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) 2022 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 * Copyright (c) 2022 Dell Inc, or its subsidiaries. All rights reserved. 6 */ 7 8 #ifndef SPDK_BDEV_NVME_H 9 #define SPDK_BDEV_NVME_H 10 11 #include "spdk/stdinc.h" 12 13 #include "spdk/queue.h" 14 #include "spdk/nvme.h" 15 #include "spdk/bdev_module.h" 16 #include "spdk/jsonrpc.h" 17 18 TAILQ_HEAD(nvme_bdev_ctrlrs, nvme_bdev_ctrlr); 19 extern struct nvme_bdev_ctrlrs g_nvme_bdev_ctrlrs; 20 extern pthread_mutex_t g_bdev_nvme_mutex; 21 extern bool g_bdev_nvme_module_finish; 22 extern struct spdk_thread *g_bdev_nvme_init_thread; 23 24 #define NVME_MAX_CONTROLLERS 1024 25 26 enum bdev_nvme_multipath_policy { 27 BDEV_NVME_MP_POLICY_ACTIVE_PASSIVE, 28 BDEV_NVME_MP_POLICY_ACTIVE_ACTIVE, 29 }; 30 31 enum bdev_nvme_multipath_selector { 32 BDEV_NVME_MP_SELECTOR_ROUND_ROBIN = 1, 33 BDEV_NVME_MP_SELECTOR_QUEUE_DEPTH, 34 }; 35 36 typedef void (*spdk_bdev_create_nvme_fn)(void *ctx, size_t bdev_count, int rc); 37 typedef void (*spdk_bdev_nvme_start_discovery_fn)(void *ctx, int status); 38 typedef void (*spdk_bdev_nvme_stop_discovery_fn)(void *ctx); 39 40 struct nvme_ctrlr_opts { 41 uint32_t prchk_flags; 42 int32_t ctrlr_loss_timeout_sec; 43 uint32_t reconnect_delay_sec; 44 uint32_t fast_io_fail_timeout_sec; 45 bool from_discovery_service; 46 /* Path to the file containing PSK, used for dumping configuration. */ 47 char psk_path[PATH_MAX]; 48 }; 49 50 struct nvme_async_probe_ctx { 51 struct spdk_nvme_probe_ctx *probe_ctx; 52 const char *base_name; 53 const char **names; 54 uint32_t max_bdevs; 55 uint32_t reported_bdevs; 56 struct spdk_poller *poller; 57 struct spdk_nvme_transport_id trid; 58 struct nvme_ctrlr_opts bdev_opts; 59 struct spdk_nvme_ctrlr_opts drv_opts; 60 spdk_bdev_create_nvme_fn cb_fn; 61 void *cb_ctx; 62 uint32_t populates_in_progress; 63 bool ctrlr_attached; 64 bool probe_done; 65 bool namespaces_populated; 66 }; 67 68 struct nvme_ns { 69 uint32_t id; 70 struct spdk_nvme_ns *ns; 71 struct nvme_ctrlr *ctrlr; 72 struct nvme_bdev *bdev; 73 uint32_t ana_group_id; 74 enum spdk_nvme_ana_state ana_state; 75 bool ana_state_updating; 76 bool ana_transition_timedout; 77 struct spdk_poller *anatt_timer; 78 struct nvme_async_probe_ctx *probe_ctx; 79 TAILQ_ENTRY(nvme_ns) tailq; 80 RB_ENTRY(nvme_ns) node; 81 82 /** 83 * record io path stat before destroyed. Allocation of stat is 84 * decided by option io_path_stat of RPC 85 * bdev_nvme_set_options 86 */ 87 struct spdk_bdev_io_stat *stat; 88 }; 89 90 struct nvme_bdev_io; 91 struct nvme_bdev_ctrlr; 92 struct nvme_bdev; 93 struct nvme_io_path; 94 95 struct nvme_path_id { 96 struct spdk_nvme_transport_id trid; 97 struct spdk_nvme_host_id hostid; 98 TAILQ_ENTRY(nvme_path_id) link; 99 uint64_t last_failed_tsc; 100 }; 101 102 typedef void (*bdev_nvme_ctrlr_op_cb)(void *cb_arg, int rc); 103 typedef void (*nvme_ctrlr_disconnected_cb)(struct nvme_ctrlr *nvme_ctrlr); 104 105 struct nvme_ctrlr { 106 /** 107 * points to pinned, physically contiguous memory region; 108 * contains 4KB IDENTIFY structure for controller which is 109 * target for CONTROLLER IDENTIFY command during initialization 110 */ 111 struct spdk_nvme_ctrlr *ctrlr; 112 struct nvme_path_id *active_path_id; 113 int ref; 114 115 uint32_t resetting : 1; 116 uint32_t reconnect_is_delayed : 1; 117 uint32_t in_failover : 1; 118 uint32_t pending_failover : 1; 119 uint32_t fast_io_fail_timedout : 1; 120 uint32_t destruct : 1; 121 uint32_t ana_log_page_updating : 1; 122 uint32_t io_path_cache_clearing : 1; 123 uint32_t dont_retry : 1; 124 uint32_t disabled : 1; 125 126 struct nvme_ctrlr_opts opts; 127 128 RB_HEAD(nvme_ns_tree, nvme_ns) namespaces; 129 130 struct spdk_opal_dev *opal_dev; 131 132 struct spdk_poller *adminq_timer_poller; 133 struct spdk_thread *thread; 134 135 bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn; 136 void *ctrlr_op_cb_arg; 137 /* Poller used to check for reset/detach completion */ 138 struct spdk_poller *reset_detach_poller; 139 struct spdk_nvme_detach_ctx *detach_ctx; 140 141 uint64_t reset_start_tsc; 142 struct spdk_poller *reconnect_delay_timer; 143 144 nvme_ctrlr_disconnected_cb disconnected_cb; 145 146 /** linked list pointer for device list */ 147 TAILQ_ENTRY(nvme_ctrlr) tailq; 148 struct nvme_bdev_ctrlr *nbdev_ctrlr; 149 150 TAILQ_HEAD(nvme_paths, nvme_path_id) trids; 151 152 uint32_t max_ana_log_page_size; 153 struct spdk_nvme_ana_page *ana_log_page; 154 struct spdk_nvme_ana_group_descriptor *copied_ana_desc; 155 156 struct nvme_async_probe_ctx *probe_ctx; 157 158 pthread_mutex_t mutex; 159 }; 160 161 struct nvme_bdev_ctrlr { 162 char *name; 163 TAILQ_HEAD(, nvme_ctrlr) ctrlrs; 164 TAILQ_HEAD(, nvme_bdev) bdevs; 165 TAILQ_ENTRY(nvme_bdev_ctrlr) tailq; 166 }; 167 168 struct nvme_error_stat { 169 uint32_t status_type[8]; 170 uint32_t status[4][256]; 171 }; 172 173 struct nvme_bdev { 174 struct spdk_bdev disk; 175 uint32_t nsid; 176 struct nvme_bdev_ctrlr *nbdev_ctrlr; 177 pthread_mutex_t mutex; 178 int ref; 179 enum bdev_nvme_multipath_policy mp_policy; 180 enum bdev_nvme_multipath_selector mp_selector; 181 uint32_t rr_min_io; 182 TAILQ_HEAD(, nvme_ns) nvme_ns_list; 183 bool opal; 184 TAILQ_ENTRY(nvme_bdev) tailq; 185 struct nvme_error_stat *err_stat; 186 }; 187 188 struct nvme_qpair { 189 struct nvme_ctrlr *ctrlr; 190 struct spdk_nvme_qpair *qpair; 191 struct nvme_poll_group *group; 192 struct nvme_ctrlr_channel *ctrlr_ch; 193 194 /* The following is used to update io_path cache of nvme_bdev_channels. */ 195 TAILQ_HEAD(, nvme_io_path) io_path_list; 196 197 TAILQ_ENTRY(nvme_qpair) tailq; 198 }; 199 200 struct nvme_ctrlr_channel { 201 struct nvme_qpair *qpair; 202 TAILQ_HEAD(, spdk_bdev_io) pending_resets; 203 204 struct spdk_io_channel_iter *reset_iter; 205 struct spdk_poller *connect_poller; 206 }; 207 208 struct nvme_io_path { 209 struct nvme_ns *nvme_ns; 210 struct nvme_qpair *qpair; 211 STAILQ_ENTRY(nvme_io_path) stailq; 212 213 /* The following are used to update io_path cache of the nvme_bdev_channel. */ 214 struct nvme_bdev_channel *nbdev_ch; 215 TAILQ_ENTRY(nvme_io_path) tailq; 216 217 /* allocation of stat is decided by option io_path_stat of RPC bdev_nvme_set_options */ 218 struct spdk_bdev_io_stat *stat; 219 }; 220 221 struct nvme_bdev_channel { 222 struct nvme_io_path *current_io_path; 223 enum bdev_nvme_multipath_policy mp_policy; 224 enum bdev_nvme_multipath_selector mp_selector; 225 uint32_t rr_min_io; 226 uint32_t rr_counter; 227 STAILQ_HEAD(, nvme_io_path) io_path_list; 228 TAILQ_HEAD(retry_io_head, spdk_bdev_io) retry_io_list; 229 struct spdk_poller *retry_io_poller; 230 }; 231 232 struct nvme_poll_group { 233 struct spdk_nvme_poll_group *group; 234 struct spdk_io_channel *accel_channel; 235 struct spdk_poller *poller; 236 bool collect_spin_stat; 237 uint64_t spin_ticks; 238 uint64_t start_ticks; 239 uint64_t end_ticks; 240 TAILQ_HEAD(, nvme_qpair) qpair_list; 241 }; 242 243 void nvme_io_path_info_json(struct spdk_json_write_ctx *w, struct nvme_io_path *io_path); 244 245 struct nvme_ctrlr *nvme_ctrlr_get_by_name(const char *name); 246 247 struct nvme_ctrlr *nvme_bdev_ctrlr_get_ctrlr_by_id(struct nvme_bdev_ctrlr *nbdev_ctrlr, 248 uint16_t cntlid); 249 250 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr_get_by_name(const char *name); 251 252 typedef void (*nvme_bdev_ctrlr_for_each_fn)(struct nvme_bdev_ctrlr *nbdev_ctrlr, void *ctx); 253 254 void nvme_bdev_ctrlr_for_each(nvme_bdev_ctrlr_for_each_fn fn, void *ctx); 255 256 void nvme_bdev_dump_trid_json(const struct spdk_nvme_transport_id *trid, 257 struct spdk_json_write_ctx *w); 258 259 void nvme_ctrlr_info_json(struct spdk_json_write_ctx *w, struct nvme_ctrlr *nvme_ctrlr); 260 261 struct nvme_ns *nvme_ctrlr_get_ns(struct nvme_ctrlr *nvme_ctrlr, uint32_t nsid); 262 struct nvme_ns *nvme_ctrlr_get_first_active_ns(struct nvme_ctrlr *nvme_ctrlr); 263 struct nvme_ns *nvme_ctrlr_get_next_active_ns(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *ns); 264 265 enum spdk_bdev_timeout_action { 266 SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE = 0, 267 SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET, 268 SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT, 269 }; 270 271 struct spdk_bdev_nvme_opts { 272 enum spdk_bdev_timeout_action action_on_timeout; 273 uint64_t timeout_us; 274 uint64_t timeout_admin_us; 275 uint32_t keep_alive_timeout_ms; 276 /* The number of attempts per I/O in the transport layer before an I/O fails. */ 277 uint32_t transport_retry_count; 278 uint32_t arbitration_burst; 279 uint32_t low_priority_weight; 280 uint32_t medium_priority_weight; 281 uint32_t high_priority_weight; 282 uint64_t nvme_adminq_poll_period_us; 283 uint64_t nvme_ioq_poll_period_us; 284 uint32_t io_queue_requests; 285 bool delay_cmd_submit; 286 /* The number of attempts per I/O in the bdev layer before an I/O fails. */ 287 int32_t bdev_retry_count; 288 uint8_t transport_ack_timeout; 289 int32_t ctrlr_loss_timeout_sec; 290 uint32_t reconnect_delay_sec; 291 uint32_t fast_io_fail_timeout_sec; 292 bool disable_auto_failback; 293 bool generate_uuids; 294 /* Type of Service - RDMA only */ 295 uint8_t transport_tos; 296 bool nvme_error_stat; 297 uint32_t rdma_srq_size; 298 bool io_path_stat; 299 bool allow_accel_sequence; 300 }; 301 302 struct spdk_nvme_qpair *bdev_nvme_get_io_qpair(struct spdk_io_channel *ctrlr_io_ch); 303 void bdev_nvme_get_opts(struct spdk_bdev_nvme_opts *opts); 304 int bdev_nvme_set_opts(const struct spdk_bdev_nvme_opts *opts); 305 int bdev_nvme_set_hotplug(bool enabled, uint64_t period_us, spdk_msg_fn cb, void *cb_ctx); 306 307 void bdev_nvme_get_default_ctrlr_opts(struct nvme_ctrlr_opts *opts); 308 309 int bdev_nvme_create(struct spdk_nvme_transport_id *trid, 310 const char *base_name, 311 const char **names, 312 uint32_t count, 313 spdk_bdev_create_nvme_fn cb_fn, 314 void *cb_ctx, 315 struct spdk_nvme_ctrlr_opts *drv_opts, 316 struct nvme_ctrlr_opts *bdev_opts, 317 bool multipath); 318 319 int bdev_nvme_start_discovery(struct spdk_nvme_transport_id *trid, const char *base_name, 320 struct spdk_nvme_ctrlr_opts *drv_opts, struct nvme_ctrlr_opts *bdev_opts, 321 uint64_t timeout, bool from_mdns, 322 spdk_bdev_nvme_start_discovery_fn cb_fn, void *cb_ctx); 323 int bdev_nvme_stop_discovery(const char *name, spdk_bdev_nvme_stop_discovery_fn cb_fn, 324 void *cb_ctx); 325 void bdev_nvme_get_discovery_info(struct spdk_json_write_ctx *w); 326 327 int bdev_nvme_start_mdns_discovery(const char *base_name, 328 const char *svcname, 329 struct spdk_nvme_ctrlr_opts *drv_opts, 330 struct nvme_ctrlr_opts *bdev_opts); 331 int bdev_nvme_stop_mdns_discovery(const char *name); 332 void bdev_nvme_get_mdns_discovery_info(struct spdk_jsonrpc_request *request); 333 void bdev_nvme_mdns_discovery_config_json(struct spdk_json_write_ctx *w); 334 335 struct spdk_nvme_ctrlr *bdev_nvme_get_ctrlr(struct spdk_bdev *bdev); 336 337 typedef void (*bdev_nvme_delete_done_fn)(void *ctx, int rc); 338 339 /** 340 * Delete NVMe controller with all bdevs on top of it, or delete the specified path 341 * if there is any alternative path. Requires to pass name of NVMe controller. 342 * 343 * \param name NVMe controller name 344 * \param path_id The specified path to remove (optional) 345 * \param delete_done Callback function on delete complete (optional) 346 * \param delete_done_ctx Context passed to callback (optional) 347 * \return zero on success, 348 * -EINVAL on wrong parameters or 349 * -ENODEV if controller is not found or 350 * -ENOMEM on no memory 351 */ 352 int bdev_nvme_delete(const char *name, const struct nvme_path_id *path_id, 353 bdev_nvme_delete_done_fn delete_done, void *delete_done_ctx); 354 355 enum nvme_ctrlr_op { 356 NVME_CTRLR_OP_RESET = 1, 357 NVME_CTRLR_OP_ENABLE, 358 NVME_CTRLR_OP_DISABLE, 359 }; 360 361 /** 362 * Perform specified operation on an NVMe controller. 363 * 364 * NOTE: The callback function is always called after this function returns except for 365 * out of memory cases. 366 * 367 * \param nvme_ctrlr The specified NVMe controller to operate 368 * \param op Operation code 369 * \param cb_fn Function to be called back after operation completes 370 * \param cb_arg Argument for callback function 371 */ 372 void nvme_ctrlr_op_rpc(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op, 373 bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg); 374 375 /** 376 * Perform specified operation on all NVMe controllers in an NVMe bdev controller. 377 * 378 * NOTE: The callback function is always called after this function returns except for 379 * out of memory cases. 380 * 381 * \param nbdev_ctrlr The specified NVMe bdev controller to operate 382 * \param op Operation code 383 * \param cb_fn Function to be called back after operation completes 384 * \param cb_arg Argument for callback function 385 */ 386 void nvme_bdev_ctrlr_op_rpc(struct nvme_bdev_ctrlr *nbdev_ctrlr, enum nvme_ctrlr_op op, 387 bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg); 388 389 typedef void (*bdev_nvme_set_preferred_path_cb)(void *cb_arg, int rc); 390 391 /** 392 * Set the preferred I/O path for an NVMe bdev in multipath mode. 393 * 394 * NOTE: This function does not support NVMe bdevs in failover mode. 395 * 396 * \param name NVMe bdev name 397 * \param cntlid NVMe-oF controller ID 398 * \param cb_fn Function to be called back after completion. 399 * \param cb_arg Argument for callback function. 400 */ 401 void bdev_nvme_set_preferred_path(const char *name, uint16_t cntlid, 402 bdev_nvme_set_preferred_path_cb cb_fn, void *cb_arg); 403 404 typedef void (*bdev_nvme_set_multipath_policy_cb)(void *cb_arg, int rc); 405 406 /** 407 * Set multipath policy of the NVMe bdev. 408 * 409 * \param name NVMe bdev name 410 * \param policy Multipath policy (active-passive or active-active) 411 * \param selector Multipath selector (round_robin, queue_depth) 412 * \param rr_min_io Number of IO to route to a path before switching to another for round-robin 413 * \param cb_fn Function to be called back after completion. 414 */ 415 void bdev_nvme_set_multipath_policy(const char *name, 416 enum bdev_nvme_multipath_policy policy, 417 enum bdev_nvme_multipath_selector selector, 418 uint32_t rr_min_io, 419 bdev_nvme_set_multipath_policy_cb cb_fn, 420 void *cb_arg); 421 422 #endif /* SPDK_BDEV_NVME_H */ 423