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