428 INIT_LIST_HEAD(&subsys->ctrls); 2429 INIT_LIST_HEAD(&subsys->nsheads); 2430 nvme_init_subnqn(subsys, ctrl, id); 2431 memcpy(subsys->serial, id->sn, sizeof(subsys->serial)); 2432 memcpy(subsys->model, id->mn, sizeof(subsys->model)); 2433 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev)); 2434 subsys->vendor_id = le16_to_cpu(id->vid); 2435 subsys->cmic = id->cmic; 2436 #ifdef CONFIG_NVME_MULTIPATH 2437 subsys->iopolicy = NVME_IOPOLICY_NUMA; 2438 #endif 2439 2440 subsys->dev.class = nvme_subsys_class; 2441 subsys->dev.release = nvme_release_subsystem; 2442 subsys->dev.groups = nvme_subsys_attrs_groups; 2443 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance); 2444 device_initialize(&subsys->dev); 2445 2446 mutex_lock(&nvme_subsystems_lock); 2447 found = __nvme_find_get_subsystem(subsys->subnqn); 2448 if (found) { 2449 __nvme_release_subsystem(subsys); 2450 subsys = found; 2451 2452 if (!nvme_validate_cntlid(subsys, ctrl, id)) { 2453 ret = -EINVAL; 2454 goto out_put_subsystem; 2455 } 2456 } else { 2457 ret = device_add(&subsys->dev); 2458 if (ret) { 2459 dev_err(ctrl->device, 2460 "failed to register subsystem device.\n"); 2461 goto out_unlock; 2462 } 2463 ida_init(&subsys->ns_ida); 2464 list_add_tail(&subsys->entry, &nvme_subsystems); 2465 } 2466 2467 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj, 2468 dev_name(ctrl->device))) { 2469 dev_err(ctrl->device, 2470 "failed to create sysfs link from subsystem.\n"); 2471 goto out_put_subsystem; 2472 } 2473 2474 ctrl->subsys = subsys; 2475 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls); 2476 mutex_unlock(&nvme_subsystems_lock); 2477 return 0; 2478 2479 out_put_subsystem: 2480 nvme_put_subsystem(subsys); 2481 out_unlock: 2482 mutex_unlock(&nvme_subsystems_lock); 2483 put_device(&subsys->dev); 2484 return ret; 2485 } 2486 2487 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, 2488 void *log, size_t size, u64 offset) 2489 { 2490 struct nvme_command c = { }; 2491 unsigned long dwlen = size / 4 - 1; 2492 2493 c.get_log_page.opcode = nvme_admin_get_log_page; 2494 c.get_log_page.nsid = cpu_to_le32(nsid); 2495 c.get_log_page.lid = log_page; 2496 c.get_log_page.lsp = lsp; 2497 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1)); 2498 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16); 2499 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset)); 2500 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset)); 2501 2502 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size); 2503 } 2504 2505 static int nvme_get_effects_log(struct nvme_ctrl *ctrl) 2506 { 2507 int ret; 2508 2509 if (!ctrl->effects) 2510 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL); 2511 2512 if (!ctrl->effects) 2513 return 0; 2514 2515 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0, 2516 ctrl->effects, sizeof(*ctrl->effects), 0); 2517 if (ret) { 2518 kfree(ctrl->effects); 2519 ctrl->effects = NULL; 2520 } 2521 return ret; 2522 } 2523 2524 /* 2525 * Initialize the cached copies of the Identify data and various controller 2526 * register in our nvme_ctrl structure. This should be called as soon as 2527 * the admin queue is fully up and running. 2528 */ 2529 int nvme_init_identify(struct nvme_ctrl *ctrl) 2530 { 2531 struct nvme_id_ctrl *id; 2532 u64 cap; 2533 int ret, page_shift; 2534 u32 max_hw_sectors; 2535 bool prev_apst_enabled; 2536 2537 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs); 2538 if (ret) { 2539 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret); 2540 return ret; 2541 } 2542 2543 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap); 2544 if (ret) { 2545 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret); 2546 return ret; 2547 } 2548 page_shift = NVME_CAP_MPSMIN(cap) + 12; 2549 2550 if (ctrl->vs >= NVME_VS(1, 1, 0)) 2551 ctrl->subsystem = NVME_CAP_NSSRC(cap); 2552 2553 ret = nvme_identify_ctrl(ctrl, &id); 2554 if (ret) { 2555 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret); 2556 return -EIO; 2557 } 2558 2559 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) { 2560 ret = nvme_get_effects_log(ctrl); 2561 if (ret < 0) 2562 goto out_free; 2563 } 2564 2565 if (!ctrl->identified) { 2566 int i; 2567 2568 ret = nvme_init_subsystem(ctrl, id); 2569 if (ret) 2570 goto out_free; 2571 2572 /* 2573 * Check for quirks. Quirk can depend on firmware version, 2574 * so, in principle, the set of quirks present can change 2575 * across a reset. As a possible future enhancement, we 2576 * could re-scan for quirks every time we reinitialize 2577 * the device, but we'd have to make sure that the driver 2578 * behaves intelligently if the quirks change. 2579 */ 2580 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) { 2581 if (quirk_matches(id, &core_quirks[i])) 2582 ctrl->quirks |= core_quirks[i].quirks; 2583 } 2584 } 2585 2586 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) { 2587 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n"); 2588 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS; 2589 } 2590 2591 ctrl->crdt[0] = le16_to_cpu(id->crdt1); 2592 ctrl->crdt[1] = le16_to_cpu(id->crdt2); 2593 ctrl->crdt[2] = le16_to_cpu(id->crdt3); 2594 2595 ctrl->oacs = le16_to_cpu(id->oacs); 2596 ctrl->oncs = le16_to_cpu(id->oncs); 2597 ctrl->mtfa = le16_to_cpu(id->mtfa); 2598 ctrl->oaes = le32_to_cpu(id->oaes); 2599 atomic_set(&ctrl->abort_limit, id->acl + 1); 2600 ctrl->vwc = id->vwc; 2601 if (id->mdts) 2602 max_hw_sectors = 1 << (id->mdts + page_shift - 9); 2603 else 2604 max_hw_sectors = UINT_MAX; 2605 ctrl->max_hw_sectors = 2606 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors); 2607 2608 nvme_set_queue_limits(ctrl, ctrl->admin_q); 2609 ctrl->sgls = le32_to_cpu(id->sgls); 2610 ctrl->kas = le16_to_cpu(id->kas); 2611 ctrl->max_namespaces = le32_to_cpu(id->mnan); 2612 ctrl->ctratt = le32_to_cpu(id->ctratt); 2613 2614 if (id->rtd3e) { 2615 /* us -> s */ 2616 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000; 2617 2618 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time, 2619 shutdown_timeout, 60); 2620 2621 if (ctrl->shutdown_timeout != shutdown_timeout) 2622 dev_info(ctrl->device, 2623 "Shutdown timeout set to %u seconds\n", 2624 ctrl->shutdown_timeout); 2625 } else 2626 ctrl->shutdown_timeout = shutdown_timeout; 2627 2628 ctrl->npss = id->npss; 2629 ctrl->apsta = id->apsta; 2630 prev_apst_enabled = ctrl->apst_enabled; 2631 if (ctrl->quirks & NVME_QUIRK_NO_APST) { 2632 if (force_apst && id->apsta) { 2633 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n"); 2634 ctrl->apst_enabled = true; 2635 } else { 2636 ctrl->apst_enabled = false; 2637 } 2638 } else { 2639 ctrl->apst_enabled = id->apsta; 2640 } 2641 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd)); 2642 2643 if (ctrl->ops->flags & NVME_F_FABRICS) { 2644 ctrl->icdoff = le16_to_cpu(id->icdoff); 2645 ctrl->ioccsz = le32_to_cpu(id->ioccsz); 2646 ctrl->iorcsz = le32_to_cpu(id->iorcsz); 2647 ctrl->maxcmd = le16_to_cpu(id->maxcmd); 2648 2649 /* 2650 * In fabrics we need to verify the cntlid matches the 2651 * admin connect 2652 */ 2653 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) { 2654 ret = -EINVAL; 2655 goto out_free; 2656 } 2657 2658 if (!ctrl->opts->discovery_nqn && !ctrl->kas) { 2659 dev_err(ctrl->device, 2660 "keep-alive support is mandatory for fabrics\n"); 2661 ret = -EINVAL; 2662 goto out_free; 2663 } 2664 } else { 2665 ctrl->cntlid = le16_to_cpu(id->cntlid); 2666 ctrl->hmpre = le32_to_cpu(id->hmpre); 2667 ctrl->hmmin = le32_to_cpu(id->hmmin); 2668 ctrl->hmminds = le32_to_cpu(id->hmminds); 2669 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd); 2670 } 2671 2672 ret = nvme_mpath_init(ctrl, id); 2673 kfree(id); 2674 2675 if (ret < 0) 2676 return ret; 2677 2678 if (ctrl->apst_enabled && !prev_apst_enabled) 2679 dev_pm_qos_expose_latency_tolerance(ctrl->device); 2680 else if (!ctrl->apst_enabled && prev_apst_enabled) 2681 dev_pm_qos_hide_latency_tolerance(ctrl->device); 2682 2683 ret = nvme_configure_apst(ctrl); 2684 if (ret < 0) 2685 return ret; 2686 2687 ret = nvme_configure_timestamp(ctrl); 2688 if (ret < 0) 2689 return ret; 2690 2691 ret = nvme_configure_directives(ctrl); 2692 if (ret < 0) 2693 return ret; 2694 2695 ret = nvme_configure_acre(ctrl); 2696 if (ret < 0) 2697 return ret; 2698 2699 ctrl->identified = true; 2700 2701 return 0; 2702 2703 out_free: 2704 kfree(id); 2705 return ret; 2706 } 2707 EXPORT_SYMBOL_GPL(nvme_init_identify); 2708 2709 static int nvme_dev_open(struct inode *inode, struct file *file) 2710 { 2711 struct nvme_ctrl *ctrl = 2712 container_of(inode->i_cdev, struct nvme_ctrl, cdev); 2713 2714 switch (ctrl->state) { 2715 case NVME_CTRL_LIVE: 2716 case NVME_CTRL_ADMIN_ONLY: 2717 break; 2718 default: 2719 return -EWOULDBLOCK; 2720 } 2721 2722 file->private_data = ctrl; 2723 return 0; 2724 } 2725 2726 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp) 2727 { 2728 struct nvme_ns *ns; 2729 int ret; 2730 2731 down_read(&ctrl->namespaces_rwsem); 2732 if (list_empty(&ctrl->namespaces)) { 2733 ret = -ENOTTY; 2734 goto out_unlock; 2735 } 2736 2737 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list); 2738 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) { 2739 dev_warn(ctrl->device, 2740 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n"); 2741 ret = -EINVAL; 2742 goto out_unlock; 2743 } 2744 2745 dev_warn(ctrl->device, 2746 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n"); 2747 kref_get(&ns->kref); 2748 up_read(&ctrl->namespaces_rwsem); 2749 2750 ret = nvme_user_cmd(ctrl, ns, argp); 2751 nvme_put_ns(ns); 2752 return ret; 2753 2754 out_unlock: 2755 up_read(&ctrl->namespaces_rwsem); 2756 return ret; 2757 } 2758 2759 static long nvme_dev_ioctl(struct file *file, unsigned int cmd, 2760 unsigned long arg) 2761 { 2762 struct nvme_ctrl *ctrl = file->private_data; 2763 void __user *argp = (void __user *)arg; 2764 2765 switch (cmd) { 2766 case NVME_IOCTL_ADMIN_CMD: 2767 return nvme_user_cmd(ctrl, NULL, argp); 2768 case NVME_IOCTL_IO_CMD: 2769 return nvme_dev_user_cmd(ctrl, argp); 2770 case NVME_IOCTL_RESET: 2771 dev_warn(ctrl->device, "resetting controller\n"); 2772 return nvme_reset_ctrl_sync(ctrl); 2773 case NVME_IOCTL_SUBSYS_RESET: 2774 return nvme_reset_subsystem(ctrl); 2775 case NVME_IOCTL_RESCAN: 2776 nvme_queue_scan(ctrl); 2777 return 0; 2778 default: 2779 return -ENOTTY; 2780 } 2781 } 2782 2783 static const struct file_operations nvme_dev_fops = { 2784 .owner = THIS_MODULE, 2785 .open = nvme_dev_open, 2786 .unlocked_ioctl = nvme_dev_ioctl, 2787 .compat_ioctl = nvme_dev_ioctl, 2788 }; 2789 2790 static ssize_t nvme_sysfs_reset(struct device *dev, 2791 struct device_attribute *attr, const char *buf, 2792 size_t count) 2793 { 2794 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); 2795 int ret; 2796 2797 ret = nvme_reset_ctrl_sync(ctrl); 2798 if (ret < 0) 2799 return ret; 2800 return count; 2801 } 2802 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset); 2803 2804 static ssize_t nvme_sysfs_rescan(struct device *dev, 2805 struct device_attribute *attr, const char *buf, 2806 size_t count) 2807 { 2808 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); 2809 2810 nvme_queue_scan(ctrl); 2811 return count; 2812 } 2813 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan); 2814 2815 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev) 2816 { 2817 struct gendisk *disk = dev_to_disk(dev); 2818 2819 if (disk->fops == &nvme_fops) 2820 return nvme_get_ns_from_dev(dev)->head; 2821 else 2822 return disk->private_data; 2823 } 2824 2825 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr, 2826 char *buf) 2827 { 2828 struct nvme_ns_head *head = dev_to_ns_head(dev); 2829 struct nvme_ns_ids *ids = &head->ids; 2830 struct nvme_subsystem *subsys = head->subsys; 2831 int serial_len = sizeof(subsys->serial); 2832 int model_len = sizeof(subsys->model); 2833 2834 if (!uuid_is_null(&ids->uuid)) 2835 return sprintf(buf, "uuid.%pU\n", &ids->uuid); 2836 2837 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid))) 2838 return sprintf(buf, "eui.%16phN\n", ids->nguid); 2839 2840 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64))) 2841 return sprintf(buf, "eui.%8phN\n", ids->eui64); 2842 2843 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' || 2844 subsys->serial[serial_len - 1] == '\0')) 2845 serial_len--; 2846 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' || 2847 subsys->model[model_len - 1] == '\0')) 2848 model_len--; 2849 2850 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id, 2851 serial_len, subsys->serial, model_len, subsys->model, 2852 head->ns_id); 2853 } 2854 static DEVICE_ATTR_RO(wwid); 2855 2856 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr, 2857 char *buf) 2858 { 2859 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid); 2860 } 2861 static DEVICE_ATTR_RO(nguid); 2862 2863 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr, 2864 char *buf) 2865 { 2866 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids; 2867 2868 /* For backward compatibility expose the NGUID to userspace if 2869 * we have no UUID set 2870 */ 2871 if (uuid_is_null(&ids->uuid)) { 2872 printk_ratelimited(KERN_WARNING 2873 "No UUID available providing old NGUID\n"); 2874 return sprintf(buf, "%pU\n", ids->nguid); 2875 } 2876 return sprintf(buf, "%pU\n", &ids->uuid); 2877 } 2878 static DEVICE_ATTR_RO(uuid); 2879 2880 static ssize_t eui_show(struct device *dev, struct device_attribute *attr, 2881 char *buf) 2882 { 2883 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64); 2884 } 2885 static DEVICE_ATTR_RO(eui); 2886 2887 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr, 2888 char *buf) 2889 { 2890 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id); 2891 } 2892 static DEVICE_ATTR_RO(nsid); 2893 2894 static struct attribute *nvme_ns_id_attrs[] = { 2895 &dev_attr_wwid.attr, 2896 &dev_attr_uuid.attr, 2897 &dev_attr_nguid.attr, 2898 &dev_attr_eui.attr, 2899 &dev_attr_nsid.attr, 2900 #ifdef CONFIG_NVME_MULTIPATH 2901 &dev_attr_ana_grpid.attr, 2902 &dev_attr_ana_state.attr, 2903 #endif 2904 NULL, 2905 }; 2906 2907 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj, 2908 struct attribute *a, int n) 2909 { 2910 struct device *dev = container_of(kobj, struct device, kobj); 2911 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids; 2912 2913 if (a == &dev_attr_uuid.attr) { 2914 if (uuid_is_null(&ids->uuid) && 2915 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid))) 2916 return 0; 2917 } 2918 if (a == &dev_attr_nguid.attr) { 2919 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid))) 2920 return 0; 2921 } 2922 if (a == &dev_attr_eui.attr) { 2923 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64))) 2924 return 0; 2925 } 2926 #ifdef CONFIG_NVME_MULTIPATH 2927 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) { 2928 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */ 2929 return 0; 2930 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl)) 2931 return 0; 2932 } 2933 #endif 2934 return a->mode; 2935 } 2936 2937 static const struct attribute_group nvme_ns_id_attr_group = { 2938 .attrs = nvme_ns_id_attrs, 2939 .is_visible = nvme_ns_id_attrs_are_visible, 2940 }; 2941 2942 const struct attribute_group *nvme_ns_id_attr_groups[] = { 2943 &nvme_ns_id_attr_group, 2944 #ifdef CONFIG_NVM 2945 &nvme_nvm_attr_group, 2946 #endif 2947 NULL, 2948 }; 2949 2950 #define nvme_show_str_function(field) \ 2951 static ssize_t field##_show(struct device *dev, \ 2952 struct device_attribute *attr, char *buf) \ 2953 { \ 2954 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \ 2955 return sprintf(buf, "%.*s\n", \ 2956 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \ 2957 } \ 2958 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL); 2959 2960 nvme_show_str_function(model); 2961 nvme_show_str_function(serial); 2962 nvme_show_str_function(firmware_rev); 2963 2964 #define nvme_show_int_function(field) \ 2965 static ssize_t field##_show(struct device *dev, \ 2966 struct device_attribute *attr, char *buf) \ 2967 { \ 2968 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \ 2969 return sprintf(buf, "%d\n", ctrl->field); \ 2970 } \ 2971 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL); 2972 2973 nvme_show_int_function(cntlid); 2974 nvme_show_int_function(numa_node); 2975 2976 static ssize_t nvme_sysfs_delete(struct device *dev, 2977 struct device_attribute *attr, const char *buf, 2978 size_t count) 2979 { 2980 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); 2981 2982 if (device_remove_file_self(dev, attr)) 2983 nvme_delete_ctrl_sync(ctrl); 2984 return count; 2985 } 2986 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete); 2987 2988 static ssize_t nvme_sysfs_show_transport(struct device *dev, 2989 struct device_attribute *attr, 2990 char *buf) 2991 { 2992 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); 2993 2994 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name); 2995 } 2996 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL); 2997 2998 static ssize_t nvme_sysfs_show_state(struct device *dev, 2999 struct device_attribute *attr, 3000 char *buf) 3001 { 3002 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); 3003 static const char *const state_name[] = { 3004 [NVME_CTRL_NEW] = "new", 3005 [NVME_CTRL_LIVE] = "live", 3006 [NVME_CTRL_ADMIN_ONLY] = "only-admin", 3007 [NVME_CTRL_RESETTING] = "resetting", 3008 [NVME_CTRL_CONNECTING] = "connecting", 3009 [NVME_CTRL_DELETING] = "deleting", 3010 [NVME_CTRL_DEAD] = "dead", 3011 }; 3012 3013 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) && 3014 state_name[ctrl->state]) 3015 return sprintf(buf, "%s\n", state_name[ctrl->state]); 3016 3017 return sprintf(buf, "unknown state\n"); 3018 } 3019 3020 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL); 3021 3022 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev, 3023 struct device_attribute *attr, 3024 char *buf) 3025 { 3026 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); 3027 3028 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn); 3029 } 3030 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL); 3031 3032 static ssize_t nvme_sysfs_show_address(struct device *dev, 3033 struct device_attribute *attr, 3034 char *buf) 3035 { 3036 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); 3037 3038 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE); 3039 } 3040 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL); 3041 3042 static struct attribute *nvme_dev_attrs[] = { 3043 &dev_attr_reset_controller.attr, 3044 &dev_attr_rescan_controller.attr, 3045 &dev_attr_model.attr, 3046 &dev_attr_serial.attr, 3047 &dev_attr_firmware_rev.attr, 3048 &dev_attr_cntlid.attr, 3049 &dev_attr_delete_controller.attr, 3050 &dev_attr_transport.attr, 3051 &dev_attr_subsysnqn.attr, 3052 &dev_attr_address.attr, 3053 &dev_attr_state.attr, 3054 &dev_attr_numa_node.attr, 3055 NULL 3056 }; 3057 3058 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj, 3059 struct attribute *a, int n) 3060 { 3061 struct device *dev = container_of(kobj, struct device, kobj); 3062 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); 3063 3064 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl) 3065 return 0; 3066 if (a == &dev_attr_address.attr && !ctrl->ops->get_address) 3067 return 0; 3068 3069 return a->mode; 3070 } 3071 3072 static struct attribute_group nvme_dev_attrs_group = { 3073 .attrs = nvme_dev_attrs, 3074 .is_visible = nvme_dev_attrs_are_visible, 3075 }; 3076 3077 static const struct attribute_group *nvme_dev_attr_groups[] = { 3078 &nvme_dev_attrs_group, 3079 NULL, 3080 }; 3081 3082 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys, 3083 unsigned nsid) 3084 { 3085 struct nvme_ns_head *h; 3086 3087 lockdep_assert_held(&subsys->lock); 3088 3089 list_for_each_entry(h, &subsys->nsheads, entry) { 3090 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref)) 3091 return h; 3092 } 3093 3094 return NULL; 3095 } 3096 3097 static int __nvme_check_ids(struct nvme_subsystem *subsys, 3098 struct nvme_ns_head *new) 3099 { 3100 struct nvme_ns_head *h; 3101 3102 lockdep_assert_held(&subsys->lock); 3103 3104 list_for_each_entry(h, &subsys->nsheads, entry) { 3105 if (nvme_ns_ids_valid(&new->ids) && 3106 !list_empty(&h->list) && 3107 nvme_ns_ids_equal(&new->ids, &h->ids)) 3108 return -EINVAL; 3109 } 3110 3111 return 0; 3112 } 3113 3114 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl, 3115 unsigned nsid, struct nvme_id_ns *id) 3116 { 3117 struct nvme_ns_head *head; 3118 size_t size = sizeof(*head); 3119 int ret = -ENOMEM; 3120 3121 #ifdef CONFIG_NVME_MULTIPATH 3122 size += num_possible_nodes() * sizeof(struct nvme_ns *); 3123 #endif 3124 3125 head = kzalloc(size, GFP_KERNEL); 3126 if (!head) 3127 goto out; 3128 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL); 3129 if (ret < 0) 3130 goto out_free_head; 3131 head->instance = ret; 3132 INIT_LIST_HEAD(&head->list); 3133 ret = init_srcu_struct(&head->srcu); 3134 if (ret) 3135 goto out_ida_remove; 3136 head->subsys = ctrl->subsys; 3137 head->ns_id = nsid; 3138 kref_init(&head->ref); 3139 3140 nvme_report_ns_ids(ctrl, nsid, id, &head->ids); 3141 3142 ret = __nvme_check_ids(ctrl->subsys, head); 3143 if (ret) { 3144 dev_err(ctrl->device, 3145 "duplicate IDs for nsid %d\n", nsid); 3146 goto out_cleanup_srcu; 3147 } 3148 3149 ret = nvme_mpath_alloc_disk(ctrl, head); 3150 if (ret) 3151 goto out_cleanup_srcu; 3152 3153 list_add_tail(&head->entry, &ctrl->subsys->nsheads); 3154 3155 kref_get(&ctrl->subsys->ref); 3156 3157 return head; 3158 out_cleanup_srcu: 3159 cleanup_srcu_struct(&head->srcu); 3160 out_ida_remove: 3161 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance); 3162 out_free_head: 3163 kfree(head); 3164 out: 3165 return ERR_PTR(ret); 3166 } 3167 3168 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid, 3169 struct nvme_id_ns *id) 3170 { 3171 struct nvme_ctrl *ctrl = ns->ctrl; 3172 bool is_shared = id->nmic & (1 << 0); 3173 struct nvme_ns_head *head = NULL; 3174 int ret = 0; 3175 3176 mutex_lock(&ctrl->subsys->lock); 3177 if (is_shared) 3178 head = __nvme_find_ns_head(ctrl->subsys, nsid); 3179 if (!head) { 3180 head = nvme_alloc_ns_head(ctrl, nsid, id); 3181 if (IS_ERR(head)) { 3182 ret = PTR_ERR(head); 3183 goto out_unlock; 3184 } 3185 } else { 3186 struct nvme_ns_ids ids; 3187 3188 nvme_report_ns_ids(ctrl, nsid, id, &ids); 3189 if (!nvme_ns_ids_equal(&head->ids, &ids)) { 3190 dev_err(ctrl->device, 3191 "IDs don't match for shared namespace %d\n", 3192 nsid); 3193 ret = -EINVAL; 3194 goto out_unlock; 3195 } 3196 } 3197 3198 list_add_tail(&ns->siblings, &head->list); 3199 ns->head = head; 3200 3201 out_unlock: 3202 mutex_unlock(&ctrl->subsys->lock); 3203 return ret; 3204 } 3205 3206 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b) 3207 { 3208 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list); 3209 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list); 3210 3211 return nsa->head->ns_id - nsb->head->ns_id; 3212 } 3213 3214 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid) 3215 { 3216 struct nvme_ns *ns, *ret = NULL; 3217 3218 down_read(&ctrl->namespaces_rwsem); 3219 list_for_each_entry(ns, &ctrl->namespaces, list) { 3220 if (ns->head->ns_id == nsid) { 3221 if (!kref_get_unless_zero(&ns->kref)) 3222 continue; 3223 ret = ns; 3224 break; 3225 } 3226 if (ns->head->ns_id > nsid) 3227 break; 3228 } 3229 up_read(&ctrl->namespaces_rwsem); 3230 return ret; 3231 } 3232 3233 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns) 3234 { 3235 struct streams_directive_params s; 3236 int ret; 3237 3238 if (!ctrl->nr_streams) 3239 return 0; 3240 3241 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id); 3242 if (ret) 3243 return ret; 3244 3245 ns->sws = le32_to_cpu(s.sws); 3246 ns->sgs = le16_to_cpu(s.sgs); 3247 3248 if (ns->sws) { 3249 unsigned int bs = 1 << ns->lba_shift; 3250 3251 blk_queue_io_min(ns->queue, bs * ns->sws); 3252 if (ns->sgs) 3253 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs); 3254 } 3255 3256 return 0; 3257 } 3258 3259 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid) 3260 { 3261 struct nvme_ns *ns; 3262 struct gendisk *disk; 3263 struct nvme_id_ns *id; 3264 char disk_name[DISK_NAME_LEN]; 3265 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret; 3266 3267 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node); 3268 if (!ns) 3269 return -ENOMEM; 3270 3271 ns->queue = blk_mq_init_queue(ctrl->tagset); 3272 if (IS_ERR(ns->queue)) { 3273 ret = PTR_ERR(ns->queue); 3274 goto out_free_ns; 3275 } 3276 3277 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue); 3278 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA) 3279 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue); 3280 3281 ns->queue->queuedata = ns; 3282 ns->ctrl = ctrl; 3283 3284 kref_init(&ns->kref); 3285 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */ 3286 3287 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift); 3288 nvme_set_queue_limits(ctrl, ns->queue); 3289 3290 id = nvme_identify_ns(ctrl, nsid); 3291 if (!id) { 3292 ret = -EIO; 3293 goto out_free_queue; 3294 } 3295 3296 if (id->ncap == 0) { 3297 ret = -EINVAL; 3298 goto out_free_id; 3299 } 3300 3301 ret = nvme_init_ns_head(ns, nsid, id); 3302 if (ret) 3303 goto out_free_id; 3304 nvme_setup_streams_ns(ctrl, ns); 3305 nvme_set_disk_name(disk_name, ns, ctrl, &flags); 3306 3307 disk = alloc_disk_node(0, node); 3308 if (!disk) { 3309 ret = -ENOMEM; 3310 goto out_unlink_ns; 3311 } 3312 3313 disk->fops = &nvme_fops; 3314 disk->private_data = ns; 3315 disk->queue = ns->queue; 3316 disk->flags = flags; 3317 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN); 3318 ns->disk = disk; 3319 3320 __nvme_revalidate_disk(disk, id); 3321 3322 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) { 3323 ret = nvme_nvm_register(ns, disk_name, node); 3324 if (ret) { 3325 dev_warn(ctrl->device, "LightNVM init failure\n"); 3326 goto out_put_disk; 3327 } 3328 } 3329 3330 down_write(&ctrl->namespaces_rwsem); 3331 list_add_tail(&ns->list, &ctrl->namespaces); 3332 up_write(&ctrl->namespaces_rwsem); 3333 3334 nvme_get_ctrl(ctrl); 3335 3336 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups); 3337 3338 nvme_mpath_add_disk(ns, id); 3339 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name); 3340 kfree(id); 3341 3342 return 0; 3343 out_put_disk: 3344 put_disk(ns->disk); 3345 out_unlink_ns: 3346 mutex_lock(&ctrl->subsys->lock); 3347 list_del_rcu(&ns->siblings); 3348 mutex_unlock(&ctrl->subsys->lock); 3349 nvme_put_ns_head(ns->head); 3350 out_free_id: 3351 kfree(id); 3352 out_free_queue: 3353 blk_cleanup_queue(ns->queue); 3354 out_free_ns: 3355 kfree(ns); 3356 return ret; 3357 } 3358 3359 static void nvme_ns_remove(struct nvme_ns *ns) 3360 { 3361 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags)) 3362 return; 3363 3364 nvme_fault_inject_fini(&ns->fault_inject); 3365 3366 mutex_lock(&ns->ctrl->subsys->lock); 3367 list_del_rcu(&ns->siblings); 3368 mutex_unlock(&ns->ctrl->subsys->lock); 3369 synchronize_rcu(); /* guarantee not available in head->list */ 3370 nvme_mpath_clear_current_path(ns); 3371 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */ 3372 3373 if (ns->disk && ns->disk->flags & GENHD_FL_UP) { 3374 del_gendisk(ns->disk); 3375 blk_cleanup_queue(ns->queue); 3376 if (blk_get_integrity(ns->disk)) 3377 blk_integrity_unregister(ns->disk); 3378 } 3379 3380 down_write(&ns->ctrl->namespaces_rwsem); 3381 list_del_init(&ns->list); 3382 up_write(&ns->ctrl->namespaces_rwsem); 3383 3384 nvme_mpath_check_last_path(ns); 3385 nvme_put_ns(ns); 3386 } 3387 3388 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid) 3389 { 3390 struct nvme_ns *ns; 3391 3392 ns = nvme_find_get_ns(ctrl, nsid); 3393 if (ns) { 3394 if (ns->disk && revalidate_disk(ns->disk)) 3395 nvme_ns_remove(ns); 3396 nvme_put_ns(ns); 3397 } else 3398 nvme_alloc_ns(ctrl, nsid); 3399 } 3400 3401 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl, 3402 unsigned nsid) 3403 { 3404 struct nvme_ns *ns, *next; 3405 LIST_HEAD(rm_list); 3406 3407 down_write(&ctrl->namespaces_rwsem); 3408 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) { 3409 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags)) 3410 list_move_tail(&ns->list, &rm_list); 3411 } 3412 up_write(&ctrl->namespaces_rwsem); 3413 3414 list_for_each_entry_safe(ns, next, &rm_list, list) 3415 nvme_ns_remove(ns); 3416 3417 } 3418 3419 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn) 3420 { 3421 struct nvme_ns *ns; 3422 __le32 *ns_list; 3423 unsigned i, j, nsid, prev = 0; 3424 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024); 3425 int ret = 0; 3426 3427 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL); 3428 if (!ns_list) 3429 return -ENOMEM; 3430 3431 for (i = 0; i < num_lists; i++) { 3432 ret = nvme_identify_ns_list(ctrl, prev, ns_list); 3433 if (ret) 3434 goto free; 3435 3436 for (j = 0; j < min(nn, 1024U); j++) { 3437 nsid = le32_to_cpu(ns_list[j]); 3438 if (!nsid) 3439 goto out; 3440 3441 nvme_validate_ns(ctrl, nsid); 3442 3443 while (++prev < nsid) { 3444 ns = nvme_find_get_ns(ctrl, prev); 3445 if (ns) { 3446 nvme_ns_remove(ns); 3447 nvme_put_ns(ns); 3448 } 3449 } 3450 } 3451 nn -= j; 3452 } 3453 out: 3454 nvme_remove_invalid_namespaces(ctrl, prev); 3455 free: 3456 kfree(ns_list); 3457 return ret; 3458 } 3459 3460 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn) 3461 { 3462 unsigned i; 3463 3464 for (i = 1; i <= nn; i++) 3465 nvme_validate_ns(ctrl, i); 3466 3467 nvme_remove_invalid_namespaces(ctrl, nn); 3468 } 3469 3470 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl) 3471 { 3472 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32); 3473 __le32 *log; 3474 int error; 3475 3476 log = kzalloc(log_size, GFP_KERNEL); 3477 if (!log) 3478 return; 3479 3480 /* 3481 * We need to read the log to clear the AEN, but we don't want to rely 3482 * on it for the changed namespace information as userspace could have 3483 * raced with us in reading the log page, which could cause us to miss 3484 * updates. 3485 */ 3486 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log, 3487 log_size, 0); 3488 if (error) 3489 dev_warn(ctrl->device, 3490 "reading changed ns log failed: %d\n", error); 3491 3492 kfree(log); 3493 } 3494 3495 static void nvme_scan_work(struct work_struct *work) 3496 { 3497 struct nvme_ctrl *ctrl = 3498 container_of(work, struct nvme_ctrl, scan_work); 3499 struct nvme_id_ctrl *id; 3500 unsigned nn; 3501 3502 if (ctrl->state != NVME_CTRL_LIVE) 3503 return; 3504 3505 WARN_ON_ONCE(!ctrl->tagset); 3506 3507 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) { 3508 dev_info(ctrl->device, "rescanning namespaces.\n"); 3509 nvme_clear_changed_ns_log(ctrl); 3510 } 3511 3512 if (nvme_identify_ctrl(ctrl, &id)) 3513 return; 3514 3515 mutex_lock(&ctrl->scan_lock); 3516 nn = le32_to_cpu(id->nn); 3517 if (ctrl->vs >= NVME_VS(1, 1, 0) && 3518 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) { 3519 if (!nvme_scan_ns_list(ctrl, nn)) 3520 goto out_free_id; 3521 } 3522 nvme_scan_ns_sequential(ctrl, nn); 3523 out_free_id: 3524 mutex_unlock(&ctrl->scan_lock); 3525 kfree(id); 3526 down_write(&ctrl->namespaces_rwsem); 3527 list_sort(NULL, &ctrl->namespaces, ns_cmp); 3528 up_write(&ctrl->namespaces_rwsem); 3529 } 3530 3531 /* 3532 * This function iterates the namespace list unlocked to allow recovery from 3533 * controller failure. It is up to the caller to ensure the namespace list is 3534 * not modified by scan work while this function is executing. 3535 */ 3536 void nvme_remove_namespaces(struct nvme_ctrl *ctrl) 3537 { 3538 struct nvme_ns *ns, *next; 3539 LIST_HEAD(ns_list); 3540 3541 /* prevent racing with ns scanning */ 3542 flush_work(&ctrl->scan_work); 3543 3544 /* 3545 * The dead states indicates the controller was not gracefully 3546 * disconnected. In that case, we won't be able to flush any data while 3547 * removing the namespaces' disks; fail all the queues now to avoid 3548 * potentially having to clean up the failed sync later. 3549 */ 3550 if (ctrl->state == NVME_CTRL_DEAD) 3551 nvme_kill_queues(ctrl); 3552 3553 down_write(&ctrl->namespaces_rwsem); 3554 list_splice_init(&ctrl->namespaces, &ns_list); 3555 up_write(&ctrl->namespaces_rwsem); 3556 3557 list_for_each_entry_safe(ns, next, &ns_list, list) 3558 nvme_ns_remove(ns); 3559 } 3560 EXPORT_SYMBOL_GPL(nvme_remove_namespaces); 3561 3562 static void nvme_aen_uevent(struct nvme_ctrl *ctrl) 3563 { 3564 char *envp[2] = { NULL, NULL }; 3565 u32 aen_result = ctrl->aen_result; 3566 3567 ctrl->aen_result = 0; 3568 if (!aen_result) 3569 return; 3570 3571 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result); 3572 if (!envp[0]) 3573 return; 3574 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp); 3575 kfree(envp[0]); 3576 } 3577 3578 static void nvme_async_event_work(struct work_struct *work) 3579 { 3580 struct nvme_ctrl *ctrl = 3581 container_of(work, struct nvme_ctrl, async_event_work); 3582 3583 nvme_aen_uevent(ctrl); 3584 ctrl->ops->submit_async_event(ctrl); 3585 } 3586 3587 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl) 3588 { 3589 3590 u32 csts; 3591 3592 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) 3593 return false; 3594 3595 if (csts == ~0) 3596 return false; 3597 3598 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP)); 3599 } 3600 3601 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl) 3602 { 3603 struct nvme_fw_slot_info_log *log; 3604 3605 log = kmalloc(sizeof(*log), GFP_KERNEL); 3606 if (!log) 3607 return; 3608 3609 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log, 3610 sizeof(*log), 0)) 3611 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n"); 3612 kfree(log); 3613 } 3614 3615 static void nvme_fw_act_work(struct work_struct *work) 3616 { 3617 struct nvme_ctrl *ctrl = container_of(work, 3618 struct nvme_ctrl, fw_act_work); 3619 unsigned long fw_act_timeout; 3620 3621 if (ctrl->mtfa) 3622 fw_act_timeout = jiffies + 3623 msecs_to_jiffies(ctrl->mtfa * 100); 3624 else 3625 fw_act_timeout = jiffies + 3626 msecs_to_jiffies(admin_timeout * 1000); 3627 3628 nvme_stop_queues(ctrl); 3629 while (nvme_ctrl_pp_status(ctrl)) { 3630 if (time_after(jiffies, fw_act_timeout)) { 3631 dev_warn(ctrl->device, 3632 "Fw activation timeout, reset controller\n"); 3633 nvme_reset_ctrl(ctrl); 3634 break; 3635 } 3636 msleep(100); 3637 } 3638 3639 if (ctrl->state != NVME_CTRL_LIVE) 3640 return; 3641 3642 nvme_start_queues(ctrl); 3643 /* read FW slot information to clear the AER */ 3644 nvme_get_fw_slot_info(ctrl); 3645 } 3646 3647 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result) 3648 { 3649 u32 aer_notice_type = (result & 0xff00) >> 8; 3650 3651 trace_nvme_async_event(ctrl, aer_notice_type); 3652 3653 switch (aer_notice_type) { 3654 case NVME_AER_NOTICE_NS_CHANGED: 3655 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events); 3656 nvme_queue_scan(ctrl); 3657 break; 3658 case NVME_AER_NOTICE_FW_ACT_STARTING: 3659 queue_work(nvme_wq, &ctrl->fw_act_work); 3660 break; 3661 #ifdef CONFIG_NVME_MULTIPATH 3662 case NVME_AER_NOTICE_ANA: 3663 if (!ctrl->ana_log_buf) 3664 break; 3665 queue_work(nvme_wq, &ctrl->ana_work); 3666 break; 3667 #endif 3668 default: 3669 dev_warn(ctrl->device, "async event result %08x\n", result); 3670 } 3671 } 3672 3673 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status, 3674 volatile union nvme_result *res) 3675 { 3676 u32 result = le32_to_cpu(res->u32); 3677 u32 aer_type = result & 0x07; 3678 3679 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS) 3680 return; 3681 3682 switch (aer_type) { 3683 case NVME_AER_NOTICE: 3684 nvme_handle_aen_notice(ctrl, result); 3685 break; 3686 case NVME_AER_ERROR: 3687 case NVME_AER_SMART: 3688 case NVME_AER_CSS: 3689 case NVME_AER_VS: 3690 trace_nvme_async_event(ctrl, aer_type); 3691 ctrl->aen_result = result; 3692 break; 3693 default: 3694 break; 3695 } 3696 queue_work(nvme_wq, &ctrl->async_event_work); 3697 } 3698 EXPORT_SYMBOL_GPL(nvme_complete_async_event); 3699 3700 void nvme_stop_ctrl(struct nvme_ctrl *ctrl) 3701 { 3702 nvme_mpath_stop(ctrl); 3703 nvme_stop_keep_alive(ctrl); 3704 flush_work(&ctrl->async_event_work); 3705 cancel_work_sync(&ctrl->fw_act_work); 3706 } 3707 EXPORT_SYMBOL_GPL(nvme_stop_ctrl); 3708 3709 void nvme_start_ctrl(struct nvme_ctrl *ctrl) 3710 { 3711 if (ctrl->kato) 3712 nvme_start_keep_alive(ctrl); 3713 3714 if (ctrl->queue_count > 1) { 3715 nvme_queue_scan(ctrl); 3716 nvme_enable_aen(ctrl); 3717 queue_work(nvme_wq, &ctrl->async_event_work); 3718 nvme_start_queues(ctrl); 3719 } 3720 } 3721 EXPORT_SYMBOL_GPL(nvme_start_ctrl); 3722 3723 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl) 3724 { 3725 nvme_fault_inject_fini(&ctrl->fault_inject); 3726 dev_pm_qos_hide_latency_tolerance(ctrl->device); 3727 cdev_device_del(&ctrl->cdev, ctrl->device); 3728 } 3729 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl); 3730 3731 static void nvme_free_ctrl(struct device *dev) 3732 { 3733 struct nvme_ctrl *ctrl = 3734 container_of(dev, struct nvme_ctrl, ctrl_device); 3735 struct nvme_subsystem *subsys = ctrl->subsys; 3736 3737 ida_simple_remove(&nvme_instance_ida, ctrl->instance); 3738 kfree(ctrl->effects); 3739 nvme_mpath_uninit(ctrl); 3740 __free_page(ctrl->discard_page); 3741 3742 if (subsys) { 3743 mutex_lock(&nvme_subsystems_lock); 3744 list_del(&ctrl->subsys_entry); 3745 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device)); 3746 mutex_unlock(&nvme_subsystems_lock); 3747 } 3748 3749 ctrl->ops->free_ctrl(ctrl); 3750 3751 if (subsys) 3752 nvme_put_subsystem(subsys); 3753 } 3754 3755 /* 3756 * Initialize a NVMe controller structures. This needs to be called during 3757 * earliest initialization so that we have the initialized structured around 3758 * during probing. 3759 */ 3760 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, 3761 const struct nvme_ctrl_ops *ops, unsigned long quirks) 3762 { 3763 int ret; 3764 3765 ctrl->state = NVME_CTRL_NEW; 3766 spin_lock_init(&ctrl->lock); 3767 mutex_init(&ctrl->scan_lock); 3768 INIT_LIST_HEAD(&ctrl->namespaces); 3769 init_rwsem(&ctrl->namespaces_rwsem); 3770 ctrl->dev = dev; 3771 ctrl->ops = ops; 3772 ctrl->quirks = quirks; 3773 INIT_WORK(&ctrl->scan_work, nvme_scan_work); 3774 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work); 3775 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work); 3776 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work); 3777 3778 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work); 3779 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd)); 3780 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive; 3781 3782 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) > 3783 PAGE_SIZE); 3784 ctrl->discard_page = alloc_page(GFP_KERNEL); 3785 if (!ctrl->discard_page) { 3786 ret = -ENOMEM; 3787 goto out; 3788 } 3789 3790 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL); 3791 if (ret < 0) 3792 goto out; 3793 ctrl->instance = ret; 3794 3795 device_initialize(&ctrl->ctrl_device); 3796 ctrl->device = &ctrl->ctrl_device; 3797 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance); 3798 ctrl->device->class = nvme_class; 3799 ctrl->device->parent = ctrl->dev; 3800 ctrl->device->groups = nvme_dev_attr_groups; 3801 ctrl->device->release = nvme_free_ctrl; 3802 dev_set_drvdata(ctrl->device, ctrl); 3803 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance); 3804 if (ret) 3805 goto out_release_instance; 3806 3807 cdev_init(&ctrl->cdev, &nvme_dev_fops); 3808 ctrl->cdev.owner = ops->module; 3809 ret = cdev_device_add(&ctrl->cdev, ctrl->device); 3810 if (ret) 3811 goto out_free_name; 3812 3813 /* 3814 * Initialize latency tolerance controls. The sysfs files won't 3815 * be visible to userspace unless the device actually supports APST. 3816 */ 3817 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance; 3818 dev_pm_qos_update_user_latency_tolerance(ctrl->device, 3819 min(default_ps_max_latency_us, (unsigned long)S32_MAX)); 3820 3821 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device)); 3822 3823 return 0; 3824 out_free_name: 3825 kfree_const(ctrl->device->kobj.name); 3826 out_release_instance: 3827 ida_simple_remove(&nvme_instance_ida, ctrl->instance); 3828 out: 3829 if (ctrl->discard_page) 3830 __free_page(ctrl->discard_page); 3831 return ret; 3832 } 3833 EXPORT_SYMBOL_GPL(nvme_init_ctrl); 3834 3835 /** 3836 * nvme_kill_queues(): Ends all namespace queues 3837 * @ctrl: the dead controller that needs to end 3838 * 3839 * Call this function when the driver determines it is unable to get the 3840 * controller in a state capable of servicing IO. 3841 */ 3842 void nvme_kill_queues(struct nvme_ctrl *ctrl) 3843 { 3844 struct nvme_ns *ns; 3845 3846 down_read(&ctrl->namespaces_rwsem); 3847 3848 /* Forcibly unquiesce queues to avoid blocking dispatch */ 3849 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q)) 3850 blk_mq_unquiesce_queue(ctrl->admin_q); 3851 3852 list_for_each_entry(ns, &ctrl->namespaces, list) 3853 nvme_set_queue_dying(ns); 3854 3855 up_read(&ctrl->namespaces_rwsem); 3856 } 3857 EXPORT_SYMBOL_GPL(nvme_kill_queues); 3858 3859 void nvme_unfreeze(struct nvme_ctrl *ctrl) 3860 { 3861 struct nvme_ns *ns; 3862 3863 down_read(&ctrl->namespaces_rwsem); 3864 list_for_each_entry(ns, &ctrl->namespaces, list) 3865 blk_mq_unfreeze_queue(ns->queue); 3866 up_read(&ctrl->namespaces_rwsem); 3867 } 3868 EXPORT_SYMBOL_GPL(nvme_unfreeze); 3869 3870 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout) 3871 { 3872 struct nvme_ns *ns; 3873 3874 down_read(&ctrl->namespaces_rwsem); 3875 list_for_each_entry(ns, &ctrl->namespaces, list) { 3876 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout); 3877 if (timeout <= 0) 3878 break; 3879 } 3880 up_read(&ctrl->namespaces_rwsem); 3881 } 3882 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout); 3883 3884 void nvme_wait_freeze(struct nvme_ctrl *ctrl) 3885 { 3886 struct nvme_ns *ns; 3887 3888 down_read(&ctrl->namespaces_rwsem); 3889 list_for_each_entry(ns, &ctrl->namespaces, list) 3890 blk_mq_freeze_queue_wait(ns->queue); 3891 up_read(&ctrl->namespaces_rwsem); 3892 } 3893 EXPORT_SYMBOL_GPL(nvme_wait_freeze); 3894 3895 void nvme_start_freeze(struct nvme_ctrl *ctrl) 3896 { 3897 struct nvme_ns *ns; 3898 3899 down_read(&ctrl->namespaces_rwsem); 3900 list_for_each_entry(ns, &ctrl->namespaces, list) 3901 blk_freeze_queue_start(ns->queue); 3902 up_read(&ctrl->namespaces_rwsem); 3903 } 3904 EXPORT_SYMBOL_GPL(nvme_start_freeze); 3905 3906 void nvme_stop_queues(struct nvme_ctrl *ctrl) 3907 { 3908 struct nvme_ns *ns; 3909 3910 down_read(&ctrl->namespaces_rwsem); 3911 list_for_each_entry(ns, &ctrl->namespaces, list) 3912 blk_mq_quiesce_queue(ns->queue); 3913 up_read(&ctrl->namespaces_rwsem); 3914 } 3915 EXPORT_SYMBOL_GPL(nvme_stop_queues); 3916 3917 void nvme_start_queues(struct nvme_ctrl *ctrl) 3918 { 3919 struct nvme_ns *ns; 3920 3921 down_read(&ctrl->namespaces_rwsem); 3922 list_for_each_entry(ns, &ctrl->namespaces, list) 3923 blk_mq_unquiesce_queue(ns->queue); 3924 up_read(&ctrl->namespaces_rwsem); 3925 } 3926 EXPORT_SYMBOL_GPL(nvme_start_queues); 3927 3928 3929 void nvme_sync_queues(struct nvme_ctrl *ctrl) 3930 { 3931 struct nvme_ns *ns; 3932 3933 down_read(&ctrl->namespaces_rwsem); 3934 list_for_each_entry(ns, &ctrl->namespaces, list) 3935 blk_sync_queue(ns->queue); 3936 up_read(&ctrl->namespaces_rwsem); 3937 } 3938 EXPORT_SYMBOL_GPL(nvme_sync_queues); 3939 3940 /* 3941 * Check we didn't inadvertently grow the command structure sizes: 3942 */ 3943 static inline void _nvme_check_size(void) 3944 { 3945 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64); 3946 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); 3947 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64); 3948 BUILD_BUG_ON(sizeof(struct nvme_features) != 64); 3949 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64); 3950 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64); 3951 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64); 3952 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64); 3953 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64); 3954 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64); 3955 BUILD_BUG_ON(sizeof(struct nvme_command) != 64); 3956 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE); 3957 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE); 3958 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); 3959 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512); 3960 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64); 3961 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64); 3962 } 3963 3964 3965 static int __init nvme_core_init(void) 3966 { 3967 int result = -ENOMEM; 3968 3969 _nvme_check_size(); 3970 3971 nvme_wq = alloc_workqueue("nvme-wq", 3972 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0); 3973 if (!nvme_wq) 3974 goto out; 3975 3976 nvme_reset_wq = alloc_workqueue("nvme-reset-wq", 3977 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0); 3978 if (!nvme_reset_wq) 3979 goto destroy_wq; 3980 3981 nvme_delete_wq = alloc_workqueue("nvme-delete-wq", 3982 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0); 3983 if (!nvme_delete_wq) 3984 goto destroy_reset_wq; 3985 3986 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme"); 3987 if (result < 0) 3988 goto destroy_delete_wq; 3989 3990 nvme_class = class_create(THIS_MODULE, "nvme"); 3991 if (IS_ERR(nvme_class)) { 3992 result = PTR_ERR(nvme_class); 3993 goto unregister_chrdev; 3994 } 3995 3996 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem"); 3997 if (IS_ERR(nvme_subsys_class)) { 3998 result = PTR_ERR(nvme_subsys_class); 3999 goto destroy_class; 4000 } 4001 return 0; 4002 4003 destroy_class: 4004 class_destroy(nvme_class); 4005 unregister_chrdev: 4006 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS); 4007 destroy_delete_wq: 4008 destroy_workqueue(nvme_delete_wq); 4009 destroy_reset_wq: 4010 destroy_workqueue(nvme_reset_wq); 4011 destroy_wq: 4012 destroy_workqueue(nvme_wq); 4013 out: 4014 return result; 4015 } 4016 4017 static void __exit nvme_core_exit(void) 4018 { 4019 ida_destroy(&nvme_subsystems_ida); 4020 class_destroy(nvme_subsys_class); 4021 class_destroy(nvme_class); 4022 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS); 4023 destroy_workqueue(nvme_delete_wq); 4024 destroy_workqueue(nvme_reset_wq); 4025 destroy_workqueue(nvme_wq); 4026 } 4027 4028 MODULE_LICENSE("GPL"); 4029 MODULE_VERSION("1.0"); 4030 module_init(nvme_core_init); 4031 module_exit(nvme_core_exit); 4032