1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2018 Intel Corporation. 3 * All rights reserved. 4 * Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 /* 8 * This is a simple example of a virtual block device module that passes IO 9 * down to a bdev (or bdevs) that its configured to attach to. 10 */ 11 12 #include "spdk/stdinc.h" 13 14 #include "vbdev_passthru.h" 15 #include "spdk/rpc.h" 16 #include "spdk/env.h" 17 #include "spdk/endian.h" 18 #include "spdk/string.h" 19 #include "spdk/thread.h" 20 #include "spdk/util.h" 21 22 #include "spdk/bdev_module.h" 23 #include "spdk/log.h" 24 25 26 static int vbdev_passthru_init(void); 27 static int vbdev_passthru_get_ctx_size(void); 28 static void vbdev_passthru_examine(struct spdk_bdev *bdev); 29 static void vbdev_passthru_finish(void); 30 static int vbdev_passthru_config_json(struct spdk_json_write_ctx *w); 31 32 static struct spdk_bdev_module passthru_if = { 33 .name = "passthru", 34 .module_init = vbdev_passthru_init, 35 .get_ctx_size = vbdev_passthru_get_ctx_size, 36 .examine_config = vbdev_passthru_examine, 37 .module_fini = vbdev_passthru_finish, 38 .config_json = vbdev_passthru_config_json 39 }; 40 41 SPDK_BDEV_MODULE_REGISTER(passthru, &passthru_if) 42 43 /* List of pt_bdev names and their base bdevs via configuration file. 44 * Used so we can parse the conf once at init and use this list in examine(). 45 */ 46 struct bdev_names { 47 char *vbdev_name; 48 char *bdev_name; 49 TAILQ_ENTRY(bdev_names) link; 50 }; 51 static TAILQ_HEAD(, bdev_names) g_bdev_names = TAILQ_HEAD_INITIALIZER(g_bdev_names); 52 53 /* List of virtual bdevs and associated info for each. */ 54 struct vbdev_passthru { 55 struct spdk_bdev *base_bdev; /* the thing we're attaching to */ 56 struct spdk_bdev_desc *base_desc; /* its descriptor we get from open */ 57 struct spdk_bdev pt_bdev; /* the PT virtual bdev */ 58 TAILQ_ENTRY(vbdev_passthru) link; 59 struct spdk_thread *thread; /* thread where base device is opened */ 60 }; 61 static TAILQ_HEAD(, vbdev_passthru) g_pt_nodes = TAILQ_HEAD_INITIALIZER(g_pt_nodes); 62 63 /* The pt vbdev channel struct. It is allocated and freed on my behalf by the io channel code. 64 * If this vbdev needed to implement a poller or a queue for IO, this is where those things 65 * would be defined. This passthru bdev doesn't actually need to allocate a channel, it could 66 * simply pass back the channel of the bdev underneath it but for example purposes we will 67 * present its own to the upper layers. 68 */ 69 struct pt_io_channel { 70 struct spdk_io_channel *base_ch; /* IO channel of base device */ 71 }; 72 73 /* Just for fun, this pt_bdev module doesn't need it but this is essentially a per IO 74 * context that we get handed by the bdev layer. 75 */ 76 struct passthru_bdev_io { 77 uint8_t test; 78 79 /* bdev related */ 80 struct spdk_io_channel *ch; 81 82 /* for bdev_io_wait */ 83 struct spdk_bdev_io_wait_entry bdev_io_wait; 84 }; 85 86 static void vbdev_passthru_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io); 87 88 89 /* Callback for unregistering the IO device. */ 90 static void 91 _device_unregister_cb(void *io_device) 92 { 93 struct vbdev_passthru *pt_node = io_device; 94 95 /* Done with this pt_node. */ 96 free(pt_node->pt_bdev.name); 97 free(pt_node); 98 } 99 100 /* Wrapper for the bdev close operation. */ 101 static void 102 _vbdev_passthru_destruct(void *ctx) 103 { 104 struct spdk_bdev_desc *desc = ctx; 105 106 spdk_bdev_close(desc); 107 } 108 109 /* Called after we've unregistered following a hot remove callback. 110 * Our finish entry point will be called next. 111 */ 112 static int 113 vbdev_passthru_destruct(void *ctx) 114 { 115 struct vbdev_passthru *pt_node = (struct vbdev_passthru *)ctx; 116 117 /* It is important to follow this exact sequence of steps for destroying 118 * a vbdev... 119 */ 120 121 TAILQ_REMOVE(&g_pt_nodes, pt_node, link); 122 123 /* Unclaim the underlying bdev. */ 124 spdk_bdev_module_release_bdev(pt_node->base_bdev); 125 126 /* Close the underlying bdev on its same opened thread. */ 127 if (pt_node->thread && pt_node->thread != spdk_get_thread()) { 128 spdk_thread_send_msg(pt_node->thread, _vbdev_passthru_destruct, pt_node->base_desc); 129 } else { 130 spdk_bdev_close(pt_node->base_desc); 131 } 132 133 /* Unregister the io_device. */ 134 spdk_io_device_unregister(pt_node, _device_unregister_cb); 135 136 return 0; 137 } 138 139 /* Completion callback for IO that were issued from this bdev. The original bdev_io 140 * is passed in as an arg so we'll complete that one with the appropriate status 141 * and then free the one that this module issued. 142 */ 143 static void 144 _pt_complete_io(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg) 145 { 146 struct spdk_bdev_io *orig_io = cb_arg; 147 int status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED; 148 struct passthru_bdev_io *io_ctx = (struct passthru_bdev_io *)orig_io->driver_ctx; 149 150 /* We setup this value in the submission routine, just showing here that it is 151 * passed back to us. 152 */ 153 if (io_ctx->test != 0x5a) { 154 SPDK_ERRLOG("Error, original IO device_ctx is wrong! 0x%x\n", 155 io_ctx->test); 156 } 157 158 /* Complete the original IO and then free the one that we created here 159 * as a result of issuing an IO via submit_request. 160 */ 161 spdk_bdev_io_complete(orig_io, status); 162 spdk_bdev_free_io(bdev_io); 163 } 164 165 static void 166 _pt_complete_zcopy_io(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg) 167 { 168 struct spdk_bdev_io *orig_io = cb_arg; 169 int status = success ? SPDK_BDEV_IO_STATUS_SUCCESS : SPDK_BDEV_IO_STATUS_FAILED; 170 struct passthru_bdev_io *io_ctx = (struct passthru_bdev_io *)orig_io->driver_ctx; 171 172 /* We setup this value in the submission routine, just showing here that it is 173 * passed back to us. 174 */ 175 if (io_ctx->test != 0x5a) { 176 SPDK_ERRLOG("Error, original IO device_ctx is wrong! 0x%x\n", 177 io_ctx->test); 178 } 179 180 /* Complete the original IO and then free the one that we created here 181 * as a result of issuing an IO via submit_request. 182 */ 183 spdk_bdev_io_set_buf(orig_io, bdev_io->u.bdev.iovs[0].iov_base, bdev_io->u.bdev.iovs[0].iov_len); 184 spdk_bdev_io_complete(orig_io, status); 185 spdk_bdev_free_io(bdev_io); 186 } 187 188 static void 189 vbdev_passthru_resubmit_io(void *arg) 190 { 191 struct spdk_bdev_io *bdev_io = (struct spdk_bdev_io *)arg; 192 struct passthru_bdev_io *io_ctx = (struct passthru_bdev_io *)bdev_io->driver_ctx; 193 194 vbdev_passthru_submit_request(io_ctx->ch, bdev_io); 195 } 196 197 static void 198 vbdev_passthru_queue_io(struct spdk_bdev_io *bdev_io) 199 { 200 struct passthru_bdev_io *io_ctx = (struct passthru_bdev_io *)bdev_io->driver_ctx; 201 struct pt_io_channel *pt_ch = spdk_io_channel_get_ctx(io_ctx->ch); 202 int rc; 203 204 io_ctx->bdev_io_wait.bdev = bdev_io->bdev; 205 io_ctx->bdev_io_wait.cb_fn = vbdev_passthru_resubmit_io; 206 io_ctx->bdev_io_wait.cb_arg = bdev_io; 207 208 /* Queue the IO using the channel of the base device. */ 209 rc = spdk_bdev_queue_io_wait(bdev_io->bdev, pt_ch->base_ch, &io_ctx->bdev_io_wait); 210 if (rc != 0) { 211 SPDK_ERRLOG("Queue io failed in vbdev_passthru_queue_io, rc=%d.\n", rc); 212 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 213 } 214 } 215 216 /* Callback for getting a buf from the bdev pool in the event that the caller passed 217 * in NULL, we need to own the buffer so it doesn't get freed by another vbdev module 218 * beneath us before we're done with it. That won't happen in this example but it could 219 * if this example were used as a template for something more complex. 220 */ 221 static void 222 pt_read_get_buf_cb(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io, bool success) 223 { 224 struct vbdev_passthru *pt_node = SPDK_CONTAINEROF(bdev_io->bdev, struct vbdev_passthru, 225 pt_bdev); 226 struct pt_io_channel *pt_ch = spdk_io_channel_get_ctx(ch); 227 struct passthru_bdev_io *io_ctx = (struct passthru_bdev_io *)bdev_io->driver_ctx; 228 int rc; 229 230 if (!success) { 231 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 232 return; 233 } 234 235 if (bdev_io->u.bdev.ext_opts) { 236 rc = spdk_bdev_readv_blocks_ext(pt_node->base_desc, pt_ch->base_ch, bdev_io->u.bdev.iovs, 237 bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, 238 bdev_io->u.bdev.num_blocks, _pt_complete_io, 239 bdev_io, bdev_io->u.bdev.ext_opts); 240 } else { 241 rc = spdk_bdev_readv_blocks_with_md(pt_node->base_desc, pt_ch->base_ch, 242 bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, 243 bdev_io->u.bdev.md_buf, 244 bdev_io->u.bdev.offset_blocks, 245 bdev_io->u.bdev.num_blocks, 246 _pt_complete_io, bdev_io); 247 } 248 249 if (rc != 0) { 250 if (rc == -ENOMEM) { 251 SPDK_ERRLOG("No memory, start to queue io for passthru.\n"); 252 io_ctx->ch = ch; 253 vbdev_passthru_queue_io(bdev_io); 254 } else { 255 SPDK_ERRLOG("ERROR on bdev_io submission!\n"); 256 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 257 } 258 } 259 } 260 261 /* Called when someone above submits IO to this pt vbdev. We're simply passing it on here 262 * via SPDK IO calls which in turn allocate another bdev IO and call our cpl callback provided 263 * below along with the original bdev_io so that we can complete it once this IO completes. 264 */ 265 static void 266 vbdev_passthru_submit_request(struct spdk_io_channel *ch, struct spdk_bdev_io *bdev_io) 267 { 268 struct vbdev_passthru *pt_node = SPDK_CONTAINEROF(bdev_io->bdev, struct vbdev_passthru, pt_bdev); 269 struct pt_io_channel *pt_ch = spdk_io_channel_get_ctx(ch); 270 struct passthru_bdev_io *io_ctx = (struct passthru_bdev_io *)bdev_io->driver_ctx; 271 int rc = 0; 272 273 /* Setup a per IO context value; we don't do anything with it in the vbdev other 274 * than confirm we get the same thing back in the completion callback just to 275 * demonstrate. 276 */ 277 io_ctx->test = 0x5a; 278 279 switch (bdev_io->type) { 280 case SPDK_BDEV_IO_TYPE_READ: 281 spdk_bdev_io_get_buf(bdev_io, pt_read_get_buf_cb, 282 bdev_io->u.bdev.num_blocks * bdev_io->bdev->blocklen); 283 break; 284 case SPDK_BDEV_IO_TYPE_WRITE: 285 if (bdev_io->u.bdev.ext_opts) { 286 rc = spdk_bdev_writev_blocks_ext(pt_node->base_desc, pt_ch->base_ch, bdev_io->u.bdev.iovs, 287 bdev_io->u.bdev.iovcnt, bdev_io->u.bdev.offset_blocks, 288 bdev_io->u.bdev.num_blocks, _pt_complete_io, 289 bdev_io, bdev_io->u.bdev.ext_opts); 290 } else { 291 rc = spdk_bdev_writev_blocks_with_md(pt_node->base_desc, pt_ch->base_ch, 292 bdev_io->u.bdev.iovs, bdev_io->u.bdev.iovcnt, 293 bdev_io->u.bdev.md_buf, 294 bdev_io->u.bdev.offset_blocks, 295 bdev_io->u.bdev.num_blocks, 296 _pt_complete_io, bdev_io); 297 } 298 break; 299 case SPDK_BDEV_IO_TYPE_WRITE_ZEROES: 300 rc = spdk_bdev_write_zeroes_blocks(pt_node->base_desc, pt_ch->base_ch, 301 bdev_io->u.bdev.offset_blocks, 302 bdev_io->u.bdev.num_blocks, 303 _pt_complete_io, bdev_io); 304 break; 305 case SPDK_BDEV_IO_TYPE_UNMAP: 306 rc = spdk_bdev_unmap_blocks(pt_node->base_desc, pt_ch->base_ch, 307 bdev_io->u.bdev.offset_blocks, 308 bdev_io->u.bdev.num_blocks, 309 _pt_complete_io, bdev_io); 310 break; 311 case SPDK_BDEV_IO_TYPE_FLUSH: 312 rc = spdk_bdev_flush_blocks(pt_node->base_desc, pt_ch->base_ch, 313 bdev_io->u.bdev.offset_blocks, 314 bdev_io->u.bdev.num_blocks, 315 _pt_complete_io, bdev_io); 316 break; 317 case SPDK_BDEV_IO_TYPE_RESET: 318 rc = spdk_bdev_reset(pt_node->base_desc, pt_ch->base_ch, 319 _pt_complete_io, bdev_io); 320 break; 321 case SPDK_BDEV_IO_TYPE_ZCOPY: 322 rc = spdk_bdev_zcopy_start(pt_node->base_desc, pt_ch->base_ch, NULL, 0, 323 bdev_io->u.bdev.offset_blocks, 324 bdev_io->u.bdev.num_blocks, bdev_io->u.bdev.zcopy.populate, 325 _pt_complete_zcopy_io, bdev_io); 326 break; 327 case SPDK_BDEV_IO_TYPE_ABORT: 328 rc = spdk_bdev_abort(pt_node->base_desc, pt_ch->base_ch, bdev_io->u.abort.bio_to_abort, 329 _pt_complete_io, bdev_io); 330 break; 331 case SPDK_BDEV_IO_TYPE_COPY: 332 rc = spdk_bdev_copy_blocks(pt_node->base_desc, pt_ch->base_ch, 333 bdev_io->u.bdev.offset_blocks, 334 bdev_io->u.bdev.copy.src_offset_blocks, 335 bdev_io->u.bdev.num_blocks, 336 _pt_complete_io, bdev_io); 337 break; 338 default: 339 SPDK_ERRLOG("passthru: unknown I/O type %d\n", bdev_io->type); 340 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 341 return; 342 } 343 if (rc != 0) { 344 if (rc == -ENOMEM) { 345 SPDK_ERRLOG("No memory, start to queue io for passthru.\n"); 346 io_ctx->ch = ch; 347 vbdev_passthru_queue_io(bdev_io); 348 } else { 349 SPDK_ERRLOG("ERROR on bdev_io submission!\n"); 350 spdk_bdev_io_complete(bdev_io, SPDK_BDEV_IO_STATUS_FAILED); 351 } 352 } 353 } 354 355 /* We'll just call the base bdev and let it answer however if we were more 356 * restrictive for some reason (or less) we could get the response back 357 * and modify according to our purposes. 358 */ 359 static bool 360 vbdev_passthru_io_type_supported(void *ctx, enum spdk_bdev_io_type io_type) 361 { 362 struct vbdev_passthru *pt_node = (struct vbdev_passthru *)ctx; 363 364 return spdk_bdev_io_type_supported(pt_node->base_bdev, io_type); 365 } 366 367 /* We supplied this as an entry point for upper layers who want to communicate to this 368 * bdev. This is how they get a channel. We are passed the same context we provided when 369 * we created our PT vbdev in examine() which, for this bdev, is the address of one of 370 * our context nodes. From here we'll ask the SPDK channel code to fill out our channel 371 * struct and we'll keep it in our PT node. 372 */ 373 static struct spdk_io_channel * 374 vbdev_passthru_get_io_channel(void *ctx) 375 { 376 struct vbdev_passthru *pt_node = (struct vbdev_passthru *)ctx; 377 struct spdk_io_channel *pt_ch = NULL; 378 379 /* The IO channel code will allocate a channel for us which consists of 380 * the SPDK channel structure plus the size of our pt_io_channel struct 381 * that we passed in when we registered our IO device. It will then call 382 * our channel create callback to populate any elements that we need to 383 * update. 384 */ 385 pt_ch = spdk_get_io_channel(pt_node); 386 387 return pt_ch; 388 } 389 390 /* This is the output for bdev_get_bdevs() for this vbdev */ 391 static int 392 vbdev_passthru_dump_info_json(void *ctx, struct spdk_json_write_ctx *w) 393 { 394 struct vbdev_passthru *pt_node = (struct vbdev_passthru *)ctx; 395 396 spdk_json_write_name(w, "passthru"); 397 spdk_json_write_object_begin(w); 398 spdk_json_write_named_string(w, "name", spdk_bdev_get_name(&pt_node->pt_bdev)); 399 spdk_json_write_named_string(w, "base_bdev_name", spdk_bdev_get_name(pt_node->base_bdev)); 400 spdk_json_write_object_end(w); 401 402 return 0; 403 } 404 405 /* This is used to generate JSON that can configure this module to its current state. */ 406 static int 407 vbdev_passthru_config_json(struct spdk_json_write_ctx *w) 408 { 409 struct vbdev_passthru *pt_node; 410 411 TAILQ_FOREACH(pt_node, &g_pt_nodes, link) { 412 spdk_json_write_object_begin(w); 413 spdk_json_write_named_string(w, "method", "bdev_passthru_create"); 414 spdk_json_write_named_object_begin(w, "params"); 415 spdk_json_write_named_string(w, "base_bdev_name", spdk_bdev_get_name(pt_node->base_bdev)); 416 spdk_json_write_named_string(w, "name", spdk_bdev_get_name(&pt_node->pt_bdev)); 417 spdk_json_write_object_end(w); 418 spdk_json_write_object_end(w); 419 } 420 return 0; 421 } 422 423 /* We provide this callback for the SPDK channel code to create a channel using 424 * the channel struct we provided in our module get_io_channel() entry point. Here 425 * we get and save off an underlying base channel of the device below us so that 426 * we can communicate with the base bdev on a per channel basis. If we needed 427 * our own poller for this vbdev, we'd register it here. 428 */ 429 static int 430 pt_bdev_ch_create_cb(void *io_device, void *ctx_buf) 431 { 432 struct pt_io_channel *pt_ch = ctx_buf; 433 struct vbdev_passthru *pt_node = io_device; 434 435 pt_ch->base_ch = spdk_bdev_get_io_channel(pt_node->base_desc); 436 437 return 0; 438 } 439 440 /* We provide this callback for the SPDK channel code to destroy a channel 441 * created with our create callback. We just need to undo anything we did 442 * when we created. If this bdev used its own poller, we'd unregister it here. 443 */ 444 static void 445 pt_bdev_ch_destroy_cb(void *io_device, void *ctx_buf) 446 { 447 struct pt_io_channel *pt_ch = ctx_buf; 448 449 spdk_put_io_channel(pt_ch->base_ch); 450 } 451 452 /* Create the passthru association from the bdev and vbdev name and insert 453 * on the global list. */ 454 static int 455 vbdev_passthru_insert_name(const char *bdev_name, const char *vbdev_name) 456 { 457 struct bdev_names *name; 458 459 TAILQ_FOREACH(name, &g_bdev_names, link) { 460 if (strcmp(vbdev_name, name->vbdev_name) == 0) { 461 SPDK_ERRLOG("passthru bdev %s already exists\n", vbdev_name); 462 return -EEXIST; 463 } 464 } 465 466 name = calloc(1, sizeof(struct bdev_names)); 467 if (!name) { 468 SPDK_ERRLOG("could not allocate bdev_names\n"); 469 return -ENOMEM; 470 } 471 472 name->bdev_name = strdup(bdev_name); 473 if (!name->bdev_name) { 474 SPDK_ERRLOG("could not allocate name->bdev_name\n"); 475 free(name); 476 return -ENOMEM; 477 } 478 479 name->vbdev_name = strdup(vbdev_name); 480 if (!name->vbdev_name) { 481 SPDK_ERRLOG("could not allocate name->vbdev_name\n"); 482 free(name->bdev_name); 483 free(name); 484 return -ENOMEM; 485 } 486 487 TAILQ_INSERT_TAIL(&g_bdev_names, name, link); 488 489 return 0; 490 } 491 492 /* On init, just perform bdev module specific initialization. */ 493 static int 494 vbdev_passthru_init(void) 495 { 496 return 0; 497 } 498 499 /* Called when the entire module is being torn down. */ 500 static void 501 vbdev_passthru_finish(void) 502 { 503 struct bdev_names *name; 504 505 while ((name = TAILQ_FIRST(&g_bdev_names))) { 506 TAILQ_REMOVE(&g_bdev_names, name, link); 507 free(name->bdev_name); 508 free(name->vbdev_name); 509 free(name); 510 } 511 } 512 513 /* During init we'll be asked how much memory we'd like passed to us 514 * in bev_io structures as context. Here's where we specify how 515 * much context we want per IO. 516 */ 517 static int 518 vbdev_passthru_get_ctx_size(void) 519 { 520 return sizeof(struct passthru_bdev_io); 521 } 522 523 /* Where vbdev_passthru_config_json() is used to generate per module JSON config data, this 524 * function is called to output any per bdev specific methods. For the PT module, there are 525 * none. 526 */ 527 static void 528 vbdev_passthru_write_config_json(struct spdk_bdev *bdev, struct spdk_json_write_ctx *w) 529 { 530 /* No config per bdev needed */ 531 } 532 533 static int 534 vbdev_passthru_get_memory_domains(void *ctx, struct spdk_memory_domain **domains, int array_size) 535 { 536 struct vbdev_passthru *pt_node = (struct vbdev_passthru *)ctx; 537 538 /* Passthru bdev doesn't work with data buffers, so it supports any memory domain used by base_bdev */ 539 return spdk_bdev_get_memory_domains(pt_node->base_bdev, domains, array_size); 540 } 541 542 /* When we register our bdev this is how we specify our entry points. */ 543 static const struct spdk_bdev_fn_table vbdev_passthru_fn_table = { 544 .destruct = vbdev_passthru_destruct, 545 .submit_request = vbdev_passthru_submit_request, 546 .io_type_supported = vbdev_passthru_io_type_supported, 547 .get_io_channel = vbdev_passthru_get_io_channel, 548 .dump_info_json = vbdev_passthru_dump_info_json, 549 .write_config_json = vbdev_passthru_write_config_json, 550 .get_memory_domains = vbdev_passthru_get_memory_domains, 551 }; 552 553 static void 554 vbdev_passthru_base_bdev_hotremove_cb(struct spdk_bdev *bdev_find) 555 { 556 struct vbdev_passthru *pt_node, *tmp; 557 558 TAILQ_FOREACH_SAFE(pt_node, &g_pt_nodes, link, tmp) { 559 if (bdev_find == pt_node->base_bdev) { 560 spdk_bdev_unregister(&pt_node->pt_bdev, NULL, NULL); 561 } 562 } 563 } 564 565 /* Called when the underlying base bdev triggers asynchronous event such as bdev removal. */ 566 static void 567 vbdev_passthru_base_bdev_event_cb(enum spdk_bdev_event_type type, struct spdk_bdev *bdev, 568 void *event_ctx) 569 { 570 switch (type) { 571 case SPDK_BDEV_EVENT_REMOVE: 572 vbdev_passthru_base_bdev_hotremove_cb(bdev); 573 break; 574 default: 575 SPDK_NOTICELOG("Unsupported bdev event: type %d\n", type); 576 break; 577 } 578 } 579 580 /* Create and register the passthru vbdev if we find it in our list of bdev names. 581 * This can be called either by the examine path or RPC method. 582 */ 583 static int 584 vbdev_passthru_register(const char *bdev_name) 585 { 586 struct bdev_names *name; 587 struct vbdev_passthru *pt_node; 588 struct spdk_bdev *bdev; 589 int rc = 0; 590 591 /* Check our list of names from config versus this bdev and if 592 * there's a match, create the pt_node & bdev accordingly. 593 */ 594 TAILQ_FOREACH(name, &g_bdev_names, link) { 595 if (strcmp(name->bdev_name, bdev_name) != 0) { 596 continue; 597 } 598 599 SPDK_NOTICELOG("Match on %s\n", bdev_name); 600 pt_node = calloc(1, sizeof(struct vbdev_passthru)); 601 if (!pt_node) { 602 rc = -ENOMEM; 603 SPDK_ERRLOG("could not allocate pt_node\n"); 604 break; 605 } 606 607 pt_node->pt_bdev.name = strdup(name->vbdev_name); 608 if (!pt_node->pt_bdev.name) { 609 rc = -ENOMEM; 610 SPDK_ERRLOG("could not allocate pt_bdev name\n"); 611 free(pt_node); 612 break; 613 } 614 pt_node->pt_bdev.product_name = "passthru"; 615 616 /* The base bdev that we're attaching to. */ 617 rc = spdk_bdev_open_ext(bdev_name, true, vbdev_passthru_base_bdev_event_cb, 618 NULL, &pt_node->base_desc); 619 if (rc) { 620 if (rc != -ENODEV) { 621 SPDK_ERRLOG("could not open bdev %s\n", bdev_name); 622 } 623 free(pt_node->pt_bdev.name); 624 free(pt_node); 625 break; 626 } 627 SPDK_NOTICELOG("base bdev opened\n"); 628 629 bdev = spdk_bdev_desc_get_bdev(pt_node->base_desc); 630 pt_node->base_bdev = bdev; 631 632 /* Copy some properties from the underlying base bdev. */ 633 pt_node->pt_bdev.write_cache = bdev->write_cache; 634 pt_node->pt_bdev.required_alignment = bdev->required_alignment; 635 pt_node->pt_bdev.optimal_io_boundary = bdev->optimal_io_boundary; 636 pt_node->pt_bdev.blocklen = bdev->blocklen; 637 pt_node->pt_bdev.blockcnt = bdev->blockcnt; 638 639 pt_node->pt_bdev.md_interleave = bdev->md_interleave; 640 pt_node->pt_bdev.md_len = bdev->md_len; 641 pt_node->pt_bdev.dif_type = bdev->dif_type; 642 pt_node->pt_bdev.dif_is_head_of_md = bdev->dif_is_head_of_md; 643 pt_node->pt_bdev.dif_check_flags = bdev->dif_check_flags; 644 645 /* This is the context that is passed to us when the bdev 646 * layer calls in so we'll save our pt_bdev node here. 647 */ 648 pt_node->pt_bdev.ctxt = pt_node; 649 pt_node->pt_bdev.fn_table = &vbdev_passthru_fn_table; 650 pt_node->pt_bdev.module = &passthru_if; 651 TAILQ_INSERT_TAIL(&g_pt_nodes, pt_node, link); 652 653 spdk_io_device_register(pt_node, pt_bdev_ch_create_cb, pt_bdev_ch_destroy_cb, 654 sizeof(struct pt_io_channel), 655 name->vbdev_name); 656 SPDK_NOTICELOG("io_device created at: 0x%p\n", pt_node); 657 658 /* Save the thread where the base device is opened */ 659 pt_node->thread = spdk_get_thread(); 660 661 rc = spdk_bdev_module_claim_bdev(bdev, pt_node->base_desc, pt_node->pt_bdev.module); 662 if (rc) { 663 SPDK_ERRLOG("could not claim bdev %s\n", bdev_name); 664 spdk_bdev_close(pt_node->base_desc); 665 TAILQ_REMOVE(&g_pt_nodes, pt_node, link); 666 spdk_io_device_unregister(pt_node, NULL); 667 free(pt_node->pt_bdev.name); 668 free(pt_node); 669 break; 670 } 671 SPDK_NOTICELOG("bdev claimed\n"); 672 673 rc = spdk_bdev_register(&pt_node->pt_bdev); 674 if (rc) { 675 SPDK_ERRLOG("could not register pt_bdev\n"); 676 spdk_bdev_module_release_bdev(&pt_node->pt_bdev); 677 spdk_bdev_close(pt_node->base_desc); 678 TAILQ_REMOVE(&g_pt_nodes, pt_node, link); 679 spdk_io_device_unregister(pt_node, NULL); 680 free(pt_node->pt_bdev.name); 681 free(pt_node); 682 break; 683 } 684 SPDK_NOTICELOG("pt_bdev registered\n"); 685 SPDK_NOTICELOG("created pt_bdev for: %s\n", name->vbdev_name); 686 } 687 688 return rc; 689 } 690 691 /* Create the passthru disk from the given bdev and vbdev name. */ 692 int 693 bdev_passthru_create_disk(const char *bdev_name, const char *vbdev_name) 694 { 695 int rc; 696 697 /* Insert the bdev name into our global name list even if it doesn't exist yet, 698 * it may show up soon... 699 */ 700 rc = vbdev_passthru_insert_name(bdev_name, vbdev_name); 701 if (rc) { 702 return rc; 703 } 704 705 rc = vbdev_passthru_register(bdev_name); 706 if (rc == -ENODEV) { 707 /* This is not an error, we tracked the name above and it still 708 * may show up later. 709 */ 710 SPDK_NOTICELOG("vbdev creation deferred pending base bdev arrival\n"); 711 rc = 0; 712 } 713 714 return rc; 715 } 716 717 void 718 bdev_passthru_delete_disk(const char *bdev_name, spdk_bdev_unregister_cb cb_fn, void *cb_arg) 719 { 720 struct bdev_names *name; 721 int rc; 722 723 /* Some cleanup happens in the destruct callback. */ 724 rc = spdk_bdev_unregister_by_name(bdev_name, &passthru_if, cb_fn, cb_arg); 725 if (rc == 0) { 726 /* Remove the association (vbdev, bdev) from g_bdev_names. This is required so that the 727 * vbdev does not get re-created if the same bdev is constructed at some other time, 728 * unless the underlying bdev was hot-removed. 729 */ 730 TAILQ_FOREACH(name, &g_bdev_names, link) { 731 if (strcmp(name->vbdev_name, bdev_name) == 0) { 732 TAILQ_REMOVE(&g_bdev_names, name, link); 733 free(name->bdev_name); 734 free(name->vbdev_name); 735 free(name); 736 break; 737 } 738 } 739 } else { 740 cb_fn(cb_arg, rc); 741 } 742 } 743 744 /* Because we specified this function in our pt bdev function table when we 745 * registered our pt bdev, we'll get this call anytime a new bdev shows up. 746 * Here we need to decide if we care about it and if so what to do. We 747 * parsed the config file at init so we check the new bdev against the list 748 * we built up at that time and if the user configured us to attach to this 749 * bdev, here's where we do it. 750 */ 751 static void 752 vbdev_passthru_examine(struct spdk_bdev *bdev) 753 { 754 vbdev_passthru_register(bdev->name); 755 756 spdk_bdev_module_examine_done(&passthru_if); 757 } 758 759 SPDK_LOG_REGISTER_COMPONENT(vbdev_passthru) 760