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