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