1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2016 Intel Corporation. All rights reserved. 3 * Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 #ifndef __NVMF_INTERNAL_H__ 8 #define __NVMF_INTERNAL_H__ 9 10 #include "spdk/stdinc.h" 11 12 #include "spdk/likely.h" 13 #include "spdk/nvmf.h" 14 #include "spdk/nvmf_cmd.h" 15 #include "spdk/nvmf_transport.h" 16 #include "spdk/nvmf_spec.h" 17 #include "spdk/assert.h" 18 #include "spdk/bdev.h" 19 #include "spdk/queue.h" 20 #include "spdk/util.h" 21 #include "spdk/thread.h" 22 #include "spdk/tree.h" 23 24 /* The spec reserves cntlid values in the range FFF0h to FFFFh. */ 25 #define NVMF_MIN_CNTLID 1 26 #define NVMF_MAX_CNTLID 0xFFEF 27 28 enum spdk_nvmf_tgt_state { 29 NVMF_TGT_IDLE = 0, 30 NVMF_TGT_RUNNING, 31 NVMF_TGT_PAUSING, 32 NVMF_TGT_PAUSED, 33 NVMF_TGT_RESUMING, 34 }; 35 36 enum spdk_nvmf_subsystem_state { 37 SPDK_NVMF_SUBSYSTEM_INACTIVE = 0, 38 SPDK_NVMF_SUBSYSTEM_ACTIVATING, 39 SPDK_NVMF_SUBSYSTEM_ACTIVE, 40 SPDK_NVMF_SUBSYSTEM_PAUSING, 41 SPDK_NVMF_SUBSYSTEM_PAUSED, 42 SPDK_NVMF_SUBSYSTEM_RESUMING, 43 SPDK_NVMF_SUBSYSTEM_DEACTIVATING, 44 SPDK_NVMF_SUBSYSTEM_NUM_STATES, 45 }; 46 47 RB_HEAD(subsystem_tree, spdk_nvmf_subsystem); 48 49 struct spdk_nvmf_tgt { 50 char name[NVMF_TGT_NAME_MAX_LENGTH]; 51 52 pthread_mutex_t mutex; 53 54 uint64_t discovery_genctr; 55 56 uint32_t max_subsystems; 57 58 enum spdk_nvmf_tgt_discovery_filter discovery_filter; 59 60 enum spdk_nvmf_tgt_state state; 61 62 struct spdk_bit_array *subsystem_ids; 63 64 struct subsystem_tree subsystems; 65 66 TAILQ_HEAD(, spdk_nvmf_transport) transports; 67 TAILQ_HEAD(, spdk_nvmf_poll_group) poll_groups; 68 TAILQ_HEAD(, spdk_nvmf_referral) referrals; 69 70 /* Used for round-robin assignment of connections to poll groups */ 71 struct spdk_nvmf_poll_group *next_poll_group; 72 73 spdk_nvmf_tgt_destroy_done_fn *destroy_cb_fn; 74 void *destroy_cb_arg; 75 76 uint16_t crdt[3]; 77 uint16_t num_poll_groups; 78 79 TAILQ_ENTRY(spdk_nvmf_tgt) link; 80 }; 81 82 struct spdk_nvmf_host { 83 char nqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 84 TAILQ_ENTRY(spdk_nvmf_host) link; 85 }; 86 87 struct spdk_nvmf_subsystem_listener { 88 struct spdk_nvmf_subsystem *subsystem; 89 spdk_nvmf_tgt_subsystem_listen_done_fn cb_fn; 90 void *cb_arg; 91 struct spdk_nvme_transport_id *trid; 92 struct spdk_nvmf_transport *transport; 93 enum spdk_nvme_ana_state *ana_state; 94 uint64_t ana_state_change_count; 95 uint16_t id; 96 struct spdk_nvmf_listener_opts opts; 97 TAILQ_ENTRY(spdk_nvmf_subsystem_listener) link; 98 }; 99 100 struct spdk_nvmf_referral { 101 /* Discovery Log Page Entry for this referral */ 102 struct spdk_nvmf_discovery_log_page_entry entry; 103 /* Transport ID */ 104 struct spdk_nvme_transport_id trid; 105 TAILQ_ENTRY(spdk_nvmf_referral) link; 106 }; 107 108 struct spdk_nvmf_subsystem_pg_ns_info { 109 struct spdk_io_channel *channel; 110 struct spdk_uuid uuid; 111 /* current reservation key, no reservation if the value is 0 */ 112 uint64_t crkey; 113 /* reservation type */ 114 enum spdk_nvme_reservation_type rtype; 115 /* Host ID which holds the reservation */ 116 struct spdk_uuid holder_id; 117 /* Host ID for the registrants with the namespace */ 118 struct spdk_uuid reg_hostid[SPDK_NVMF_MAX_NUM_REGISTRANTS]; 119 uint64_t num_blocks; 120 121 /* I/O outstanding to this namespace */ 122 uint64_t io_outstanding; 123 enum spdk_nvmf_subsystem_state state; 124 }; 125 126 typedef void(*spdk_nvmf_poll_group_mod_done)(void *cb_arg, int status); 127 128 struct spdk_nvmf_subsystem_poll_group { 129 /* Array of namespace information for each namespace indexed by nsid - 1 */ 130 struct spdk_nvmf_subsystem_pg_ns_info *ns_info; 131 uint32_t num_ns; 132 enum spdk_nvmf_subsystem_state state; 133 134 /* Number of ADMIN and FABRICS requests outstanding */ 135 uint64_t mgmt_io_outstanding; 136 spdk_nvmf_poll_group_mod_done cb_fn; 137 void *cb_arg; 138 139 TAILQ_HEAD(, spdk_nvmf_request) queued; 140 }; 141 142 struct spdk_nvmf_registrant { 143 TAILQ_ENTRY(spdk_nvmf_registrant) link; 144 struct spdk_uuid hostid; 145 /* Registration key */ 146 uint64_t rkey; 147 }; 148 149 struct spdk_nvmf_ns { 150 uint32_t nsid; 151 uint32_t anagrpid; 152 struct spdk_nvmf_subsystem *subsystem; 153 struct spdk_bdev *bdev; 154 struct spdk_bdev_desc *desc; 155 struct spdk_nvmf_ns_opts opts; 156 /* reservation notification mask */ 157 uint32_t mask; 158 /* generation code */ 159 uint32_t gen; 160 /* registrants head */ 161 TAILQ_HEAD(, spdk_nvmf_registrant) registrants; 162 /* current reservation key */ 163 uint64_t crkey; 164 /* reservation type */ 165 enum spdk_nvme_reservation_type rtype; 166 /* current reservation holder, only valid if reservation type can only have one holder */ 167 struct spdk_nvmf_registrant *holder; 168 /* Persist Through Power Loss file which contains the persistent reservation */ 169 char *ptpl_file; 170 /* Persist Through Power Loss feature is enabled */ 171 bool ptpl_activated; 172 /* ZCOPY supported on bdev device */ 173 bool zcopy; 174 /* Command Set Identifier */ 175 enum spdk_nvme_csi csi; 176 }; 177 178 /* 179 * NVMf reservation notification log page. 180 */ 181 struct spdk_nvmf_reservation_log { 182 struct spdk_nvme_reservation_notification_log log; 183 TAILQ_ENTRY(spdk_nvmf_reservation_log) link; 184 struct spdk_nvmf_ctrlr *ctrlr; 185 }; 186 187 /* 188 * NVMf async event completion. 189 */ 190 struct spdk_nvmf_async_event_completion { 191 union spdk_nvme_async_event_completion event; 192 STAILQ_ENTRY(spdk_nvmf_async_event_completion) link; 193 }; 194 195 /* 196 * This structure represents an NVMe-oF controller, 197 * which is like a "session" in networking terms. 198 */ 199 struct spdk_nvmf_ctrlr { 200 uint16_t cntlid; 201 char hostnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 202 struct spdk_nvmf_subsystem *subsys; 203 204 struct spdk_nvmf_ctrlr_data cdata; 205 206 struct spdk_nvmf_registers vcprop; 207 208 struct spdk_nvmf_ctrlr_feat feat; 209 210 struct spdk_nvmf_qpair *admin_qpair; 211 struct spdk_thread *thread; 212 struct spdk_bit_array *qpair_mask; 213 214 const struct spdk_nvmf_subsystem_listener *listener; 215 216 struct spdk_nvmf_request *aer_req[SPDK_NVMF_MAX_ASYNC_EVENTS]; 217 STAILQ_HEAD(, spdk_nvmf_async_event_completion) async_events; 218 uint64_t notice_aen_mask; 219 uint8_t nr_aer_reqs; 220 struct spdk_uuid hostid; 221 222 uint32_t association_timeout; /* in milliseconds */ 223 uint16_t changed_ns_list_count; 224 struct spdk_nvme_ns_list changed_ns_list; 225 uint64_t log_page_count; 226 uint8_t num_avail_log_pages; 227 TAILQ_HEAD(log_page_head, spdk_nvmf_reservation_log) log_head; 228 229 /* Time to trigger keep-alive--poller_time = now_tick + period */ 230 uint64_t last_keep_alive_tick; 231 struct spdk_poller *keep_alive_poller; 232 233 struct spdk_poller *association_timer; 234 235 struct spdk_poller *cc_timer; 236 uint64_t cc_timeout_tsc; 237 struct spdk_poller *cc_timeout_timer; 238 239 bool dif_insert_or_strip; 240 bool in_destruct; 241 bool disconnect_in_progress; 242 /* valid only when disconnect_in_progress is true */ 243 bool disconnect_is_shn; 244 bool acre_enabled; 245 bool dynamic_ctrlr; 246 247 TAILQ_ENTRY(spdk_nvmf_ctrlr) link; 248 }; 249 250 #define NVMF_MAX_LISTENERS_PER_SUBSYSTEM 16 251 252 struct spdk_nvmf_subsystem { 253 struct spdk_thread *thread; 254 255 uint32_t id; 256 257 enum spdk_nvmf_subsystem_state state; 258 enum spdk_nvmf_subtype subtype; 259 260 uint16_t next_cntlid; 261 struct { 262 uint8_t allow_any_host : 1; 263 uint8_t allow_any_listener : 1; 264 uint8_t ana_reporting : 1; 265 uint8_t reserved : 5; 266 } flags; 267 268 /* boolean for state change synchronization */ 269 bool changing_state; 270 271 bool destroying; 272 bool async_destroy; 273 274 /* Zoned storage related fields */ 275 bool zone_append_supported; 276 uint64_t max_zone_append_size_kib; 277 278 struct spdk_nvmf_tgt *tgt; 279 RB_ENTRY(spdk_nvmf_subsystem) link; 280 281 /* Array of pointers to namespaces of size max_nsid indexed by nsid - 1 */ 282 struct spdk_nvmf_ns **ns; 283 uint32_t max_nsid; 284 285 uint16_t min_cntlid; 286 uint16_t max_cntlid; 287 288 uint64_t max_discard_size_kib; 289 uint64_t max_write_zeroes_size_kib; 290 291 TAILQ_HEAD(, spdk_nvmf_ctrlr) ctrlrs; 292 293 /* A mutex used to protect the hosts list and allow_any_host flag. Unlike the namespace 294 * array, this list is not used on the I/O path (it's needed for handling things like 295 * the CONNECT command), so use a mutex to protect it instead of requiring the subsystem 296 * state to be paused. This removes the requirement to pause the subsystem when hosts 297 * are added or removed dynamically. */ 298 pthread_mutex_t mutex; 299 TAILQ_HEAD(, spdk_nvmf_host) hosts; 300 TAILQ_HEAD(, spdk_nvmf_subsystem_listener) listeners; 301 struct spdk_bit_array *used_listener_ids; 302 303 TAILQ_ENTRY(spdk_nvmf_subsystem) entries; 304 305 nvmf_subsystem_destroy_cb async_destroy_cb; 306 void *async_destroy_cb_arg; 307 308 char sn[SPDK_NVME_CTRLR_SN_LEN + 1]; 309 char mn[SPDK_NVME_CTRLR_MN_LEN + 1]; 310 char subnqn[SPDK_NVMF_NQN_MAX_LEN + 1]; 311 312 /* Array of namespace count per ANA group of size max_nsid indexed anagrpid - 1 313 * It will be enough for ANA group to use the same size as namespaces. 314 */ 315 uint32_t *ana_group; 316 }; 317 318 static int 319 subsystem_cmp(struct spdk_nvmf_subsystem *subsystem1, struct spdk_nvmf_subsystem *subsystem2) 320 { 321 return strncmp(subsystem1->subnqn, subsystem2->subnqn, sizeof(subsystem1->subnqn)); 322 } 323 324 RB_GENERATE_STATIC(subsystem_tree, spdk_nvmf_subsystem, link, subsystem_cmp); 325 326 int nvmf_poll_group_update_subsystem(struct spdk_nvmf_poll_group *group, 327 struct spdk_nvmf_subsystem *subsystem); 328 int nvmf_poll_group_add_subsystem(struct spdk_nvmf_poll_group *group, 329 struct spdk_nvmf_subsystem *subsystem, 330 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg); 331 void nvmf_poll_group_remove_subsystem(struct spdk_nvmf_poll_group *group, 332 struct spdk_nvmf_subsystem *subsystem, spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg); 333 void nvmf_poll_group_pause_subsystem(struct spdk_nvmf_poll_group *group, 334 struct spdk_nvmf_subsystem *subsystem, 335 uint32_t nsid, 336 spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg); 337 void nvmf_poll_group_resume_subsystem(struct spdk_nvmf_poll_group *group, 338 struct spdk_nvmf_subsystem *subsystem, spdk_nvmf_poll_group_mod_done cb_fn, void *cb_arg); 339 340 void nvmf_update_discovery_log(struct spdk_nvmf_tgt *tgt, const char *hostnqn); 341 void nvmf_get_discovery_log_page(struct spdk_nvmf_tgt *tgt, const char *hostnqn, struct iovec *iov, 342 uint32_t iovcnt, uint64_t offset, uint32_t length, 343 struct spdk_nvme_transport_id *cmd_source_trid); 344 345 void nvmf_ctrlr_destruct(struct spdk_nvmf_ctrlr *ctrlr); 346 int nvmf_ctrlr_process_admin_cmd(struct spdk_nvmf_request *req); 347 int nvmf_ctrlr_process_io_cmd(struct spdk_nvmf_request *req); 348 bool nvmf_ctrlr_dsm_supported(struct spdk_nvmf_ctrlr *ctrlr); 349 bool nvmf_ctrlr_write_zeroes_supported(struct spdk_nvmf_ctrlr *ctrlr); 350 bool nvmf_ctrlr_copy_supported(struct spdk_nvmf_ctrlr *ctrlr); 351 void nvmf_ctrlr_ns_changed(struct spdk_nvmf_ctrlr *ctrlr, uint32_t nsid); 352 bool nvmf_ctrlr_use_zcopy(struct spdk_nvmf_request *req); 353 354 void nvmf_bdev_ctrlr_identify_ns(struct spdk_nvmf_ns *ns, struct spdk_nvme_ns_data *nsdata, 355 bool dif_insert_or_strip); 356 int nvmf_bdev_ctrlr_read_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 357 struct spdk_io_channel *ch, struct spdk_nvmf_request *req); 358 int nvmf_bdev_ctrlr_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 359 struct spdk_io_channel *ch, struct spdk_nvmf_request *req); 360 int nvmf_bdev_ctrlr_compare_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 361 struct spdk_io_channel *ch, struct spdk_nvmf_request *req); 362 int nvmf_bdev_ctrlr_compare_and_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 363 struct spdk_io_channel *ch, struct spdk_nvmf_request *cmp_req, struct spdk_nvmf_request *write_req); 364 int nvmf_bdev_ctrlr_write_zeroes_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 365 struct spdk_io_channel *ch, struct spdk_nvmf_request *req); 366 int nvmf_bdev_ctrlr_flush_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 367 struct spdk_io_channel *ch, struct spdk_nvmf_request *req); 368 int nvmf_bdev_ctrlr_dsm_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 369 struct spdk_io_channel *ch, struct spdk_nvmf_request *req); 370 int nvmf_bdev_ctrlr_copy_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 371 struct spdk_io_channel *ch, struct spdk_nvmf_request *req); 372 int nvmf_bdev_ctrlr_nvme_passthru_io(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 373 struct spdk_io_channel *ch, struct spdk_nvmf_request *req); 374 bool nvmf_bdev_ctrlr_get_dif_ctx(struct spdk_bdev *bdev, struct spdk_nvme_cmd *cmd, 375 struct spdk_dif_ctx *dif_ctx); 376 bool nvmf_bdev_zcopy_enabled(struct spdk_bdev *bdev); 377 378 int nvmf_subsystem_add_ctrlr(struct spdk_nvmf_subsystem *subsystem, 379 struct spdk_nvmf_ctrlr *ctrlr); 380 void nvmf_subsystem_remove_ctrlr(struct spdk_nvmf_subsystem *subsystem, 381 struct spdk_nvmf_ctrlr *ctrlr); 382 void nvmf_subsystem_remove_all_listeners(struct spdk_nvmf_subsystem *subsystem, 383 bool stop); 384 struct spdk_nvmf_ctrlr *nvmf_subsystem_get_ctrlr(struct spdk_nvmf_subsystem *subsystem, 385 uint16_t cntlid); 386 struct spdk_nvmf_subsystem_listener *nvmf_subsystem_find_listener( 387 struct spdk_nvmf_subsystem *subsystem, 388 const struct spdk_nvme_transport_id *trid); 389 struct spdk_nvmf_listener *nvmf_transport_find_listener( 390 struct spdk_nvmf_transport *transport, 391 const struct spdk_nvme_transport_id *trid); 392 void nvmf_transport_dump_opts(struct spdk_nvmf_transport *transport, struct spdk_json_write_ctx *w, 393 bool named); 394 void nvmf_transport_listen_dump_trid(const struct spdk_nvme_transport_id *trid, 395 struct spdk_json_write_ctx *w); 396 397 /** 398 * Sets the controller ID range for a subsystem. 399 * Valid range is [1, 0xFFEF]. 400 * 401 * May only be performed on subsystems in the INACTIVE state. 402 * 403 * \param subsystem Subsystem to modify. 404 * \param min_cntlid Minimum controller ID. 405 * \param max_cntlid Maximum controller ID. 406 * 407 * \return 0 on success, or negated errno value on failure. 408 */ 409 int nvmf_subsystem_set_cntlid_range(struct spdk_nvmf_subsystem *subsystem, 410 uint16_t min_cntlid, uint16_t max_cntlid); 411 412 int nvmf_ctrlr_async_event_ns_notice(struct spdk_nvmf_ctrlr *ctrlr); 413 int nvmf_ctrlr_async_event_ana_change_notice(struct spdk_nvmf_ctrlr *ctrlr); 414 void nvmf_ctrlr_async_event_discovery_log_change_notice(void *ctx); 415 void nvmf_ctrlr_async_event_reservation_notification(struct spdk_nvmf_ctrlr *ctrlr); 416 417 void nvmf_ns_reservation_request(void *ctx); 418 void nvmf_ctrlr_reservation_notice_log(struct spdk_nvmf_ctrlr *ctrlr, 419 struct spdk_nvmf_ns *ns, 420 enum spdk_nvme_reservation_notification_log_page_type type); 421 422 bool nvmf_ns_is_ptpl_capable(const struct spdk_nvmf_ns *ns); 423 424 /* 425 * Abort zero-copy requests that already got the buffer (received zcopy_start cb), but haven't 426 * started zcopy_end. These requests are kept on the outstanding queue, but are not waiting for a 427 * completion from the bdev layer, so, when a qpair is being disconnected, we need to kick them to 428 * force their completion. 429 */ 430 void nvmf_qpair_abort_pending_zcopy_reqs(struct spdk_nvmf_qpair *qpair); 431 432 /* 433 * Free aer simply frees the rdma resources for the aer without informing the host. 434 * This function should be called when deleting a qpair when one wants to make sure 435 * the qpair is completely empty before freeing the request. The reason we free the 436 * AER without sending a completion is to prevent the host from sending another AER. 437 */ 438 void nvmf_qpair_free_aer(struct spdk_nvmf_qpair *qpair); 439 440 int nvmf_ctrlr_abort_request(struct spdk_nvmf_request *req); 441 442 void nvmf_ctrlr_set_fatal_status(struct spdk_nvmf_ctrlr *ctrlr); 443 444 static inline struct spdk_nvmf_ns * 445 _nvmf_subsystem_get_ns(struct spdk_nvmf_subsystem *subsystem, uint32_t nsid) 446 { 447 /* NOTE: This implicitly also checks for 0, since 0 - 1 wraps around to UINT32_MAX. */ 448 if (spdk_unlikely(nsid - 1 >= subsystem->max_nsid)) { 449 return NULL; 450 } 451 452 return subsystem->ns[nsid - 1]; 453 } 454 455 static inline bool 456 nvmf_qpair_is_admin_queue(struct spdk_nvmf_qpair *qpair) 457 { 458 return qpair->qid == 0; 459 } 460 461 /* 462 * Tests whether a given string represents a valid NQN. 463 */ 464 bool nvmf_nqn_is_valid(const char *nqn); 465 466 /* 467 * Tests whether a given NQN describes a discovery subsystem. 468 */ 469 bool nvmf_nqn_is_discovery(const char *nqn); 470 471 /** 472 * Initiates a zcopy start operation 473 * 474 * \param bdev The \ref spdk_bdev 475 * \param desc The \ref spdk_bdev_desc 476 * \param ch The \ref spdk_io_channel 477 * \param req The \ref spdk_nvmf_request passed to the bdev for processing 478 * 479 * \return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE if the command was completed immediately or 480 * SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS if the command was submitted and will be 481 * completed asynchronously. Asynchronous completions are notified through 482 * spdk_nvmf_request_complete(). 483 */ 484 int nvmf_bdev_ctrlr_zcopy_start(struct spdk_bdev *bdev, 485 struct spdk_bdev_desc *desc, 486 struct spdk_io_channel *ch, 487 struct spdk_nvmf_request *req); 488 489 /** 490 * Ends a zcopy operation 491 * 492 * \param req The NVMe-oF request 493 * \param commit Flag indicating whether the buffers should be committed 494 */ 495 void nvmf_bdev_ctrlr_zcopy_end(struct spdk_nvmf_request *req, bool commit); 496 497 #endif /* __NVMF_INTERNAL_H__ */ 498