1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. All rights reserved. 5 * Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * * Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * * Neither the name of Intel Corporation nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 32 */ 33 34 #include "spdk/stdinc.h" 35 36 #include "bdev_nvme.h" 37 #include "bdev_ocssd.h" 38 39 #include "spdk/accel_engine.h" 40 #include "spdk/config.h" 41 #include "spdk/endian.h" 42 #include "spdk/bdev.h" 43 #include "spdk/json.h" 44 #include "spdk/nvme.h" 45 #include "spdk/nvme_ocssd.h" 46 #include "spdk/nvme_zns.h" 47 #include "spdk/thread.h" 48 #include "spdk/string.h" 49 #include "spdk/util.h" 50 51 #include "spdk/bdev_module.h" 52 #include "spdk/log.h" 53 54 #define SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT true 55 #define SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS (10000) 56 57 static int bdev_nvme_config_json(struct spdk_json_write_ctx *w); 58 59 struct nvme_bdev_io { 60 /** array of iovecs to transfer. */ 61 struct iovec *iovs; 62 63 /** Number of iovecs in iovs array. */ 64 int iovcnt; 65 66 /** Current iovec position. */ 67 int iovpos; 68 69 /** Offset in current iovec. */ 70 uint32_t iov_offset; 71 72 /** array of iovecs to transfer. */ 73 struct iovec *fused_iovs; 74 75 /** Number of iovecs in iovs array. */ 76 int fused_iovcnt; 77 78 /** Current iovec position. */ 79 int fused_iovpos; 80 81 /** Offset in current iovec. */ 82 uint32_t fused_iov_offset; 83 84 /** Saved status for admin passthru completion event, PI error verification, or intermediate compare-and-write status */ 85 struct spdk_nvme_cpl cpl; 86 87 /** Originating thread */ 88 struct spdk_thread *orig_thread; 89 90 /** Keeps track if first of fused commands was submitted */ 91 bool first_fused_submitted; 92 93 /** Temporary pointer to zone report buffer */ 94 struct spdk_nvme_zns_zone_report *zone_report_buf; 95 96 /** Keep track of how many zones that have been copied to the spdk_bdev_zone_info struct */ 97 uint64_t handled_zones; 98 }; 99 100 struct nvme_probe_ctx { 101 size_t count; 102 struct spdk_nvme_transport_id trids[NVME_MAX_CONTROLLERS]; 103 struct spdk_nvme_host_id hostids[NVME_MAX_CONTROLLERS]; 104 const char *names[NVME_MAX_CONTROLLERS]; 105 uint32_t prchk_flags[NVME_MAX_CONTROLLERS]; 106 const char *hostnqn; 107 }; 108 109 struct nvme_probe_skip_entry { 110 struct spdk_nvme_transport_id trid; 111 TAILQ_ENTRY(nvme_probe_skip_entry) tailq; 112 }; 113 /* All the controllers deleted by users via RPC are skipped by hotplug monitor */ 114 static TAILQ_HEAD(, nvme_probe_skip_entry) g_skipped_nvme_ctrlrs = TAILQ_HEAD_INITIALIZER( 115 g_skipped_nvme_ctrlrs); 116 117 static struct spdk_bdev_nvme_opts g_opts = { 118 .action_on_timeout = SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE, 119 .timeout_us = 0, 120 .keep_alive_timeout_ms = SPDK_BDEV_NVME_DEFAULT_KEEP_ALIVE_TIMEOUT_IN_MS, 121 .retry_count = 4, 122 .arbitration_burst = 0, 123 .low_priority_weight = 0, 124 .medium_priority_weight = 0, 125 .high_priority_weight = 0, 126 .nvme_adminq_poll_period_us = 10000ULL, 127 .nvme_ioq_poll_period_us = 0, 128 .io_queue_requests = 0, 129 .delay_cmd_submit = SPDK_BDEV_NVME_DEFAULT_DELAY_CMD_SUBMIT, 130 }; 131 132 #define NVME_HOTPLUG_POLL_PERIOD_MAX 10000000ULL 133 #define NVME_HOTPLUG_POLL_PERIOD_DEFAULT 100000ULL 134 135 static int g_hot_insert_nvme_controller_index = 0; 136 static uint64_t g_nvme_hotplug_poll_period_us = NVME_HOTPLUG_POLL_PERIOD_DEFAULT; 137 static bool g_nvme_hotplug_enabled = false; 138 static struct spdk_thread *g_bdev_nvme_init_thread; 139 static struct spdk_poller *g_hotplug_poller; 140 static struct spdk_poller *g_hotplug_probe_poller; 141 static struct spdk_nvme_probe_ctx *g_hotplug_probe_ctx; 142 143 static void nvme_ctrlr_populate_namespaces(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 144 struct nvme_async_probe_ctx *ctx); 145 static void nvme_ctrlr_populate_namespaces_done(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 146 struct nvme_async_probe_ctx *ctx); 147 static int bdev_nvme_library_init(void); 148 static void bdev_nvme_library_fini(void); 149 static int bdev_nvme_readv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 150 struct nvme_bdev_io *bio, 151 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba, 152 uint32_t flags); 153 static int bdev_nvme_no_pi_readv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 154 struct nvme_bdev_io *bio, 155 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba); 156 static int bdev_nvme_writev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 157 struct nvme_bdev_io *bio, 158 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba, 159 uint32_t flags); 160 static int bdev_nvme_zone_appendv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 161 struct nvme_bdev_io *bio, 162 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, 163 uint64_t zslba, uint32_t flags); 164 static int bdev_nvme_comparev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 165 struct nvme_bdev_io *bio, 166 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba, 167 uint32_t flags); 168 static int bdev_nvme_comparev_and_writev(struct spdk_nvme_ns *ns, 169 struct spdk_nvme_qpair *qpair, 170 struct nvme_bdev_io *bio, struct iovec *cmp_iov, int cmp_iovcnt, struct iovec *write_iov, 171 int write_iovcnt, void *md, uint64_t lba_count, uint64_t lba, 172 uint32_t flags); 173 static int bdev_nvme_get_zone_info(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 174 struct nvme_bdev_io *bio, uint64_t zone_id, uint32_t num_zones, 175 struct spdk_bdev_zone_info *info); 176 static int bdev_nvme_zone_management(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 177 struct nvme_bdev_io *bio, uint64_t zone_id, 178 enum spdk_bdev_zone_action action); 179 static int bdev_nvme_admin_passthru(struct nvme_io_channel *nvme_ch, 180 struct nvme_bdev_io *bio, 181 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes); 182 static int bdev_nvme_io_passthru(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 183 struct nvme_bdev_io *bio, 184 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes); 185 static int bdev_nvme_io_passthru_md(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 186 struct nvme_bdev_io *bio, 187 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len); 188 static int bdev_nvme_abort(struct nvme_io_channel *nvme_ch, 189 struct nvme_bdev_io *bio, struct nvme_bdev_io *bio_to_abort); 190 static int bdev_nvme_reset(struct nvme_io_channel *nvme_ch, struct nvme_bdev_io *bio); 191 static int bdev_nvme_failover(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, bool remove); 192 static void remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr); 193 194 typedef void (*populate_namespace_fn)(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 195 struct nvme_bdev_ns *nvme_ns, struct nvme_async_probe_ctx *ctx); 196 static void nvme_ctrlr_populate_standard_namespace(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 197 struct nvme_bdev_ns *nvme_ns, struct nvme_async_probe_ctx *ctx); 198 199 static populate_namespace_fn g_populate_namespace_fn[] = { 200 NULL, 201 nvme_ctrlr_populate_standard_namespace, 202 bdev_ocssd_populate_namespace, 203 }; 204 205 typedef void (*depopulate_namespace_fn)(struct nvme_bdev_ns *nvme_ns); 206 static void nvme_ctrlr_depopulate_standard_namespace(struct nvme_bdev_ns *nvme_ns); 207 208 static depopulate_namespace_fn g_depopulate_namespace_fn[] = { 209 NULL, 210 nvme_ctrlr_depopulate_standard_namespace, 211 bdev_ocssd_depopulate_namespace, 212 }; 213 214 typedef void (*config_json_namespace_fn)(struct spdk_json_write_ctx *w, 215 struct nvme_bdev_ns *nvme_ns); 216 static void nvme_ctrlr_config_json_standard_namespace(struct spdk_json_write_ctx *w, 217 struct nvme_bdev_ns *nvme_ns); 218 219 static config_json_namespace_fn g_config_json_namespace_fn[] = { 220 NULL, 221 nvme_ctrlr_config_json_standard_namespace, 222 bdev_ocssd_namespace_config_json, 223 }; 224 225 struct spdk_nvme_qpair * 226 bdev_nvme_get_io_qpair(struct spdk_io_channel *ctrlr_io_ch) 227 { 228 struct nvme_io_channel *nvme_ch; 229 230 assert(ctrlr_io_ch != NULL); 231 232 nvme_ch = spdk_io_channel_get_ctx(ctrlr_io_ch); 233 234 return nvme_ch->qpair; 235 } 236 237 static int 238 bdev_nvme_get_ctx_size(void) 239 { 240 return sizeof(struct nvme_bdev_io); 241 } 242 243 static struct spdk_bdev_module nvme_if = { 244 .name = "nvme", 245 .async_fini = true, 246 .module_init = bdev_nvme_library_init, 247 .module_fini = bdev_nvme_library_fini, 248 .config_json = bdev_nvme_config_json, 249 .get_ctx_size = bdev_nvme_get_ctx_size, 250 251 }; 252 SPDK_BDEV_MODULE_REGISTER(nvme, &nvme_if) 253 254 static void 255 bdev_nvme_disconnected_qpair_cb(struct spdk_nvme_qpair *qpair, void *poll_group_ctx) 256 { 257 int rc; 258 259 SPDK_DEBUGLOG(bdev_nvme, "qpair %p is disconnected, attempting reconnect.\n", qpair); 260 /* 261 * Currently, just try to reconnect indefinitely. If we are doing a reset, the reset will 262 * reconnect a qpair and we will stop getting a callback for this one. 263 */ 264 rc = spdk_nvme_ctrlr_reconnect_io_qpair(qpair); 265 if (rc != 0) { 266 SPDK_WARNLOG("Failed to reconnect to qpair %p, errno %d\n", qpair, -rc); 267 } 268 } 269 270 static int 271 bdev_nvme_poll(void *arg) 272 { 273 struct nvme_bdev_poll_group *group = arg; 274 int64_t num_completions; 275 276 if (group->collect_spin_stat && group->start_ticks == 0) { 277 group->start_ticks = spdk_get_ticks(); 278 } 279 280 num_completions = spdk_nvme_poll_group_process_completions(group->group, 0, 281 bdev_nvme_disconnected_qpair_cb); 282 if (group->collect_spin_stat) { 283 if (num_completions > 0) { 284 if (group->end_ticks != 0) { 285 group->spin_ticks += (group->end_ticks - group->start_ticks); 286 group->end_ticks = 0; 287 } 288 group->start_ticks = 0; 289 } else { 290 group->end_ticks = spdk_get_ticks(); 291 } 292 } 293 294 return num_completions > 0 ? SPDK_POLLER_BUSY : SPDK_POLLER_IDLE; 295 } 296 297 static int 298 bdev_nvme_poll_adminq(void *arg) 299 { 300 int32_t rc; 301 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = arg; 302 303 assert(nvme_bdev_ctrlr != NULL); 304 305 rc = spdk_nvme_ctrlr_process_admin_completions(nvme_bdev_ctrlr->ctrlr); 306 if (rc < 0) { 307 bdev_nvme_failover(nvme_bdev_ctrlr, false); 308 } 309 310 return rc == 0 ? SPDK_POLLER_IDLE : SPDK_POLLER_BUSY; 311 } 312 313 static int 314 bdev_nvme_destruct(void *ctx) 315 { 316 struct nvme_bdev *nvme_disk = ctx; 317 struct nvme_bdev_ns *nvme_ns = nvme_disk->nvme_ns; 318 319 pthread_mutex_lock(&nvme_ns->ctrlr->mutex); 320 321 nvme_ns->bdev = NULL; 322 323 if (!nvme_ns->populated) { 324 pthread_mutex_unlock(&nvme_ns->ctrlr->mutex); 325 326 nvme_bdev_ctrlr_destruct(nvme_ns->ctrlr); 327 } else { 328 pthread_mutex_unlock(&nvme_ns->ctrlr->mutex); 329 } 330 331 free(nvme_disk->disk.name); 332 free(nvme_disk); 333 334 return 0; 335 } 336 337 static int 338 bdev_nvme_flush(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 339 struct nvme_bdev_io *bio, uint64_t offset, uint64_t nbytes) 340 { 341 spdk_bdev_io_complete(spdk_bdev_io_from_ctx(bio), SPDK_BDEV_IO_STATUS_SUCCESS); 342 343 return 0; 344 } 345 346 static int 347 bdev_nvme_create_qpair(struct nvme_io_channel *nvme_ch) 348 { 349 struct spdk_nvme_ctrlr *ctrlr = nvme_ch->ctrlr->ctrlr; 350 struct spdk_nvme_io_qpair_opts opts; 351 struct spdk_nvme_qpair *qpair; 352 int rc; 353 354 spdk_nvme_ctrlr_get_default_io_qpair_opts(ctrlr, &opts, sizeof(opts)); 355 opts.delay_cmd_submit = g_opts.delay_cmd_submit; 356 opts.create_only = true; 357 opts.io_queue_requests = spdk_max(g_opts.io_queue_requests, opts.io_queue_requests); 358 g_opts.io_queue_requests = opts.io_queue_requests; 359 360 qpair = spdk_nvme_ctrlr_alloc_io_qpair(ctrlr, &opts, sizeof(opts)); 361 if (qpair == NULL) { 362 return -1; 363 } 364 365 assert(nvme_ch->group != NULL); 366 367 rc = spdk_nvme_poll_group_add(nvme_ch->group->group, qpair); 368 if (rc != 0) { 369 SPDK_ERRLOG("Unable to begin polling on NVMe Channel.\n"); 370 goto err; 371 } 372 373 rc = spdk_nvme_ctrlr_connect_io_qpair(ctrlr, qpair); 374 if (rc != 0) { 375 SPDK_ERRLOG("Unable to connect I/O qpair.\n"); 376 goto err; 377 } 378 379 nvme_ch->qpair = qpair; 380 381 return 0; 382 383 err: 384 spdk_nvme_ctrlr_free_io_qpair(qpair); 385 386 return rc; 387 } 388 389 static int 390 bdev_nvme_destroy_qpair(struct nvme_io_channel *nvme_ch) 391 { 392 int rc; 393 394 if (nvme_ch->qpair == NULL) { 395 return 0; 396 } 397 398 rc = spdk_nvme_ctrlr_free_io_qpair(nvme_ch->qpair); 399 if (!rc) { 400 nvme_ch->qpair = NULL; 401 } 402 return rc; 403 } 404 405 static void 406 _bdev_nvme_check_pending_destruct(struct spdk_io_channel_iter *i, int status) 407 { 408 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = spdk_io_channel_iter_get_ctx(i); 409 410 pthread_mutex_lock(&nvme_bdev_ctrlr->mutex); 411 if (nvme_bdev_ctrlr->destruct_after_reset) { 412 assert(nvme_bdev_ctrlr->ref == 0 && nvme_bdev_ctrlr->destruct); 413 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 414 415 spdk_thread_send_msg(nvme_bdev_ctrlr->thread, nvme_bdev_ctrlr_unregister, 416 nvme_bdev_ctrlr); 417 } else { 418 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 419 } 420 } 421 422 static void 423 _bdev_nvme_complete_pending_resets(struct nvme_io_channel *nvme_ch, 424 enum spdk_bdev_io_status status) 425 { 426 struct spdk_bdev_io *bdev_io; 427 428 while (!TAILQ_EMPTY(&nvme_ch->pending_resets)) { 429 bdev_io = TAILQ_FIRST(&nvme_ch->pending_resets); 430 TAILQ_REMOVE(&nvme_ch->pending_resets, bdev_io, module_link); 431 spdk_bdev_io_complete(bdev_io, status); 432 } 433 } 434 435 static void 436 bdev_nvme_complete_pending_resets(struct spdk_io_channel_iter *i) 437 { 438 struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i); 439 struct nvme_io_channel *nvme_ch = spdk_io_channel_get_ctx(_ch); 440 441 _bdev_nvme_complete_pending_resets(nvme_ch, SPDK_BDEV_IO_STATUS_SUCCESS); 442 443 spdk_for_each_channel_continue(i, 0); 444 } 445 446 static void 447 bdev_nvme_abort_pending_resets(struct spdk_io_channel_iter *i) 448 { 449 struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i); 450 struct nvme_io_channel *nvme_ch = spdk_io_channel_get_ctx(_ch); 451 452 _bdev_nvme_complete_pending_resets(nvme_ch, SPDK_BDEV_IO_STATUS_FAILED); 453 454 spdk_for_each_channel_continue(i, 0); 455 } 456 457 static void 458 _bdev_nvme_reset_complete(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, int rc) 459 { 460 struct nvme_bdev_ctrlr_trid *curr_trid; 461 struct nvme_bdev_io *bio = nvme_bdev_ctrlr->reset_bio; 462 enum spdk_bdev_io_status io_status = SPDK_BDEV_IO_STATUS_SUCCESS; 463 464 nvme_bdev_ctrlr->reset_bio = NULL; 465 466 if (rc) { 467 SPDK_ERRLOG("Resetting controller failed.\n"); 468 io_status = SPDK_BDEV_IO_STATUS_FAILED; 469 } else { 470 SPDK_NOTICELOG("Resetting controller successful.\n"); 471 } 472 473 pthread_mutex_lock(&nvme_bdev_ctrlr->mutex); 474 nvme_bdev_ctrlr->resetting = false; 475 nvme_bdev_ctrlr->failover_in_progress = false; 476 477 curr_trid = TAILQ_FIRST(&nvme_bdev_ctrlr->trids); 478 assert(curr_trid != NULL); 479 assert(&curr_trid->trid == nvme_bdev_ctrlr->connected_trid); 480 481 curr_trid->is_failed = rc != 0 ? true : false; 482 483 if (nvme_bdev_ctrlr->ref == 0 && nvme_bdev_ctrlr->destruct) { 484 /* Destruct ctrlr after clearing pending resets. */ 485 nvme_bdev_ctrlr->destruct_after_reset = true; 486 } 487 488 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 489 490 if (bio) { 491 spdk_bdev_io_complete(spdk_bdev_io_from_ctx(bio), io_status); 492 } 493 494 /* Make sure we clear any pending resets before returning. */ 495 spdk_for_each_channel(nvme_bdev_ctrlr, 496 rc == 0 ? bdev_nvme_complete_pending_resets : 497 bdev_nvme_abort_pending_resets, 498 nvme_bdev_ctrlr, 499 _bdev_nvme_check_pending_destruct); 500 } 501 502 static void 503 _bdev_nvme_reset_create_qpairs_done(struct spdk_io_channel_iter *i, int status) 504 { 505 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = spdk_io_channel_iter_get_ctx(i); 506 507 _bdev_nvme_reset_complete(nvme_bdev_ctrlr, status); 508 } 509 510 static void 511 _bdev_nvme_reset_create_qpair(struct spdk_io_channel_iter *i) 512 { 513 struct spdk_io_channel *_ch = spdk_io_channel_iter_get_channel(i); 514 struct nvme_io_channel *nvme_ch = spdk_io_channel_get_ctx(_ch); 515 int rc; 516 517 rc = bdev_nvme_create_qpair(nvme_ch); 518 519 spdk_for_each_channel_continue(i, rc); 520 } 521 522 static void 523 _bdev_nvme_reset_ctrlr(struct spdk_io_channel_iter *i, int status) 524 { 525 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = spdk_io_channel_iter_get_ctx(i); 526 int rc; 527 528 if (status) { 529 rc = status; 530 goto err; 531 } 532 533 rc = spdk_nvme_ctrlr_reset(nvme_bdev_ctrlr->ctrlr); 534 if (rc != 0) { 535 goto err; 536 } 537 538 /* Recreate all of the I/O queue pairs */ 539 spdk_for_each_channel(nvme_bdev_ctrlr, 540 _bdev_nvme_reset_create_qpair, 541 nvme_bdev_ctrlr, 542 _bdev_nvme_reset_create_qpairs_done); 543 return; 544 545 err: 546 _bdev_nvme_reset_complete(nvme_bdev_ctrlr, rc); 547 } 548 549 static void 550 _bdev_nvme_reset_destroy_qpair(struct spdk_io_channel_iter *i) 551 { 552 struct spdk_io_channel *ch = spdk_io_channel_iter_get_channel(i); 553 struct nvme_io_channel *nvme_ch = spdk_io_channel_get_ctx(ch); 554 int rc; 555 556 rc = bdev_nvme_destroy_qpair(nvme_ch); 557 558 spdk_for_each_channel_continue(i, rc); 559 } 560 561 static int 562 _bdev_nvme_reset(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr) 563 { 564 pthread_mutex_lock(&nvme_bdev_ctrlr->mutex); 565 if (nvme_bdev_ctrlr->destruct) { 566 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 567 return -EBUSY; 568 } 569 570 if (nvme_bdev_ctrlr->resetting) { 571 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 572 SPDK_NOTICELOG("Unable to perform reset, already in progress.\n"); 573 return -EAGAIN; 574 } 575 576 nvme_bdev_ctrlr->resetting = true; 577 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 578 579 /* First, delete all NVMe I/O queue pairs. */ 580 spdk_for_each_channel(nvme_bdev_ctrlr, 581 _bdev_nvme_reset_destroy_qpair, 582 nvme_bdev_ctrlr, 583 _bdev_nvme_reset_ctrlr); 584 585 return 0; 586 } 587 588 static int 589 bdev_nvme_reset(struct nvme_io_channel *nvme_ch, struct nvme_bdev_io *bio) 590 { 591 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio); 592 int rc; 593 594 rc = _bdev_nvme_reset(nvme_ch->ctrlr); 595 if (rc == 0) { 596 assert(nvme_ch->ctrlr->reset_bio == NULL); 597 nvme_ch->ctrlr->reset_bio = bio; 598 } else if (rc == -EBUSY) { 599 /* Don't bother resetting if the controller is in the process of being destructed. */ 600 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 601 } else if (rc == -EAGAIN) { 602 /* 603 * Reset call is queued only if it is from the app framework. This is on purpose so that 604 * we don't interfere with the app framework reset strategy. i.e. we are deferring to the 605 * upper level. If they are in the middle of a reset, we won't try to schedule another one. 606 */ 607 TAILQ_INSERT_TAIL(&nvme_ch->pending_resets, bdev_io, module_link); 608 } else { 609 return rc; 610 } 611 612 return 0; 613 } 614 615 static int 616 _bdev_nvme_failover_start(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, bool remove) 617 { 618 struct nvme_bdev_ctrlr_trid *curr_trid = NULL, *next_trid = NULL; 619 int rc; 620 621 pthread_mutex_lock(&nvme_bdev_ctrlr->mutex); 622 if (nvme_bdev_ctrlr->destruct) { 623 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 624 /* Don't bother resetting if the controller is in the process of being destructed. */ 625 return -EBUSY; 626 } 627 628 curr_trid = TAILQ_FIRST(&nvme_bdev_ctrlr->trids); 629 assert(curr_trid); 630 assert(&curr_trid->trid == nvme_bdev_ctrlr->connected_trid); 631 next_trid = TAILQ_NEXT(curr_trid, link); 632 633 if (nvme_bdev_ctrlr->resetting) { 634 if (next_trid && !nvme_bdev_ctrlr->failover_in_progress) { 635 rc = -EAGAIN; 636 } else { 637 rc = -EBUSY; 638 } 639 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 640 SPDK_NOTICELOG("Unable to perform reset, already in progress.\n"); 641 return rc; 642 } 643 644 nvme_bdev_ctrlr->resetting = true; 645 curr_trid->is_failed = true; 646 647 if (next_trid) { 648 assert(curr_trid->trid.trtype != SPDK_NVME_TRANSPORT_PCIE); 649 650 SPDK_NOTICELOG("Start failover from %s:%s to %s:%s\n", curr_trid->trid.traddr, 651 curr_trid->trid.trsvcid, next_trid->trid.traddr, next_trid->trid.trsvcid); 652 653 nvme_bdev_ctrlr->failover_in_progress = true; 654 spdk_nvme_ctrlr_fail(nvme_bdev_ctrlr->ctrlr); 655 nvme_bdev_ctrlr->connected_trid = &next_trid->trid; 656 rc = spdk_nvme_ctrlr_set_trid(nvme_bdev_ctrlr->ctrlr, &next_trid->trid); 657 assert(rc == 0); 658 TAILQ_REMOVE(&nvme_bdev_ctrlr->trids, curr_trid, link); 659 if (!remove) { 660 /** Shuffle the old trid to the end of the list and use the new one. 661 * Allows for round robin through multiple connections. 662 */ 663 TAILQ_INSERT_TAIL(&nvme_bdev_ctrlr->trids, curr_trid, link); 664 } else { 665 free(curr_trid); 666 } 667 } 668 669 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 670 return 0; 671 } 672 673 static int 674 bdev_nvme_failover(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, bool remove) 675 { 676 int rc; 677 678 rc = _bdev_nvme_failover_start(nvme_bdev_ctrlr, remove); 679 if (rc == 0) { 680 /* First, delete all NVMe I/O queue pairs. */ 681 spdk_for_each_channel(nvme_bdev_ctrlr, 682 _bdev_nvme_reset_destroy_qpair, 683 nvme_bdev_ctrlr, 684 _bdev_nvme_reset_ctrlr); 685 } else if (rc != -EBUSY) { 686 return rc; 687 } 688 689 return 0; 690 } 691 692 static int 693 bdev_nvme_unmap(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 694 struct nvme_bdev_io *bio, 695 uint64_t offset_blocks, 696 uint64_t num_blocks); 697 698 static void 699 bdev_nvme_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, 700 bool success) 701 { 702 struct spdk_bdev *bdev = bdev_io->bdev; 703 struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev->ctxt; 704 struct nvme_io_channel *nvme_ch = spdk_io_channel_get_ctx(ch); 705 struct nvme_bdev_ns *nvme_ns; 706 struct spdk_nvme_qpair *qpair; 707 int ret; 708 709 if (!success) { 710 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 711 return; 712 } 713 714 if (spdk_unlikely(!bdev_nvme_find_io_path(nbdev, nvme_ch, &nvme_ns, &qpair))) { 715 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 716 return; 717 } 718 719 ret = bdev_nvme_readv(nvme_ns->ns, 720 qpair, 721 (struct nvme_bdev_io *)bdev_io->driver_ctx, 722 bdev_io->u.bdev.iovs, 723 bdev_io->u.bdev.iovcnt, 724 bdev_io->u.bdev.md_buf, 725 bdev_io->u.bdev.num_blocks, 726 bdev_io->u.bdev.offset_blocks, 727 bdev->dif_check_flags); 728 729 if (spdk_likely(ret == 0)) { 730 return; 731 } else if (ret == -ENOMEM) { 732 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_NOMEM); 733 } else { 734 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 735 } 736 } 737 738 static int 739 _bdev_nvme_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) 740 { 741 struct nvme_io_channel *nvme_ch = spdk_io_channel_get_ctx(ch); 742 struct spdk_bdev *bdev = bdev_io->bdev; 743 struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev->ctxt; 744 struct nvme_bdev_io *nbdev_io = (struct nvme_bdev_io *)bdev_io->driver_ctx; 745 struct nvme_bdev_io *nbdev_io_to_abort; 746 struct nvme_bdev_ns *nvme_ns; 747 struct spdk_nvme_qpair *qpair; 748 749 if (spdk_unlikely(!bdev_nvme_find_io_path(nbdev, nvme_ch, &nvme_ns, &qpair))) { 750 return -1; 751 } 752 753 switch (bdev_io->type) { 754 case SPDK_BDEV_IO_TYPE_READ: 755 if (bdev_io->u.bdev.iovs && bdev_io->u.bdev.iovs[0].iov_base) { 756 return bdev_nvme_readv(nvme_ns->ns, 757 qpair, 758 nbdev_io, 759 bdev_io->u.bdev.iovs, 760 bdev_io->u.bdev.iovcnt, 761 bdev_io->u.bdev.md_buf, 762 bdev_io->u.bdev.num_blocks, 763 bdev_io->u.bdev.offset_blocks, 764 bdev->dif_check_flags); 765 } else { 766 spdk_bdev_io_get_buf(bdev_io, bdev_nvme_get_buf_cb, 767 bdev_io->u.bdev.num_blocks * bdev->blocklen); 768 return 0; 769 } 770 771 case SPDK_BDEV_IO_TYPE_WRITE: 772 return bdev_nvme_writev(nvme_ns->ns, 773 qpair, 774 nbdev_io, 775 bdev_io->u.bdev.iovs, 776 bdev_io->u.bdev.iovcnt, 777 bdev_io->u.bdev.md_buf, 778 bdev_io->u.bdev.num_blocks, 779 bdev_io->u.bdev.offset_blocks, 780 bdev->dif_check_flags); 781 782 case SPDK_BDEV_IO_TYPE_COMPARE: 783 return bdev_nvme_comparev(nvme_ns->ns, 784 qpair, 785 nbdev_io, 786 bdev_io->u.bdev.iovs, 787 bdev_io->u.bdev.iovcnt, 788 bdev_io->u.bdev.md_buf, 789 bdev_io->u.bdev.num_blocks, 790 bdev_io->u.bdev.offset_blocks, 791 bdev->dif_check_flags); 792 793 case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE: 794 return bdev_nvme_comparev_and_writev(nvme_ns->ns, 795 qpair, 796 nbdev_io, 797 bdev_io->u.bdev.iovs, 798 bdev_io->u.bdev.iovcnt, 799 bdev_io->u.bdev.fused_iovs, 800 bdev_io->u.bdev.fused_iovcnt, 801 bdev_io->u.bdev.md_buf, 802 bdev_io->u.bdev.num_blocks, 803 bdev_io->u.bdev.offset_blocks, 804 bdev->dif_check_flags); 805 806 case SPDK_BDEV_IO_TYPE_UNMAP: 807 return bdev_nvme_unmap(nvme_ns->ns, 808 qpair, 809 nbdev_io, 810 bdev_io->u.bdev.offset_blocks, 811 bdev_io->u.bdev.num_blocks); 812 813 case SPDK_BDEV_IO_TYPE_RESET: 814 return bdev_nvme_reset(nvme_ch, nbdev_io); 815 816 case SPDK_BDEV_IO_TYPE_FLUSH: 817 return bdev_nvme_flush(nvme_ns->ns, 818 qpair, 819 nbdev_io, 820 bdev_io->u.bdev.offset_blocks, 821 bdev_io->u.bdev.num_blocks); 822 823 case SPDK_BDEV_IO_TYPE_ZONE_APPEND: 824 return bdev_nvme_zone_appendv(nvme_ns->ns, 825 qpair, 826 nbdev_io, 827 bdev_io->u.bdev.iovs, 828 bdev_io->u.bdev.iovcnt, 829 bdev_io->u.bdev.md_buf, 830 bdev_io->u.bdev.num_blocks, 831 bdev_io->u.bdev.offset_blocks, 832 bdev->dif_check_flags); 833 834 case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO: 835 return bdev_nvme_get_zone_info(nvme_ns->ns, 836 qpair, 837 nbdev_io, 838 bdev_io->u.zone_mgmt.zone_id, 839 bdev_io->u.zone_mgmt.num_zones, 840 bdev_io->u.zone_mgmt.buf); 841 842 case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT: 843 return bdev_nvme_zone_management(nvme_ns->ns, 844 qpair, 845 nbdev_io, 846 bdev_io->u.zone_mgmt.zone_id, 847 bdev_io->u.zone_mgmt.zone_action); 848 849 case SPDK_BDEV_IO_TYPE_NVME_ADMIN: 850 return bdev_nvme_admin_passthru(nvme_ch, 851 nbdev_io, 852 &bdev_io->u.nvme_passthru.cmd, 853 bdev_io->u.nvme_passthru.buf, 854 bdev_io->u.nvme_passthru.nbytes); 855 856 case SPDK_BDEV_IO_TYPE_NVME_IO: 857 return bdev_nvme_io_passthru(nvme_ns->ns, 858 qpair, 859 nbdev_io, 860 &bdev_io->u.nvme_passthru.cmd, 861 bdev_io->u.nvme_passthru.buf, 862 bdev_io->u.nvme_passthru.nbytes); 863 864 case SPDK_BDEV_IO_TYPE_NVME_IO_MD: 865 return bdev_nvme_io_passthru_md(nvme_ns->ns, 866 qpair, 867 nbdev_io, 868 &bdev_io->u.nvme_passthru.cmd, 869 bdev_io->u.nvme_passthru.buf, 870 bdev_io->u.nvme_passthru.nbytes, 871 bdev_io->u.nvme_passthru.md_buf, 872 bdev_io->u.nvme_passthru.md_len); 873 874 case SPDK_BDEV_IO_TYPE_ABORT: 875 nbdev_io_to_abort = (struct nvme_bdev_io *)bdev_io->u.abort.bio_to_abort->driver_ctx; 876 return bdev_nvme_abort(nvme_ch, 877 nbdev_io, 878 nbdev_io_to_abort); 879 880 default: 881 return -EINVAL; 882 } 883 return 0; 884 } 885 886 static void 887 bdev_nvme_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) 888 { 889 int rc = _bdev_nvme_submit_request(ch, bdev_io); 890 891 if (spdk_unlikely(rc != 0)) { 892 if (rc == -ENOMEM) { 893 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_NOMEM); 894 } else { 895 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 896 } 897 } 898 } 899 900 static bool 901 bdev_nvme_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type) 902 { 903 struct nvme_bdev *nbdev = ctx; 904 struct nvme_bdev_ns *nvme_ns; 905 struct spdk_nvme_ns *ns; 906 struct spdk_nvme_ctrlr *ctrlr; 907 const struct spdk_nvme_ctrlr_data *cdata; 908 909 nvme_ns = nvme_bdev_to_bdev_ns(nbdev); 910 assert(nvme_ns != NULL); 911 ns = nvme_ns->ns; 912 ctrlr = spdk_nvme_ns_get_ctrlr(ns); 913 914 switch (io_type) { 915 case SPDK_BDEV_IO_TYPE_READ: 916 case SPDK_BDEV_IO_TYPE_WRITE: 917 case SPDK_BDEV_IO_TYPE_RESET: 918 case SPDK_BDEV_IO_TYPE_FLUSH: 919 case SPDK_BDEV_IO_TYPE_NVME_ADMIN: 920 case SPDK_BDEV_IO_TYPE_NVME_IO: 921 case SPDK_BDEV_IO_TYPE_ABORT: 922 return true; 923 924 case SPDK_BDEV_IO_TYPE_COMPARE: 925 return spdk_nvme_ns_supports_compare(ns); 926 927 case SPDK_BDEV_IO_TYPE_NVME_IO_MD: 928 return spdk_nvme_ns_get_md_size(ns) ? true : false; 929 930 case SPDK_BDEV_IO_TYPE_UNMAP: 931 cdata = spdk_nvme_ctrlr_get_data(ctrlr); 932 return cdata->oncs.dsm; 933 934 case SPDK_BDEV_IO_TYPE_WRITE_ZEROES: 935 /* 936 * The NVMe controller write_zeroes function is currently not used by our driver. 937 * NVMe write zeroes is limited to 16-bit block count, and the bdev layer currently 938 * has no mechanism for reporting a max write zeroes block count, nor ability to 939 * split a write zeroes request. 940 */ 941 return false; 942 943 case SPDK_BDEV_IO_TYPE_COMPARE_AND_WRITE: 944 if (spdk_nvme_ctrlr_get_flags(ctrlr) & 945 SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED) { 946 return true; 947 } 948 return false; 949 950 case SPDK_BDEV_IO_TYPE_GET_ZONE_INFO: 951 case SPDK_BDEV_IO_TYPE_ZONE_MANAGEMENT: 952 return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS; 953 954 case SPDK_BDEV_IO_TYPE_ZONE_APPEND: 955 return spdk_nvme_ns_get_csi(ns) == SPDK_NVME_CSI_ZNS && 956 spdk_nvme_ctrlr_get_flags(ctrlr) & SPDK_NVME_CTRLR_ZONE_APPEND_SUPPORTED; 957 958 default: 959 return false; 960 } 961 } 962 963 static int 964 bdev_nvme_create_cb(void *io_device, void *ctx_buf) 965 { 966 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = io_device; 967 struct nvme_io_channel *nvme_ch = ctx_buf; 968 struct spdk_io_channel *pg_ch = NULL; 969 int rc; 970 971 if (spdk_nvme_ctrlr_is_ocssd_supported(nvme_bdev_ctrlr->ctrlr)) { 972 rc = bdev_ocssd_create_io_channel(nvme_ch); 973 if (rc != 0) { 974 return rc; 975 } 976 } 977 978 pg_ch = spdk_get_io_channel(&g_nvme_bdev_ctrlrs); 979 if (!pg_ch) { 980 rc = -1; 981 goto err_pg_ch; 982 } 983 984 nvme_ch->group = spdk_io_channel_get_ctx(pg_ch); 985 986 #ifdef SPDK_CONFIG_VTUNE 987 nvme_ch->group->collect_spin_stat = true; 988 #else 989 nvme_ch->group->collect_spin_stat = false; 990 #endif 991 992 TAILQ_INIT(&nvme_ch->pending_resets); 993 994 nvme_ch->ctrlr = nvme_bdev_ctrlr; 995 996 rc = bdev_nvme_create_qpair(nvme_ch); 997 if (rc != 0) { 998 goto err_qpair; 999 } 1000 1001 return 0; 1002 1003 err_qpair: 1004 spdk_put_io_channel(pg_ch); 1005 err_pg_ch: 1006 if (nvme_ch->ocssd_ch) { 1007 bdev_ocssd_destroy_io_channel(nvme_ch); 1008 } 1009 1010 return rc; 1011 } 1012 1013 static void 1014 bdev_nvme_destroy_cb(void *io_device, void *ctx_buf) 1015 { 1016 struct nvme_io_channel *nvme_ch = ctx_buf; 1017 1018 assert(nvme_ch->group != NULL); 1019 1020 if (nvme_ch->ocssd_ch != NULL) { 1021 bdev_ocssd_destroy_io_channel(nvme_ch); 1022 } 1023 1024 bdev_nvme_destroy_qpair(nvme_ch); 1025 1026 spdk_put_io_channel(spdk_io_channel_from_ctx(nvme_ch->group)); 1027 } 1028 1029 static void 1030 bdev_nvme_poll_group_submit_accel_crc32c(void *ctx, uint32_t *dst, struct iovec *iov, 1031 uint32_t iov_cnt, uint32_t seed, 1032 spdk_nvme_accel_completion_cb cb_fn, void *cb_arg) 1033 { 1034 struct nvme_bdev_poll_group *group = ctx; 1035 int rc; 1036 1037 assert(group->accel_channel != NULL); 1038 assert(cb_fn != NULL); 1039 1040 rc = spdk_accel_submit_crc32cv(group->accel_channel, dst, iov, iov_cnt, seed, cb_fn, cb_arg); 1041 if (rc) { 1042 /* For the two cases, spdk_accel_submit_crc32cv does not call the user's cb_fn */ 1043 if (rc == -ENOMEM || rc == -EINVAL) { 1044 cb_fn(cb_arg, rc); 1045 } 1046 SPDK_ERRLOG("Cannot complete the accelerated crc32c operation with iov=%p\n", iov); 1047 } 1048 } 1049 1050 static struct spdk_nvme_accel_fn_table g_bdev_nvme_accel_fn_table = { 1051 .table_size = sizeof(struct spdk_nvme_accel_fn_table), 1052 .submit_accel_crc32c = bdev_nvme_poll_group_submit_accel_crc32c, 1053 }; 1054 1055 static int 1056 bdev_nvme_poll_group_create_cb(void *io_device, void *ctx_buf) 1057 { 1058 struct nvme_bdev_poll_group *group = ctx_buf; 1059 1060 group->group = spdk_nvme_poll_group_create(group, &g_bdev_nvme_accel_fn_table); 1061 if (group->group == NULL) { 1062 return -1; 1063 } 1064 1065 group->accel_channel = spdk_accel_engine_get_io_channel(); 1066 if (!group->accel_channel) { 1067 spdk_nvme_poll_group_destroy(group->group); 1068 SPDK_ERRLOG("Cannot get the accel_channel for bdev nvme polling group=%p\n", 1069 group); 1070 return -1; 1071 } 1072 1073 group->poller = SPDK_POLLER_REGISTER(bdev_nvme_poll, group, g_opts.nvme_ioq_poll_period_us); 1074 1075 if (group->poller == NULL) { 1076 spdk_put_io_channel(group->accel_channel); 1077 spdk_nvme_poll_group_destroy(group->group); 1078 return -1; 1079 } 1080 1081 return 0; 1082 } 1083 1084 static void 1085 bdev_nvme_poll_group_destroy_cb(void *io_device, void *ctx_buf) 1086 { 1087 struct nvme_bdev_poll_group *group = ctx_buf; 1088 1089 if (group->accel_channel) { 1090 spdk_put_io_channel(group->accel_channel); 1091 } 1092 1093 spdk_poller_unregister(&group->poller); 1094 if (spdk_nvme_poll_group_destroy(group->group)) { 1095 SPDK_ERRLOG("Unable to destroy a poll group for the NVMe bdev module.\n"); 1096 assert(false); 1097 } 1098 } 1099 1100 static struct spdk_io_channel * 1101 bdev_nvme_get_io_channel(void *ctx) 1102 { 1103 struct nvme_bdev *nvme_bdev = ctx; 1104 1105 return spdk_get_io_channel(nvme_bdev->nvme_ns->ctrlr); 1106 } 1107 1108 static void * 1109 bdev_nvme_get_module_ctx(void *ctx) 1110 { 1111 struct nvme_bdev *nvme_bdev = ctx; 1112 1113 return bdev_nvme_get_ctrlr(&nvme_bdev->disk); 1114 } 1115 1116 static int 1117 bdev_nvme_dump_info_json(void *ctx, struct spdk_json_write_ctx *w) 1118 { 1119 struct nvme_bdev *nvme_bdev = ctx; 1120 struct nvme_bdev_ns *nvme_ns; 1121 struct spdk_nvme_ns *ns; 1122 struct spdk_nvme_ctrlr *ctrlr; 1123 const struct spdk_nvme_ctrlr_data *cdata; 1124 const struct spdk_nvme_transport_id *trid; 1125 union spdk_nvme_vs_register vs; 1126 union spdk_nvme_csts_register csts; 1127 char buf[128]; 1128 1129 nvme_ns = nvme_bdev_to_bdev_ns(nvme_bdev); 1130 assert(nvme_ns != NULL); 1131 ns = nvme_ns->ns; 1132 ctrlr = spdk_nvme_ns_get_ctrlr(ns); 1133 1134 cdata = spdk_nvme_ctrlr_get_data(ctrlr); 1135 trid = spdk_nvme_ctrlr_get_transport_id(ctrlr); 1136 vs = spdk_nvme_ctrlr_get_regs_vs(ctrlr); 1137 csts = spdk_nvme_ctrlr_get_regs_csts(ctrlr); 1138 1139 spdk_json_write_named_object_begin(w, "nvme"); 1140 1141 if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) { 1142 spdk_json_write_named_string(w, "pci_address", trid->traddr); 1143 } 1144 1145 spdk_json_write_named_object_begin(w, "trid"); 1146 1147 nvme_bdev_dump_trid_json(trid, w); 1148 1149 spdk_json_write_object_end(w); 1150 1151 #ifdef SPDK_CONFIG_NVME_CUSE 1152 size_t cuse_name_size = 128; 1153 char cuse_name[cuse_name_size]; 1154 1155 int rc = spdk_nvme_cuse_get_ns_name(ctrlr, spdk_nvme_ns_get_id(ns), 1156 cuse_name, &cuse_name_size); 1157 if (rc == 0) { 1158 spdk_json_write_named_string(w, "cuse_device", cuse_name); 1159 } 1160 #endif 1161 1162 spdk_json_write_named_object_begin(w, "ctrlr_data"); 1163 1164 spdk_json_write_named_string_fmt(w, "vendor_id", "0x%04x", cdata->vid); 1165 1166 snprintf(buf, sizeof(cdata->mn) + 1, "%s", cdata->mn); 1167 spdk_str_trim(buf); 1168 spdk_json_write_named_string(w, "model_number", buf); 1169 1170 snprintf(buf, sizeof(cdata->sn) + 1, "%s", cdata->sn); 1171 spdk_str_trim(buf); 1172 spdk_json_write_named_string(w, "serial_number", buf); 1173 1174 snprintf(buf, sizeof(cdata->fr) + 1, "%s", cdata->fr); 1175 spdk_str_trim(buf); 1176 spdk_json_write_named_string(w, "firmware_revision", buf); 1177 1178 if (cdata->subnqn[0] != '\0') { 1179 spdk_json_write_named_string(w, "subnqn", cdata->subnqn); 1180 } 1181 1182 spdk_json_write_named_object_begin(w, "oacs"); 1183 1184 spdk_json_write_named_uint32(w, "security", cdata->oacs.security); 1185 spdk_json_write_named_uint32(w, "format", cdata->oacs.format); 1186 spdk_json_write_named_uint32(w, "firmware", cdata->oacs.firmware); 1187 spdk_json_write_named_uint32(w, "ns_manage", cdata->oacs.ns_manage); 1188 1189 spdk_json_write_object_end(w); 1190 1191 spdk_json_write_object_end(w); 1192 1193 spdk_json_write_named_object_begin(w, "vs"); 1194 1195 spdk_json_write_name(w, "nvme_version"); 1196 if (vs.bits.ter) { 1197 spdk_json_write_string_fmt(w, "%u.%u.%u", vs.bits.mjr, vs.bits.mnr, vs.bits.ter); 1198 } else { 1199 spdk_json_write_string_fmt(w, "%u.%u", vs.bits.mjr, vs.bits.mnr); 1200 } 1201 1202 spdk_json_write_object_end(w); 1203 1204 spdk_json_write_named_object_begin(w, "csts"); 1205 1206 spdk_json_write_named_uint32(w, "rdy", csts.bits.rdy); 1207 spdk_json_write_named_uint32(w, "cfs", csts.bits.cfs); 1208 1209 spdk_json_write_object_end(w); 1210 1211 spdk_json_write_named_object_begin(w, "ns_data"); 1212 1213 spdk_json_write_named_uint32(w, "id", spdk_nvme_ns_get_id(ns)); 1214 1215 spdk_json_write_object_end(w); 1216 1217 if (cdata->oacs.security) { 1218 spdk_json_write_named_object_begin(w, "security"); 1219 1220 spdk_json_write_named_bool(w, "opal", nvme_bdev->opal); 1221 1222 spdk_json_write_object_end(w); 1223 } 1224 1225 spdk_json_write_object_end(w); 1226 1227 return 0; 1228 } 1229 1230 static void 1231 bdev_nvme_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w) 1232 { 1233 /* No config per bdev needed */ 1234 } 1235 1236 static uint64_t 1237 bdev_nvme_get_spin_time(struct spdk_io_channel *ch) 1238 { 1239 struct nvme_io_channel *nvme_ch = spdk_io_channel_get_ctx(ch); 1240 struct nvme_bdev_poll_group *group = nvme_ch->group; 1241 uint64_t spin_time; 1242 1243 if (!group || !group->collect_spin_stat) { 1244 return 0; 1245 } 1246 1247 if (group->end_ticks != 0) { 1248 group->spin_ticks += (group->end_ticks - group->start_ticks); 1249 group->end_ticks = 0; 1250 } 1251 1252 spin_time = (group->spin_ticks * 1000000ULL) / spdk_get_ticks_hz(); 1253 group->start_ticks = 0; 1254 group->spin_ticks = 0; 1255 1256 return spin_time; 1257 } 1258 1259 static const struct spdk_bdev_fn_table nvmelib_fn_table = { 1260 .destruct = bdev_nvme_destruct, 1261 .submit_request = bdev_nvme_submit_request, 1262 .io_type_supported = bdev_nvme_io_type_supported, 1263 .get_io_channel = bdev_nvme_get_io_channel, 1264 .dump_info_json = bdev_nvme_dump_info_json, 1265 .write_config_json = bdev_nvme_write_config_json, 1266 .get_spin_time = bdev_nvme_get_spin_time, 1267 .get_module_ctx = bdev_nvme_get_module_ctx, 1268 }; 1269 1270 static int 1271 nvme_disk_create(struct spdk_bdev *disk, const char *base_name, 1272 struct spdk_nvme_ctrlr *ctrlr, struct spdk_nvme_ns *ns, 1273 uint32_t prchk_flags, void *ctx) 1274 { 1275 const struct spdk_uuid *uuid; 1276 const struct spdk_nvme_ctrlr_data *cdata; 1277 const struct spdk_nvme_ns_data *nsdata; 1278 int rc; 1279 enum spdk_nvme_csi csi; 1280 1281 cdata = spdk_nvme_ctrlr_get_data(ctrlr); 1282 csi = spdk_nvme_ns_get_csi(ns); 1283 1284 switch (csi) { 1285 case SPDK_NVME_CSI_NVM: 1286 disk->product_name = "NVMe disk"; 1287 break; 1288 case SPDK_NVME_CSI_ZNS: 1289 disk->product_name = "NVMe ZNS disk"; 1290 disk->zoned = true; 1291 disk->zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns); 1292 disk->max_zone_append_size = spdk_nvme_zns_ctrlr_get_max_zone_append_size(ctrlr) / 1293 spdk_nvme_ns_get_extended_sector_size(ns); 1294 disk->max_open_zones = spdk_nvme_zns_ns_get_max_open_zones(ns); 1295 disk->max_active_zones = spdk_nvme_zns_ns_get_max_active_zones(ns); 1296 break; 1297 default: 1298 SPDK_ERRLOG("unsupported CSI: %u\n", csi); 1299 return -ENOTSUP; 1300 } 1301 1302 disk->name = spdk_sprintf_alloc("%sn%d", base_name, spdk_nvme_ns_get_id(ns)); 1303 if (!disk->name) { 1304 return -ENOMEM; 1305 } 1306 1307 disk->write_cache = 0; 1308 if (cdata->vwc.present) { 1309 /* Enable if the Volatile Write Cache exists */ 1310 disk->write_cache = 1; 1311 } 1312 disk->blocklen = spdk_nvme_ns_get_extended_sector_size(ns); 1313 disk->blockcnt = spdk_nvme_ns_get_num_sectors(ns); 1314 disk->optimal_io_boundary = spdk_nvme_ns_get_optimal_io_boundary(ns); 1315 1316 uuid = spdk_nvme_ns_get_uuid(ns); 1317 if (uuid != NULL) { 1318 disk->uuid = *uuid; 1319 } 1320 1321 nsdata = spdk_nvme_ns_get_data(ns); 1322 1323 disk->md_len = spdk_nvme_ns_get_md_size(ns); 1324 if (disk->md_len != 0) { 1325 disk->md_interleave = nsdata->flbas.extended; 1326 disk->dif_type = (enum spdk_dif_type)spdk_nvme_ns_get_pi_type(ns); 1327 if (disk->dif_type != SPDK_DIF_DISABLE) { 1328 disk->dif_is_head_of_md = nsdata->dps.md_start; 1329 disk->dif_check_flags = prchk_flags; 1330 } 1331 } 1332 1333 if (!(spdk_nvme_ctrlr_get_flags(ctrlr) & 1334 SPDK_NVME_CTRLR_COMPARE_AND_WRITE_SUPPORTED)) { 1335 disk->acwu = 0; 1336 } else if (nsdata->nsfeat.ns_atomic_write_unit) { 1337 disk->acwu = nsdata->nacwu; 1338 } else { 1339 disk->acwu = cdata->acwu; 1340 } 1341 1342 disk->ctxt = ctx; 1343 disk->fn_table = &nvmelib_fn_table; 1344 disk->module = &nvme_if; 1345 rc = spdk_bdev_register(disk); 1346 if (rc) { 1347 SPDK_ERRLOG("spdk_bdev_register() failed\n"); 1348 free(disk->name); 1349 return rc; 1350 } 1351 1352 return 0; 1353 } 1354 1355 static int 1356 nvme_bdev_create(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, struct nvme_bdev_ns *nvme_ns) 1357 { 1358 struct nvme_bdev *bdev; 1359 int rc; 1360 1361 bdev = calloc(1, sizeof(*bdev)); 1362 if (!bdev) { 1363 SPDK_ERRLOG("bdev calloc() failed\n"); 1364 return -ENOMEM; 1365 } 1366 1367 bdev->nvme_ns = nvme_ns; 1368 bdev->opal = nvme_bdev_ctrlr->opal_dev != NULL; 1369 1370 rc = nvme_disk_create(&bdev->disk, nvme_bdev_ctrlr->name, nvme_bdev_ctrlr->ctrlr, 1371 nvme_ns->ns, nvme_bdev_ctrlr->prchk_flags, bdev); 1372 if (rc != 0) { 1373 SPDK_ERRLOG("Failed to create NVMe disk\n"); 1374 free(bdev); 1375 return rc; 1376 } 1377 1378 nvme_ns->bdev = bdev; 1379 1380 return 0; 1381 } 1382 1383 static void 1384 nvme_ctrlr_populate_standard_namespace(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 1385 struct nvme_bdev_ns *nvme_ns, struct nvme_async_probe_ctx *ctx) 1386 { 1387 struct spdk_nvme_ctrlr *ctrlr = nvme_bdev_ctrlr->ctrlr; 1388 struct spdk_nvme_ns *ns; 1389 int rc = 0; 1390 1391 ns = spdk_nvme_ctrlr_get_ns(ctrlr, nvme_ns->id); 1392 if (!ns) { 1393 SPDK_DEBUGLOG(bdev_nvme, "Invalid NS %d\n", nvme_ns->id); 1394 rc = -EINVAL; 1395 goto done; 1396 } 1397 1398 nvme_ns->ns = ns; 1399 nvme_ns->populated = true; 1400 1401 rc = nvme_bdev_create(nvme_bdev_ctrlr, nvme_ns); 1402 done: 1403 nvme_ctrlr_populate_namespace_done(ctx, nvme_ns, rc); 1404 } 1405 1406 static bool 1407 hotplug_probe_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid, 1408 struct spdk_nvme_ctrlr_opts *opts) 1409 { 1410 struct nvme_probe_skip_entry *entry; 1411 1412 TAILQ_FOREACH(entry, &g_skipped_nvme_ctrlrs, tailq) { 1413 if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) { 1414 return false; 1415 } 1416 } 1417 1418 opts->arbitration_burst = (uint8_t)g_opts.arbitration_burst; 1419 opts->low_priority_weight = (uint8_t)g_opts.low_priority_weight; 1420 opts->medium_priority_weight = (uint8_t)g_opts.medium_priority_weight; 1421 opts->high_priority_weight = (uint8_t)g_opts.high_priority_weight; 1422 1423 SPDK_DEBUGLOG(bdev_nvme, "Attaching to %s\n", trid->traddr); 1424 1425 return true; 1426 } 1427 1428 static void 1429 nvme_abort_cpl(void *ctx, const struct spdk_nvme_cpl *cpl) 1430 { 1431 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = ctx; 1432 1433 if (spdk_nvme_cpl_is_error(cpl)) { 1434 SPDK_WARNLOG("Abort failed. Resetting controller. sc is %u, sct is %u.\n", cpl->status.sc, 1435 cpl->status.sct); 1436 _bdev_nvme_reset(nvme_bdev_ctrlr); 1437 } 1438 } 1439 1440 static void 1441 timeout_cb(void *cb_arg, struct spdk_nvme_ctrlr *ctrlr, 1442 struct spdk_nvme_qpair *qpair, uint16_t cid) 1443 { 1444 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = cb_arg; 1445 union spdk_nvme_csts_register csts; 1446 int rc; 1447 1448 assert(nvme_bdev_ctrlr->ctrlr == ctrlr); 1449 1450 SPDK_WARNLOG("Warning: Detected a timeout. ctrlr=%p qpair=%p cid=%u\n", ctrlr, qpair, cid); 1451 1452 /* Only try to read CSTS if it's a PCIe controller or we have a timeout on an I/O 1453 * queue. (Note: qpair == NULL when there's an admin cmd timeout.) Otherwise we 1454 * would submit another fabrics cmd on the admin queue to read CSTS and check for its 1455 * completion recursively. 1456 */ 1457 if (nvme_bdev_ctrlr->connected_trid->trtype == SPDK_NVME_TRANSPORT_PCIE || qpair != NULL) { 1458 csts = spdk_nvme_ctrlr_get_regs_csts(ctrlr); 1459 if (csts.bits.cfs) { 1460 SPDK_ERRLOG("Controller Fatal Status, reset required\n"); 1461 _bdev_nvme_reset(nvme_bdev_ctrlr); 1462 return; 1463 } 1464 } 1465 1466 switch (g_opts.action_on_timeout) { 1467 case SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT: 1468 if (qpair) { 1469 /* Don't send abort to ctrlr when reset is running. */ 1470 pthread_mutex_lock(&nvme_bdev_ctrlr->mutex); 1471 if (nvme_bdev_ctrlr->resetting) { 1472 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 1473 SPDK_NOTICELOG("Quit abort. Ctrlr is in the process of reseting.\n"); 1474 return; 1475 } 1476 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 1477 1478 rc = spdk_nvme_ctrlr_cmd_abort(ctrlr, qpair, cid, 1479 nvme_abort_cpl, nvme_bdev_ctrlr); 1480 if (rc == 0) { 1481 return; 1482 } 1483 1484 SPDK_ERRLOG("Unable to send abort. Resetting, rc is %d.\n", rc); 1485 } 1486 1487 /* FALLTHROUGH */ 1488 case SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET: 1489 _bdev_nvme_reset(nvme_bdev_ctrlr); 1490 break; 1491 case SPDK_BDEV_NVME_TIMEOUT_ACTION_NONE: 1492 SPDK_DEBUGLOG(bdev_nvme, "No action for nvme controller timeout.\n"); 1493 break; 1494 default: 1495 SPDK_ERRLOG("An invalid timeout action value is found.\n"); 1496 break; 1497 } 1498 } 1499 1500 static void 1501 nvme_ctrlr_depopulate_standard_namespace(struct nvme_bdev_ns *nvme_ns) 1502 { 1503 struct nvme_bdev *bdev; 1504 1505 bdev = nvme_ns->bdev; 1506 if (bdev != NULL) { 1507 spdk_bdev_unregister(&bdev->disk, NULL, NULL); 1508 } 1509 1510 nvme_ctrlr_depopulate_namespace_done(nvme_ns); 1511 } 1512 1513 static void 1514 nvme_ctrlr_populate_namespace(struct nvme_bdev_ctrlr *ctrlr, struct nvme_bdev_ns *nvme_ns, 1515 struct nvme_async_probe_ctx *ctx) 1516 { 1517 g_populate_namespace_fn[nvme_ns->type](ctrlr, nvme_ns, ctx); 1518 } 1519 1520 static void 1521 nvme_ctrlr_depopulate_namespace(struct nvme_bdev_ctrlr *ctrlr, struct nvme_bdev_ns *nvme_ns) 1522 { 1523 g_depopulate_namespace_fn[nvme_ns->type](nvme_ns); 1524 } 1525 1526 void 1527 nvme_ctrlr_populate_namespace_done(struct nvme_async_probe_ctx *ctx, 1528 struct nvme_bdev_ns *nvme_ns, int rc) 1529 { 1530 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = nvme_ns->ctrlr; 1531 1532 assert(nvme_bdev_ctrlr != NULL); 1533 1534 if (rc == 0) { 1535 pthread_mutex_lock(&nvme_bdev_ctrlr->mutex); 1536 nvme_bdev_ctrlr->ref++; 1537 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 1538 } else { 1539 memset(nvme_ns, 0, sizeof(*nvme_ns)); 1540 } 1541 1542 if (ctx) { 1543 ctx->populates_in_progress--; 1544 if (ctx->populates_in_progress == 0) { 1545 nvme_ctrlr_populate_namespaces_done(nvme_bdev_ctrlr, ctx); 1546 } 1547 } 1548 } 1549 1550 static void 1551 nvme_ctrlr_populate_namespaces(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 1552 struct nvme_async_probe_ctx *ctx) 1553 { 1554 struct spdk_nvme_ctrlr *ctrlr = nvme_bdev_ctrlr->ctrlr; 1555 struct nvme_bdev_ns *nvme_ns; 1556 struct spdk_nvme_ns *ns; 1557 struct nvme_bdev *bdev; 1558 uint32_t i; 1559 int rc; 1560 uint64_t num_sectors; 1561 bool ns_is_active; 1562 1563 if (ctx) { 1564 /* Initialize this count to 1 to handle the populate functions 1565 * calling nvme_ctrlr_populate_namespace_done() immediately. 1566 */ 1567 ctx->populates_in_progress = 1; 1568 } 1569 1570 for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) { 1571 uint32_t nsid = i + 1; 1572 1573 nvme_ns = nvme_bdev_ctrlr->namespaces[i]; 1574 ns_is_active = spdk_nvme_ctrlr_is_active_ns(ctrlr, nsid); 1575 1576 if (nvme_ns->populated && ns_is_active && nvme_ns->type == NVME_BDEV_NS_STANDARD) { 1577 /* NS is still there but attributes may have changed */ 1578 ns = spdk_nvme_ctrlr_get_ns(ctrlr, nsid); 1579 num_sectors = spdk_nvme_ns_get_num_sectors(ns); 1580 bdev = nvme_ns->bdev; 1581 assert(bdev != NULL); 1582 if (bdev->disk.blockcnt != num_sectors) { 1583 SPDK_NOTICELOG("NSID %u is resized: bdev name %s, old size %" PRIu64 ", new size %" PRIu64 "\n", 1584 nsid, 1585 bdev->disk.name, 1586 bdev->disk.blockcnt, 1587 num_sectors); 1588 rc = spdk_bdev_notify_blockcnt_change(&bdev->disk, num_sectors); 1589 if (rc != 0) { 1590 SPDK_ERRLOG("Could not change num blocks for nvme bdev: name %s, errno: %d.\n", 1591 bdev->disk.name, rc); 1592 } 1593 } 1594 } 1595 1596 if (!nvme_ns->populated && ns_is_active) { 1597 nvme_ns->id = nsid; 1598 nvme_ns->ctrlr = nvme_bdev_ctrlr; 1599 if (spdk_nvme_ctrlr_is_ocssd_supported(ctrlr)) { 1600 nvme_ns->type = NVME_BDEV_NS_OCSSD; 1601 } else { 1602 nvme_ns->type = NVME_BDEV_NS_STANDARD; 1603 } 1604 1605 nvme_ns->bdev = NULL; 1606 1607 if (ctx) { 1608 ctx->populates_in_progress++; 1609 } 1610 nvme_ctrlr_populate_namespace(nvme_bdev_ctrlr, nvme_ns, ctx); 1611 } 1612 1613 if (nvme_ns->populated && !ns_is_active) { 1614 nvme_ctrlr_depopulate_namespace(nvme_bdev_ctrlr, nvme_ns); 1615 } 1616 } 1617 1618 if (ctx) { 1619 /* Decrement this count now that the loop is over to account 1620 * for the one we started with. If the count is then 0, we 1621 * know any populate_namespace functions completed immediately, 1622 * so we'll kick the callback here. 1623 */ 1624 ctx->populates_in_progress--; 1625 if (ctx->populates_in_progress == 0) { 1626 nvme_ctrlr_populate_namespaces_done(nvme_bdev_ctrlr, ctx); 1627 } 1628 } 1629 1630 } 1631 1632 static void 1633 nvme_ctrlr_depopulate_namespaces(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr) 1634 { 1635 uint32_t i; 1636 struct nvme_bdev_ns *nvme_ns; 1637 1638 for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) { 1639 uint32_t nsid = i + 1; 1640 1641 nvme_ns = nvme_bdev_ctrlr->namespaces[nsid - 1]; 1642 if (nvme_ns->populated) { 1643 assert(nvme_ns->id == nsid); 1644 nvme_ctrlr_depopulate_namespace(nvme_bdev_ctrlr, nvme_ns); 1645 } 1646 } 1647 } 1648 1649 static void 1650 aer_cb(void *arg, const struct spdk_nvme_cpl *cpl) 1651 { 1652 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = arg; 1653 union spdk_nvme_async_event_completion event; 1654 1655 if (spdk_nvme_cpl_is_error(cpl)) { 1656 SPDK_WARNLOG("AER request execute failed"); 1657 return; 1658 } 1659 1660 event.raw = cpl->cdw0; 1661 if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_NOTICE) && 1662 (event.bits.async_event_info == SPDK_NVME_ASYNC_EVENT_NS_ATTR_CHANGED)) { 1663 nvme_ctrlr_populate_namespaces(nvme_bdev_ctrlr, NULL); 1664 } else if ((event.bits.async_event_type == SPDK_NVME_ASYNC_EVENT_TYPE_VENDOR) && 1665 (event.bits.log_page_identifier == SPDK_OCSSD_LOG_CHUNK_NOTIFICATION) && 1666 spdk_nvme_ctrlr_is_ocssd_supported(nvme_bdev_ctrlr->ctrlr)) { 1667 bdev_ocssd_handle_chunk_notification(nvme_bdev_ctrlr); 1668 } 1669 } 1670 1671 static void 1672 populate_namespaces_cb(struct nvme_async_probe_ctx *ctx, size_t count, int rc) 1673 { 1674 if (ctx->cb_fn) { 1675 ctx->cb_fn(ctx->cb_ctx, count, rc); 1676 } 1677 1678 ctx->namespaces_populated = true; 1679 if (ctx->probe_done) { 1680 /* The probe was already completed, so we need to free the context 1681 * here. This can happen for cases like OCSSD, where we need to 1682 * send additional commands to the SSD after attach. 1683 */ 1684 free(ctx); 1685 } 1686 } 1687 1688 static int 1689 _nvme_bdev_ctrlr_create(struct spdk_nvme_ctrlr *ctrlr, 1690 const char *name, 1691 const struct spdk_nvme_transport_id *trid, 1692 uint32_t prchk_flags, 1693 struct nvme_bdev_ctrlr **_nvme_bdev_ctrlr) 1694 { 1695 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr; 1696 struct nvme_bdev_ctrlr_trid *trid_entry; 1697 uint32_t i; 1698 int rc; 1699 1700 nvme_bdev_ctrlr = calloc(1, sizeof(*nvme_bdev_ctrlr)); 1701 if (nvme_bdev_ctrlr == NULL) { 1702 SPDK_ERRLOG("Failed to allocate device struct\n"); 1703 return -ENOMEM; 1704 } 1705 1706 rc = pthread_mutex_init(&nvme_bdev_ctrlr->mutex, NULL); 1707 if (rc != 0) { 1708 goto err_init_mutex; 1709 } 1710 1711 TAILQ_INIT(&nvme_bdev_ctrlr->trids); 1712 nvme_bdev_ctrlr->num_ns = spdk_nvme_ctrlr_get_num_ns(ctrlr); 1713 if (nvme_bdev_ctrlr->num_ns != 0) { 1714 nvme_bdev_ctrlr->namespaces = calloc(nvme_bdev_ctrlr->num_ns, sizeof(struct nvme_bdev_ns *)); 1715 if (!nvme_bdev_ctrlr->namespaces) { 1716 SPDK_ERRLOG("Failed to allocate block namespaces pointer\n"); 1717 rc = -ENOMEM; 1718 goto err_alloc_namespaces; 1719 } 1720 } 1721 1722 trid_entry = calloc(1, sizeof(*trid_entry)); 1723 if (trid_entry == NULL) { 1724 SPDK_ERRLOG("Failed to allocate trid entry pointer\n"); 1725 rc = -ENOMEM; 1726 goto err_alloc_trid; 1727 } 1728 1729 trid_entry->trid = *trid; 1730 1731 for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) { 1732 nvme_bdev_ctrlr->namespaces[i] = calloc(1, sizeof(struct nvme_bdev_ns)); 1733 if (nvme_bdev_ctrlr->namespaces[i] == NULL) { 1734 SPDK_ERRLOG("Failed to allocate block namespace struct\n"); 1735 rc = -ENOMEM; 1736 goto err_alloc_namespace; 1737 } 1738 } 1739 1740 nvme_bdev_ctrlr->thread = spdk_get_thread(); 1741 nvme_bdev_ctrlr->adminq_timer_poller = NULL; 1742 nvme_bdev_ctrlr->ctrlr = ctrlr; 1743 nvme_bdev_ctrlr->ref = 1; 1744 nvme_bdev_ctrlr->connected_trid = &trid_entry->trid; 1745 nvme_bdev_ctrlr->name = strdup(name); 1746 if (nvme_bdev_ctrlr->name == NULL) { 1747 rc = -ENOMEM; 1748 goto err_alloc_name; 1749 } 1750 1751 if (spdk_nvme_ctrlr_is_ocssd_supported(nvme_bdev_ctrlr->ctrlr)) { 1752 rc = bdev_ocssd_init_ctrlr(nvme_bdev_ctrlr); 1753 if (spdk_unlikely(rc != 0)) { 1754 SPDK_ERRLOG("Unable to initialize OCSSD controller\n"); 1755 goto err_init_ocssd; 1756 } 1757 } 1758 1759 nvme_bdev_ctrlr->prchk_flags = prchk_flags; 1760 1761 spdk_io_device_register(nvme_bdev_ctrlr, bdev_nvme_create_cb, bdev_nvme_destroy_cb, 1762 sizeof(struct nvme_io_channel), 1763 name); 1764 1765 nvme_bdev_ctrlr->adminq_timer_poller = SPDK_POLLER_REGISTER(bdev_nvme_poll_adminq, nvme_bdev_ctrlr, 1766 g_opts.nvme_adminq_poll_period_us); 1767 1768 TAILQ_INSERT_TAIL(&g_nvme_bdev_ctrlrs, nvme_bdev_ctrlr, tailq); 1769 1770 if (g_opts.timeout_us > 0) { 1771 spdk_nvme_ctrlr_register_timeout_callback(ctrlr, g_opts.timeout_us, 1772 timeout_cb, nvme_bdev_ctrlr); 1773 } 1774 1775 spdk_nvme_ctrlr_register_aer_callback(ctrlr, aer_cb, nvme_bdev_ctrlr); 1776 spdk_nvme_ctrlr_set_remove_cb(ctrlr, remove_cb, nvme_bdev_ctrlr); 1777 1778 if (spdk_nvme_ctrlr_get_flags(nvme_bdev_ctrlr->ctrlr) & 1779 SPDK_NVME_CTRLR_SECURITY_SEND_RECV_SUPPORTED) { 1780 nvme_bdev_ctrlr->opal_dev = spdk_opal_dev_construct(nvme_bdev_ctrlr->ctrlr); 1781 if (nvme_bdev_ctrlr->opal_dev == NULL) { 1782 SPDK_ERRLOG("Failed to initialize Opal\n"); 1783 } 1784 } 1785 1786 TAILQ_INSERT_HEAD(&nvme_bdev_ctrlr->trids, trid_entry, link); 1787 1788 if (_nvme_bdev_ctrlr != NULL) { 1789 *_nvme_bdev_ctrlr = nvme_bdev_ctrlr; 1790 } 1791 return 0; 1792 1793 err_init_ocssd: 1794 free(nvme_bdev_ctrlr->name); 1795 err_alloc_name: 1796 err_alloc_namespace: 1797 for (; i > 0; i--) { 1798 free(nvme_bdev_ctrlr->namespaces[i - 1]); 1799 } 1800 free(trid_entry); 1801 err_alloc_trid: 1802 free(nvme_bdev_ctrlr->namespaces); 1803 err_alloc_namespaces: 1804 pthread_mutex_destroy(&nvme_bdev_ctrlr->mutex); 1805 err_init_mutex: 1806 free(nvme_bdev_ctrlr); 1807 return rc; 1808 } 1809 1810 static void 1811 nvme_bdev_ctrlr_create(struct spdk_nvme_ctrlr *ctrlr, 1812 const char *name, 1813 const struct spdk_nvme_transport_id *trid, 1814 uint32_t prchk_flags, 1815 struct nvme_async_probe_ctx *ctx) 1816 { 1817 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = NULL; 1818 int rc; 1819 1820 rc = _nvme_bdev_ctrlr_create(ctrlr, name, trid, prchk_flags, &nvme_bdev_ctrlr); 1821 if (rc != 0) { 1822 SPDK_ERRLOG("Failed to create new NVMe controller\n"); 1823 goto err; 1824 } 1825 1826 nvme_ctrlr_populate_namespaces(nvme_bdev_ctrlr, ctx); 1827 return; 1828 1829 err: 1830 if (ctx != NULL) { 1831 populate_namespaces_cb(ctx, 0, rc); 1832 } 1833 } 1834 1835 static void 1836 attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid, 1837 struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts) 1838 { 1839 struct nvme_probe_ctx *ctx = cb_ctx; 1840 char *name = NULL; 1841 uint32_t prchk_flags = 0; 1842 size_t i; 1843 1844 if (ctx) { 1845 for (i = 0; i < ctx->count; i++) { 1846 if (spdk_nvme_transport_id_compare(trid, &ctx->trids[i]) == 0) { 1847 prchk_flags = ctx->prchk_flags[i]; 1848 name = strdup(ctx->names[i]); 1849 break; 1850 } 1851 } 1852 } else { 1853 name = spdk_sprintf_alloc("HotInNvme%d", g_hot_insert_nvme_controller_index++); 1854 } 1855 if (!name) { 1856 SPDK_ERRLOG("Failed to assign name to NVMe device\n"); 1857 return; 1858 } 1859 1860 SPDK_DEBUGLOG(bdev_nvme, "Attached to %s (%s)\n", trid->traddr, name); 1861 1862 nvme_bdev_ctrlr_create(ctrlr, name, trid, prchk_flags, NULL); 1863 1864 free(name); 1865 } 1866 1867 static void 1868 _nvme_bdev_ctrlr_destruct(void *ctx) 1869 { 1870 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = ctx; 1871 1872 nvme_ctrlr_depopulate_namespaces(nvme_bdev_ctrlr); 1873 nvme_bdev_ctrlr_destruct(nvme_bdev_ctrlr); 1874 } 1875 1876 static int 1877 _bdev_nvme_delete(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, bool hotplug) 1878 { 1879 struct nvme_probe_skip_entry *entry; 1880 1881 pthread_mutex_lock(&nvme_bdev_ctrlr->mutex); 1882 1883 /* The controller's destruction was already started */ 1884 if (nvme_bdev_ctrlr->destruct) { 1885 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 1886 return 0; 1887 } 1888 1889 if (!hotplug && 1890 nvme_bdev_ctrlr->connected_trid->trtype == SPDK_NVME_TRANSPORT_PCIE) { 1891 entry = calloc(1, sizeof(*entry)); 1892 if (!entry) { 1893 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 1894 return -ENOMEM; 1895 } 1896 entry->trid = *nvme_bdev_ctrlr->connected_trid; 1897 TAILQ_INSERT_TAIL(&g_skipped_nvme_ctrlrs, entry, tailq); 1898 } 1899 1900 nvme_bdev_ctrlr->destruct = true; 1901 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 1902 1903 _nvme_bdev_ctrlr_destruct(nvme_bdev_ctrlr); 1904 1905 return 0; 1906 } 1907 1908 static void 1909 remove_cb(void *cb_ctx, struct spdk_nvme_ctrlr *ctrlr) 1910 { 1911 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr = cb_ctx; 1912 1913 _bdev_nvme_delete(nvme_bdev_ctrlr, true); 1914 } 1915 1916 static int 1917 bdev_nvme_hotplug_probe(void *arg) 1918 { 1919 if (g_hotplug_probe_ctx == NULL) { 1920 spdk_poller_unregister(&g_hotplug_probe_poller); 1921 return SPDK_POLLER_IDLE; 1922 } 1923 1924 if (spdk_nvme_probe_poll_async(g_hotplug_probe_ctx) != -EAGAIN) { 1925 g_hotplug_probe_ctx = NULL; 1926 spdk_poller_unregister(&g_hotplug_probe_poller); 1927 } 1928 1929 return SPDK_POLLER_BUSY; 1930 } 1931 1932 static int 1933 bdev_nvme_hotplug(void *arg) 1934 { 1935 struct spdk_nvme_transport_id trid_pcie; 1936 1937 if (g_hotplug_probe_ctx) { 1938 return SPDK_POLLER_BUSY; 1939 } 1940 1941 memset(&trid_pcie, 0, sizeof(trid_pcie)); 1942 spdk_nvme_trid_populate_transport(&trid_pcie, SPDK_NVME_TRANSPORT_PCIE); 1943 1944 g_hotplug_probe_ctx = spdk_nvme_probe_async(&trid_pcie, NULL, 1945 hotplug_probe_cb, attach_cb, NULL); 1946 1947 if (g_hotplug_probe_ctx) { 1948 assert(g_hotplug_probe_poller == NULL); 1949 g_hotplug_probe_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug_probe, NULL, 1000); 1950 } 1951 1952 return SPDK_POLLER_BUSY; 1953 } 1954 1955 void 1956 bdev_nvme_get_opts(struct spdk_bdev_nvme_opts *opts) 1957 { 1958 *opts = g_opts; 1959 } 1960 1961 int 1962 bdev_nvme_set_opts(const struct spdk_bdev_nvme_opts *opts) 1963 { 1964 if (g_bdev_nvme_init_thread != NULL) { 1965 if (!TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) { 1966 return -EPERM; 1967 } 1968 } 1969 1970 g_opts = *opts; 1971 1972 return 0; 1973 } 1974 1975 struct set_nvme_hotplug_ctx { 1976 uint64_t period_us; 1977 bool enabled; 1978 spdk_msg_fn fn; 1979 void *fn_ctx; 1980 }; 1981 1982 static void 1983 set_nvme_hotplug_period_cb(void *_ctx) 1984 { 1985 struct set_nvme_hotplug_ctx *ctx = _ctx; 1986 1987 spdk_poller_unregister(&g_hotplug_poller); 1988 if (ctx->enabled) { 1989 g_hotplug_poller = SPDK_POLLER_REGISTER(bdev_nvme_hotplug, NULL, ctx->period_us); 1990 } 1991 1992 g_nvme_hotplug_poll_period_us = ctx->period_us; 1993 g_nvme_hotplug_enabled = ctx->enabled; 1994 if (ctx->fn) { 1995 ctx->fn(ctx->fn_ctx); 1996 } 1997 1998 free(ctx); 1999 } 2000 2001 int 2002 bdev_nvme_set_hotplug(bool enabled, uint64_t period_us, spdk_msg_fn cb, void *cb_ctx) 2003 { 2004 struct set_nvme_hotplug_ctx *ctx; 2005 2006 if (enabled == true && !spdk_process_is_primary()) { 2007 return -EPERM; 2008 } 2009 2010 ctx = calloc(1, sizeof(*ctx)); 2011 if (ctx == NULL) { 2012 return -ENOMEM; 2013 } 2014 2015 period_us = period_us == 0 ? NVME_HOTPLUG_POLL_PERIOD_DEFAULT : period_us; 2016 ctx->period_us = spdk_min(period_us, NVME_HOTPLUG_POLL_PERIOD_MAX); 2017 ctx->enabled = enabled; 2018 ctx->fn = cb; 2019 ctx->fn_ctx = cb_ctx; 2020 2021 spdk_thread_send_msg(g_bdev_nvme_init_thread, set_nvme_hotplug_period_cb, ctx); 2022 return 0; 2023 } 2024 2025 static void 2026 nvme_ctrlr_populate_namespaces_done(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 2027 struct nvme_async_probe_ctx *ctx) 2028 { 2029 struct nvme_bdev_ns *nvme_ns; 2030 struct nvme_bdev *nvme_bdev; 2031 uint32_t i, nsid; 2032 size_t j; 2033 2034 assert(nvme_bdev_ctrlr != NULL); 2035 2036 /* 2037 * Report the new bdevs that were created in this call. 2038 * There can be more than one bdev per NVMe controller. 2039 */ 2040 j = 0; 2041 for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) { 2042 nsid = i + 1; 2043 nvme_ns = nvme_bdev_ctrlr->namespaces[nsid - 1]; 2044 if (!nvme_ns->populated) { 2045 continue; 2046 } 2047 assert(nvme_ns->id == nsid); 2048 nvme_bdev = nvme_ns->bdev; 2049 if (nvme_bdev == NULL) { 2050 assert(nvme_ns->type == NVME_BDEV_NS_OCSSD); 2051 continue; 2052 } 2053 if (j < ctx->count) { 2054 ctx->names[j] = nvme_bdev->disk.name; 2055 j++; 2056 } else { 2057 SPDK_ERRLOG("Maximum number of namespaces supported per NVMe controller is %du. Unable to return all names of created bdevs\n", 2058 ctx->count); 2059 populate_namespaces_cb(ctx, 0, -ERANGE); 2060 return; 2061 } 2062 } 2063 2064 populate_namespaces_cb(ctx, j, 0); 2065 } 2066 2067 static int 2068 bdev_nvme_compare_trids(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 2069 struct spdk_nvme_ctrlr *new_ctrlr, 2070 struct spdk_nvme_transport_id *trid) 2071 { 2072 struct nvme_bdev_ctrlr_trid *tmp_trid; 2073 2074 if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) { 2075 SPDK_ERRLOG("PCIe failover is not supported.\n"); 2076 return -ENOTSUP; 2077 } 2078 2079 /* Currently we only support failover to the same transport type. */ 2080 if (nvme_bdev_ctrlr->connected_trid->trtype != trid->trtype) { 2081 return -EINVAL; 2082 } 2083 2084 /* Currently we only support failover to the same NQN. */ 2085 if (strncmp(trid->subnqn, nvme_bdev_ctrlr->connected_trid->subnqn, SPDK_NVMF_NQN_MAX_LEN)) { 2086 return -EINVAL; 2087 } 2088 2089 /* Skip all the other checks if we've already registered this path. */ 2090 TAILQ_FOREACH(tmp_trid, &nvme_bdev_ctrlr->trids, link) { 2091 if (!spdk_nvme_transport_id_compare(&tmp_trid->trid, trid)) { 2092 return -EEXIST; 2093 } 2094 } 2095 2096 return 0; 2097 } 2098 2099 static bool 2100 bdev_nvme_compare_ns(struct spdk_nvme_ns *ns1, struct spdk_nvme_ns *ns2) 2101 { 2102 const struct spdk_nvme_ns_data *nsdata1, *nsdata2; 2103 2104 nsdata1 = spdk_nvme_ns_get_data(ns1); 2105 nsdata2 = spdk_nvme_ns_get_data(ns2); 2106 2107 return memcmp(nsdata1->nguid, nsdata2->nguid, sizeof(nsdata1->nguid)); 2108 } 2109 2110 static int 2111 bdev_nvme_compare_namespaces(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 2112 struct spdk_nvme_ctrlr *new_ctrlr) 2113 { 2114 uint32_t i, nsid; 2115 struct nvme_bdev_ns *nvme_ns; 2116 struct spdk_nvme_ns *new_ns; 2117 2118 if (spdk_nvme_ctrlr_get_num_ns(new_ctrlr) != nvme_bdev_ctrlr->num_ns) { 2119 return -EINVAL; 2120 } 2121 2122 for (i = 0; i < nvme_bdev_ctrlr->num_ns; i++) { 2123 nsid = i + 1; 2124 2125 nvme_ns = nvme_bdev_ctrlr->namespaces[i]; 2126 if (!nvme_ns->populated) { 2127 continue; 2128 } 2129 2130 new_ns = spdk_nvme_ctrlr_get_ns(new_ctrlr, nsid); 2131 assert(new_ns != NULL); 2132 2133 if (bdev_nvme_compare_ns(nvme_ns->ns, new_ns) != 0) { 2134 return -EINVAL; 2135 } 2136 } 2137 2138 return 0; 2139 } 2140 2141 static int 2142 _bdev_nvme_add_secondary_trid(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 2143 struct spdk_nvme_transport_id *trid) 2144 { 2145 struct nvme_bdev_ctrlr_trid *new_trid, *tmp_trid; 2146 2147 new_trid = calloc(1, sizeof(*new_trid)); 2148 if (new_trid == NULL) { 2149 return -ENOMEM; 2150 } 2151 new_trid->trid = *trid; 2152 new_trid->is_failed = false; 2153 2154 TAILQ_FOREACH(tmp_trid, &nvme_bdev_ctrlr->trids, link) { 2155 if (tmp_trid->is_failed) { 2156 TAILQ_INSERT_BEFORE(tmp_trid, new_trid, link); 2157 return 0; 2158 } 2159 } 2160 2161 TAILQ_INSERT_TAIL(&nvme_bdev_ctrlr->trids, new_trid, link); 2162 return 0; 2163 } 2164 2165 /* This is the case that a secondary path is added to an existing 2166 * nvme_bdev_ctrlr for failover. After checking if it can access the same 2167 * namespaces as the primary path, it is disconnected until failover occurs. 2168 */ 2169 static void 2170 bdev_nvme_add_secondary_trid(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 2171 struct spdk_nvme_ctrlr *new_ctrlr, 2172 struct spdk_nvme_transport_id *trid, 2173 struct nvme_async_probe_ctx *ctx) 2174 { 2175 int rc; 2176 2177 assert(nvme_bdev_ctrlr != NULL); 2178 2179 pthread_mutex_lock(&nvme_bdev_ctrlr->mutex); 2180 2181 rc = bdev_nvme_compare_trids(nvme_bdev_ctrlr, new_ctrlr, trid); 2182 if (rc != 0) { 2183 goto exit; 2184 } 2185 2186 rc = bdev_nvme_compare_namespaces(nvme_bdev_ctrlr, new_ctrlr); 2187 if (rc != 0) { 2188 goto exit; 2189 } 2190 2191 rc = _bdev_nvme_add_secondary_trid(nvme_bdev_ctrlr, trid); 2192 2193 exit: 2194 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 2195 2196 spdk_nvme_detach(new_ctrlr); 2197 2198 if (ctx != NULL) { 2199 populate_namespaces_cb(ctx, 0, rc); 2200 } 2201 } 2202 2203 static void 2204 connect_attach_cb(void *cb_ctx, const struct spdk_nvme_transport_id *trid, 2205 struct spdk_nvme_ctrlr *ctrlr, const struct spdk_nvme_ctrlr_opts *opts) 2206 { 2207 struct spdk_nvme_ctrlr_opts *user_opts = cb_ctx; 2208 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr; 2209 struct nvme_async_probe_ctx *ctx; 2210 2211 ctx = SPDK_CONTAINEROF(user_opts, struct nvme_async_probe_ctx, opts); 2212 ctx->ctrlr_attached = true; 2213 2214 nvme_bdev_ctrlr = nvme_bdev_ctrlr_get_by_name(ctx->base_name); 2215 if (nvme_bdev_ctrlr) { 2216 bdev_nvme_add_secondary_trid(nvme_bdev_ctrlr, ctrlr, &ctx->trid, ctx); 2217 return; 2218 } 2219 2220 nvme_bdev_ctrlr_create(ctrlr, ctx->base_name, &ctx->trid, ctx->prchk_flags, ctx); 2221 } 2222 2223 static int 2224 bdev_nvme_async_poll(void *arg) 2225 { 2226 struct nvme_async_probe_ctx *ctx = arg; 2227 int rc; 2228 2229 rc = spdk_nvme_probe_poll_async(ctx->probe_ctx); 2230 if (spdk_unlikely(rc != -EAGAIN)) { 2231 ctx->probe_done = true; 2232 spdk_poller_unregister(&ctx->poller); 2233 if (!ctx->ctrlr_attached) { 2234 /* The probe is done, but no controller was attached. 2235 * That means we had a failure, so report -EIO back to 2236 * the caller (usually the RPC). populate_namespaces_cb() 2237 * will take care of freeing the nvme_async_probe_ctx. 2238 */ 2239 populate_namespaces_cb(ctx, 0, -EIO); 2240 } else if (ctx->namespaces_populated) { 2241 /* The namespaces for the attached controller were all 2242 * populated and the response was already sent to the 2243 * caller (usually the RPC). So free the context here. 2244 */ 2245 free(ctx); 2246 } 2247 } 2248 2249 return SPDK_POLLER_BUSY; 2250 } 2251 2252 int 2253 bdev_nvme_create(struct spdk_nvme_transport_id *trid, 2254 struct spdk_nvme_host_id *hostid, 2255 const char *base_name, 2256 const char **names, 2257 uint32_t count, 2258 const char *hostnqn, 2259 uint32_t prchk_flags, 2260 spdk_bdev_create_nvme_fn cb_fn, 2261 void *cb_ctx, 2262 struct spdk_nvme_ctrlr_opts *opts) 2263 { 2264 struct nvme_probe_skip_entry *entry, *tmp; 2265 struct nvme_async_probe_ctx *ctx; 2266 2267 /* TODO expand this check to include both the host and target TRIDs. 2268 * Only if both are the same should we fail. 2269 */ 2270 if (nvme_bdev_ctrlr_get(trid) != NULL) { 2271 SPDK_ERRLOG("A controller with the provided trid (traddr: %s) already exists.\n", trid->traddr); 2272 return -EEXIST; 2273 } 2274 2275 ctx = calloc(1, sizeof(*ctx)); 2276 if (!ctx) { 2277 return -ENOMEM; 2278 } 2279 ctx->base_name = base_name; 2280 ctx->names = names; 2281 ctx->count = count; 2282 ctx->cb_fn = cb_fn; 2283 ctx->cb_ctx = cb_ctx; 2284 ctx->prchk_flags = prchk_flags; 2285 ctx->trid = *trid; 2286 2287 if (trid->trtype == SPDK_NVME_TRANSPORT_PCIE) { 2288 TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, tmp) { 2289 if (spdk_nvme_transport_id_compare(trid, &entry->trid) == 0) { 2290 TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq); 2291 free(entry); 2292 break; 2293 } 2294 } 2295 } 2296 2297 if (opts) { 2298 memcpy(&ctx->opts, opts, sizeof(*opts)); 2299 } else { 2300 spdk_nvme_ctrlr_get_default_ctrlr_opts(&ctx->opts, sizeof(ctx->opts)); 2301 } 2302 2303 ctx->opts.transport_retry_count = g_opts.retry_count; 2304 ctx->opts.keep_alive_timeout_ms = g_opts.keep_alive_timeout_ms; 2305 2306 if (hostnqn) { 2307 snprintf(ctx->opts.hostnqn, sizeof(ctx->opts.hostnqn), "%s", hostnqn); 2308 } 2309 2310 if (hostid->hostaddr[0] != '\0') { 2311 snprintf(ctx->opts.src_addr, sizeof(ctx->opts.src_addr), "%s", hostid->hostaddr); 2312 } 2313 2314 if (hostid->hostsvcid[0] != '\0') { 2315 snprintf(ctx->opts.src_svcid, sizeof(ctx->opts.src_svcid), "%s", hostid->hostsvcid); 2316 } 2317 2318 ctx->probe_ctx = spdk_nvme_connect_async(trid, &ctx->opts, connect_attach_cb); 2319 if (ctx->probe_ctx == NULL) { 2320 SPDK_ERRLOG("No controller was found with provided trid (traddr: %s)\n", trid->traddr); 2321 free(ctx); 2322 return -ENODEV; 2323 } 2324 ctx->poller = SPDK_POLLER_REGISTER(bdev_nvme_async_poll, ctx, 1000); 2325 2326 return 0; 2327 } 2328 2329 static int 2330 bdev_nvme_delete_secondary_trid(struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, 2331 const struct spdk_nvme_transport_id *trid) 2332 { 2333 struct nvme_bdev_ctrlr_trid *ctrlr_trid, *tmp_trid; 2334 2335 if (!spdk_nvme_transport_id_compare(trid, nvme_bdev_ctrlr->connected_trid)) { 2336 return -EBUSY; 2337 } 2338 2339 TAILQ_FOREACH_SAFE(ctrlr_trid, &nvme_bdev_ctrlr->trids, link, tmp_trid) { 2340 if (!spdk_nvme_transport_id_compare(&ctrlr_trid->trid, trid)) { 2341 TAILQ_REMOVE(&nvme_bdev_ctrlr->trids, ctrlr_trid, link); 2342 free(ctrlr_trid); 2343 return 0; 2344 } 2345 } 2346 2347 return -ENXIO; 2348 } 2349 2350 int 2351 bdev_nvme_delete(const char *name, const struct spdk_nvme_transport_id *trid) 2352 { 2353 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr; 2354 struct nvme_bdev_ctrlr_trid *ctrlr_trid; 2355 2356 if (name == NULL) { 2357 return -EINVAL; 2358 } 2359 2360 nvme_bdev_ctrlr = nvme_bdev_ctrlr_get_by_name(name); 2361 if (nvme_bdev_ctrlr == NULL) { 2362 SPDK_ERRLOG("Failed to find NVMe controller\n"); 2363 return -ENODEV; 2364 } 2365 2366 /* case 1: remove the controller itself. */ 2367 if (trid == NULL) { 2368 return _bdev_nvme_delete(nvme_bdev_ctrlr, false); 2369 } 2370 2371 /* case 2: we are currently using the path to be removed. */ 2372 if (!spdk_nvme_transport_id_compare(trid, nvme_bdev_ctrlr->connected_trid)) { 2373 ctrlr_trid = TAILQ_FIRST(&nvme_bdev_ctrlr->trids); 2374 assert(nvme_bdev_ctrlr->connected_trid == &ctrlr_trid->trid); 2375 /* case 2A: the current path is the only path. */ 2376 if (!TAILQ_NEXT(ctrlr_trid, link)) { 2377 return _bdev_nvme_delete(nvme_bdev_ctrlr, false); 2378 } 2379 2380 /* case 2B: there is an alternative path. */ 2381 return bdev_nvme_failover(nvme_bdev_ctrlr, true); 2382 } 2383 2384 /* case 3: We are not using the specified path. */ 2385 return bdev_nvme_delete_secondary_trid(nvme_bdev_ctrlr, trid); 2386 } 2387 2388 static int 2389 bdev_nvme_library_init(void) 2390 { 2391 g_bdev_nvme_init_thread = spdk_get_thread(); 2392 2393 spdk_io_device_register(&g_nvme_bdev_ctrlrs, bdev_nvme_poll_group_create_cb, 2394 bdev_nvme_poll_group_destroy_cb, 2395 sizeof(struct nvme_bdev_poll_group), "bdev_nvme_poll_groups"); 2396 2397 return 0; 2398 } 2399 2400 static void 2401 bdev_nvme_library_fini(void) 2402 { 2403 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr, *tmp; 2404 struct nvme_probe_skip_entry *entry, *entry_tmp; 2405 2406 spdk_poller_unregister(&g_hotplug_poller); 2407 free(g_hotplug_probe_ctx); 2408 g_hotplug_probe_ctx = NULL; 2409 2410 TAILQ_FOREACH_SAFE(entry, &g_skipped_nvme_ctrlrs, tailq, entry_tmp) { 2411 TAILQ_REMOVE(&g_skipped_nvme_ctrlrs, entry, tailq); 2412 free(entry); 2413 } 2414 2415 pthread_mutex_lock(&g_bdev_nvme_mutex); 2416 TAILQ_FOREACH_SAFE(nvme_bdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq, tmp) { 2417 pthread_mutex_lock(&nvme_bdev_ctrlr->mutex); 2418 if (nvme_bdev_ctrlr->destruct) { 2419 /* This controller's destruction was already started 2420 * before the application started shutting down 2421 */ 2422 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 2423 continue; 2424 } 2425 nvme_bdev_ctrlr->destruct = true; 2426 pthread_mutex_unlock(&nvme_bdev_ctrlr->mutex); 2427 2428 spdk_thread_send_msg(nvme_bdev_ctrlr->thread, _nvme_bdev_ctrlr_destruct, 2429 nvme_bdev_ctrlr); 2430 } 2431 2432 g_bdev_nvme_module_finish = true; 2433 if (TAILQ_EMPTY(&g_nvme_bdev_ctrlrs)) { 2434 pthread_mutex_unlock(&g_bdev_nvme_mutex); 2435 spdk_io_device_unregister(&g_nvme_bdev_ctrlrs, NULL); 2436 spdk_bdev_module_finish_done(); 2437 return; 2438 } 2439 2440 pthread_mutex_unlock(&g_bdev_nvme_mutex); 2441 } 2442 2443 static void 2444 bdev_nvme_verify_pi_error(struct spdk_bdev_io *bdev_io) 2445 { 2446 struct spdk_bdev *bdev = bdev_io->bdev; 2447 struct spdk_dif_ctx dif_ctx; 2448 struct spdk_dif_error err_blk = {}; 2449 int rc; 2450 2451 rc = spdk_dif_ctx_init(&dif_ctx, 2452 bdev->blocklen, bdev->md_len, bdev->md_interleave, 2453 bdev->dif_is_head_of_md, bdev->dif_type, bdev->dif_check_flags, 2454 bdev_io->u.bdev.offset_blocks, 0, 0, 0, 0); 2455 if (rc != 0) { 2456 SPDK_ERRLOG("Initialization of DIF context failed\n"); 2457 return; 2458 } 2459 2460 if (bdev->md_interleave) { 2461 rc = spdk_dif_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, 2462 bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk); 2463 } else { 2464 struct iovec md_iov = { 2465 .iov_base = bdev_io->u.bdev.md_buf, 2466 .iov_len = bdev_io->u.bdev.num_blocks * bdev->md_len, 2467 }; 2468 2469 rc = spdk_dix_verify(bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, 2470 &md_iov, bdev_io->u.bdev.num_blocks, &dif_ctx, &err_blk); 2471 } 2472 2473 if (rc != 0) { 2474 SPDK_ERRLOG("DIF error detected. type=%d, offset=%" PRIu32 "\n", 2475 err_blk.err_type, err_blk.err_offset); 2476 } else { 2477 SPDK_ERRLOG("Hardware reported PI error but SPDK could not find any.\n"); 2478 } 2479 } 2480 2481 static void 2482 bdev_nvme_no_pi_readv_done(void *ref, const struct spdk_nvme_cpl *cpl) 2483 { 2484 struct nvme_bdev_io *bio = ref; 2485 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio); 2486 2487 if (spdk_nvme_cpl_is_success(cpl)) { 2488 /* Run PI verification for read data buffer. */ 2489 bdev_nvme_verify_pi_error(bdev_io); 2490 } 2491 2492 /* Return original completion status */ 2493 spdk_bdev_io_complete_nvme_status(bdev_io, bio->cpl.cdw0, bio->cpl.status.sct, 2494 bio->cpl.status.sc); 2495 } 2496 2497 static void 2498 bdev_nvme_readv_done(void *ref, const struct spdk_nvme_cpl *cpl) 2499 { 2500 struct nvme_bdev_io *bio = ref; 2501 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio); 2502 struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt; 2503 struct nvme_io_channel *nvme_ch; 2504 struct nvme_bdev_ns *nvme_ns; 2505 struct spdk_nvme_qpair *qpair; 2506 int ret; 2507 2508 if (spdk_unlikely(spdk_nvme_cpl_is_pi_error(cpl))) { 2509 SPDK_ERRLOG("readv completed with PI error (sct=%d, sc=%d)\n", 2510 cpl->status.sct, cpl->status.sc); 2511 2512 /* Save completion status to use after verifying PI error. */ 2513 bio->cpl = *cpl; 2514 2515 nvme_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io)); 2516 2517 if (spdk_likely(bdev_nvme_find_io_path(nbdev, nvme_ch, &nvme_ns, &qpair))) { 2518 /* Read without PI checking to verify PI error. */ 2519 ret = bdev_nvme_no_pi_readv(nvme_ns->ns, 2520 qpair, 2521 bio, 2522 bdev_io->u.bdev.iovs, 2523 bdev_io->u.bdev.iovcnt, 2524 bdev_io->u.bdev.md_buf, 2525 bdev_io->u.bdev.num_blocks, 2526 bdev_io->u.bdev.offset_blocks); 2527 if (ret == 0) { 2528 return; 2529 } 2530 } 2531 } 2532 2533 spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc); 2534 } 2535 2536 static void 2537 bdev_nvme_writev_done(void *ref, const struct spdk_nvme_cpl *cpl) 2538 { 2539 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx((struct nvme_bdev_io *)ref); 2540 2541 if (spdk_nvme_cpl_is_pi_error(cpl)) { 2542 SPDK_ERRLOG("writev completed with PI error (sct=%d, sc=%d)\n", 2543 cpl->status.sct, cpl->status.sc); 2544 /* Run PI verification for write data buffer if PI error is detected. */ 2545 bdev_nvme_verify_pi_error(bdev_io); 2546 } 2547 2548 spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc); 2549 } 2550 2551 static void 2552 bdev_nvme_zone_appendv_done(void *ref, const struct spdk_nvme_cpl *cpl) 2553 { 2554 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx((struct nvme_bdev_io *)ref); 2555 2556 /* spdk_bdev_io_get_append_location() requires that the ALBA is stored in offset_blocks. 2557 * Additionally, offset_blocks has to be set before calling bdev_nvme_verify_pi_error(). 2558 */ 2559 bdev_io->u.bdev.offset_blocks = *(uint64_t *)&cpl->cdw0; 2560 2561 if (spdk_nvme_cpl_is_pi_error(cpl)) { 2562 SPDK_ERRLOG("zone append completed with PI error (sct=%d, sc=%d)\n", 2563 cpl->status.sct, cpl->status.sc); 2564 /* Run PI verification for zone append data buffer if PI error is detected. */ 2565 bdev_nvme_verify_pi_error(bdev_io); 2566 } 2567 2568 spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc); 2569 } 2570 2571 static void 2572 bdev_nvme_comparev_done(void *ref, const struct spdk_nvme_cpl *cpl) 2573 { 2574 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx((struct nvme_bdev_io *)ref); 2575 2576 if (spdk_nvme_cpl_is_pi_error(cpl)) { 2577 SPDK_ERRLOG("comparev completed with PI error (sct=%d, sc=%d)\n", 2578 cpl->status.sct, cpl->status.sc); 2579 /* Run PI verification for compare data buffer if PI error is detected. */ 2580 bdev_nvme_verify_pi_error(bdev_io); 2581 } 2582 2583 spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc); 2584 } 2585 2586 static void 2587 bdev_nvme_comparev_and_writev_done(void *ref, const struct spdk_nvme_cpl *cpl) 2588 { 2589 struct nvme_bdev_io *bio = ref; 2590 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio); 2591 2592 /* Compare operation completion */ 2593 if ((cpl->cdw0 & 0xFF) == SPDK_NVME_OPC_COMPARE) { 2594 /* Save compare result for write callback */ 2595 bio->cpl = *cpl; 2596 return; 2597 } 2598 2599 /* Write operation completion */ 2600 if (spdk_nvme_cpl_is_error(&bio->cpl)) { 2601 /* If bio->cpl is already an error, it means the compare operation failed. In that case, 2602 * complete the IO with the compare operation's status. 2603 */ 2604 if (!spdk_nvme_cpl_is_error(cpl)) { 2605 SPDK_ERRLOG("Unexpected write success after compare failure.\n"); 2606 } 2607 2608 spdk_bdev_io_complete_nvme_status(bdev_io, bio->cpl.cdw0, bio->cpl.status.sct, bio->cpl.status.sc); 2609 } else { 2610 spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc); 2611 } 2612 } 2613 2614 static void 2615 bdev_nvme_queued_done(void *ref, const struct spdk_nvme_cpl *cpl) 2616 { 2617 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx((struct nvme_bdev_io *)ref); 2618 2619 spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc); 2620 } 2621 2622 static int 2623 fill_zone_from_report(struct spdk_bdev_zone_info *info, struct spdk_nvme_zns_zone_desc *desc) 2624 { 2625 switch (desc->zs) { 2626 case SPDK_NVME_ZONE_STATE_EMPTY: 2627 info->state = SPDK_BDEV_ZONE_STATE_EMPTY; 2628 break; 2629 case SPDK_NVME_ZONE_STATE_IOPEN: 2630 info->state = SPDK_BDEV_ZONE_STATE_IMP_OPEN; 2631 break; 2632 case SPDK_NVME_ZONE_STATE_EOPEN: 2633 info->state = SPDK_BDEV_ZONE_STATE_EXP_OPEN; 2634 break; 2635 case SPDK_NVME_ZONE_STATE_CLOSED: 2636 info->state = SPDK_BDEV_ZONE_STATE_CLOSED; 2637 break; 2638 case SPDK_NVME_ZONE_STATE_RONLY: 2639 info->state = SPDK_BDEV_ZONE_STATE_READ_ONLY; 2640 break; 2641 case SPDK_NVME_ZONE_STATE_FULL: 2642 info->state = SPDK_BDEV_ZONE_STATE_FULL; 2643 break; 2644 case SPDK_NVME_ZONE_STATE_OFFLINE: 2645 info->state = SPDK_BDEV_ZONE_STATE_OFFLINE; 2646 break; 2647 default: 2648 SPDK_ERRLOG("Invalid zone state: %#x in zone report\n", desc->zs); 2649 return -EIO; 2650 } 2651 2652 info->zone_id = desc->zslba; 2653 info->write_pointer = desc->wp; 2654 info->capacity = desc->zcap; 2655 2656 return 0; 2657 } 2658 2659 static void 2660 bdev_nvme_get_zone_info_done(void *ref, const struct spdk_nvme_cpl *cpl) 2661 { 2662 struct nvme_bdev_io *bio = ref; 2663 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio); 2664 struct nvme_bdev *nbdev = (struct nvme_bdev *)bdev_io->bdev->ctxt; 2665 struct spdk_io_channel *ch = spdk_bdev_io_get_io_channel(bdev_io); 2666 struct nvme_io_channel *nvme_ch = spdk_io_channel_get_ctx(ch); 2667 uint64_t zone_id = bdev_io->u.zone_mgmt.zone_id; 2668 uint32_t zones_to_copy = bdev_io->u.zone_mgmt.num_zones; 2669 struct spdk_bdev_zone_info *info = bdev_io->u.zone_mgmt.buf; 2670 enum spdk_bdev_io_status status; 2671 uint64_t max_zones_per_buf, i; 2672 uint32_t zone_report_bufsize; 2673 struct nvme_bdev_ns *nvme_ns; 2674 struct spdk_nvme_qpair *qpair; 2675 int ret; 2676 2677 if (spdk_nvme_cpl_is_error(cpl)) { 2678 goto out_complete_io_nvme_cpl; 2679 } 2680 2681 if (!bdev_nvme_find_io_path(nbdev, nvme_ch, &nvme_ns, &qpair)) { 2682 status = SPDK_BDEV_IO_STATUS_FAILED; 2683 goto out_complete_io_status; 2684 } 2685 2686 zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(nvme_ns->ns); 2687 max_zones_per_buf = (zone_report_bufsize - sizeof(*bio->zone_report_buf)) / 2688 sizeof(bio->zone_report_buf->descs[0]); 2689 2690 if (bio->zone_report_buf->nr_zones > max_zones_per_buf) { 2691 status = SPDK_BDEV_IO_STATUS_FAILED; 2692 goto out_complete_io_status; 2693 } 2694 2695 if (!bio->zone_report_buf->nr_zones) { 2696 status = SPDK_BDEV_IO_STATUS_FAILED; 2697 goto out_complete_io_status; 2698 } 2699 2700 for (i = 0; i < bio->zone_report_buf->nr_zones && bio->handled_zones < zones_to_copy; i++) { 2701 ret = fill_zone_from_report(&info[bio->handled_zones], 2702 &bio->zone_report_buf->descs[i]); 2703 if (ret) { 2704 status = SPDK_BDEV_IO_STATUS_FAILED; 2705 goto out_complete_io_status; 2706 } 2707 bio->handled_zones++; 2708 } 2709 2710 if (bio->handled_zones < zones_to_copy) { 2711 uint64_t zone_size_lba = spdk_nvme_zns_ns_get_zone_size_sectors(nvme_ns->ns); 2712 uint64_t slba = zone_id + (zone_size_lba * bio->handled_zones); 2713 2714 memset(bio->zone_report_buf, 0, zone_report_bufsize); 2715 ret = spdk_nvme_zns_report_zones(nvme_ns->ns, qpair, 2716 bio->zone_report_buf, zone_report_bufsize, 2717 slba, SPDK_NVME_ZRA_LIST_ALL, true, 2718 bdev_nvme_get_zone_info_done, bio); 2719 if (!ret) { 2720 return; 2721 } else if (ret == -ENOMEM) { 2722 status = SPDK_BDEV_IO_STATUS_NOMEM; 2723 goto out_complete_io_status; 2724 } else { 2725 status = SPDK_BDEV_IO_STATUS_FAILED; 2726 goto out_complete_io_status; 2727 } 2728 } 2729 2730 out_complete_io_nvme_cpl: 2731 free(bio->zone_report_buf); 2732 bio->zone_report_buf = NULL; 2733 spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc); 2734 return; 2735 2736 out_complete_io_status: 2737 free(bio->zone_report_buf); 2738 bio->zone_report_buf = NULL; 2739 spdk_bdev_io_complete(bdev_io, status); 2740 } 2741 2742 static void 2743 bdev_nvme_zone_management_done(void *ref, const struct spdk_nvme_cpl *cpl) 2744 { 2745 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx((struct nvme_bdev_io *)ref); 2746 2747 spdk_bdev_io_complete_nvme_status(bdev_io, cpl->cdw0, cpl->status.sct, cpl->status.sc); 2748 } 2749 2750 static void 2751 bdev_nvme_admin_passthru_completion(void *ctx) 2752 { 2753 struct nvme_bdev_io *bio = ctx; 2754 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio); 2755 2756 spdk_bdev_io_complete_nvme_status(bdev_io, 2757 bio->cpl.cdw0, bio->cpl.status.sct, bio->cpl.status.sc); 2758 } 2759 2760 static void 2761 bdev_nvme_abort_completion(void *ctx) 2762 { 2763 struct nvme_bdev_io *bio = ctx; 2764 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio); 2765 2766 if (spdk_nvme_cpl_is_abort_success(&bio->cpl)) { 2767 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_SUCCESS); 2768 } else { 2769 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 2770 } 2771 } 2772 2773 static void 2774 bdev_nvme_abort_done(void *ref, const struct spdk_nvme_cpl *cpl) 2775 { 2776 struct nvme_bdev_io *bio = ref; 2777 2778 bio->cpl = *cpl; 2779 spdk_thread_send_msg(bio->orig_thread, bdev_nvme_abort_completion, bio); 2780 } 2781 2782 static void 2783 bdev_nvme_admin_passthru_done(void *ref, const struct spdk_nvme_cpl *cpl) 2784 { 2785 struct nvme_bdev_io *bio = ref; 2786 2787 bio->cpl = *cpl; 2788 spdk_thread_send_msg(bio->orig_thread, bdev_nvme_admin_passthru_completion, bio); 2789 } 2790 2791 static void 2792 bdev_nvme_queued_reset_sgl(void *ref, uint32_t sgl_offset) 2793 { 2794 struct nvme_bdev_io *bio = ref; 2795 struct iovec *iov; 2796 2797 bio->iov_offset = sgl_offset; 2798 for (bio->iovpos = 0; bio->iovpos < bio->iovcnt; bio->iovpos++) { 2799 iov = &bio->iovs[bio->iovpos]; 2800 if (bio->iov_offset < iov->iov_len) { 2801 break; 2802 } 2803 2804 bio->iov_offset -= iov->iov_len; 2805 } 2806 } 2807 2808 static int 2809 bdev_nvme_queued_next_sge(void *ref, void **address, uint32_t *length) 2810 { 2811 struct nvme_bdev_io *bio = ref; 2812 struct iovec *iov; 2813 2814 assert(bio->iovpos < bio->iovcnt); 2815 2816 iov = &bio->iovs[bio->iovpos]; 2817 2818 *address = iov->iov_base; 2819 *length = iov->iov_len; 2820 2821 if (bio->iov_offset) { 2822 assert(bio->iov_offset <= iov->iov_len); 2823 *address += bio->iov_offset; 2824 *length -= bio->iov_offset; 2825 } 2826 2827 bio->iov_offset += *length; 2828 if (bio->iov_offset == iov->iov_len) { 2829 bio->iovpos++; 2830 bio->iov_offset = 0; 2831 } 2832 2833 return 0; 2834 } 2835 2836 static void 2837 bdev_nvme_queued_reset_fused_sgl(void *ref, uint32_t sgl_offset) 2838 { 2839 struct nvme_bdev_io *bio = ref; 2840 struct iovec *iov; 2841 2842 bio->fused_iov_offset = sgl_offset; 2843 for (bio->fused_iovpos = 0; bio->fused_iovpos < bio->fused_iovcnt; bio->fused_iovpos++) { 2844 iov = &bio->fused_iovs[bio->fused_iovpos]; 2845 if (bio->fused_iov_offset < iov->iov_len) { 2846 break; 2847 } 2848 2849 bio->fused_iov_offset -= iov->iov_len; 2850 } 2851 } 2852 2853 static int 2854 bdev_nvme_queued_next_fused_sge(void *ref, void **address, uint32_t *length) 2855 { 2856 struct nvme_bdev_io *bio = ref; 2857 struct iovec *iov; 2858 2859 assert(bio->fused_iovpos < bio->fused_iovcnt); 2860 2861 iov = &bio->fused_iovs[bio->fused_iovpos]; 2862 2863 *address = iov->iov_base; 2864 *length = iov->iov_len; 2865 2866 if (bio->fused_iov_offset) { 2867 assert(bio->fused_iov_offset <= iov->iov_len); 2868 *address += bio->fused_iov_offset; 2869 *length -= bio->fused_iov_offset; 2870 } 2871 2872 bio->fused_iov_offset += *length; 2873 if (bio->fused_iov_offset == iov->iov_len) { 2874 bio->fused_iovpos++; 2875 bio->fused_iov_offset = 0; 2876 } 2877 2878 return 0; 2879 } 2880 2881 static int 2882 bdev_nvme_no_pi_readv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 2883 struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt, 2884 void *md, uint64_t lba_count, uint64_t lba) 2885 { 2886 int rc; 2887 2888 SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 " without PI check\n", 2889 lba_count, lba); 2890 2891 bio->iovs = iov; 2892 bio->iovcnt = iovcnt; 2893 bio->iovpos = 0; 2894 bio->iov_offset = 0; 2895 2896 rc = spdk_nvme_ns_cmd_readv_with_md(ns, qpair, lba, lba_count, 2897 bdev_nvme_no_pi_readv_done, bio, 0, 2898 bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, 2899 md, 0, 0); 2900 2901 if (rc != 0 && rc != -ENOMEM) { 2902 SPDK_ERRLOG("no_pi_readv failed: rc = %d\n", rc); 2903 } 2904 return rc; 2905 } 2906 2907 static int 2908 bdev_nvme_readv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 2909 struct nvme_bdev_io *bio, struct iovec *iov, int iovcnt, 2910 void *md, uint64_t lba_count, uint64_t lba, uint32_t flags) 2911 { 2912 int rc; 2913 2914 SPDK_DEBUGLOG(bdev_nvme, "read %" PRIu64 " blocks with offset %#" PRIx64 "\n", 2915 lba_count, lba); 2916 2917 bio->iovs = iov; 2918 bio->iovcnt = iovcnt; 2919 bio->iovpos = 0; 2920 bio->iov_offset = 0; 2921 2922 if (iovcnt == 1) { 2923 rc = spdk_nvme_ns_cmd_read_with_md(ns, qpair, iov[0].iov_base, md, lba, 2924 lba_count, 2925 bdev_nvme_readv_done, bio, 2926 flags, 2927 0, 0); 2928 } else { 2929 rc = spdk_nvme_ns_cmd_readv_with_md(ns, qpair, lba, lba_count, 2930 bdev_nvme_readv_done, bio, flags, 2931 bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, 2932 md, 0, 0); 2933 } 2934 2935 if (rc != 0 && rc != -ENOMEM) { 2936 SPDK_ERRLOG("readv failed: rc = %d\n", rc); 2937 } 2938 return rc; 2939 } 2940 2941 static int 2942 bdev_nvme_writev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 2943 struct nvme_bdev_io *bio, 2944 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba, 2945 uint32_t flags) 2946 { 2947 int rc; 2948 2949 SPDK_DEBUGLOG(bdev_nvme, "write %" PRIu64 " blocks with offset %#" PRIx64 "\n", 2950 lba_count, lba); 2951 2952 bio->iovs = iov; 2953 bio->iovcnt = iovcnt; 2954 bio->iovpos = 0; 2955 bio->iov_offset = 0; 2956 2957 if (iovcnt == 1) { 2958 rc = spdk_nvme_ns_cmd_write_with_md(ns, qpair, iov[0].iov_base, md, lba, 2959 lba_count, 2960 bdev_nvme_writev_done, bio, 2961 flags, 2962 0, 0); 2963 } else { 2964 rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count, 2965 bdev_nvme_writev_done, bio, flags, 2966 bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, 2967 md, 0, 0); 2968 } 2969 2970 if (rc != 0 && rc != -ENOMEM) { 2971 SPDK_ERRLOG("writev failed: rc = %d\n", rc); 2972 } 2973 return rc; 2974 } 2975 2976 static int 2977 bdev_nvme_zone_appendv(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 2978 struct nvme_bdev_io *bio, 2979 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t zslba, 2980 uint32_t flags) 2981 { 2982 int rc; 2983 2984 SPDK_DEBUGLOG(bdev_nvme, "zone append %" PRIu64 " blocks to zone start lba %#" PRIx64 "\n", 2985 lba_count, zslba); 2986 2987 bio->iovs = iov; 2988 bio->iovcnt = iovcnt; 2989 bio->iovpos = 0; 2990 bio->iov_offset = 0; 2991 2992 if (iovcnt == 1) { 2993 rc = spdk_nvme_zns_zone_append_with_md(ns, qpair, iov[0].iov_base, md, zslba, 2994 lba_count, 2995 bdev_nvme_zone_appendv_done, bio, 2996 flags, 2997 0, 0); 2998 } else { 2999 rc = spdk_nvme_zns_zone_appendv_with_md(ns, qpair, zslba, lba_count, 3000 bdev_nvme_zone_appendv_done, bio, flags, 3001 bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, 3002 md, 0, 0); 3003 } 3004 3005 if (rc != 0 && rc != -ENOMEM) { 3006 SPDK_ERRLOG("zone append failed: rc = %d\n", rc); 3007 } 3008 return rc; 3009 } 3010 3011 static int 3012 bdev_nvme_comparev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 3013 struct nvme_bdev_io *bio, 3014 struct iovec *iov, int iovcnt, void *md, uint64_t lba_count, uint64_t lba, 3015 uint32_t flags) 3016 { 3017 int rc; 3018 3019 SPDK_DEBUGLOG(bdev_nvme, "compare %" PRIu64 " blocks with offset %#" PRIx64 "\n", 3020 lba_count, lba); 3021 3022 bio->iovs = iov; 3023 bio->iovcnt = iovcnt; 3024 bio->iovpos = 0; 3025 bio->iov_offset = 0; 3026 3027 rc = spdk_nvme_ns_cmd_comparev_with_md(ns, qpair, lba, lba_count, 3028 bdev_nvme_comparev_done, bio, flags, 3029 bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, 3030 md, 0, 0); 3031 3032 if (rc != 0 && rc != -ENOMEM) { 3033 SPDK_ERRLOG("comparev failed: rc = %d\n", rc); 3034 } 3035 return rc; 3036 } 3037 3038 static int 3039 bdev_nvme_comparev_and_writev(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 3040 struct nvme_bdev_io *bio, struct iovec *cmp_iov, int cmp_iovcnt, 3041 struct iovec *write_iov, int write_iovcnt, 3042 void *md, uint64_t lba_count, uint64_t lba, uint32_t flags) 3043 { 3044 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio); 3045 int rc; 3046 3047 SPDK_DEBUGLOG(bdev_nvme, "compare and write %" PRIu64 " blocks with offset %#" PRIx64 "\n", 3048 lba_count, lba); 3049 3050 bio->iovs = cmp_iov; 3051 bio->iovcnt = cmp_iovcnt; 3052 bio->iovpos = 0; 3053 bio->iov_offset = 0; 3054 bio->fused_iovs = write_iov; 3055 bio->fused_iovcnt = write_iovcnt; 3056 bio->fused_iovpos = 0; 3057 bio->fused_iov_offset = 0; 3058 3059 if (bdev_io->num_retries == 0) { 3060 bio->first_fused_submitted = false; 3061 } 3062 3063 if (!bio->first_fused_submitted) { 3064 flags |= SPDK_NVME_IO_FLAGS_FUSE_FIRST; 3065 memset(&bio->cpl, 0, sizeof(bio->cpl)); 3066 3067 rc = spdk_nvme_ns_cmd_comparev_with_md(ns, qpair, lba, lba_count, 3068 bdev_nvme_comparev_and_writev_done, bio, flags, 3069 bdev_nvme_queued_reset_sgl, bdev_nvme_queued_next_sge, md, 0, 0); 3070 if (rc == 0) { 3071 bio->first_fused_submitted = true; 3072 flags &= ~SPDK_NVME_IO_FLAGS_FUSE_FIRST; 3073 } else { 3074 if (rc != -ENOMEM) { 3075 SPDK_ERRLOG("compare failed: rc = %d\n", rc); 3076 } 3077 return rc; 3078 } 3079 } 3080 3081 flags |= SPDK_NVME_IO_FLAGS_FUSE_SECOND; 3082 3083 rc = spdk_nvme_ns_cmd_writev_with_md(ns, qpair, lba, lba_count, 3084 bdev_nvme_comparev_and_writev_done, bio, flags, 3085 bdev_nvme_queued_reset_fused_sgl, bdev_nvme_queued_next_fused_sge, md, 0, 0); 3086 if (rc != 0 && rc != -ENOMEM) { 3087 SPDK_ERRLOG("write failed: rc = %d\n", rc); 3088 rc = 0; 3089 } 3090 3091 return rc; 3092 } 3093 3094 static int 3095 bdev_nvme_unmap(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 3096 struct nvme_bdev_io *bio, 3097 uint64_t offset_blocks, 3098 uint64_t num_blocks) 3099 { 3100 struct spdk_nvme_dsm_range dsm_ranges[SPDK_NVME_DATASET_MANAGEMENT_MAX_RANGES]; 3101 struct spdk_nvme_dsm_range *range; 3102 uint64_t offset, remaining; 3103 uint64_t num_ranges_u64; 3104 uint16_t num_ranges; 3105 int rc; 3106 3107 num_ranges_u64 = (num_blocks + SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS - 1) / 3108 SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS; 3109 if (num_ranges_u64 > SPDK_COUNTOF(dsm_ranges)) { 3110 SPDK_ERRLOG("Unmap request for %" PRIu64 " blocks is too large\n", num_blocks); 3111 return -EINVAL; 3112 } 3113 num_ranges = (uint16_t)num_ranges_u64; 3114 3115 offset = offset_blocks; 3116 remaining = num_blocks; 3117 range = &dsm_ranges[0]; 3118 3119 /* Fill max-size ranges until the remaining blocks fit into one range */ 3120 while (remaining > SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS) { 3121 range->attributes.raw = 0; 3122 range->length = SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS; 3123 range->starting_lba = offset; 3124 3125 offset += SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS; 3126 remaining -= SPDK_NVME_DATASET_MANAGEMENT_RANGE_MAX_BLOCKS; 3127 range++; 3128 } 3129 3130 /* Final range describes the remaining blocks */ 3131 range->attributes.raw = 0; 3132 range->length = remaining; 3133 range->starting_lba = offset; 3134 3135 rc = spdk_nvme_ns_cmd_dataset_management(ns, qpair, 3136 SPDK_NVME_DSM_ATTR_DEALLOCATE, 3137 dsm_ranges, num_ranges, 3138 bdev_nvme_queued_done, bio); 3139 3140 return rc; 3141 } 3142 3143 static int 3144 bdev_nvme_get_zone_info(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 3145 struct nvme_bdev_io *bio, uint64_t zone_id, uint32_t num_zones, 3146 struct spdk_bdev_zone_info *info) 3147 { 3148 uint32_t zone_report_bufsize = spdk_nvme_ns_get_max_io_xfer_size(ns); 3149 uint64_t zone_size = spdk_nvme_zns_ns_get_zone_size_sectors(ns); 3150 uint64_t total_zones = spdk_nvme_zns_ns_get_num_zones(ns); 3151 3152 if (zone_id % zone_size != 0) { 3153 return -EINVAL; 3154 } 3155 3156 if (num_zones > total_zones || !num_zones) { 3157 return -EINVAL; 3158 } 3159 3160 assert(!bio->zone_report_buf); 3161 bio->zone_report_buf = calloc(1, zone_report_bufsize); 3162 if (!bio->zone_report_buf) { 3163 return -ENOMEM; 3164 } 3165 3166 bio->handled_zones = 0; 3167 3168 return spdk_nvme_zns_report_zones(ns, qpair, bio->zone_report_buf, zone_report_bufsize, 3169 zone_id, SPDK_NVME_ZRA_LIST_ALL, true, 3170 bdev_nvme_get_zone_info_done, bio); 3171 } 3172 3173 static int 3174 bdev_nvme_zone_management(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 3175 struct nvme_bdev_io *bio, uint64_t zone_id, 3176 enum spdk_bdev_zone_action action) 3177 { 3178 switch (action) { 3179 case SPDK_BDEV_ZONE_CLOSE: 3180 return spdk_nvme_zns_close_zone(ns, qpair, zone_id, false, 3181 bdev_nvme_zone_management_done, bio); 3182 case SPDK_BDEV_ZONE_FINISH: 3183 return spdk_nvme_zns_finish_zone(ns, qpair, zone_id, false, 3184 bdev_nvme_zone_management_done, bio); 3185 case SPDK_BDEV_ZONE_OPEN: 3186 return spdk_nvme_zns_open_zone(ns, qpair, zone_id, false, 3187 bdev_nvme_zone_management_done, bio); 3188 case SPDK_BDEV_ZONE_RESET: 3189 return spdk_nvme_zns_reset_zone(ns, qpair, zone_id, false, 3190 bdev_nvme_zone_management_done, bio); 3191 case SPDK_BDEV_ZONE_OFFLINE: 3192 return spdk_nvme_zns_offline_zone(ns, qpair, zone_id, false, 3193 bdev_nvme_zone_management_done, bio); 3194 default: 3195 return -EINVAL; 3196 } 3197 } 3198 3199 static int 3200 bdev_nvme_admin_passthru(struct nvme_io_channel *nvme_ch, struct nvme_bdev_io *bio, 3201 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes) 3202 { 3203 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr; 3204 uint32_t max_xfer_size; 3205 3206 if (!bdev_nvme_find_admin_path(nvme_ch, &nvme_bdev_ctrlr)) { 3207 return -EINVAL; 3208 } 3209 3210 max_xfer_size = spdk_nvme_ctrlr_get_max_xfer_size(nvme_bdev_ctrlr->ctrlr); 3211 3212 if (nbytes > max_xfer_size) { 3213 SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size); 3214 return -EINVAL; 3215 } 3216 3217 bio->orig_thread = spdk_get_thread(); 3218 3219 return spdk_nvme_ctrlr_cmd_admin_raw(nvme_bdev_ctrlr->ctrlr, cmd, buf, 3220 (uint32_t)nbytes, bdev_nvme_admin_passthru_done, bio); 3221 } 3222 3223 static int 3224 bdev_nvme_io_passthru(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 3225 struct nvme_bdev_io *bio, 3226 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes) 3227 { 3228 uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns); 3229 struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns); 3230 3231 if (nbytes > max_xfer_size) { 3232 SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size); 3233 return -EINVAL; 3234 } 3235 3236 /* 3237 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid, 3238 * so fill it out automatically. 3239 */ 3240 cmd->nsid = spdk_nvme_ns_get_id(ns); 3241 3242 return spdk_nvme_ctrlr_cmd_io_raw(ctrlr, qpair, cmd, buf, 3243 (uint32_t)nbytes, bdev_nvme_queued_done, bio); 3244 } 3245 3246 static int 3247 bdev_nvme_io_passthru_md(struct spdk_nvme_ns *ns, struct spdk_nvme_qpair *qpair, 3248 struct nvme_bdev_io *bio, 3249 struct spdk_nvme_cmd *cmd, void *buf, size_t nbytes, void *md_buf, size_t md_len) 3250 { 3251 size_t nr_sectors = nbytes / spdk_nvme_ns_get_extended_sector_size(ns); 3252 uint32_t max_xfer_size = spdk_nvme_ns_get_max_io_xfer_size(ns); 3253 struct spdk_nvme_ctrlr *ctrlr = spdk_nvme_ns_get_ctrlr(ns); 3254 3255 if (nbytes > max_xfer_size) { 3256 SPDK_ERRLOG("nbytes is greater than MDTS %" PRIu32 ".\n", max_xfer_size); 3257 return -EINVAL; 3258 } 3259 3260 if (md_len != nr_sectors * spdk_nvme_ns_get_md_size(ns)) { 3261 SPDK_ERRLOG("invalid meta data buffer size\n"); 3262 return -EINVAL; 3263 } 3264 3265 /* 3266 * Each NVMe bdev is a specific namespace, and all NVMe I/O commands require a nsid, 3267 * so fill it out automatically. 3268 */ 3269 cmd->nsid = spdk_nvme_ns_get_id(ns); 3270 3271 return spdk_nvme_ctrlr_cmd_io_raw_with_md(ctrlr, qpair, cmd, buf, 3272 (uint32_t)nbytes, md_buf, bdev_nvme_queued_done, bio); 3273 } 3274 3275 static void 3276 bdev_nvme_abort_admin_cmd(void *ctx) 3277 { 3278 struct nvme_bdev_io *bio = ctx; 3279 struct spdk_bdev_io *bdev_io = spdk_bdev_io_from_ctx(bio); 3280 struct nvme_io_channel *nvme_ch; 3281 struct nvme_bdev_io *bio_to_abort; 3282 int rc; 3283 3284 nvme_ch = spdk_io_channel_get_ctx(spdk_bdev_io_get_io_channel(bdev_io)); 3285 bio_to_abort = (struct nvme_bdev_io *)bdev_io->u.abort.bio_to_abort->driver_ctx; 3286 3287 rc = spdk_nvme_ctrlr_cmd_abort_ext(nvme_ch->ctrlr->ctrlr, 3288 NULL, 3289 bio_to_abort, 3290 bdev_nvme_abort_done, bio); 3291 if (rc == -ENOENT) { 3292 /* If no admin command was found in admin qpair, complete the abort 3293 * request with failure. 3294 */ 3295 bio->cpl.cdw0 |= 1U; 3296 bio->cpl.status.sc = SPDK_NVME_SC_SUCCESS; 3297 bio->cpl.status.sct = SPDK_NVME_SCT_GENERIC; 3298 3299 spdk_thread_send_msg(bio->orig_thread, bdev_nvme_abort_completion, bio); 3300 } 3301 } 3302 3303 static int 3304 bdev_nvme_abort(struct nvme_io_channel *nvme_ch, struct nvme_bdev_io *bio, 3305 struct nvme_bdev_io *bio_to_abort) 3306 { 3307 int rc; 3308 3309 bio->orig_thread = spdk_get_thread(); 3310 3311 rc = spdk_nvme_ctrlr_cmd_abort_ext(nvme_ch->ctrlr->ctrlr, 3312 nvme_ch->qpair, 3313 bio_to_abort, 3314 bdev_nvme_abort_done, bio); 3315 if (rc == -ENOENT) { 3316 /* If no command was found in I/O qpair, the target command may be 3317 * admin command. Only a single thread tries aborting admin command 3318 * to clean I/O flow. 3319 */ 3320 spdk_thread_send_msg(nvme_ch->ctrlr->thread, 3321 bdev_nvme_abort_admin_cmd, bio); 3322 rc = 0; 3323 } 3324 3325 return rc; 3326 } 3327 3328 static void 3329 nvme_ctrlr_config_json_standard_namespace(struct spdk_json_write_ctx *w, 3330 struct nvme_bdev_ns *nvme_ns) 3331 { 3332 /* nop */ 3333 } 3334 3335 static void 3336 nvme_namespace_config_json(struct spdk_json_write_ctx *w, struct nvme_bdev_ns *nvme_ns) 3337 { 3338 g_config_json_namespace_fn[nvme_ns->type](w, nvme_ns); 3339 } 3340 3341 static void 3342 bdev_nvme_opts_config_json(struct spdk_json_write_ctx *w) 3343 { 3344 const char *action; 3345 3346 if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_RESET) { 3347 action = "reset"; 3348 } else if (g_opts.action_on_timeout == SPDK_BDEV_NVME_TIMEOUT_ACTION_ABORT) { 3349 action = "abort"; 3350 } else { 3351 action = "none"; 3352 } 3353 3354 spdk_json_write_object_begin(w); 3355 3356 spdk_json_write_named_string(w, "method", "bdev_nvme_set_options"); 3357 3358 spdk_json_write_named_object_begin(w, "params"); 3359 spdk_json_write_named_string(w, "action_on_timeout", action); 3360 spdk_json_write_named_uint64(w, "timeout_us", g_opts.timeout_us); 3361 spdk_json_write_named_uint32(w, "keep_alive_timeout_ms", g_opts.keep_alive_timeout_ms); 3362 spdk_json_write_named_uint32(w, "retry_count", g_opts.retry_count); 3363 spdk_json_write_named_uint32(w, "arbitration_burst", g_opts.arbitration_burst); 3364 spdk_json_write_named_uint32(w, "low_priority_weight", g_opts.low_priority_weight); 3365 spdk_json_write_named_uint32(w, "medium_priority_weight", g_opts.medium_priority_weight); 3366 spdk_json_write_named_uint32(w, "high_priority_weight", g_opts.high_priority_weight); 3367 spdk_json_write_named_uint64(w, "nvme_adminq_poll_period_us", g_opts.nvme_adminq_poll_period_us); 3368 spdk_json_write_named_uint64(w, "nvme_ioq_poll_period_us", g_opts.nvme_ioq_poll_period_us); 3369 spdk_json_write_named_uint32(w, "io_queue_requests", g_opts.io_queue_requests); 3370 spdk_json_write_named_bool(w, "delay_cmd_submit", g_opts.delay_cmd_submit); 3371 spdk_json_write_object_end(w); 3372 3373 spdk_json_write_object_end(w); 3374 } 3375 3376 static void 3377 nvme_bdev_ctrlr_config_json(struct spdk_json_write_ctx *w, 3378 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr) 3379 { 3380 struct spdk_nvme_transport_id *trid; 3381 3382 trid = nvme_bdev_ctrlr->connected_trid; 3383 3384 spdk_json_write_object_begin(w); 3385 3386 spdk_json_write_named_string(w, "method", "bdev_nvme_attach_controller"); 3387 3388 spdk_json_write_named_object_begin(w, "params"); 3389 spdk_json_write_named_string(w, "name", nvme_bdev_ctrlr->name); 3390 nvme_bdev_dump_trid_json(trid, w); 3391 spdk_json_write_named_bool(w, "prchk_reftag", 3392 (nvme_bdev_ctrlr->prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_REFTAG) != 0); 3393 spdk_json_write_named_bool(w, "prchk_guard", 3394 (nvme_bdev_ctrlr->prchk_flags & SPDK_NVME_IO_FLAGS_PRCHK_GUARD) != 0); 3395 3396 spdk_json_write_object_end(w); 3397 3398 spdk_json_write_object_end(w); 3399 } 3400 3401 static void 3402 bdev_nvme_hotplug_config_json(struct spdk_json_write_ctx *w) 3403 { 3404 spdk_json_write_object_begin(w); 3405 spdk_json_write_named_string(w, "method", "bdev_nvme_set_hotplug"); 3406 3407 spdk_json_write_named_object_begin(w, "params"); 3408 spdk_json_write_named_uint64(w, "period_us", g_nvme_hotplug_poll_period_us); 3409 spdk_json_write_named_bool(w, "enable", g_nvme_hotplug_enabled); 3410 spdk_json_write_object_end(w); 3411 3412 spdk_json_write_object_end(w); 3413 } 3414 3415 static int 3416 bdev_nvme_config_json(struct spdk_json_write_ctx *w) 3417 { 3418 struct nvme_bdev_ctrlr *nvme_bdev_ctrlr; 3419 uint32_t nsid; 3420 3421 bdev_nvme_opts_config_json(w); 3422 3423 pthread_mutex_lock(&g_bdev_nvme_mutex); 3424 3425 TAILQ_FOREACH(nvme_bdev_ctrlr, &g_nvme_bdev_ctrlrs, tailq) { 3426 nvme_bdev_ctrlr_config_json(w, nvme_bdev_ctrlr); 3427 3428 for (nsid = 0; nsid < nvme_bdev_ctrlr->num_ns; ++nsid) { 3429 if (!nvme_bdev_ctrlr->namespaces[nsid]->populated) { 3430 continue; 3431 } 3432 3433 nvme_namespace_config_json(w, nvme_bdev_ctrlr->namespaces[nsid]); 3434 } 3435 } 3436 3437 /* Dump as last parameter to give all NVMe bdevs chance to be constructed 3438 * before enabling hotplug poller. 3439 */ 3440 bdev_nvme_hotplug_config_json(w); 3441 3442 pthread_mutex_unlock(&g_bdev_nvme_mutex); 3443 return 0; 3444 } 3445 3446 struct spdk_nvme_ctrlr * 3447 bdev_nvme_get_ctrlr(struct spdk_bdev *bdev) 3448 { 3449 if (!bdev || bdev->module != &nvme_if) { 3450 return NULL; 3451 } 3452 3453 return SPDK_CONTAINEROF(bdev, struct nvme_bdev, disk)->nvme_ns->ctrlr->ctrlr; 3454 } 3455 3456 SPDK_LOG_REGISTER_COMPONENT(bdev_nvme) 3457