at v5.0 1481 lines 36 kB view raw
1/* 2 * Common code for the NVMe target. 3 * Copyright (c) 2015-2016 HGST, a Western Digital Company. 4 * 5 * This program is free software; you can redistribute it and/or modify it 6 * under the terms and conditions of the GNU General Public License, 7 * version 2, as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 12 * more details. 13 */ 14#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 15#include <linux/module.h> 16#include <linux/random.h> 17#include <linux/rculist.h> 18#include <linux/pci-p2pdma.h> 19 20#include "nvmet.h" 21 22struct workqueue_struct *buffered_io_wq; 23static const struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX]; 24static DEFINE_IDA(cntlid_ida); 25 26/* 27 * This read/write semaphore is used to synchronize access to configuration 28 * information on a target system that will result in discovery log page 29 * information change for at least one host. 30 * The full list of resources to protected by this semaphore is: 31 * 32 * - subsystems list 33 * - per-subsystem allowed hosts list 34 * - allow_any_host subsystem attribute 35 * - nvmet_genctr 36 * - the nvmet_transports array 37 * 38 * When updating any of those lists/structures write lock should be obtained, 39 * while when reading (popolating discovery log page or checking host-subsystem 40 * link) read lock is obtained to allow concurrent reads. 41 */ 42DECLARE_RWSEM(nvmet_config_sem); 43 44u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1]; 45u64 nvmet_ana_chgcnt; 46DECLARE_RWSEM(nvmet_ana_sem); 47 48inline u16 errno_to_nvme_status(struct nvmet_req *req, int errno) 49{ 50 u16 status; 51 52 switch (errno) { 53 case -ENOSPC: 54 req->error_loc = offsetof(struct nvme_rw_command, length); 55 status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR; 56 break; 57 case -EREMOTEIO: 58 req->error_loc = offsetof(struct nvme_rw_command, slba); 59 status = NVME_SC_LBA_RANGE | NVME_SC_DNR; 60 break; 61 case -EOPNOTSUPP: 62 req->error_loc = offsetof(struct nvme_common_command, opcode); 63 switch (req->cmd->common.opcode) { 64 case nvme_cmd_dsm: 65 case nvme_cmd_write_zeroes: 66 status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR; 67 break; 68 default: 69 status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR; 70 } 71 break; 72 case -ENODATA: 73 req->error_loc = offsetof(struct nvme_rw_command, nsid); 74 status = NVME_SC_ACCESS_DENIED; 75 break; 76 case -EIO: 77 /* FALLTHRU */ 78 default: 79 req->error_loc = offsetof(struct nvme_common_command, opcode); 80 status = NVME_SC_INTERNAL | NVME_SC_DNR; 81 } 82 83 return status; 84} 85 86static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port, 87 const char *subsysnqn); 88 89u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf, 90 size_t len) 91{ 92 if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len) { 93 req->error_loc = offsetof(struct nvme_common_command, dptr); 94 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR; 95 } 96 return 0; 97} 98 99u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, size_t len) 100{ 101 if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len) { 102 req->error_loc = offsetof(struct nvme_common_command, dptr); 103 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR; 104 } 105 return 0; 106} 107 108u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len) 109{ 110 if (sg_zero_buffer(req->sg, req->sg_cnt, len, off) != len) { 111 req->error_loc = offsetof(struct nvme_common_command, dptr); 112 return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR; 113 } 114 return 0; 115} 116 117static unsigned int nvmet_max_nsid(struct nvmet_subsys *subsys) 118{ 119 struct nvmet_ns *ns; 120 121 if (list_empty(&subsys->namespaces)) 122 return 0; 123 124 ns = list_last_entry(&subsys->namespaces, struct nvmet_ns, dev_link); 125 return ns->nsid; 126} 127 128static u32 nvmet_async_event_result(struct nvmet_async_event *aen) 129{ 130 return aen->event_type | (aen->event_info << 8) | (aen->log_page << 16); 131} 132 133static void nvmet_async_events_free(struct nvmet_ctrl *ctrl) 134{ 135 struct nvmet_req *req; 136 137 while (1) { 138 mutex_lock(&ctrl->lock); 139 if (!ctrl->nr_async_event_cmds) { 140 mutex_unlock(&ctrl->lock); 141 return; 142 } 143 144 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds]; 145 mutex_unlock(&ctrl->lock); 146 nvmet_req_complete(req, NVME_SC_INTERNAL | NVME_SC_DNR); 147 } 148} 149 150static void nvmet_async_event_work(struct work_struct *work) 151{ 152 struct nvmet_ctrl *ctrl = 153 container_of(work, struct nvmet_ctrl, async_event_work); 154 struct nvmet_async_event *aen; 155 struct nvmet_req *req; 156 157 while (1) { 158 mutex_lock(&ctrl->lock); 159 aen = list_first_entry_or_null(&ctrl->async_events, 160 struct nvmet_async_event, entry); 161 if (!aen || !ctrl->nr_async_event_cmds) { 162 mutex_unlock(&ctrl->lock); 163 return; 164 } 165 166 req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds]; 167 nvmet_set_result(req, nvmet_async_event_result(aen)); 168 169 list_del(&aen->entry); 170 kfree(aen); 171 172 mutex_unlock(&ctrl->lock); 173 nvmet_req_complete(req, 0); 174 } 175} 176 177void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type, 178 u8 event_info, u8 log_page) 179{ 180 struct nvmet_async_event *aen; 181 182 aen = kmalloc(sizeof(*aen), GFP_KERNEL); 183 if (!aen) 184 return; 185 186 aen->event_type = event_type; 187 aen->event_info = event_info; 188 aen->log_page = log_page; 189 190 mutex_lock(&ctrl->lock); 191 list_add_tail(&aen->entry, &ctrl->async_events); 192 mutex_unlock(&ctrl->lock); 193 194 schedule_work(&ctrl->async_event_work); 195} 196 197static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid) 198{ 199 u32 i; 200 201 mutex_lock(&ctrl->lock); 202 if (ctrl->nr_changed_ns > NVME_MAX_CHANGED_NAMESPACES) 203 goto out_unlock; 204 205 for (i = 0; i < ctrl->nr_changed_ns; i++) { 206 if (ctrl->changed_ns_list[i] == nsid) 207 goto out_unlock; 208 } 209 210 if (ctrl->nr_changed_ns == NVME_MAX_CHANGED_NAMESPACES) { 211 ctrl->changed_ns_list[0] = cpu_to_le32(0xffffffff); 212 ctrl->nr_changed_ns = U32_MAX; 213 goto out_unlock; 214 } 215 216 ctrl->changed_ns_list[ctrl->nr_changed_ns++] = nsid; 217out_unlock: 218 mutex_unlock(&ctrl->lock); 219} 220 221void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid) 222{ 223 struct nvmet_ctrl *ctrl; 224 225 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) { 226 nvmet_add_to_changed_ns_log(ctrl, cpu_to_le32(nsid)); 227 if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_NS_ATTR)) 228 continue; 229 nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE, 230 NVME_AER_NOTICE_NS_CHANGED, 231 NVME_LOG_CHANGED_NS); 232 } 233} 234 235void nvmet_send_ana_event(struct nvmet_subsys *subsys, 236 struct nvmet_port *port) 237{ 238 struct nvmet_ctrl *ctrl; 239 240 mutex_lock(&subsys->lock); 241 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) { 242 if (port && ctrl->port != port) 243 continue; 244 if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_ANA_CHANGE)) 245 continue; 246 nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE, 247 NVME_AER_NOTICE_ANA, NVME_LOG_ANA); 248 } 249 mutex_unlock(&subsys->lock); 250} 251 252void nvmet_port_send_ana_event(struct nvmet_port *port) 253{ 254 struct nvmet_subsys_link *p; 255 256 down_read(&nvmet_config_sem); 257 list_for_each_entry(p, &port->subsystems, entry) 258 nvmet_send_ana_event(p->subsys, port); 259 up_read(&nvmet_config_sem); 260} 261 262int nvmet_register_transport(const struct nvmet_fabrics_ops *ops) 263{ 264 int ret = 0; 265 266 down_write(&nvmet_config_sem); 267 if (nvmet_transports[ops->type]) 268 ret = -EINVAL; 269 else 270 nvmet_transports[ops->type] = ops; 271 up_write(&nvmet_config_sem); 272 273 return ret; 274} 275EXPORT_SYMBOL_GPL(nvmet_register_transport); 276 277void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops) 278{ 279 down_write(&nvmet_config_sem); 280 nvmet_transports[ops->type] = NULL; 281 up_write(&nvmet_config_sem); 282} 283EXPORT_SYMBOL_GPL(nvmet_unregister_transport); 284 285int nvmet_enable_port(struct nvmet_port *port) 286{ 287 const struct nvmet_fabrics_ops *ops; 288 int ret; 289 290 lockdep_assert_held(&nvmet_config_sem); 291 292 ops = nvmet_transports[port->disc_addr.trtype]; 293 if (!ops) { 294 up_write(&nvmet_config_sem); 295 request_module("nvmet-transport-%d", port->disc_addr.trtype); 296 down_write(&nvmet_config_sem); 297 ops = nvmet_transports[port->disc_addr.trtype]; 298 if (!ops) { 299 pr_err("transport type %d not supported\n", 300 port->disc_addr.trtype); 301 return -EINVAL; 302 } 303 } 304 305 if (!try_module_get(ops->owner)) 306 return -EINVAL; 307 308 ret = ops->add_port(port); 309 if (ret) { 310 module_put(ops->owner); 311 return ret; 312 } 313 314 /* If the transport didn't set inline_data_size, then disable it. */ 315 if (port->inline_data_size < 0) 316 port->inline_data_size = 0; 317 318 port->enabled = true; 319 return 0; 320} 321 322void nvmet_disable_port(struct nvmet_port *port) 323{ 324 const struct nvmet_fabrics_ops *ops; 325 326 lockdep_assert_held(&nvmet_config_sem); 327 328 port->enabled = false; 329 330 ops = nvmet_transports[port->disc_addr.trtype]; 331 ops->remove_port(port); 332 module_put(ops->owner); 333} 334 335static void nvmet_keep_alive_timer(struct work_struct *work) 336{ 337 struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work), 338 struct nvmet_ctrl, ka_work); 339 bool cmd_seen = ctrl->cmd_seen; 340 341 ctrl->cmd_seen = false; 342 if (cmd_seen) { 343 pr_debug("ctrl %d reschedule traffic based keep-alive timer\n", 344 ctrl->cntlid); 345 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); 346 return; 347 } 348 349 pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n", 350 ctrl->cntlid, ctrl->kato); 351 352 nvmet_ctrl_fatal_error(ctrl); 353} 354 355static void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl) 356{ 357 pr_debug("ctrl %d start keep-alive timer for %d secs\n", 358 ctrl->cntlid, ctrl->kato); 359 360 INIT_DELAYED_WORK(&ctrl->ka_work, nvmet_keep_alive_timer); 361 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); 362} 363 364static void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl) 365{ 366 pr_debug("ctrl %d stop keep-alive\n", ctrl->cntlid); 367 368 cancel_delayed_work_sync(&ctrl->ka_work); 369} 370 371static struct nvmet_ns *__nvmet_find_namespace(struct nvmet_ctrl *ctrl, 372 __le32 nsid) 373{ 374 struct nvmet_ns *ns; 375 376 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) { 377 if (ns->nsid == le32_to_cpu(nsid)) 378 return ns; 379 } 380 381 return NULL; 382} 383 384struct nvmet_ns *nvmet_find_namespace(struct nvmet_ctrl *ctrl, __le32 nsid) 385{ 386 struct nvmet_ns *ns; 387 388 rcu_read_lock(); 389 ns = __nvmet_find_namespace(ctrl, nsid); 390 if (ns) 391 percpu_ref_get(&ns->ref); 392 rcu_read_unlock(); 393 394 return ns; 395} 396 397static void nvmet_destroy_namespace(struct percpu_ref *ref) 398{ 399 struct nvmet_ns *ns = container_of(ref, struct nvmet_ns, ref); 400 401 complete(&ns->disable_done); 402} 403 404void nvmet_put_namespace(struct nvmet_ns *ns) 405{ 406 percpu_ref_put(&ns->ref); 407} 408 409static void nvmet_ns_dev_disable(struct nvmet_ns *ns) 410{ 411 nvmet_bdev_ns_disable(ns); 412 nvmet_file_ns_disable(ns); 413} 414 415static int nvmet_p2pmem_ns_enable(struct nvmet_ns *ns) 416{ 417 int ret; 418 struct pci_dev *p2p_dev; 419 420 if (!ns->use_p2pmem) 421 return 0; 422 423 if (!ns->bdev) { 424 pr_err("peer-to-peer DMA is not supported by non-block device namespaces\n"); 425 return -EINVAL; 426 } 427 428 if (!blk_queue_pci_p2pdma(ns->bdev->bd_queue)) { 429 pr_err("peer-to-peer DMA is not supported by the driver of %s\n", 430 ns->device_path); 431 return -EINVAL; 432 } 433 434 if (ns->p2p_dev) { 435 ret = pci_p2pdma_distance(ns->p2p_dev, nvmet_ns_dev(ns), true); 436 if (ret < 0) 437 return -EINVAL; 438 } else { 439 /* 440 * Right now we just check that there is p2pmem available so 441 * we can report an error to the user right away if there 442 * is not. We'll find the actual device to use once we 443 * setup the controller when the port's device is available. 444 */ 445 446 p2p_dev = pci_p2pmem_find(nvmet_ns_dev(ns)); 447 if (!p2p_dev) { 448 pr_err("no peer-to-peer memory is available for %s\n", 449 ns->device_path); 450 return -EINVAL; 451 } 452 453 pci_dev_put(p2p_dev); 454 } 455 456 return 0; 457} 458 459/* 460 * Note: ctrl->subsys->lock should be held when calling this function 461 */ 462static void nvmet_p2pmem_ns_add_p2p(struct nvmet_ctrl *ctrl, 463 struct nvmet_ns *ns) 464{ 465 struct device *clients[2]; 466 struct pci_dev *p2p_dev; 467 int ret; 468 469 if (!ctrl->p2p_client || !ns->use_p2pmem) 470 return; 471 472 if (ns->p2p_dev) { 473 ret = pci_p2pdma_distance(ns->p2p_dev, ctrl->p2p_client, true); 474 if (ret < 0) 475 return; 476 477 p2p_dev = pci_dev_get(ns->p2p_dev); 478 } else { 479 clients[0] = ctrl->p2p_client; 480 clients[1] = nvmet_ns_dev(ns); 481 482 p2p_dev = pci_p2pmem_find_many(clients, ARRAY_SIZE(clients)); 483 if (!p2p_dev) { 484 pr_err("no peer-to-peer memory is available that's supported by %s and %s\n", 485 dev_name(ctrl->p2p_client), ns->device_path); 486 return; 487 } 488 } 489 490 ret = radix_tree_insert(&ctrl->p2p_ns_map, ns->nsid, p2p_dev); 491 if (ret < 0) 492 pci_dev_put(p2p_dev); 493 494 pr_info("using p2pmem on %s for nsid %d\n", pci_name(p2p_dev), 495 ns->nsid); 496} 497 498int nvmet_ns_enable(struct nvmet_ns *ns) 499{ 500 struct nvmet_subsys *subsys = ns->subsys; 501 struct nvmet_ctrl *ctrl; 502 int ret; 503 504 mutex_lock(&subsys->lock); 505 ret = -EMFILE; 506 if (subsys->nr_namespaces == NVMET_MAX_NAMESPACES) 507 goto out_unlock; 508 ret = 0; 509 if (ns->enabled) 510 goto out_unlock; 511 512 ret = nvmet_bdev_ns_enable(ns); 513 if (ret == -ENOTBLK) 514 ret = nvmet_file_ns_enable(ns); 515 if (ret) 516 goto out_unlock; 517 518 ret = nvmet_p2pmem_ns_enable(ns); 519 if (ret) 520 goto out_unlock; 521 522 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) 523 nvmet_p2pmem_ns_add_p2p(ctrl, ns); 524 525 ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace, 526 0, GFP_KERNEL); 527 if (ret) 528 goto out_dev_put; 529 530 if (ns->nsid > subsys->max_nsid) 531 subsys->max_nsid = ns->nsid; 532 533 /* 534 * The namespaces list needs to be sorted to simplify the implementation 535 * of the Identify Namepace List subcommand. 536 */ 537 if (list_empty(&subsys->namespaces)) { 538 list_add_tail_rcu(&ns->dev_link, &subsys->namespaces); 539 } else { 540 struct nvmet_ns *old; 541 542 list_for_each_entry_rcu(old, &subsys->namespaces, dev_link) { 543 BUG_ON(ns->nsid == old->nsid); 544 if (ns->nsid < old->nsid) 545 break; 546 } 547 548 list_add_tail_rcu(&ns->dev_link, &old->dev_link); 549 } 550 subsys->nr_namespaces++; 551 552 nvmet_ns_changed(subsys, ns->nsid); 553 ns->enabled = true; 554 ret = 0; 555out_unlock: 556 mutex_unlock(&subsys->lock); 557 return ret; 558out_dev_put: 559 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) 560 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid)); 561 562 nvmet_ns_dev_disable(ns); 563 goto out_unlock; 564} 565 566void nvmet_ns_disable(struct nvmet_ns *ns) 567{ 568 struct nvmet_subsys *subsys = ns->subsys; 569 struct nvmet_ctrl *ctrl; 570 571 mutex_lock(&subsys->lock); 572 if (!ns->enabled) 573 goto out_unlock; 574 575 ns->enabled = false; 576 list_del_rcu(&ns->dev_link); 577 if (ns->nsid == subsys->max_nsid) 578 subsys->max_nsid = nvmet_max_nsid(subsys); 579 580 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) 581 pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid)); 582 583 mutex_unlock(&subsys->lock); 584 585 /* 586 * Now that we removed the namespaces from the lookup list, we 587 * can kill the per_cpu ref and wait for any remaining references 588 * to be dropped, as well as a RCU grace period for anyone only 589 * using the namepace under rcu_read_lock(). Note that we can't 590 * use call_rcu here as we need to ensure the namespaces have 591 * been fully destroyed before unloading the module. 592 */ 593 percpu_ref_kill(&ns->ref); 594 synchronize_rcu(); 595 wait_for_completion(&ns->disable_done); 596 percpu_ref_exit(&ns->ref); 597 598 mutex_lock(&subsys->lock); 599 600 subsys->nr_namespaces--; 601 nvmet_ns_changed(subsys, ns->nsid); 602 nvmet_ns_dev_disable(ns); 603out_unlock: 604 mutex_unlock(&subsys->lock); 605} 606 607void nvmet_ns_free(struct nvmet_ns *ns) 608{ 609 nvmet_ns_disable(ns); 610 611 down_write(&nvmet_ana_sem); 612 nvmet_ana_group_enabled[ns->anagrpid]--; 613 up_write(&nvmet_ana_sem); 614 615 kfree(ns->device_path); 616 kfree(ns); 617} 618 619struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid) 620{ 621 struct nvmet_ns *ns; 622 623 ns = kzalloc(sizeof(*ns), GFP_KERNEL); 624 if (!ns) 625 return NULL; 626 627 INIT_LIST_HEAD(&ns->dev_link); 628 init_completion(&ns->disable_done); 629 630 ns->nsid = nsid; 631 ns->subsys = subsys; 632 633 down_write(&nvmet_ana_sem); 634 ns->anagrpid = NVMET_DEFAULT_ANA_GRPID; 635 nvmet_ana_group_enabled[ns->anagrpid]++; 636 up_write(&nvmet_ana_sem); 637 638 uuid_gen(&ns->uuid); 639 ns->buffered_io = false; 640 641 return ns; 642} 643 644static void nvmet_update_sq_head(struct nvmet_req *req) 645{ 646 if (req->sq->size) { 647 u32 old_sqhd, new_sqhd; 648 649 do { 650 old_sqhd = req->sq->sqhd; 651 new_sqhd = (old_sqhd + 1) % req->sq->size; 652 } while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) != 653 old_sqhd); 654 } 655 req->rsp->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF); 656} 657 658static void nvmet_set_error(struct nvmet_req *req, u16 status) 659{ 660 struct nvmet_ctrl *ctrl = req->sq->ctrl; 661 struct nvme_error_slot *new_error_slot; 662 unsigned long flags; 663 664 req->rsp->status = cpu_to_le16(status << 1); 665 666 if (!ctrl || req->error_loc == NVMET_NO_ERROR_LOC) 667 return; 668 669 spin_lock_irqsave(&ctrl->error_lock, flags); 670 ctrl->err_counter++; 671 new_error_slot = 672 &ctrl->slots[ctrl->err_counter % NVMET_ERROR_LOG_SLOTS]; 673 674 new_error_slot->error_count = cpu_to_le64(ctrl->err_counter); 675 new_error_slot->sqid = cpu_to_le16(req->sq->qid); 676 new_error_slot->cmdid = cpu_to_le16(req->cmd->common.command_id); 677 new_error_slot->status_field = cpu_to_le16(status << 1); 678 new_error_slot->param_error_location = cpu_to_le16(req->error_loc); 679 new_error_slot->lba = cpu_to_le64(req->error_slba); 680 new_error_slot->nsid = req->cmd->common.nsid; 681 spin_unlock_irqrestore(&ctrl->error_lock, flags); 682 683 /* set the more bit for this request */ 684 req->rsp->status |= cpu_to_le16(1 << 14); 685} 686 687static void __nvmet_req_complete(struct nvmet_req *req, u16 status) 688{ 689 if (!req->sq->sqhd_disabled) 690 nvmet_update_sq_head(req); 691 req->rsp->sq_id = cpu_to_le16(req->sq->qid); 692 req->rsp->command_id = req->cmd->common.command_id; 693 694 if (unlikely(status)) 695 nvmet_set_error(req, status); 696 if (req->ns) 697 nvmet_put_namespace(req->ns); 698 req->ops->queue_response(req); 699} 700 701void nvmet_req_complete(struct nvmet_req *req, u16 status) 702{ 703 __nvmet_req_complete(req, status); 704 percpu_ref_put(&req->sq->ref); 705} 706EXPORT_SYMBOL_GPL(nvmet_req_complete); 707 708void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq, 709 u16 qid, u16 size) 710{ 711 cq->qid = qid; 712 cq->size = size; 713 714 ctrl->cqs[qid] = cq; 715} 716 717void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq, 718 u16 qid, u16 size) 719{ 720 sq->sqhd = 0; 721 sq->qid = qid; 722 sq->size = size; 723 724 ctrl->sqs[qid] = sq; 725} 726 727static void nvmet_confirm_sq(struct percpu_ref *ref) 728{ 729 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref); 730 731 complete(&sq->confirm_done); 732} 733 734void nvmet_sq_destroy(struct nvmet_sq *sq) 735{ 736 /* 737 * If this is the admin queue, complete all AERs so that our 738 * queue doesn't have outstanding requests on it. 739 */ 740 if (sq->ctrl && sq->ctrl->sqs && sq->ctrl->sqs[0] == sq) 741 nvmet_async_events_free(sq->ctrl); 742 percpu_ref_kill_and_confirm(&sq->ref, nvmet_confirm_sq); 743 wait_for_completion(&sq->confirm_done); 744 wait_for_completion(&sq->free_done); 745 percpu_ref_exit(&sq->ref); 746 747 if (sq->ctrl) { 748 nvmet_ctrl_put(sq->ctrl); 749 sq->ctrl = NULL; /* allows reusing the queue later */ 750 } 751} 752EXPORT_SYMBOL_GPL(nvmet_sq_destroy); 753 754static void nvmet_sq_free(struct percpu_ref *ref) 755{ 756 struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref); 757 758 complete(&sq->free_done); 759} 760 761int nvmet_sq_init(struct nvmet_sq *sq) 762{ 763 int ret; 764 765 ret = percpu_ref_init(&sq->ref, nvmet_sq_free, 0, GFP_KERNEL); 766 if (ret) { 767 pr_err("percpu_ref init failed!\n"); 768 return ret; 769 } 770 init_completion(&sq->free_done); 771 init_completion(&sq->confirm_done); 772 773 return 0; 774} 775EXPORT_SYMBOL_GPL(nvmet_sq_init); 776 777static inline u16 nvmet_check_ana_state(struct nvmet_port *port, 778 struct nvmet_ns *ns) 779{ 780 enum nvme_ana_state state = port->ana_state[ns->anagrpid]; 781 782 if (unlikely(state == NVME_ANA_INACCESSIBLE)) 783 return NVME_SC_ANA_INACCESSIBLE; 784 if (unlikely(state == NVME_ANA_PERSISTENT_LOSS)) 785 return NVME_SC_ANA_PERSISTENT_LOSS; 786 if (unlikely(state == NVME_ANA_CHANGE)) 787 return NVME_SC_ANA_TRANSITION; 788 return 0; 789} 790 791static inline u16 nvmet_io_cmd_check_access(struct nvmet_req *req) 792{ 793 if (unlikely(req->ns->readonly)) { 794 switch (req->cmd->common.opcode) { 795 case nvme_cmd_read: 796 case nvme_cmd_flush: 797 break; 798 default: 799 return NVME_SC_NS_WRITE_PROTECTED; 800 } 801 } 802 803 return 0; 804} 805 806static u16 nvmet_parse_io_cmd(struct nvmet_req *req) 807{ 808 struct nvme_command *cmd = req->cmd; 809 u16 ret; 810 811 ret = nvmet_check_ctrl_status(req, cmd); 812 if (unlikely(ret)) 813 return ret; 814 815 req->ns = nvmet_find_namespace(req->sq->ctrl, cmd->rw.nsid); 816 if (unlikely(!req->ns)) { 817 req->error_loc = offsetof(struct nvme_common_command, nsid); 818 return NVME_SC_INVALID_NS | NVME_SC_DNR; 819 } 820 ret = nvmet_check_ana_state(req->port, req->ns); 821 if (unlikely(ret)) { 822 req->error_loc = offsetof(struct nvme_common_command, nsid); 823 return ret; 824 } 825 ret = nvmet_io_cmd_check_access(req); 826 if (unlikely(ret)) { 827 req->error_loc = offsetof(struct nvme_common_command, nsid); 828 return ret; 829 } 830 831 if (req->ns->file) 832 return nvmet_file_parse_io_cmd(req); 833 else 834 return nvmet_bdev_parse_io_cmd(req); 835} 836 837bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq, 838 struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops) 839{ 840 u8 flags = req->cmd->common.flags; 841 u16 status; 842 843 req->cq = cq; 844 req->sq = sq; 845 req->ops = ops; 846 req->sg = NULL; 847 req->sg_cnt = 0; 848 req->transfer_len = 0; 849 req->rsp->status = 0; 850 req->rsp->sq_head = 0; 851 req->ns = NULL; 852 req->error_loc = NVMET_NO_ERROR_LOC; 853 req->error_slba = 0; 854 855 /* no support for fused commands yet */ 856 if (unlikely(flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND))) { 857 req->error_loc = offsetof(struct nvme_common_command, flags); 858 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 859 goto fail; 860 } 861 862 /* 863 * For fabrics, PSDT field shall describe metadata pointer (MPTR) that 864 * contains an address of a single contiguous physical buffer that is 865 * byte aligned. 866 */ 867 if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF)) { 868 req->error_loc = offsetof(struct nvme_common_command, flags); 869 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 870 goto fail; 871 } 872 873 if (unlikely(!req->sq->ctrl)) 874 /* will return an error for any Non-connect command: */ 875 status = nvmet_parse_connect_cmd(req); 876 else if (likely(req->sq->qid != 0)) 877 status = nvmet_parse_io_cmd(req); 878 else if (req->cmd->common.opcode == nvme_fabrics_command) 879 status = nvmet_parse_fabrics_cmd(req); 880 else if (req->sq->ctrl->subsys->type == NVME_NQN_DISC) 881 status = nvmet_parse_discovery_cmd(req); 882 else 883 status = nvmet_parse_admin_cmd(req); 884 885 if (status) 886 goto fail; 887 888 if (unlikely(!percpu_ref_tryget_live(&sq->ref))) { 889 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 890 goto fail; 891 } 892 893 if (sq->ctrl) 894 sq->ctrl->cmd_seen = true; 895 896 return true; 897 898fail: 899 __nvmet_req_complete(req, status); 900 return false; 901} 902EXPORT_SYMBOL_GPL(nvmet_req_init); 903 904void nvmet_req_uninit(struct nvmet_req *req) 905{ 906 percpu_ref_put(&req->sq->ref); 907 if (req->ns) 908 nvmet_put_namespace(req->ns); 909} 910EXPORT_SYMBOL_GPL(nvmet_req_uninit); 911 912void nvmet_req_execute(struct nvmet_req *req) 913{ 914 if (unlikely(req->data_len != req->transfer_len)) { 915 req->error_loc = offsetof(struct nvme_common_command, dptr); 916 nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR); 917 } else 918 req->execute(req); 919} 920EXPORT_SYMBOL_GPL(nvmet_req_execute); 921 922int nvmet_req_alloc_sgl(struct nvmet_req *req) 923{ 924 struct pci_dev *p2p_dev = NULL; 925 926 if (IS_ENABLED(CONFIG_PCI_P2PDMA)) { 927 if (req->sq->ctrl && req->ns) 928 p2p_dev = radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, 929 req->ns->nsid); 930 931 req->p2p_dev = NULL; 932 if (req->sq->qid && p2p_dev) { 933 req->sg = pci_p2pmem_alloc_sgl(p2p_dev, &req->sg_cnt, 934 req->transfer_len); 935 if (req->sg) { 936 req->p2p_dev = p2p_dev; 937 return 0; 938 } 939 } 940 941 /* 942 * If no P2P memory was available we fallback to using 943 * regular memory 944 */ 945 } 946 947 req->sg = sgl_alloc(req->transfer_len, GFP_KERNEL, &req->sg_cnt); 948 if (!req->sg) 949 return -ENOMEM; 950 951 return 0; 952} 953EXPORT_SYMBOL_GPL(nvmet_req_alloc_sgl); 954 955void nvmet_req_free_sgl(struct nvmet_req *req) 956{ 957 if (req->p2p_dev) 958 pci_p2pmem_free_sgl(req->p2p_dev, req->sg); 959 else 960 sgl_free(req->sg); 961 962 req->sg = NULL; 963 req->sg_cnt = 0; 964} 965EXPORT_SYMBOL_GPL(nvmet_req_free_sgl); 966 967static inline bool nvmet_cc_en(u32 cc) 968{ 969 return (cc >> NVME_CC_EN_SHIFT) & 0x1; 970} 971 972static inline u8 nvmet_cc_css(u32 cc) 973{ 974 return (cc >> NVME_CC_CSS_SHIFT) & 0x7; 975} 976 977static inline u8 nvmet_cc_mps(u32 cc) 978{ 979 return (cc >> NVME_CC_MPS_SHIFT) & 0xf; 980} 981 982static inline u8 nvmet_cc_ams(u32 cc) 983{ 984 return (cc >> NVME_CC_AMS_SHIFT) & 0x7; 985} 986 987static inline u8 nvmet_cc_shn(u32 cc) 988{ 989 return (cc >> NVME_CC_SHN_SHIFT) & 0x3; 990} 991 992static inline u8 nvmet_cc_iosqes(u32 cc) 993{ 994 return (cc >> NVME_CC_IOSQES_SHIFT) & 0xf; 995} 996 997static inline u8 nvmet_cc_iocqes(u32 cc) 998{ 999 return (cc >> NVME_CC_IOCQES_SHIFT) & 0xf; 1000} 1001 1002static void nvmet_start_ctrl(struct nvmet_ctrl *ctrl) 1003{ 1004 lockdep_assert_held(&ctrl->lock); 1005 1006 if (nvmet_cc_iosqes(ctrl->cc) != NVME_NVM_IOSQES || 1007 nvmet_cc_iocqes(ctrl->cc) != NVME_NVM_IOCQES || 1008 nvmet_cc_mps(ctrl->cc) != 0 || 1009 nvmet_cc_ams(ctrl->cc) != 0 || 1010 nvmet_cc_css(ctrl->cc) != 0) { 1011 ctrl->csts = NVME_CSTS_CFS; 1012 return; 1013 } 1014 1015 ctrl->csts = NVME_CSTS_RDY; 1016 1017 /* 1018 * Controllers that are not yet enabled should not really enforce the 1019 * keep alive timeout, but we still want to track a timeout and cleanup 1020 * in case a host died before it enabled the controller. Hence, simply 1021 * reset the keep alive timer when the controller is enabled. 1022 */ 1023 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ); 1024} 1025 1026static void nvmet_clear_ctrl(struct nvmet_ctrl *ctrl) 1027{ 1028 lockdep_assert_held(&ctrl->lock); 1029 1030 /* XXX: tear down queues? */ 1031 ctrl->csts &= ~NVME_CSTS_RDY; 1032 ctrl->cc = 0; 1033} 1034 1035void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new) 1036{ 1037 u32 old; 1038 1039 mutex_lock(&ctrl->lock); 1040 old = ctrl->cc; 1041 ctrl->cc = new; 1042 1043 if (nvmet_cc_en(new) && !nvmet_cc_en(old)) 1044 nvmet_start_ctrl(ctrl); 1045 if (!nvmet_cc_en(new) && nvmet_cc_en(old)) 1046 nvmet_clear_ctrl(ctrl); 1047 if (nvmet_cc_shn(new) && !nvmet_cc_shn(old)) { 1048 nvmet_clear_ctrl(ctrl); 1049 ctrl->csts |= NVME_CSTS_SHST_CMPLT; 1050 } 1051 if (!nvmet_cc_shn(new) && nvmet_cc_shn(old)) 1052 ctrl->csts &= ~NVME_CSTS_SHST_CMPLT; 1053 mutex_unlock(&ctrl->lock); 1054} 1055 1056static void nvmet_init_cap(struct nvmet_ctrl *ctrl) 1057{ 1058 /* command sets supported: NVMe command set: */ 1059 ctrl->cap = (1ULL << 37); 1060 /* CC.EN timeout in 500msec units: */ 1061 ctrl->cap |= (15ULL << 24); 1062 /* maximum queue entries supported: */ 1063 ctrl->cap |= NVMET_QUEUE_SIZE - 1; 1064} 1065 1066u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid, 1067 struct nvmet_req *req, struct nvmet_ctrl **ret) 1068{ 1069 struct nvmet_subsys *subsys; 1070 struct nvmet_ctrl *ctrl; 1071 u16 status = 0; 1072 1073 subsys = nvmet_find_get_subsys(req->port, subsysnqn); 1074 if (!subsys) { 1075 pr_warn("connect request for invalid subsystem %s!\n", 1076 subsysnqn); 1077 req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn); 1078 return NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR; 1079 } 1080 1081 mutex_lock(&subsys->lock); 1082 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) { 1083 if (ctrl->cntlid == cntlid) { 1084 if (strncmp(hostnqn, ctrl->hostnqn, NVMF_NQN_SIZE)) { 1085 pr_warn("hostnqn mismatch.\n"); 1086 continue; 1087 } 1088 if (!kref_get_unless_zero(&ctrl->ref)) 1089 continue; 1090 1091 *ret = ctrl; 1092 goto out; 1093 } 1094 } 1095 1096 pr_warn("could not find controller %d for subsys %s / host %s\n", 1097 cntlid, subsysnqn, hostnqn); 1098 req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid); 1099 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR; 1100 1101out: 1102 mutex_unlock(&subsys->lock); 1103 nvmet_subsys_put(subsys); 1104 return status; 1105} 1106 1107u16 nvmet_check_ctrl_status(struct nvmet_req *req, struct nvme_command *cmd) 1108{ 1109 if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) { 1110 pr_err("got cmd %d while CC.EN == 0 on qid = %d\n", 1111 cmd->common.opcode, req->sq->qid); 1112 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR; 1113 } 1114 1115 if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) { 1116 pr_err("got cmd %d while CSTS.RDY == 0 on qid = %d\n", 1117 cmd->common.opcode, req->sq->qid); 1118 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR; 1119 } 1120 return 0; 1121} 1122 1123bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn) 1124{ 1125 struct nvmet_host_link *p; 1126 1127 lockdep_assert_held(&nvmet_config_sem); 1128 1129 if (subsys->allow_any_host) 1130 return true; 1131 1132 if (subsys->type == NVME_NQN_DISC) /* allow all access to disc subsys */ 1133 return true; 1134 1135 list_for_each_entry(p, &subsys->hosts, entry) { 1136 if (!strcmp(nvmet_host_name(p->host), hostnqn)) 1137 return true; 1138 } 1139 1140 return false; 1141} 1142 1143/* 1144 * Note: ctrl->subsys->lock should be held when calling this function 1145 */ 1146static void nvmet_setup_p2p_ns_map(struct nvmet_ctrl *ctrl, 1147 struct nvmet_req *req) 1148{ 1149 struct nvmet_ns *ns; 1150 1151 if (!req->p2p_client) 1152 return; 1153 1154 ctrl->p2p_client = get_device(req->p2p_client); 1155 1156 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) 1157 nvmet_p2pmem_ns_add_p2p(ctrl, ns); 1158} 1159 1160/* 1161 * Note: ctrl->subsys->lock should be held when calling this function 1162 */ 1163static void nvmet_release_p2p_ns_map(struct nvmet_ctrl *ctrl) 1164{ 1165 struct radix_tree_iter iter; 1166 void __rcu **slot; 1167 1168 radix_tree_for_each_slot(slot, &ctrl->p2p_ns_map, &iter, 0) 1169 pci_dev_put(radix_tree_deref_slot(slot)); 1170 1171 put_device(ctrl->p2p_client); 1172} 1173 1174u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn, 1175 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp) 1176{ 1177 struct nvmet_subsys *subsys; 1178 struct nvmet_ctrl *ctrl; 1179 int ret; 1180 u16 status; 1181 1182 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR; 1183 subsys = nvmet_find_get_subsys(req->port, subsysnqn); 1184 if (!subsys) { 1185 pr_warn("connect request for invalid subsystem %s!\n", 1186 subsysnqn); 1187 req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn); 1188 goto out; 1189 } 1190 1191 status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR; 1192 down_read(&nvmet_config_sem); 1193 if (!nvmet_host_allowed(subsys, hostnqn)) { 1194 pr_info("connect by host %s for subsystem %s not allowed\n", 1195 hostnqn, subsysnqn); 1196 req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn); 1197 up_read(&nvmet_config_sem); 1198 status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR; 1199 goto out_put_subsystem; 1200 } 1201 up_read(&nvmet_config_sem); 1202 1203 status = NVME_SC_INTERNAL; 1204 ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL); 1205 if (!ctrl) 1206 goto out_put_subsystem; 1207 mutex_init(&ctrl->lock); 1208 1209 nvmet_init_cap(ctrl); 1210 1211 ctrl->port = req->port; 1212 1213 INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work); 1214 INIT_LIST_HEAD(&ctrl->async_events); 1215 INIT_RADIX_TREE(&ctrl->p2p_ns_map, GFP_KERNEL); 1216 1217 memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE); 1218 memcpy(ctrl->hostnqn, hostnqn, NVMF_NQN_SIZE); 1219 1220 kref_init(&ctrl->ref); 1221 ctrl->subsys = subsys; 1222 WRITE_ONCE(ctrl->aen_enabled, NVMET_AEN_CFG_OPTIONAL); 1223 1224 ctrl->changed_ns_list = kmalloc_array(NVME_MAX_CHANGED_NAMESPACES, 1225 sizeof(__le32), GFP_KERNEL); 1226 if (!ctrl->changed_ns_list) 1227 goto out_free_ctrl; 1228 1229 ctrl->cqs = kcalloc(subsys->max_qid + 1, 1230 sizeof(struct nvmet_cq *), 1231 GFP_KERNEL); 1232 if (!ctrl->cqs) 1233 goto out_free_changed_ns_list; 1234 1235 ctrl->sqs = kcalloc(subsys->max_qid + 1, 1236 sizeof(struct nvmet_sq *), 1237 GFP_KERNEL); 1238 if (!ctrl->sqs) 1239 goto out_free_cqs; 1240 1241 ret = ida_simple_get(&cntlid_ida, 1242 NVME_CNTLID_MIN, NVME_CNTLID_MAX, 1243 GFP_KERNEL); 1244 if (ret < 0) { 1245 status = NVME_SC_CONNECT_CTRL_BUSY | NVME_SC_DNR; 1246 goto out_free_sqs; 1247 } 1248 ctrl->cntlid = ret; 1249 1250 ctrl->ops = req->ops; 1251 1252 /* 1253 * Discovery controllers may use some arbitrary high value 1254 * in order to cleanup stale discovery sessions 1255 */ 1256 if ((ctrl->subsys->type == NVME_NQN_DISC) && !kato) 1257 kato = NVMET_DISC_KATO_MS; 1258 1259 /* keep-alive timeout in seconds */ 1260 ctrl->kato = DIV_ROUND_UP(kato, 1000); 1261 1262 ctrl->err_counter = 0; 1263 spin_lock_init(&ctrl->error_lock); 1264 1265 nvmet_start_keep_alive_timer(ctrl); 1266 1267 mutex_lock(&subsys->lock); 1268 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls); 1269 nvmet_setup_p2p_ns_map(ctrl, req); 1270 mutex_unlock(&subsys->lock); 1271 1272 *ctrlp = ctrl; 1273 return 0; 1274 1275out_free_sqs: 1276 kfree(ctrl->sqs); 1277out_free_cqs: 1278 kfree(ctrl->cqs); 1279out_free_changed_ns_list: 1280 kfree(ctrl->changed_ns_list); 1281out_free_ctrl: 1282 kfree(ctrl); 1283out_put_subsystem: 1284 nvmet_subsys_put(subsys); 1285out: 1286 return status; 1287} 1288 1289static void nvmet_ctrl_free(struct kref *ref) 1290{ 1291 struct nvmet_ctrl *ctrl = container_of(ref, struct nvmet_ctrl, ref); 1292 struct nvmet_subsys *subsys = ctrl->subsys; 1293 1294 mutex_lock(&subsys->lock); 1295 nvmet_release_p2p_ns_map(ctrl); 1296 list_del(&ctrl->subsys_entry); 1297 mutex_unlock(&subsys->lock); 1298 1299 nvmet_stop_keep_alive_timer(ctrl); 1300 1301 flush_work(&ctrl->async_event_work); 1302 cancel_work_sync(&ctrl->fatal_err_work); 1303 1304 ida_simple_remove(&cntlid_ida, ctrl->cntlid); 1305 1306 kfree(ctrl->sqs); 1307 kfree(ctrl->cqs); 1308 kfree(ctrl->changed_ns_list); 1309 kfree(ctrl); 1310 1311 nvmet_subsys_put(subsys); 1312} 1313 1314void nvmet_ctrl_put(struct nvmet_ctrl *ctrl) 1315{ 1316 kref_put(&ctrl->ref, nvmet_ctrl_free); 1317} 1318 1319static void nvmet_fatal_error_handler(struct work_struct *work) 1320{ 1321 struct nvmet_ctrl *ctrl = 1322 container_of(work, struct nvmet_ctrl, fatal_err_work); 1323 1324 pr_err("ctrl %d fatal error occurred!\n", ctrl->cntlid); 1325 ctrl->ops->delete_ctrl(ctrl); 1326} 1327 1328void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl) 1329{ 1330 mutex_lock(&ctrl->lock); 1331 if (!(ctrl->csts & NVME_CSTS_CFS)) { 1332 ctrl->csts |= NVME_CSTS_CFS; 1333 INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler); 1334 schedule_work(&ctrl->fatal_err_work); 1335 } 1336 mutex_unlock(&ctrl->lock); 1337} 1338EXPORT_SYMBOL_GPL(nvmet_ctrl_fatal_error); 1339 1340static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port, 1341 const char *subsysnqn) 1342{ 1343 struct nvmet_subsys_link *p; 1344 1345 if (!port) 1346 return NULL; 1347 1348 if (!strcmp(NVME_DISC_SUBSYS_NAME, subsysnqn)) { 1349 if (!kref_get_unless_zero(&nvmet_disc_subsys->ref)) 1350 return NULL; 1351 return nvmet_disc_subsys; 1352 } 1353 1354 down_read(&nvmet_config_sem); 1355 list_for_each_entry(p, &port->subsystems, entry) { 1356 if (!strncmp(p->subsys->subsysnqn, subsysnqn, 1357 NVMF_NQN_SIZE)) { 1358 if (!kref_get_unless_zero(&p->subsys->ref)) 1359 break; 1360 up_read(&nvmet_config_sem); 1361 return p->subsys; 1362 } 1363 } 1364 up_read(&nvmet_config_sem); 1365 return NULL; 1366} 1367 1368struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn, 1369 enum nvme_subsys_type type) 1370{ 1371 struct nvmet_subsys *subsys; 1372 1373 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL); 1374 if (!subsys) 1375 return NULL; 1376 1377 subsys->ver = NVME_VS(1, 3, 0); /* NVMe 1.3.0 */ 1378 /* generate a random serial number as our controllers are ephemeral: */ 1379 get_random_bytes(&subsys->serial, sizeof(subsys->serial)); 1380 1381 switch (type) { 1382 case NVME_NQN_NVME: 1383 subsys->max_qid = NVMET_NR_QUEUES; 1384 break; 1385 case NVME_NQN_DISC: 1386 subsys->max_qid = 0; 1387 break; 1388 default: 1389 pr_err("%s: Unknown Subsystem type - %d\n", __func__, type); 1390 kfree(subsys); 1391 return NULL; 1392 } 1393 subsys->type = type; 1394 subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE, 1395 GFP_KERNEL); 1396 if (!subsys->subsysnqn) { 1397 kfree(subsys); 1398 return NULL; 1399 } 1400 1401 kref_init(&subsys->ref); 1402 1403 mutex_init(&subsys->lock); 1404 INIT_LIST_HEAD(&subsys->namespaces); 1405 INIT_LIST_HEAD(&subsys->ctrls); 1406 INIT_LIST_HEAD(&subsys->hosts); 1407 1408 return subsys; 1409} 1410 1411static void nvmet_subsys_free(struct kref *ref) 1412{ 1413 struct nvmet_subsys *subsys = 1414 container_of(ref, struct nvmet_subsys, ref); 1415 1416 WARN_ON_ONCE(!list_empty(&subsys->namespaces)); 1417 1418 kfree(subsys->subsysnqn); 1419 kfree(subsys); 1420} 1421 1422void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys) 1423{ 1424 struct nvmet_ctrl *ctrl; 1425 1426 mutex_lock(&subsys->lock); 1427 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) 1428 ctrl->ops->delete_ctrl(ctrl); 1429 mutex_unlock(&subsys->lock); 1430} 1431 1432void nvmet_subsys_put(struct nvmet_subsys *subsys) 1433{ 1434 kref_put(&subsys->ref, nvmet_subsys_free); 1435} 1436 1437static int __init nvmet_init(void) 1438{ 1439 int error; 1440 1441 nvmet_ana_group_enabled[NVMET_DEFAULT_ANA_GRPID] = 1; 1442 1443 buffered_io_wq = alloc_workqueue("nvmet-buffered-io-wq", 1444 WQ_MEM_RECLAIM, 0); 1445 if (!buffered_io_wq) { 1446 error = -ENOMEM; 1447 goto out; 1448 } 1449 1450 error = nvmet_init_discovery(); 1451 if (error) 1452 goto out_free_work_queue; 1453 1454 error = nvmet_init_configfs(); 1455 if (error) 1456 goto out_exit_discovery; 1457 return 0; 1458 1459out_exit_discovery: 1460 nvmet_exit_discovery(); 1461out_free_work_queue: 1462 destroy_workqueue(buffered_io_wq); 1463out: 1464 return error; 1465} 1466 1467static void __exit nvmet_exit(void) 1468{ 1469 nvmet_exit_configfs(); 1470 nvmet_exit_discovery(); 1471 ida_destroy(&cntlid_ida); 1472 destroy_workqueue(buffered_io_wq); 1473 1474 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_entry) != 1024); 1475 BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_hdr) != 1024); 1476} 1477 1478module_init(nvmet_init); 1479module_exit(nvmet_exit); 1480 1481MODULE_LICENSE("GPL v2");