1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (c) Intel Corporation. All rights reserved. 3 * Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved. 4 * Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 5 */ 6 7 #include "spdk/stdinc.h" 8 9 #include "nvmf_internal.h" 10 11 #include "spdk/bdev.h" 12 #include "spdk/endian.h" 13 #include "spdk/thread.h" 14 #include "spdk/likely.h" 15 #include "spdk/nvme.h" 16 #include "spdk/nvmf_cmd.h" 17 #include "spdk/nvmf_spec.h" 18 #include "spdk/trace.h" 19 #include "spdk/scsi_spec.h" 20 #include "spdk/string.h" 21 #include "spdk/util.h" 22 23 #include "spdk/log.h" 24 25 static bool 26 nvmf_subsystem_bdev_io_type_supported(struct spdk_nvmf_subsystem *subsystem, 27 enum spdk_bdev_io_type io_type) 28 { 29 struct spdk_nvmf_ns *ns; 30 31 for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL; 32 ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) { 33 if (ns->bdev == NULL) { 34 continue; 35 } 36 37 if (!spdk_bdev_io_type_supported(ns->bdev, io_type)) { 38 SPDK_DEBUGLOG(nvmf, 39 "Subsystem %s namespace %u (%s) does not support io_type %d\n", 40 spdk_nvmf_subsystem_get_nqn(subsystem), 41 ns->opts.nsid, spdk_bdev_get_name(ns->bdev), (int)io_type); 42 return false; 43 } 44 } 45 46 SPDK_DEBUGLOG(nvmf, "All devices in Subsystem %s support io_type %d\n", 47 spdk_nvmf_subsystem_get_nqn(subsystem), (int)io_type); 48 return true; 49 } 50 51 bool 52 nvmf_ctrlr_dsm_supported(struct spdk_nvmf_ctrlr *ctrlr) 53 { 54 return nvmf_subsystem_bdev_io_type_supported(ctrlr->subsys, SPDK_BDEV_IO_TYPE_UNMAP); 55 } 56 57 bool 58 nvmf_ctrlr_write_zeroes_supported(struct spdk_nvmf_ctrlr *ctrlr) 59 { 60 return nvmf_subsystem_bdev_io_type_supported(ctrlr->subsys, SPDK_BDEV_IO_TYPE_WRITE_ZEROES); 61 } 62 63 static void 64 nvmf_bdev_ctrlr_complete_cmd(struct spdk_bdev_io *bdev_io, bool success, 65 void *cb_arg) 66 { 67 struct spdk_nvmf_request *req = cb_arg; 68 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 69 int first_sc = 0, first_sct = 0, sc = 0, sct = 0; 70 uint32_t cdw0 = 0; 71 struct spdk_nvmf_request *first_req = req->first_fused_req; 72 73 if (spdk_unlikely(first_req != NULL)) { 74 /* fused commands - get status for both operations */ 75 struct spdk_nvme_cpl *first_response = &first_req->rsp->nvme_cpl; 76 77 spdk_bdev_io_get_nvme_fused_status(bdev_io, &cdw0, &first_sct, &first_sc, &sct, &sc); 78 first_response->cdw0 = cdw0; 79 first_response->status.sc = first_sc; 80 first_response->status.sct = first_sct; 81 82 /* first request should be completed */ 83 spdk_nvmf_request_complete(first_req); 84 req->first_fused_req = NULL; 85 } else { 86 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 87 } 88 89 response->cdw0 = cdw0; 90 response->status.sc = sc; 91 response->status.sct = sct; 92 93 spdk_nvmf_request_complete(req); 94 spdk_bdev_free_io(bdev_io); 95 } 96 97 static void 98 nvmf_bdev_ctrlr_complete_admin_cmd(struct spdk_bdev_io *bdev_io, bool success, 99 void *cb_arg) 100 { 101 struct spdk_nvmf_request *req = cb_arg; 102 103 if (req->cmd_cb_fn) { 104 req->cmd_cb_fn(req); 105 } 106 107 nvmf_bdev_ctrlr_complete_cmd(bdev_io, success, req); 108 } 109 110 void 111 nvmf_bdev_ctrlr_identify_ns(struct spdk_nvmf_ns *ns, struct spdk_nvme_ns_data *nsdata, 112 bool dif_insert_or_strip) 113 { 114 struct spdk_bdev *bdev = ns->bdev; 115 uint64_t num_blocks; 116 uint32_t phys_blocklen; 117 118 num_blocks = spdk_bdev_get_num_blocks(bdev); 119 120 nsdata->nsze = num_blocks; 121 nsdata->ncap = num_blocks; 122 nsdata->nuse = num_blocks; 123 nsdata->nlbaf = 0; 124 nsdata->flbas.format = 0; 125 nsdata->nacwu = spdk_bdev_get_acwu(bdev) - 1; /* nacwu is 0-based */ 126 if (!dif_insert_or_strip) { 127 nsdata->lbaf[0].ms = spdk_bdev_get_md_size(bdev); 128 nsdata->lbaf[0].lbads = spdk_u32log2(spdk_bdev_get_block_size(bdev)); 129 if (nsdata->lbaf[0].ms != 0) { 130 nsdata->flbas.extended = 1; 131 nsdata->mc.extended = 1; 132 nsdata->mc.pointer = 0; 133 nsdata->dps.md_start = spdk_bdev_is_dif_head_of_md(bdev); 134 135 switch (spdk_bdev_get_dif_type(bdev)) { 136 case SPDK_DIF_TYPE1: 137 nsdata->dpc.pit1 = 1; 138 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE1; 139 break; 140 case SPDK_DIF_TYPE2: 141 nsdata->dpc.pit2 = 1; 142 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE2; 143 break; 144 case SPDK_DIF_TYPE3: 145 nsdata->dpc.pit3 = 1; 146 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE3; 147 break; 148 default: 149 SPDK_DEBUGLOG(nvmf, "Protection Disabled\n"); 150 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_DISABLE; 151 break; 152 } 153 } 154 } else { 155 nsdata->lbaf[0].ms = 0; 156 nsdata->lbaf[0].lbads = spdk_u32log2(spdk_bdev_get_data_block_size(bdev)); 157 } 158 159 phys_blocklen = spdk_bdev_get_physical_block_size(bdev); 160 assert(phys_blocklen > 0); 161 /* Linux driver uses min(nawupf, npwg) to set physical_block_size */ 162 nsdata->nsfeat.optperf = 1; 163 nsdata->nsfeat.ns_atomic_write_unit = 1; 164 nsdata->npwg = (phys_blocklen >> nsdata->lbaf[0].lbads) - 1; 165 nsdata->nawupf = nsdata->npwg; 166 nsdata->npwa = nsdata->npwg; 167 nsdata->npdg = nsdata->npwg; 168 nsdata->npda = nsdata->npwg; 169 170 nsdata->noiob = spdk_bdev_get_optimal_io_boundary(bdev); 171 nsdata->nmic.can_share = 1; 172 if (ns->ptpl_file != NULL) { 173 nsdata->nsrescap.rescap.persist = 1; 174 } 175 nsdata->nsrescap.rescap.write_exclusive = 1; 176 nsdata->nsrescap.rescap.exclusive_access = 1; 177 nsdata->nsrescap.rescap.write_exclusive_reg_only = 1; 178 nsdata->nsrescap.rescap.exclusive_access_reg_only = 1; 179 nsdata->nsrescap.rescap.write_exclusive_all_reg = 1; 180 nsdata->nsrescap.rescap.exclusive_access_all_reg = 1; 181 nsdata->nsrescap.rescap.ignore_existing_key = 1; 182 183 SPDK_STATIC_ASSERT(sizeof(nsdata->nguid) == sizeof(ns->opts.nguid), "size mismatch"); 184 memcpy(nsdata->nguid, ns->opts.nguid, sizeof(nsdata->nguid)); 185 186 SPDK_STATIC_ASSERT(sizeof(nsdata->eui64) == sizeof(ns->opts.eui64), "size mismatch"); 187 memcpy(&nsdata->eui64, ns->opts.eui64, sizeof(nsdata->eui64)); 188 } 189 190 static void 191 nvmf_bdev_ctrlr_get_rw_params(const struct spdk_nvme_cmd *cmd, uint64_t *start_lba, 192 uint64_t *num_blocks) 193 { 194 /* SLBA: CDW10 and CDW11 */ 195 *start_lba = from_le64(&cmd->cdw10); 196 197 /* NLB: CDW12 bits 15:00, 0's based */ 198 *num_blocks = (from_le32(&cmd->cdw12) & 0xFFFFu) + 1; 199 } 200 201 static bool 202 nvmf_bdev_ctrlr_lba_in_range(uint64_t bdev_num_blocks, uint64_t io_start_lba, 203 uint64_t io_num_blocks) 204 { 205 if (io_start_lba + io_num_blocks > bdev_num_blocks || 206 io_start_lba + io_num_blocks < io_start_lba) { 207 return false; 208 } 209 210 return true; 211 } 212 213 static void 214 nvmf_ctrlr_process_io_cmd_resubmit(void *arg) 215 { 216 struct spdk_nvmf_request *req = arg; 217 int rc; 218 219 rc = nvmf_ctrlr_process_io_cmd(req); 220 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 221 spdk_nvmf_request_complete(req); 222 } 223 } 224 225 static void 226 nvmf_ctrlr_process_admin_cmd_resubmit(void *arg) 227 { 228 struct spdk_nvmf_request *req = arg; 229 int rc; 230 231 rc = nvmf_ctrlr_process_admin_cmd(req); 232 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 233 spdk_nvmf_request_complete(req); 234 } 235 } 236 237 static void 238 nvmf_bdev_ctrl_queue_io(struct spdk_nvmf_request *req, struct spdk_bdev *bdev, 239 struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn, void *cb_arg) 240 { 241 int rc; 242 243 req->bdev_io_wait.bdev = bdev; 244 req->bdev_io_wait.cb_fn = cb_fn; 245 req->bdev_io_wait.cb_arg = cb_arg; 246 247 rc = spdk_bdev_queue_io_wait(bdev, ch, &req->bdev_io_wait); 248 if (rc != 0) { 249 assert(false); 250 } 251 req->qpair->group->stat.pending_bdev_io++; 252 } 253 254 bool 255 nvmf_bdev_zcopy_enabled(struct spdk_bdev *bdev) 256 { 257 return spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY); 258 } 259 260 int 261 nvmf_bdev_ctrlr_read_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 262 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 263 { 264 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 265 uint32_t block_size = spdk_bdev_get_block_size(bdev); 266 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 267 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 268 uint64_t start_lba; 269 uint64_t num_blocks; 270 int rc; 271 272 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 273 274 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 275 SPDK_ERRLOG("end of media\n"); 276 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 277 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 278 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 279 } 280 281 if (spdk_unlikely(num_blocks * block_size > req->length)) { 282 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 283 num_blocks, block_size, req->length); 284 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 285 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 286 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 287 } 288 289 assert(!spdk_nvmf_request_using_zcopy(req)); 290 291 rc = spdk_bdev_readv_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 292 nvmf_bdev_ctrlr_complete_cmd, req); 293 if (spdk_unlikely(rc)) { 294 if (rc == -ENOMEM) { 295 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 296 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 297 } 298 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 299 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 300 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 301 } 302 303 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 304 } 305 306 int 307 nvmf_bdev_ctrlr_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 308 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 309 { 310 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 311 uint32_t block_size = spdk_bdev_get_block_size(bdev); 312 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 313 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 314 uint64_t start_lba; 315 uint64_t num_blocks; 316 int rc; 317 318 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 319 320 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 321 SPDK_ERRLOG("end of media\n"); 322 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 323 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 324 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 325 } 326 327 if (spdk_unlikely(num_blocks * block_size > req->length)) { 328 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 329 num_blocks, block_size, req->length); 330 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 331 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 332 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 333 } 334 335 assert(!spdk_nvmf_request_using_zcopy(req)); 336 337 rc = spdk_bdev_writev_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 338 nvmf_bdev_ctrlr_complete_cmd, req); 339 if (spdk_unlikely(rc)) { 340 if (rc == -ENOMEM) { 341 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 342 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 343 } 344 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 345 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 346 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 347 } 348 349 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 350 } 351 352 int 353 nvmf_bdev_ctrlr_compare_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 354 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 355 { 356 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 357 uint32_t block_size = spdk_bdev_get_block_size(bdev); 358 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 359 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 360 uint64_t start_lba; 361 uint64_t num_blocks; 362 int rc; 363 364 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 365 366 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 367 SPDK_ERRLOG("end of media\n"); 368 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 369 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 370 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 371 } 372 373 if (spdk_unlikely(num_blocks * block_size > req->length)) { 374 SPDK_ERRLOG("Compare NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 375 num_blocks, block_size, req->length); 376 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 377 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 378 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 379 } 380 381 rc = spdk_bdev_comparev_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 382 nvmf_bdev_ctrlr_complete_cmd, req); 383 if (spdk_unlikely(rc)) { 384 if (rc == -ENOMEM) { 385 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 386 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 387 } 388 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 389 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 390 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 391 } 392 393 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 394 } 395 396 int 397 nvmf_bdev_ctrlr_compare_and_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 398 struct spdk_io_channel *ch, struct spdk_nvmf_request *cmp_req, struct spdk_nvmf_request *write_req) 399 { 400 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 401 uint32_t block_size = spdk_bdev_get_block_size(bdev); 402 struct spdk_nvme_cmd *cmp_cmd = &cmp_req->cmd->nvme_cmd; 403 struct spdk_nvme_cmd *write_cmd = &write_req->cmd->nvme_cmd; 404 struct spdk_nvme_cpl *rsp = &write_req->rsp->nvme_cpl; 405 uint64_t write_start_lba, cmp_start_lba; 406 uint64_t write_num_blocks, cmp_num_blocks; 407 int rc; 408 409 nvmf_bdev_ctrlr_get_rw_params(cmp_cmd, &cmp_start_lba, &cmp_num_blocks); 410 nvmf_bdev_ctrlr_get_rw_params(write_cmd, &write_start_lba, &write_num_blocks); 411 412 if (spdk_unlikely(write_start_lba != cmp_start_lba || write_num_blocks != cmp_num_blocks)) { 413 SPDK_ERRLOG("Fused command start lba / num blocks mismatch\n"); 414 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 415 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 416 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 417 } 418 419 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, write_start_lba, 420 write_num_blocks))) { 421 SPDK_ERRLOG("end of media\n"); 422 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 423 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 424 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 425 } 426 427 if (spdk_unlikely(write_num_blocks * block_size > write_req->length)) { 428 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 429 write_num_blocks, block_size, write_req->length); 430 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 431 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 432 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 433 } 434 435 rc = spdk_bdev_comparev_and_writev_blocks(desc, ch, cmp_req->iov, cmp_req->iovcnt, write_req->iov, 436 write_req->iovcnt, write_start_lba, write_num_blocks, nvmf_bdev_ctrlr_complete_cmd, write_req); 437 if (spdk_unlikely(rc)) { 438 if (rc == -ENOMEM) { 439 nvmf_bdev_ctrl_queue_io(cmp_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, cmp_req); 440 nvmf_bdev_ctrl_queue_io(write_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, write_req); 441 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 442 } 443 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 444 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 445 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 446 } 447 448 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 449 } 450 451 int 452 nvmf_bdev_ctrlr_write_zeroes_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 453 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 454 { 455 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 456 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 457 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 458 uint64_t start_lba; 459 uint64_t num_blocks; 460 int rc; 461 462 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 463 464 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 465 SPDK_ERRLOG("end of media\n"); 466 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 467 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 468 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 469 } 470 471 rc = spdk_bdev_write_zeroes_blocks(desc, ch, start_lba, num_blocks, 472 nvmf_bdev_ctrlr_complete_cmd, req); 473 if (spdk_unlikely(rc)) { 474 if (rc == -ENOMEM) { 475 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 476 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 477 } 478 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 479 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 480 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 481 } 482 483 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 484 } 485 486 int 487 nvmf_bdev_ctrlr_flush_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 488 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 489 { 490 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 491 int rc; 492 493 /* As for NVMeoF controller, SPDK always set volatile write 494 * cache bit to 1, return success for those block devices 495 * which can't support FLUSH command. 496 */ 497 if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_FLUSH)) { 498 response->status.sct = SPDK_NVME_SCT_GENERIC; 499 response->status.sc = SPDK_NVME_SC_SUCCESS; 500 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 501 } 502 503 rc = spdk_bdev_flush_blocks(desc, ch, 0, spdk_bdev_get_num_blocks(bdev), 504 nvmf_bdev_ctrlr_complete_cmd, req); 505 if (spdk_unlikely(rc)) { 506 if (rc == -ENOMEM) { 507 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 508 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 509 } 510 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 511 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 512 } 513 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 514 } 515 516 struct nvmf_bdev_ctrlr_unmap { 517 struct spdk_nvmf_request *req; 518 uint32_t count; 519 struct spdk_bdev_desc *desc; 520 struct spdk_bdev *bdev; 521 struct spdk_io_channel *ch; 522 uint32_t range_index; 523 }; 524 525 static void 526 nvmf_bdev_ctrlr_unmap_cpl(struct spdk_bdev_io *bdev_io, bool success, 527 void *cb_arg) 528 { 529 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = cb_arg; 530 struct spdk_nvmf_request *req = unmap_ctx->req; 531 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 532 int sc, sct; 533 uint32_t cdw0; 534 535 unmap_ctx->count--; 536 537 if (response->status.sct == SPDK_NVME_SCT_GENERIC && 538 response->status.sc == SPDK_NVME_SC_SUCCESS) { 539 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 540 response->cdw0 = cdw0; 541 response->status.sc = sc; 542 response->status.sct = sct; 543 } 544 545 if (unmap_ctx->count == 0) { 546 spdk_nvmf_request_complete(req); 547 free(unmap_ctx); 548 } 549 spdk_bdev_free_io(bdev_io); 550 } 551 552 static int nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 553 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 554 struct nvmf_bdev_ctrlr_unmap *unmap_ctx); 555 static void 556 nvmf_bdev_ctrlr_unmap_resubmit(void *arg) 557 { 558 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = arg; 559 struct spdk_nvmf_request *req = unmap_ctx->req; 560 struct spdk_bdev_desc *desc = unmap_ctx->desc; 561 struct spdk_bdev *bdev = unmap_ctx->bdev; 562 struct spdk_io_channel *ch = unmap_ctx->ch; 563 564 nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, unmap_ctx); 565 } 566 567 static int 568 nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 569 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 570 struct nvmf_bdev_ctrlr_unmap *unmap_ctx) 571 { 572 uint16_t nr, i; 573 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 574 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 575 struct spdk_nvme_dsm_range *dsm_range; 576 uint64_t lba; 577 uint32_t lba_count; 578 int rc; 579 580 nr = cmd->cdw10_bits.dsm.nr + 1; 581 if (nr * sizeof(struct spdk_nvme_dsm_range) > req->length) { 582 SPDK_ERRLOG("Dataset Management number of ranges > SGL length\n"); 583 response->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 584 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 585 } 586 587 if (unmap_ctx == NULL) { 588 unmap_ctx = calloc(1, sizeof(*unmap_ctx)); 589 if (!unmap_ctx) { 590 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 591 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 592 } 593 594 unmap_ctx->req = req; 595 unmap_ctx->desc = desc; 596 unmap_ctx->ch = ch; 597 unmap_ctx->bdev = bdev; 598 599 response->status.sct = SPDK_NVME_SCT_GENERIC; 600 response->status.sc = SPDK_NVME_SC_SUCCESS; 601 } else { 602 unmap_ctx->count--; /* dequeued */ 603 } 604 605 dsm_range = (struct spdk_nvme_dsm_range *)req->data; 606 for (i = unmap_ctx->range_index; i < nr; i++) { 607 lba = dsm_range[i].starting_lba; 608 lba_count = dsm_range[i].length; 609 610 unmap_ctx->count++; 611 612 rc = spdk_bdev_unmap_blocks(desc, ch, lba, lba_count, 613 nvmf_bdev_ctrlr_unmap_cpl, unmap_ctx); 614 if (rc) { 615 if (rc == -ENOMEM) { 616 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_bdev_ctrlr_unmap_resubmit, unmap_ctx); 617 /* Unmap was not yet submitted to bdev */ 618 /* unmap_ctx->count will be decremented when the request is dequeued */ 619 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 620 } 621 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 622 unmap_ctx->count--; 623 /* We can't return here - we may have to wait for any other 624 * unmaps already sent to complete */ 625 break; 626 } 627 unmap_ctx->range_index++; 628 } 629 630 if (unmap_ctx->count == 0) { 631 free(unmap_ctx); 632 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 633 } 634 635 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 636 } 637 638 int 639 nvmf_bdev_ctrlr_dsm_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 640 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 641 { 642 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 643 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 644 645 if (cmd->cdw11_bits.dsm.ad) { 646 return nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, NULL); 647 } 648 649 response->status.sct = SPDK_NVME_SCT_GENERIC; 650 response->status.sc = SPDK_NVME_SC_SUCCESS; 651 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 652 } 653 654 int 655 nvmf_bdev_ctrlr_nvme_passthru_io(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 656 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 657 { 658 int rc; 659 660 rc = spdk_bdev_nvme_io_passthru(desc, ch, &req->cmd->nvme_cmd, req->data, req->length, 661 nvmf_bdev_ctrlr_complete_cmd, req); 662 if (spdk_unlikely(rc)) { 663 if (rc == -ENOMEM) { 664 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 665 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 666 } 667 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 668 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 669 req->rsp->nvme_cpl.status.dnr = 1; 670 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 671 } 672 673 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 674 } 675 676 int 677 spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 678 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 679 spdk_nvmf_nvme_passthru_cmd_cb cb_fn) 680 { 681 int rc; 682 683 req->cmd_cb_fn = cb_fn; 684 685 rc = spdk_bdev_nvme_admin_passthru(desc, ch, &req->cmd->nvme_cmd, req->data, req->length, 686 nvmf_bdev_ctrlr_complete_admin_cmd, req); 687 if (spdk_unlikely(rc)) { 688 if (rc == -ENOMEM) { 689 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 690 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 691 } 692 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 693 if (rc == -ENOTSUP) { 694 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 695 } else { 696 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 697 } 698 699 req->rsp->nvme_cpl.status.dnr = 1; 700 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 701 } 702 703 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 704 } 705 706 static void 707 nvmf_bdev_ctrlr_complete_abort_cmd(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg) 708 { 709 struct spdk_nvmf_request *req = cb_arg; 710 711 if (success) { 712 req->rsp->nvme_cpl.cdw0 &= ~1U; 713 } 714 715 spdk_nvmf_request_complete(req); 716 spdk_bdev_free_io(bdev_io); 717 } 718 719 int 720 spdk_nvmf_bdev_ctrlr_abort_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 721 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 722 struct spdk_nvmf_request *req_to_abort) 723 { 724 int rc; 725 726 assert((req->rsp->nvme_cpl.cdw0 & 1U) != 0); 727 728 rc = spdk_bdev_abort(desc, ch, req_to_abort, nvmf_bdev_ctrlr_complete_abort_cmd, req); 729 if (spdk_likely(rc == 0)) { 730 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 731 } else if (rc == -ENOMEM) { 732 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 733 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 734 } else { 735 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 736 } 737 } 738 739 bool 740 nvmf_bdev_ctrlr_get_dif_ctx(struct spdk_bdev *bdev, struct spdk_nvme_cmd *cmd, 741 struct spdk_dif_ctx *dif_ctx) 742 { 743 uint32_t init_ref_tag, dif_check_flags = 0; 744 int rc; 745 746 if (spdk_bdev_get_md_size(bdev) == 0) { 747 return false; 748 } 749 750 /* Initial Reference Tag is the lower 32 bits of the start LBA. */ 751 init_ref_tag = (uint32_t)from_le64(&cmd->cdw10); 752 753 if (spdk_bdev_is_dif_check_enabled(bdev, SPDK_DIF_CHECK_TYPE_REFTAG)) { 754 dif_check_flags |= SPDK_DIF_FLAGS_REFTAG_CHECK; 755 } 756 757 if (spdk_bdev_is_dif_check_enabled(bdev, SPDK_DIF_CHECK_TYPE_GUARD)) { 758 dif_check_flags |= SPDK_DIF_FLAGS_GUARD_CHECK; 759 } 760 761 rc = spdk_dif_ctx_init(dif_ctx, 762 spdk_bdev_get_block_size(bdev), 763 spdk_bdev_get_md_size(bdev), 764 spdk_bdev_is_md_interleaved(bdev), 765 spdk_bdev_is_dif_head_of_md(bdev), 766 spdk_bdev_get_dif_type(bdev), 767 dif_check_flags, 768 init_ref_tag, 0, 0, 0, 0); 769 770 return (rc == 0) ? true : false; 771 } 772 773 static void 774 nvmf_bdev_ctrlr_zcopy_start_complete(struct spdk_bdev_io *bdev_io, bool success, 775 void *cb_arg) 776 { 777 struct spdk_nvmf_request *req = cb_arg; 778 struct iovec *iov; 779 int iovcnt; 780 781 if (spdk_unlikely(!success)) { 782 int sc = 0, sct = 0; 783 uint32_t cdw0 = 0; 784 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 785 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 786 787 response->cdw0 = cdw0; 788 response->status.sc = sc; 789 response->status.sct = sct; 790 791 spdk_bdev_free_io(bdev_io); 792 spdk_nvmf_request_complete(req); 793 return; 794 } 795 796 spdk_bdev_io_get_iovec(bdev_io, &iov, &iovcnt); 797 798 assert(iovcnt <= NVMF_REQ_MAX_BUFFERS); 799 assert(iovcnt > 0); 800 801 req->iovcnt = iovcnt; 802 803 assert(req->iov == iov); 804 805 /* backward compatible */ 806 req->data = req->iov[0].iov_base; 807 808 req->zcopy_bdev_io = bdev_io; /* Preserve the bdev_io for the end zcopy */ 809 810 spdk_nvmf_request_complete(req); 811 /* Don't free the bdev_io here as it is needed for the END ZCOPY */ 812 } 813 814 int 815 nvmf_bdev_ctrlr_zcopy_start(struct spdk_bdev *bdev, 816 struct spdk_bdev_desc *desc, 817 struct spdk_io_channel *ch, 818 struct spdk_nvmf_request *req) 819 { 820 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 821 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 822 uint32_t block_size = spdk_bdev_get_block_size(bdev); 823 uint64_t start_lba; 824 uint64_t num_blocks; 825 int rc; 826 827 nvmf_bdev_ctrlr_get_rw_params(&req->cmd->nvme_cmd, &start_lba, &num_blocks); 828 829 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 830 SPDK_ERRLOG("end of media\n"); 831 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 832 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 833 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 834 } 835 836 if (spdk_unlikely(num_blocks * block_size > req->length)) { 837 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 838 num_blocks, block_size, req->length); 839 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 840 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 841 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 842 } 843 844 bool populate = (req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_READ) ? true : false; 845 846 rc = spdk_bdev_zcopy_start(desc, ch, req->iov, req->iovcnt, start_lba, 847 num_blocks, populate, nvmf_bdev_ctrlr_zcopy_start_complete, req); 848 if (spdk_unlikely(rc != 0)) { 849 if (rc == -ENOMEM) { 850 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 851 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 852 } 853 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 854 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 855 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 856 } 857 858 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 859 } 860 861 static void 862 nvmf_bdev_ctrlr_zcopy_end_complete(struct spdk_bdev_io *bdev_io, bool success, 863 void *cb_arg) 864 { 865 struct spdk_nvmf_request *req = cb_arg; 866 867 if (spdk_unlikely(!success)) { 868 int sc = 0, sct = 0; 869 uint32_t cdw0 = 0; 870 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 871 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 872 873 response->cdw0 = cdw0; 874 response->status.sc = sc; 875 response->status.sct = sct; 876 } 877 878 spdk_bdev_free_io(bdev_io); 879 req->zcopy_bdev_io = NULL; 880 spdk_nvmf_request_complete(req); 881 } 882 883 void 884 nvmf_bdev_ctrlr_zcopy_end(struct spdk_nvmf_request *req, bool commit) 885 { 886 int rc __attribute__((unused)); 887 888 rc = spdk_bdev_zcopy_end(req->zcopy_bdev_io, commit, nvmf_bdev_ctrlr_zcopy_end_complete, req); 889 890 /* The only way spdk_bdev_zcopy_end() can fail is if we pass a bdev_io type that isn't ZCOPY */ 891 assert(rc == 0); 892 } 893