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) 2022-2024 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/module/bdev/nvme.h" 17 #include "spdk/jsonrpc.h" 18 19 TAILQ_HEAD(nvme_bdev_ctrlrs, nvme_bdev_ctrlr); 20 extern struct nvme_bdev_ctrlrs g_nvme_bdev_ctrlrs; 21 extern pthread_mutex_t g_bdev_nvme_mutex; 22 extern bool g_bdev_nvme_module_finish; 23 extern struct spdk_thread *g_bdev_nvme_init_thread; 24 25 #define NVME_MAX_CONTROLLERS 1024 26 27 typedef void (*spdk_bdev_nvme_start_discovery_fn)(void *ctx, int status); 28 typedef void (*spdk_bdev_nvme_stop_discovery_fn)(void *ctx); 29 30 struct nvme_async_probe_ctx { 31 struct spdk_nvme_probe_ctx *probe_ctx; 32 char *base_name; 33 const char **names; 34 uint32_t max_bdevs; 35 uint32_t reported_bdevs; 36 struct spdk_poller *poller; 37 struct spdk_nvme_transport_id trid; 38 struct spdk_bdev_nvme_ctrlr_opts bdev_opts; 39 struct spdk_nvme_ctrlr_opts drv_opts; 40 spdk_bdev_nvme_create_cb cb_fn; 41 void *cb_ctx; 42 uint32_t populates_in_progress; 43 bool ctrlr_attached; 44 bool probe_done; 45 bool namespaces_populated; 46 }; 47 48 struct nvme_ns { 49 uint32_t id; 50 struct spdk_nvme_ns *ns; 51 struct nvme_ctrlr *ctrlr; 52 struct nvme_bdev *bdev; 53 uint32_t ana_group_id; 54 enum spdk_nvme_ana_state ana_state; 55 bool ana_state_updating; 56 bool ana_transition_timedout; 57 bool depopulating; 58 struct spdk_poller *anatt_timer; 59 struct nvme_async_probe_ctx *probe_ctx; 60 TAILQ_ENTRY(nvme_ns) tailq; 61 RB_ENTRY(nvme_ns) node; 62 63 /** 64 * record io path stat before destroyed. Allocation of stat is 65 * decided by option io_path_stat of RPC 66 * bdev_nvme_set_options 67 */ 68 struct spdk_bdev_io_stat *stat; 69 }; 70 71 struct nvme_bdev_io; 72 struct nvme_bdev_ctrlr; 73 struct nvme_bdev; 74 struct nvme_io_path; 75 struct nvme_ctrlr_channel_iter; 76 struct nvme_bdev_channel_iter; 77 78 struct nvme_path_id { 79 struct spdk_nvme_transport_id trid; 80 struct spdk_nvme_host_id hostid; 81 TAILQ_ENTRY(nvme_path_id) link; 82 uint64_t last_failed_tsc; 83 }; 84 85 typedef void (*bdev_nvme_ctrlr_op_cb)(void *cb_arg, int rc); 86 typedef void (*nvme_ctrlr_disconnected_cb)(struct nvme_ctrlr *nvme_ctrlr); 87 88 struct nvme_ctrlr { 89 /** 90 * points to pinned, physically contiguous memory region; 91 * contains 4KB IDENTIFY structure for controller which is 92 * target for CONTROLLER IDENTIFY command during initialization 93 */ 94 struct spdk_nvme_ctrlr *ctrlr; 95 struct nvme_path_id *active_path_id; 96 int ref; 97 98 uint32_t resetting : 1; 99 uint32_t reconnect_is_delayed : 1; 100 uint32_t in_failover : 1; 101 uint32_t pending_failover : 1; 102 uint32_t fast_io_fail_timedout : 1; 103 uint32_t destruct : 1; 104 uint32_t ana_log_page_updating : 1; 105 uint32_t io_path_cache_clearing : 1; 106 uint32_t dont_retry : 1; 107 uint32_t disabled : 1; 108 109 struct spdk_bdev_nvme_ctrlr_opts opts; 110 111 RB_HEAD(nvme_ns_tree, nvme_ns) namespaces; 112 113 struct spdk_opal_dev *opal_dev; 114 115 struct spdk_poller *adminq_timer_poller; 116 struct spdk_thread *thread; 117 struct spdk_interrupt *intr; 118 119 bdev_nvme_ctrlr_op_cb ctrlr_op_cb_fn; 120 void *ctrlr_op_cb_arg; 121 /* Poller used to check for reset/detach completion */ 122 struct spdk_poller *reset_detach_poller; 123 struct spdk_nvme_detach_ctx *detach_ctx; 124 125 uint64_t reset_start_tsc; 126 struct spdk_poller *reconnect_delay_timer; 127 128 nvme_ctrlr_disconnected_cb disconnected_cb; 129 130 TAILQ_HEAD(, nvme_bdev_io) pending_resets; 131 132 /** linked list pointer for device list */ 133 TAILQ_ENTRY(nvme_ctrlr) tailq; 134 struct nvme_bdev_ctrlr *nbdev_ctrlr; 135 136 TAILQ_HEAD(nvme_paths, nvme_path_id) trids; 137 138 uint32_t max_ana_log_page_size; 139 struct spdk_nvme_ana_page *ana_log_page; 140 struct spdk_nvme_ana_group_descriptor *copied_ana_desc; 141 142 struct nvme_async_probe_ctx *probe_ctx; 143 struct spdk_key *psk; 144 struct spdk_key *dhchap_key; 145 struct spdk_key *dhchap_ctrlr_key; 146 147 pthread_mutex_t mutex; 148 }; 149 150 struct nvme_bdev_ctrlr { 151 char *name; 152 TAILQ_HEAD(, nvme_ctrlr) ctrlrs; 153 TAILQ_HEAD(, nvme_bdev) bdevs; 154 TAILQ_ENTRY(nvme_bdev_ctrlr) tailq; 155 }; 156 157 struct nvme_error_stat { 158 uint32_t status_type[8]; 159 uint32_t status[4][256]; 160 }; 161 162 struct nvme_bdev { 163 struct spdk_bdev disk; 164 uint32_t nsid; 165 struct nvme_bdev_ctrlr *nbdev_ctrlr; 166 pthread_mutex_t mutex; 167 int ref; 168 enum spdk_bdev_nvme_multipath_policy mp_policy; 169 enum spdk_bdev_nvme_multipath_selector mp_selector; 170 uint32_t rr_min_io; 171 TAILQ_HEAD(, nvme_ns) nvme_ns_list; 172 bool opal; 173 TAILQ_ENTRY(nvme_bdev) tailq; 174 struct nvme_error_stat *err_stat; 175 }; 176 177 struct nvme_qpair { 178 struct nvme_ctrlr *ctrlr; 179 struct spdk_nvme_qpair *qpair; 180 struct nvme_poll_group *group; 181 struct nvme_ctrlr_channel *ctrlr_ch; 182 183 /* The following is used to update io_path cache of nvme_bdev_channels. */ 184 TAILQ_HEAD(, nvme_io_path) io_path_list; 185 186 TAILQ_ENTRY(nvme_qpair) tailq; 187 }; 188 189 struct nvme_ctrlr_channel { 190 struct nvme_qpair *qpair; 191 192 struct nvme_ctrlr_channel_iter *reset_iter; 193 struct spdk_poller *connect_poller; 194 }; 195 196 struct nvme_io_path { 197 struct nvme_ns *nvme_ns; 198 struct nvme_qpair *qpair; 199 STAILQ_ENTRY(nvme_io_path) stailq; 200 201 /* The following are used to update io_path cache of the nvme_bdev_channel. */ 202 struct nvme_bdev_channel *nbdev_ch; 203 TAILQ_ENTRY(nvme_io_path) tailq; 204 205 /* allocation of stat is decided by option io_path_stat of RPC bdev_nvme_set_options */ 206 struct spdk_bdev_io_stat *stat; 207 }; 208 209 struct nvme_bdev_channel { 210 struct nvme_io_path *current_io_path; 211 enum spdk_bdev_nvme_multipath_policy mp_policy; 212 enum spdk_bdev_nvme_multipath_selector mp_selector; 213 uint32_t rr_min_io; 214 uint32_t rr_counter; 215 STAILQ_HEAD(, nvme_io_path) io_path_list; 216 TAILQ_HEAD(retry_io_head, nvme_bdev_io) retry_io_list; 217 struct spdk_poller *retry_io_poller; 218 bool resetting; 219 }; 220 221 struct nvme_poll_group { 222 struct spdk_nvme_poll_group *group; 223 struct spdk_io_channel *accel_channel; 224 struct spdk_poller *poller; 225 bool collect_spin_stat; 226 uint64_t spin_ticks; 227 uint64_t start_ticks; 228 uint64_t end_ticks; 229 TAILQ_HEAD(, nvme_qpair) qpair_list; 230 struct spdk_interrupt *intr; 231 }; 232 233 void nvme_io_path_info_json(struct spdk_json_write_ctx *w, struct nvme_io_path *io_path); 234 235 struct nvme_ctrlr *nvme_ctrlr_get_by_name(const char *name); 236 237 typedef void (*nvme_ctrlr_for_each_channel_msg)(struct nvme_ctrlr_channel_iter *iter, 238 struct nvme_ctrlr *nvme_ctrlr, 239 struct nvme_ctrlr_channel *ctrlr_ch, 240 void *ctx); 241 242 typedef void (*nvme_ctrlr_for_each_channel_done)(struct nvme_ctrlr *nvme_ctrlr, 243 void *ctx, int status); 244 245 void nvme_ctrlr_for_each_channel(struct nvme_ctrlr *nvme_ctrlr, 246 nvme_ctrlr_for_each_channel_msg fn, void *ctx, 247 nvme_ctrlr_for_each_channel_done cpl); 248 249 void nvme_ctrlr_for_each_channel_continue(struct nvme_ctrlr_channel_iter *iter, 250 int status); 251 252 253 typedef void (*nvme_bdev_for_each_channel_msg)(struct nvme_bdev_channel_iter *iter, 254 struct nvme_bdev *nbdev, 255 struct nvme_bdev_channel *nbdev_ch, 256 void *ctx); 257 258 typedef void (*nvme_bdev_for_each_channel_done)(struct nvme_bdev *nbdev, 259 void *ctx, int status); 260 261 void nvme_bdev_for_each_channel(struct nvme_bdev *nbdev, 262 nvme_bdev_for_each_channel_msg fn, void *ctx, 263 nvme_bdev_for_each_channel_done cpl); 264 265 void nvme_bdev_for_each_channel_continue(struct nvme_bdev_channel_iter *iter, 266 int status); 267 268 struct nvme_ctrlr *nvme_bdev_ctrlr_get_ctrlr_by_id(struct nvme_bdev_ctrlr *nbdev_ctrlr, 269 uint16_t cntlid); 270 271 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr_get_by_name(const char *name); 272 273 typedef void (*nvme_bdev_ctrlr_for_each_fn)(struct nvme_bdev_ctrlr *nbdev_ctrlr, void *ctx); 274 275 void nvme_bdev_ctrlr_for_each(nvme_bdev_ctrlr_for_each_fn fn, void *ctx); 276 277 void nvme_bdev_dump_trid_json(const struct spdk_nvme_transport_id *trid, 278 struct spdk_json_write_ctx *w); 279 280 void nvme_ctrlr_info_json(struct spdk_json_write_ctx *w, struct nvme_ctrlr *nvme_ctrlr); 281 282 struct nvme_ns *nvme_ctrlr_get_ns(struct nvme_ctrlr *nvme_ctrlr, uint32_t nsid); 283 struct nvme_ns *nvme_ctrlr_get_first_active_ns(struct nvme_ctrlr *nvme_ctrlr); 284 struct nvme_ns *nvme_ctrlr_get_next_active_ns(struct nvme_ctrlr *nvme_ctrlr, struct nvme_ns *ns); 285 286 struct spdk_nvme_qpair *bdev_nvme_get_io_qpair(struct spdk_io_channel *ctrlr_io_ch); 287 int bdev_nvme_set_hotplug(bool enabled, uint64_t period_us, spdk_msg_fn cb, void *cb_ctx); 288 289 int bdev_nvme_start_discovery(struct spdk_nvme_transport_id *trid, const char *base_name, 290 struct spdk_nvme_ctrlr_opts *drv_opts, struct spdk_bdev_nvme_ctrlr_opts *bdev_opts, 291 uint64_t timeout, bool from_mdns, 292 spdk_bdev_nvme_start_discovery_fn cb_fn, void *cb_ctx); 293 int bdev_nvme_stop_discovery(const char *name, spdk_bdev_nvme_stop_discovery_fn cb_fn, 294 void *cb_ctx); 295 void bdev_nvme_get_discovery_info(struct spdk_json_write_ctx *w); 296 297 int bdev_nvme_start_mdns_discovery(const char *base_name, 298 const char *svcname, 299 struct spdk_nvme_ctrlr_opts *drv_opts, 300 struct spdk_bdev_nvme_ctrlr_opts *bdev_opts); 301 int bdev_nvme_stop_mdns_discovery(const char *name); 302 void bdev_nvme_get_mdns_discovery_info(struct spdk_jsonrpc_request *request); 303 void bdev_nvme_mdns_discovery_config_json(struct spdk_json_write_ctx *w); 304 305 typedef void (*bdev_nvme_set_keys_cb)(void *ctx, int status); 306 307 int bdev_nvme_set_keys(const char *name, const char *dhchap_key, const char *dhchap_ctrlr_key, 308 bdev_nvme_set_keys_cb cb_fn, void *cb_ctx); 309 310 struct spdk_nvme_ctrlr *bdev_nvme_get_ctrlr(struct spdk_bdev *bdev); 311 312 typedef void (*bdev_nvme_delete_done_fn)(void *ctx, int rc); 313 314 /** 315 * Delete NVMe controller with all bdevs on top of it, or delete the specified path 316 * if there is any alternative path. Requires to pass name of NVMe controller. 317 * 318 * \param name NVMe controller name 319 * \param path_id The specified path to remove (optional) 320 * \param delete_done Callback function on delete complete (optional) 321 * \param delete_done_ctx Context passed to callback (optional) 322 * \return zero on success, 323 * -EINVAL on wrong parameters or 324 * -ENODEV if controller is not found or 325 * -ENOMEM on no memory 326 */ 327 int bdev_nvme_delete(const char *name, const struct nvme_path_id *path_id, 328 bdev_nvme_delete_done_fn delete_done, void *delete_done_ctx); 329 330 enum nvme_ctrlr_op { 331 NVME_CTRLR_OP_RESET = 1, 332 NVME_CTRLR_OP_ENABLE, 333 NVME_CTRLR_OP_DISABLE, 334 }; 335 336 /** 337 * Perform specified operation on an NVMe controller. 338 * 339 * NOTE: The callback function is always called after this function returns except for 340 * out of memory cases. 341 * 342 * \param nvme_ctrlr The specified NVMe controller to operate 343 * \param op Operation code 344 * \param cb_fn Function to be called back after operation completes 345 * \param cb_arg Argument for callback function 346 */ 347 void nvme_ctrlr_op_rpc(struct nvme_ctrlr *nvme_ctrlr, enum nvme_ctrlr_op op, 348 bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg); 349 350 /** 351 * Perform specified operation on all NVMe controllers in an NVMe bdev controller. 352 * 353 * NOTE: The callback function is always called after this function returns except for 354 * out of memory cases. 355 * 356 * \param nbdev_ctrlr The specified NVMe bdev 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_bdev_ctrlr_op_rpc(struct nvme_bdev_ctrlr *nbdev_ctrlr, enum nvme_ctrlr_op op, 362 bdev_nvme_ctrlr_op_cb cb_fn, void *cb_arg); 363 364 typedef void (*bdev_nvme_set_preferred_path_cb)(void *cb_arg, int rc); 365 366 /** 367 * Set the preferred I/O path for an NVMe bdev in multipath mode. 368 * 369 * NOTE: This function does not support NVMe bdevs in failover mode. 370 * 371 * \param name NVMe bdev name 372 * \param cntlid NVMe-oF controller ID 373 * \param cb_fn Function to be called back after completion. 374 * \param cb_arg Argument for callback function. 375 */ 376 void bdev_nvme_set_preferred_path(const char *name, uint16_t cntlid, 377 bdev_nvme_set_preferred_path_cb cb_fn, void *cb_arg); 378 379 #endif /* SPDK_BDEV_NVME_H */ 380