1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Copyright (c) 2015-2016 HGST, a Western Digital Company. 4 */ 5 6 #ifndef _NVMET_H 7 #define _NVMET_H 8 9 #include <linux/dma-mapping.h> 10 #include <linux/types.h> 11 #include <linux/device.h> 12 #include <linux/kref.h> 13 #include <linux/percpu-refcount.h> 14 #include <linux/list.h> 15 #include <linux/mutex.h> 16 #include <linux/uuid.h> 17 #include <linux/nvme.h> 18 #include <linux/configfs.h> 19 #include <linux/rcupdate.h> 20 #include <linux/blkdev.h> 21 #include <linux/radix-tree.h> 22 #include <linux/t10-pi.h> 23 24 #define NVMET_DEFAULT_VS NVME_VS(1, 3, 0) 25 26 #define NVMET_ASYNC_EVENTS 4 27 #define NVMET_ERROR_LOG_SLOTS 128 28 #define NVMET_NO_ERROR_LOC ((u16)-1) 29 #define NVMET_DEFAULT_CTRL_MODEL "Linux" 30 31 /* 32 * Supported optional AENs: 33 */ 34 #define NVMET_AEN_CFG_OPTIONAL \ 35 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_ANA_CHANGE) 36 #define NVMET_DISC_AEN_CFG_OPTIONAL \ 37 (NVME_AEN_CFG_DISC_CHANGE) 38 39 /* 40 * Plus mandatory SMART AENs (we'll never send them, but allow enabling them): 41 */ 42 #define NVMET_AEN_CFG_ALL \ 43 (NVME_SMART_CRIT_SPARE | NVME_SMART_CRIT_TEMPERATURE | \ 44 NVME_SMART_CRIT_RELIABILITY | NVME_SMART_CRIT_MEDIA | \ 45 NVME_SMART_CRIT_VOLATILE_MEMORY | NVMET_AEN_CFG_OPTIONAL) 46 47 /* Helper Macros when NVMe error is NVME_SC_CONNECT_INVALID_PARAM 48 * The 16 bit shift is to set IATTR bit to 1, which means offending 49 * offset starts in the data section of connect() 50 */ 51 #define IPO_IATTR_CONNECT_DATA(x) \ 52 (cpu_to_le32((1 << 16) | (offsetof(struct nvmf_connect_data, x)))) 53 #define IPO_IATTR_CONNECT_SQE(x) \ 54 (cpu_to_le32(offsetof(struct nvmf_connect_command, x))) 55 56 struct nvmet_ns { 57 struct percpu_ref ref; 58 struct block_device *bdev; 59 struct file *file; 60 bool readonly; 61 u32 nsid; 62 u32 blksize_shift; 63 loff_t size; 64 u8 nguid[16]; 65 uuid_t uuid; 66 u32 anagrpid; 67 68 bool buffered_io; 69 bool enabled; 70 struct nvmet_subsys *subsys; 71 const char *device_path; 72 73 struct config_group device_group; 74 struct config_group group; 75 76 struct completion disable_done; 77 mempool_t *bvec_pool; 78 struct kmem_cache *bvec_cache; 79 80 int use_p2pmem; 81 struct pci_dev *p2p_dev; 82 int pi_type; 83 int metadata_size; 84 }; 85 86 static inline struct nvmet_ns *to_nvmet_ns(struct config_item *item) 87 { 88 return container_of(to_config_group(item), struct nvmet_ns, group); 89 } 90 91 static inline struct device *nvmet_ns_dev(struct nvmet_ns *ns) 92 { 93 return ns->bdev ? disk_to_dev(ns->bdev->bd_disk) : NULL; 94 } 95 96 struct nvmet_cq { 97 u16 qid; 98 u16 size; 99 }; 100 101 struct nvmet_sq { 102 struct nvmet_ctrl *ctrl; 103 struct percpu_ref ref; 104 u16 qid; 105 u16 size; 106 u32 sqhd; 107 bool sqhd_disabled; 108 struct completion free_done; 109 struct completion confirm_done; 110 }; 111 112 struct nvmet_ana_group { 113 struct config_group group; 114 struct nvmet_port *port; 115 u32 grpid; 116 }; 117 118 static inline struct nvmet_ana_group *to_ana_group(struct config_item *item) 119 { 120 return container_of(to_config_group(item), struct nvmet_ana_group, 121 group); 122 } 123 124 /** 125 * struct nvmet_port - Common structure to keep port 126 * information for the target. 127 * @entry: Entry into referrals or transport list. 128 * @disc_addr: Address information is stored in a format defined 129 * for a discovery log page entry. 130 * @group: ConfigFS group for this element's folder. 131 * @priv: Private data for the transport. 132 */ 133 struct nvmet_port { 134 struct list_head entry; 135 struct nvmf_disc_rsp_page_entry disc_addr; 136 struct config_group group; 137 struct config_group subsys_group; 138 struct list_head subsystems; 139 struct config_group referrals_group; 140 struct list_head referrals; 141 struct list_head global_entry; 142 struct config_group ana_groups_group; 143 struct nvmet_ana_group ana_default_group; 144 enum nvme_ana_state *ana_state; 145 void *priv; 146 bool enabled; 147 int inline_data_size; 148 const struct nvmet_fabrics_ops *tr_ops; 149 bool pi_enable; 150 }; 151 152 static inline struct nvmet_port *to_nvmet_port(struct config_item *item) 153 { 154 return container_of(to_config_group(item), struct nvmet_port, 155 group); 156 } 157 158 static inline struct nvmet_port *ana_groups_to_port( 159 struct config_item *item) 160 { 161 return container_of(to_config_group(item), struct nvmet_port, 162 ana_groups_group); 163 } 164 165 struct nvmet_ctrl { 166 struct nvmet_subsys *subsys; 167 struct nvmet_cq **cqs; 168 struct nvmet_sq **sqs; 169 170 bool cmd_seen; 171 172 struct mutex lock; 173 u64 cap; 174 u32 cc; 175 u32 csts; 176 177 uuid_t hostid; 178 u16 cntlid; 179 u32 kato; 180 181 struct nvmet_port *port; 182 183 u32 aen_enabled; 184 unsigned long aen_masked; 185 struct nvmet_req *async_event_cmds[NVMET_ASYNC_EVENTS]; 186 unsigned int nr_async_event_cmds; 187 struct list_head async_events; 188 struct work_struct async_event_work; 189 190 struct list_head subsys_entry; 191 struct kref ref; 192 struct delayed_work ka_work; 193 struct work_struct fatal_err_work; 194 195 const struct nvmet_fabrics_ops *ops; 196 197 __le32 *changed_ns_list; 198 u32 nr_changed_ns; 199 200 char subsysnqn[NVMF_NQN_FIELD_LEN]; 201 char hostnqn[NVMF_NQN_FIELD_LEN]; 202 203 struct device *p2p_client; 204 struct radix_tree_root p2p_ns_map; 205 206 spinlock_t error_lock; 207 u64 err_counter; 208 struct nvme_error_slot slots[NVMET_ERROR_LOG_SLOTS]; 209 bool pi_support; 210 }; 211 212 struct nvmet_subsys_model { 213 struct rcu_head rcuhead; 214 char number[]; 215 }; 216 217 struct nvmet_subsys { 218 enum nvme_subsys_type type; 219 220 struct mutex lock; 221 struct kref ref; 222 223 struct xarray namespaces; 224 unsigned int nr_namespaces; 225 unsigned int max_nsid; 226 u16 cntlid_min; 227 u16 cntlid_max; 228 229 struct list_head ctrls; 230 231 struct list_head hosts; 232 bool allow_any_host; 233 234 u16 max_qid; 235 236 u64 ver; 237 u64 serial; 238 char *subsysnqn; 239 bool pi_support; 240 241 struct config_group group; 242 243 struct config_group namespaces_group; 244 struct config_group allowed_hosts_group; 245 246 struct nvmet_subsys_model __rcu *model; 247 248 #ifdef CONFIG_NVME_TARGET_PASSTHRU 249 struct nvme_ctrl *passthru_ctrl; 250 char *passthru_ctrl_path; 251 struct config_group passthru_group; 252 #endif /* CONFIG_NVME_TARGET_PASSTHRU */ 253 }; 254 255 static inline struct nvmet_subsys *to_subsys(struct config_item *item) 256 { 257 return container_of(to_config_group(item), struct nvmet_subsys, group); 258 } 259 260 static inline struct nvmet_subsys *namespaces_to_subsys( 261 struct config_item *item) 262 { 263 return container_of(to_config_group(item), struct nvmet_subsys, 264 namespaces_group); 265 } 266 267 struct nvmet_host { 268 struct config_group group; 269 }; 270 271 static inline struct nvmet_host *to_host(struct config_item *item) 272 { 273 return container_of(to_config_group(item), struct nvmet_host, group); 274 } 275 276 static inline char *nvmet_host_name(struct nvmet_host *host) 277 { 278 return config_item_name(&host->group.cg_item); 279 } 280 281 struct nvmet_host_link { 282 struct list_head entry; 283 struct nvmet_host *host; 284 }; 285 286 struct nvmet_subsys_link { 287 struct list_head entry; 288 struct nvmet_subsys *subsys; 289 }; 290 291 struct nvmet_req; 292 struct nvmet_fabrics_ops { 293 struct module *owner; 294 unsigned int type; 295 unsigned int msdbd; 296 unsigned int flags; 297 #define NVMF_KEYED_SGLS (1 << 0) 298 #define NVMF_METADATA_SUPPORTED (1 << 1) 299 void (*queue_response)(struct nvmet_req *req); 300 int (*add_port)(struct nvmet_port *port); 301 void (*remove_port)(struct nvmet_port *port); 302 void (*delete_ctrl)(struct nvmet_ctrl *ctrl); 303 void (*disc_traddr)(struct nvmet_req *req, 304 struct nvmet_port *port, char *traddr); 305 u16 (*install_queue)(struct nvmet_sq *nvme_sq); 306 void (*discovery_chg)(struct nvmet_port *port); 307 u8 (*get_mdts)(const struct nvmet_ctrl *ctrl); 308 }; 309 310 #define NVMET_MAX_INLINE_BIOVEC 8 311 #define NVMET_MAX_INLINE_DATA_LEN NVMET_MAX_INLINE_BIOVEC * PAGE_SIZE 312 313 struct nvmet_req { 314 struct nvme_command *cmd; 315 struct nvme_completion *cqe; 316 struct nvmet_sq *sq; 317 struct nvmet_cq *cq; 318 struct nvmet_ns *ns; 319 struct scatterlist *sg; 320 struct scatterlist *metadata_sg; 321 struct bio_vec inline_bvec[NVMET_MAX_INLINE_BIOVEC]; 322 union { 323 struct { 324 struct bio inline_bio; 325 } b; 326 struct { 327 bool mpool_alloc; 328 struct kiocb iocb; 329 struct bio_vec *bvec; 330 struct work_struct work; 331 } f; 332 struct { 333 struct request *rq; 334 struct work_struct work; 335 bool use_workqueue; 336 } p; 337 }; 338 int sg_cnt; 339 int metadata_sg_cnt; 340 /* data length as parsed from the SGL descriptor: */ 341 size_t transfer_len; 342 size_t metadata_len; 343 344 struct nvmet_port *port; 345 346 void (*execute)(struct nvmet_req *req); 347 const struct nvmet_fabrics_ops *ops; 348 349 struct pci_dev *p2p_dev; 350 struct device *p2p_client; 351 u16 error_loc; 352 u64 error_slba; 353 }; 354 355 extern struct workqueue_struct *buffered_io_wq; 356 357 static inline void nvmet_set_result(struct nvmet_req *req, u32 result) 358 { 359 req->cqe->result.u32 = cpu_to_le32(result); 360 } 361 362 /* 363 * NVMe command writes actually are DMA reads for us on the target side. 364 */ 365 static inline enum dma_data_direction 366 nvmet_data_dir(struct nvmet_req *req) 367 { 368 return nvme_is_write(req->cmd) ? DMA_FROM_DEVICE : DMA_TO_DEVICE; 369 } 370 371 struct nvmet_async_event { 372 struct list_head entry; 373 u8 event_type; 374 u8 event_info; 375 u8 log_page; 376 }; 377 378 static inline void nvmet_clear_aen_bit(struct nvmet_req *req, u32 bn) 379 { 380 int rae = le32_to_cpu(req->cmd->common.cdw10) & 1 << 15; 381 382 if (!rae) 383 clear_bit(bn, &req->sq->ctrl->aen_masked); 384 } 385 386 static inline bool nvmet_aen_bit_disabled(struct nvmet_ctrl *ctrl, u32 bn) 387 { 388 if (!(READ_ONCE(ctrl->aen_enabled) & (1 << bn))) 389 return true; 390 return test_and_set_bit(bn, &ctrl->aen_masked); 391 } 392 393 void nvmet_get_feat_kato(struct nvmet_req *req); 394 void nvmet_get_feat_async_event(struct nvmet_req *req); 395 u16 nvmet_set_feat_kato(struct nvmet_req *req); 396 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask); 397 void nvmet_execute_async_event(struct nvmet_req *req); 398 void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl); 399 void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl); 400 401 u16 nvmet_parse_connect_cmd(struct nvmet_req *req); 402 void nvmet_bdev_set_limits(struct block_device *bdev, struct nvme_id_ns *id); 403 u16 nvmet_bdev_parse_io_cmd(struct nvmet_req *req); 404 u16 nvmet_file_parse_io_cmd(struct nvmet_req *req); 405 u16 nvmet_parse_admin_cmd(struct nvmet_req *req); 406 u16 nvmet_parse_discovery_cmd(struct nvmet_req *req); 407 u16 nvmet_parse_fabrics_cmd(struct nvmet_req *req); 408 409 bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq, 410 struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops); 411 void nvmet_req_uninit(struct nvmet_req *req); 412 bool nvmet_check_transfer_len(struct nvmet_req *req, size_t len); 413 bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len); 414 void nvmet_req_complete(struct nvmet_req *req, u16 status); 415 int nvmet_req_alloc_sgls(struct nvmet_req *req); 416 void nvmet_req_free_sgls(struct nvmet_req *req); 417 418 void nvmet_execute_set_features(struct nvmet_req *req); 419 void nvmet_execute_get_features(struct nvmet_req *req); 420 void nvmet_execute_keep_alive(struct nvmet_req *req); 421 422 void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq, u16 qid, 423 u16 size); 424 void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq, u16 qid, 425 u16 size); 426 void nvmet_sq_destroy(struct nvmet_sq *sq); 427 int nvmet_sq_init(struct nvmet_sq *sq); 428 429 void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl); 430 431 void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new); 432 u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn, 433 struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp); 434 u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid, 435 struct nvmet_req *req, struct nvmet_ctrl **ret); 436 void nvmet_ctrl_put(struct nvmet_ctrl *ctrl); 437 u16 nvmet_check_ctrl_status(struct nvmet_req *req, struct nvme_command *cmd); 438 439 struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn, 440 enum nvme_subsys_type type); 441 void nvmet_subsys_put(struct nvmet_subsys *subsys); 442 void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys); 443 444 struct nvmet_ns *nvmet_find_namespace(struct nvmet_ctrl *ctrl, __le32 nsid); 445 void nvmet_put_namespace(struct nvmet_ns *ns); 446 int nvmet_ns_enable(struct nvmet_ns *ns); 447 void nvmet_ns_disable(struct nvmet_ns *ns); 448 struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid); 449 void nvmet_ns_free(struct nvmet_ns *ns); 450 451 void nvmet_send_ana_event(struct nvmet_subsys *subsys, 452 struct nvmet_port *port); 453 void nvmet_port_send_ana_event(struct nvmet_port *port); 454 455 int nvmet_register_transport(const struct nvmet_fabrics_ops *ops); 456 void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops); 457 458 void nvmet_port_del_ctrls(struct nvmet_port *port, 459 struct nvmet_subsys *subsys); 460 461 int nvmet_enable_port(struct nvmet_port *port); 462 void nvmet_disable_port(struct nvmet_port *port); 463 464 void nvmet_referral_enable(struct nvmet_port *parent, struct nvmet_port *port); 465 void nvmet_referral_disable(struct nvmet_port *parent, struct nvmet_port *port); 466 467 u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf, 468 size_t len); 469 u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, 470 size_t len); 471 u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len); 472 473 u32 nvmet_get_log_page_len(struct nvme_command *cmd); 474 u64 nvmet_get_log_page_offset(struct nvme_command *cmd); 475 476 extern struct list_head *nvmet_ports; 477 void nvmet_port_disc_changed(struct nvmet_port *port, 478 struct nvmet_subsys *subsys); 479 void nvmet_subsys_disc_changed(struct nvmet_subsys *subsys, 480 struct nvmet_host *host); 481 void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type, 482 u8 event_info, u8 log_page); 483 484 #define NVMET_QUEUE_SIZE 1024 485 #define NVMET_NR_QUEUES 128 486 #define NVMET_MAX_CMD NVMET_QUEUE_SIZE 487 488 /* 489 * Nice round number that makes a list of nsids fit into a page. 490 * Should become tunable at some point in the future. 491 */ 492 #define NVMET_MAX_NAMESPACES 1024 493 494 /* 495 * 0 is not a valid ANA group ID, so we start numbering at 1. 496 * 497 * ANA Group 1 exists without manual intervention, has namespaces assigned to it 498 * by default, and is available in an optimized state through all ports. 499 */ 500 #define NVMET_MAX_ANAGRPS 128 501 #define NVMET_DEFAULT_ANA_GRPID 1 502 503 #define NVMET_KAS 10 504 #define NVMET_DISC_KATO_MS 120000 505 506 int __init nvmet_init_configfs(void); 507 void __exit nvmet_exit_configfs(void); 508 509 int __init nvmet_init_discovery(void); 510 void nvmet_exit_discovery(void); 511 512 extern struct nvmet_subsys *nvmet_disc_subsys; 513 extern struct rw_semaphore nvmet_config_sem; 514 515 extern u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1]; 516 extern u64 nvmet_ana_chgcnt; 517 extern struct rw_semaphore nvmet_ana_sem; 518 519 bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn); 520 521 int nvmet_bdev_ns_enable(struct nvmet_ns *ns); 522 int nvmet_file_ns_enable(struct nvmet_ns *ns); 523 void nvmet_bdev_ns_disable(struct nvmet_ns *ns); 524 void nvmet_file_ns_disable(struct nvmet_ns *ns); 525 u16 nvmet_bdev_flush(struct nvmet_req *req); 526 u16 nvmet_file_flush(struct nvmet_req *req); 527 void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid); 528 void nvmet_bdev_ns_revalidate(struct nvmet_ns *ns); 529 int nvmet_file_ns_revalidate(struct nvmet_ns *ns); 530 void nvmet_ns_revalidate(struct nvmet_ns *ns); 531 532 static inline u32 nvmet_rw_data_len(struct nvmet_req *req) 533 { 534 return ((u32)le16_to_cpu(req->cmd->rw.length) + 1) << 535 req->ns->blksize_shift; 536 } 537 538 static inline u32 nvmet_rw_metadata_len(struct nvmet_req *req) 539 { 540 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) 541 return 0; 542 return ((u32)le16_to_cpu(req->cmd->rw.length) + 1) * 543 req->ns->metadata_size; 544 } 545 546 static inline u32 nvmet_dsm_len(struct nvmet_req *req) 547 { 548 return (le32_to_cpu(req->cmd->dsm.nr) + 1) * 549 sizeof(struct nvme_dsm_range); 550 } 551 552 #ifdef CONFIG_NVME_TARGET_PASSTHRU 553 void nvmet_passthru_subsys_free(struct nvmet_subsys *subsys); 554 int nvmet_passthru_ctrl_enable(struct nvmet_subsys *subsys); 555 void nvmet_passthru_ctrl_disable(struct nvmet_subsys *subsys); 556 u16 nvmet_parse_passthru_admin_cmd(struct nvmet_req *req); 557 u16 nvmet_parse_passthru_io_cmd(struct nvmet_req *req); 558 static inline struct nvme_ctrl *nvmet_passthru_ctrl(struct nvmet_subsys *subsys) 559 { 560 return subsys->passthru_ctrl; 561 } 562 #else /* CONFIG_NVME_TARGET_PASSTHRU */ 563 static inline void nvmet_passthru_subsys_free(struct nvmet_subsys *subsys) 564 { 565 } 566 static inline void nvmet_passthru_ctrl_disable(struct nvmet_subsys *subsys) 567 { 568 } 569 static inline u16 nvmet_parse_passthru_admin_cmd(struct nvmet_req *req) 570 { 571 return 0; 572 } 573 static inline u16 nvmet_parse_passthru_io_cmd(struct nvmet_req *req) 574 { 575 return 0; 576 } 577 static inline struct nvme_ctrl *nvmet_passthru_ctrl(struct nvmet_subsys *subsys) 578 { 579 return NULL; 580 } 581 #endif /* CONFIG_NVME_TARGET_PASSTHRU */ 582 583 static inline struct nvme_ctrl * 584 nvmet_req_passthru_ctrl(struct nvmet_req *req) 585 { 586 return nvmet_passthru_ctrl(req->sq->ctrl->subsys); 587 } 588 589 u16 errno_to_nvme_status(struct nvmet_req *req, int errno); 590 591 /* Convert a 32-bit number to a 16-bit 0's based number */ 592 static inline __le16 to0based(u32 a) 593 { 594 return cpu_to_le16(max(1U, min(1U << 16, a)) - 1); 595 } 596 597 static inline bool nvmet_ns_has_pi(struct nvmet_ns *ns) 598 { 599 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)) 600 return false; 601 return ns->pi_type && ns->metadata_size == sizeof(struct t10_pi_tuple); 602 } 603 604 #endif /* _NVMET_H */ 605