1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright (C) 2017 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 bool 64 nvmf_ctrlr_copy_supported(struct spdk_nvmf_ctrlr *ctrlr) 65 { 66 return nvmf_subsystem_bdev_io_type_supported(ctrlr->subsys, SPDK_BDEV_IO_TYPE_COPY); 67 } 68 69 static void 70 nvmf_bdev_ctrlr_complete_cmd(struct spdk_bdev_io *bdev_io, bool success, 71 void *cb_arg) 72 { 73 struct spdk_nvmf_request *req = cb_arg; 74 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 75 int sc = 0, sct = 0; 76 uint32_t cdw0 = 0; 77 78 if (spdk_unlikely(req->first_fused)) { 79 struct spdk_nvmf_request *first_req = req->first_fused_req; 80 struct spdk_nvme_cpl *first_response = &first_req->rsp->nvme_cpl; 81 int first_sc = 0, first_sct = 0; 82 83 /* get status for both operations */ 84 spdk_bdev_io_get_nvme_fused_status(bdev_io, &cdw0, &first_sct, &first_sc, &sct, &sc); 85 first_response->cdw0 = cdw0; 86 first_response->status.sc = first_sc; 87 first_response->status.sct = first_sct; 88 89 /* first request should be completed */ 90 spdk_nvmf_request_complete(first_req); 91 req->first_fused_req = NULL; 92 req->first_fused = false; 93 } else { 94 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 95 } 96 97 response->cdw0 = cdw0; 98 response->status.sc = sc; 99 response->status.sct = sct; 100 101 spdk_nvmf_request_complete(req); 102 spdk_bdev_free_io(bdev_io); 103 } 104 105 static void 106 nvmf_bdev_ctrlr_complete_admin_cmd(struct spdk_bdev_io *bdev_io, bool success, 107 void *cb_arg) 108 { 109 struct spdk_nvmf_request *req = cb_arg; 110 111 if (req->cmd_cb_fn) { 112 req->cmd_cb_fn(req); 113 } 114 115 nvmf_bdev_ctrlr_complete_cmd(bdev_io, success, req); 116 } 117 118 void 119 nvmf_bdev_ctrlr_identify_ns(struct spdk_nvmf_ns *ns, struct spdk_nvme_ns_data *nsdata, 120 bool dif_insert_or_strip) 121 { 122 struct spdk_bdev *bdev = ns->bdev; 123 struct spdk_bdev_desc *desc = ns->desc; 124 uint64_t num_blocks; 125 uint32_t phys_blocklen; 126 uint32_t max_copy; 127 128 num_blocks = spdk_bdev_get_num_blocks(bdev); 129 130 nsdata->nsze = num_blocks; 131 nsdata->ncap = num_blocks; 132 nsdata->nuse = num_blocks; 133 nsdata->nlbaf = 0; 134 nsdata->flbas.format = 0; 135 nsdata->flbas.msb_format = 0; 136 nsdata->nacwu = spdk_bdev_get_acwu(bdev) - 1; /* nacwu is 0-based */ 137 if (!dif_insert_or_strip) { 138 nsdata->lbaf[0].ms = spdk_bdev_desc_get_md_size(desc); 139 nsdata->lbaf[0].lbads = spdk_u32log2(spdk_bdev_desc_get_block_size(desc)); 140 if (nsdata->lbaf[0].ms != 0) { 141 nsdata->flbas.extended = 1; 142 nsdata->mc.extended = 1; 143 nsdata->mc.pointer = 0; 144 nsdata->dps.md_start = spdk_bdev_desc_is_dif_head_of_md(desc); 145 /* NVMf library doesn't process PRACT and PRCHK flags, we 146 * leave the use of extended LBA buffer to users. 147 */ 148 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_DISABLE; 149 } 150 } else { 151 nsdata->lbaf[0].ms = 0; 152 nsdata->lbaf[0].lbads = spdk_u32log2(spdk_bdev_get_data_block_size(bdev)); 153 } 154 155 phys_blocklen = spdk_bdev_get_physical_block_size(bdev); 156 assert(phys_blocklen > 0); 157 /* Linux driver uses min(nawupf, npwg) to set physical_block_size */ 158 nsdata->nsfeat.optperf = 1; 159 nsdata->nsfeat.ns_atomic_write_unit = 1; 160 nsdata->npwg = (phys_blocklen >> nsdata->lbaf[0].lbads) - 1; 161 nsdata->nawupf = nsdata->npwg; 162 nsdata->npwa = nsdata->npwg; 163 nsdata->npdg = nsdata->npwg; 164 nsdata->npda = nsdata->npwg; 165 166 if (spdk_bdev_get_write_unit_size(bdev) == 1) { 167 nsdata->noiob = spdk_bdev_get_optimal_io_boundary(bdev); 168 } 169 nsdata->nmic.can_share = 1; 170 if (nvmf_ns_is_ptpl_capable(ns)) { 171 nsdata->nsrescap.rescap.persist = 1; 172 } 173 nsdata->nsrescap.rescap.write_exclusive = 1; 174 nsdata->nsrescap.rescap.exclusive_access = 1; 175 nsdata->nsrescap.rescap.write_exclusive_reg_only = 1; 176 nsdata->nsrescap.rescap.exclusive_access_reg_only = 1; 177 nsdata->nsrescap.rescap.write_exclusive_all_reg = 1; 178 nsdata->nsrescap.rescap.exclusive_access_all_reg = 1; 179 nsdata->nsrescap.rescap.ignore_existing_key = 1; 180 181 SPDK_STATIC_ASSERT(sizeof(nsdata->nguid) == sizeof(ns->opts.nguid), "size mismatch"); 182 memcpy(nsdata->nguid, ns->opts.nguid, sizeof(nsdata->nguid)); 183 184 SPDK_STATIC_ASSERT(sizeof(nsdata->eui64) == sizeof(ns->opts.eui64), "size mismatch"); 185 memcpy(&nsdata->eui64, ns->opts.eui64, sizeof(nsdata->eui64)); 186 187 /* For now we support just one source range for copy command */ 188 nsdata->msrc = 0; 189 190 max_copy = spdk_bdev_get_max_copy(bdev); 191 if (max_copy == 0 || max_copy > UINT16_MAX) { 192 /* Zero means copy size is unlimited */ 193 nsdata->mcl = UINT16_MAX; 194 nsdata->mssrl = UINT16_MAX; 195 } else { 196 nsdata->mcl = max_copy; 197 nsdata->mssrl = max_copy; 198 } 199 } 200 201 void 202 nvmf_bdev_ctrlr_identify_iocs_nvm(struct spdk_nvmf_ns *ns, 203 struct spdk_nvme_nvm_ns_data *nsdata_nvm) 204 { 205 struct spdk_bdev_desc *desc = ns->desc; 206 uint8_t _16bpists; 207 uint32_t sts, pif; 208 209 if (spdk_bdev_desc_get_dif_type(desc) == SPDK_DIF_DISABLE) { 210 return; 211 } 212 213 pif = spdk_bdev_desc_get_dif_pi_format(desc); 214 215 /* 216 * 16BPISTS shall be 1 for 32/64b Guard PI. 217 * STCRS shall be 1 if 16BPISTS is 1. 218 * 16 is the minimum value of STS for 32b Guard PI. 219 */ 220 switch (pif) { 221 case SPDK_DIF_PI_FORMAT_16: 222 _16bpists = 0; 223 sts = 0; 224 break; 225 case SPDK_DIF_PI_FORMAT_32: 226 _16bpists = 1; 227 sts = 16; 228 break; 229 case SPDK_DIF_PI_FORMAT_64: 230 _16bpists = 1; 231 sts = 0; 232 break; 233 default: 234 SPDK_WARNLOG("PI format %u is not supported\n", pif); 235 return; 236 } 237 238 /* For 16b Guard PI, Storage Tag is not available because we set STS to 0. 239 * In this case, we do not have to set 16BPISTM to 1. For simplicity, 240 * set 16BPISTM to 0 and set LBSTM to all zeroes. 241 * 242 * We will revisit here when we find any OS uses Storage Tag. 243 */ 244 nsdata_nvm->lbstm = 0; 245 nsdata_nvm->pic._16bpistm = 0; 246 247 nsdata_nvm->pic._16bpists = _16bpists; 248 nsdata_nvm->pic.stcrs = 0; 249 nsdata_nvm->elbaf[0].sts = sts; 250 nsdata_nvm->elbaf[0].pif = pif; 251 } 252 253 static void 254 nvmf_bdev_ctrlr_get_rw_params(const struct spdk_nvme_cmd *cmd, uint64_t *start_lba, 255 uint64_t *num_blocks) 256 { 257 /* SLBA: CDW10 and CDW11 */ 258 *start_lba = from_le64(&cmd->cdw10); 259 260 /* NLB: CDW12 bits 15:00, 0's based */ 261 *num_blocks = (from_le32(&cmd->cdw12) & 0xFFFFu) + 1; 262 } 263 264 static void 265 nvmf_bdev_ctrlr_get_rw_ext_params(const struct spdk_nvme_cmd *cmd, 266 struct spdk_bdev_ext_io_opts *opts) 267 { 268 /* Get CDW12 values */ 269 opts->nvme_cdw12.raw = from_le32(&cmd->cdw12); 270 271 /* Get CDW13 values */ 272 opts->nvme_cdw13.raw = from_le32(&cmd->cdw13); 273 274 /* Bdev layer checks PRACT in CDW12 because it is NVMe specific, but 275 * it does not check DIF check flags in CDW because DIF is not NVMe 276 * specific. Hence, copy DIF check flags from CDW12 to dif_check_flags_exclude_mask. 277 */ 278 opts->dif_check_flags_exclude_mask = (~opts->nvme_cdw12.raw) & SPDK_NVME_IO_FLAGS_PRCHK_MASK; 279 } 280 281 static bool 282 nvmf_bdev_ctrlr_lba_in_range(uint64_t bdev_num_blocks, uint64_t io_start_lba, 283 uint64_t io_num_blocks) 284 { 285 if (io_start_lba + io_num_blocks > bdev_num_blocks || 286 io_start_lba + io_num_blocks < io_start_lba) { 287 return false; 288 } 289 290 return true; 291 } 292 293 static void 294 nvmf_ctrlr_process_io_cmd_resubmit(void *arg) 295 { 296 struct spdk_nvmf_request *req = arg; 297 int rc; 298 299 rc = nvmf_ctrlr_process_io_cmd(req); 300 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 301 spdk_nvmf_request_complete(req); 302 } 303 } 304 305 static void 306 nvmf_ctrlr_process_admin_cmd_resubmit(void *arg) 307 { 308 struct spdk_nvmf_request *req = arg; 309 int rc; 310 311 rc = nvmf_ctrlr_process_admin_cmd(req); 312 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 313 spdk_nvmf_request_complete(req); 314 } 315 } 316 317 static void 318 nvmf_bdev_ctrl_queue_io(struct spdk_nvmf_request *req, struct spdk_bdev *bdev, 319 struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn, void *cb_arg) 320 { 321 int rc; 322 323 req->bdev_io_wait.bdev = bdev; 324 req->bdev_io_wait.cb_fn = cb_fn; 325 req->bdev_io_wait.cb_arg = cb_arg; 326 327 rc = spdk_bdev_queue_io_wait(bdev, ch, &req->bdev_io_wait); 328 if (rc != 0) { 329 assert(false); 330 } 331 req->qpair->group->stat.pending_bdev_io++; 332 } 333 334 bool 335 nvmf_bdev_zcopy_enabled(struct spdk_bdev *bdev) 336 { 337 return spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY); 338 } 339 340 int 341 nvmf_bdev_ctrlr_read_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 342 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 343 { 344 struct spdk_bdev_ext_io_opts opts = { 345 .size = SPDK_SIZEOF(&opts, accel_sequence), 346 .memory_domain = req->memory_domain, 347 .memory_domain_ctx = req->memory_domain_ctx, 348 .accel_sequence = req->accel_sequence, 349 }; 350 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 351 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 352 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 353 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 354 uint64_t start_lba; 355 uint64_t num_blocks; 356 int rc; 357 358 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 359 nvmf_bdev_ctrlr_get_rw_ext_params(cmd, &opts); 360 361 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 362 SPDK_ERRLOG("end of media\n"); 363 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 364 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 365 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 366 } 367 368 if (spdk_unlikely(num_blocks * block_size > req->length)) { 369 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 370 num_blocks, block_size, req->length); 371 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 372 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 373 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 374 } 375 376 assert(!spdk_nvmf_request_using_zcopy(req)); 377 378 rc = spdk_bdev_readv_blocks_ext(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 379 nvmf_bdev_ctrlr_complete_cmd, req, &opts); 380 if (spdk_unlikely(rc)) { 381 if (rc == -ENOMEM) { 382 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 383 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 384 } 385 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 386 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 387 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 388 } 389 390 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 391 } 392 393 int 394 nvmf_bdev_ctrlr_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 395 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 396 { 397 struct spdk_bdev_ext_io_opts opts = { 398 .size = SPDK_SIZEOF(&opts, nvme_cdw13), 399 .memory_domain = req->memory_domain, 400 .memory_domain_ctx = req->memory_domain_ctx, 401 .accel_sequence = req->accel_sequence, 402 }; 403 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 404 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 405 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 406 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 407 uint64_t start_lba; 408 uint64_t num_blocks; 409 int rc; 410 411 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 412 nvmf_bdev_ctrlr_get_rw_ext_params(cmd, &opts); 413 414 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 415 SPDK_ERRLOG("end of media\n"); 416 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 417 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 418 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 419 } 420 421 if (spdk_unlikely(num_blocks * block_size > req->length)) { 422 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 423 num_blocks, block_size, req->length); 424 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 425 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 426 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 427 } 428 429 assert(!spdk_nvmf_request_using_zcopy(req)); 430 431 rc = spdk_bdev_writev_blocks_ext(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 432 nvmf_bdev_ctrlr_complete_cmd, req, &opts); 433 if (spdk_unlikely(rc)) { 434 if (rc == -ENOMEM) { 435 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 436 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 437 } 438 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 439 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 440 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 441 } 442 443 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 444 } 445 446 int 447 nvmf_bdev_ctrlr_compare_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 448 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 449 { 450 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 451 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 452 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 453 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 454 uint64_t start_lba; 455 uint64_t num_blocks; 456 int rc; 457 458 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 459 460 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 461 SPDK_ERRLOG("end of media\n"); 462 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 463 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 464 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 465 } 466 467 if (spdk_unlikely(num_blocks * block_size > req->length)) { 468 SPDK_ERRLOG("Compare NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 469 num_blocks, block_size, req->length); 470 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 471 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 472 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 473 } 474 475 rc = spdk_bdev_comparev_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 476 nvmf_bdev_ctrlr_complete_cmd, req); 477 if (spdk_unlikely(rc)) { 478 if (rc == -ENOMEM) { 479 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 480 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 481 } 482 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 483 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 484 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 485 } 486 487 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 488 } 489 490 int 491 nvmf_bdev_ctrlr_compare_and_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 492 struct spdk_io_channel *ch, struct spdk_nvmf_request *cmp_req, struct spdk_nvmf_request *write_req) 493 { 494 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 495 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 496 struct spdk_nvme_cmd *cmp_cmd = &cmp_req->cmd->nvme_cmd; 497 struct spdk_nvme_cmd *write_cmd = &write_req->cmd->nvme_cmd; 498 struct spdk_nvme_cpl *rsp = &write_req->rsp->nvme_cpl; 499 uint64_t write_start_lba, cmp_start_lba; 500 uint64_t write_num_blocks, cmp_num_blocks; 501 int rc; 502 503 nvmf_bdev_ctrlr_get_rw_params(cmp_cmd, &cmp_start_lba, &cmp_num_blocks); 504 nvmf_bdev_ctrlr_get_rw_params(write_cmd, &write_start_lba, &write_num_blocks); 505 506 if (spdk_unlikely(write_start_lba != cmp_start_lba || write_num_blocks != cmp_num_blocks)) { 507 SPDK_ERRLOG("Fused command start lba / num blocks mismatch\n"); 508 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 509 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 510 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 511 } 512 513 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, write_start_lba, 514 write_num_blocks))) { 515 SPDK_ERRLOG("end of media\n"); 516 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 517 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 518 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 519 } 520 521 if (spdk_unlikely(write_num_blocks * block_size > write_req->length)) { 522 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 523 write_num_blocks, block_size, write_req->length); 524 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 525 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 526 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 527 } 528 529 rc = spdk_bdev_comparev_and_writev_blocks(desc, ch, cmp_req->iov, cmp_req->iovcnt, write_req->iov, 530 write_req->iovcnt, write_start_lba, write_num_blocks, nvmf_bdev_ctrlr_complete_cmd, write_req); 531 if (spdk_unlikely(rc)) { 532 if (rc == -ENOMEM) { 533 nvmf_bdev_ctrl_queue_io(cmp_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, cmp_req); 534 nvmf_bdev_ctrl_queue_io(write_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, write_req); 535 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 536 } 537 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 538 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 539 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 540 } 541 542 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 543 } 544 545 int 546 nvmf_bdev_ctrlr_write_zeroes_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 547 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 548 { 549 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 550 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 551 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 552 uint64_t max_write_zeroes_size = req->qpair->ctrlr->subsys->max_write_zeroes_size_kib; 553 uint64_t start_lba; 554 uint64_t num_blocks; 555 int rc; 556 557 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 558 if (spdk_unlikely(max_write_zeroes_size > 0 && 559 num_blocks > (max_write_zeroes_size << 10) / spdk_bdev_desc_get_block_size(desc))) { 560 SPDK_ERRLOG("invalid write zeroes size, should not exceed %" PRIu64 "Kib\n", max_write_zeroes_size); 561 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 562 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 563 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 564 } 565 566 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 567 SPDK_ERRLOG("end of media\n"); 568 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 569 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 570 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 571 } 572 573 if (spdk_unlikely(cmd->cdw12_bits.write_zeroes.deac)) { 574 SPDK_ERRLOG("Write Zeroes Deallocate is not supported\n"); 575 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 576 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 577 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 578 } 579 580 rc = spdk_bdev_write_zeroes_blocks(desc, ch, start_lba, num_blocks, 581 nvmf_bdev_ctrlr_complete_cmd, req); 582 if (spdk_unlikely(rc)) { 583 if (rc == -ENOMEM) { 584 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 585 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 586 } 587 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 588 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 589 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 590 } 591 592 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 593 } 594 595 int 596 nvmf_bdev_ctrlr_flush_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 597 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 598 { 599 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 600 int rc; 601 602 /* As for NVMeoF controller, SPDK always set volatile write 603 * cache bit to 1, return success for those block devices 604 * which can't support FLUSH command. 605 */ 606 if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_FLUSH)) { 607 response->status.sct = SPDK_NVME_SCT_GENERIC; 608 response->status.sc = SPDK_NVME_SC_SUCCESS; 609 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 610 } 611 612 rc = spdk_bdev_flush_blocks(desc, ch, 0, spdk_bdev_get_num_blocks(bdev), 613 nvmf_bdev_ctrlr_complete_cmd, req); 614 if (spdk_unlikely(rc)) { 615 if (rc == -ENOMEM) { 616 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 617 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 618 } 619 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 620 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 621 } 622 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 623 } 624 625 struct nvmf_bdev_ctrlr_unmap { 626 struct spdk_nvmf_request *req; 627 uint32_t count; 628 struct spdk_bdev_desc *desc; 629 struct spdk_bdev *bdev; 630 struct spdk_io_channel *ch; 631 uint32_t range_index; 632 }; 633 634 static void 635 nvmf_bdev_ctrlr_unmap_cpl(struct spdk_bdev_io *bdev_io, bool success, 636 void *cb_arg) 637 { 638 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = cb_arg; 639 struct spdk_nvmf_request *req = unmap_ctx->req; 640 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 641 int sc, sct; 642 uint32_t cdw0; 643 644 unmap_ctx->count--; 645 646 if (response->status.sct == SPDK_NVME_SCT_GENERIC && 647 response->status.sc == SPDK_NVME_SC_SUCCESS) { 648 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 649 response->cdw0 = cdw0; 650 response->status.sc = sc; 651 response->status.sct = sct; 652 } 653 654 if (unmap_ctx->count == 0) { 655 spdk_nvmf_request_complete(req); 656 free(unmap_ctx); 657 } 658 spdk_bdev_free_io(bdev_io); 659 } 660 661 static int nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 662 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 663 struct nvmf_bdev_ctrlr_unmap *unmap_ctx); 664 static void 665 nvmf_bdev_ctrlr_unmap_resubmit(void *arg) 666 { 667 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = arg; 668 struct spdk_nvmf_request *req = unmap_ctx->req; 669 struct spdk_bdev_desc *desc = unmap_ctx->desc; 670 struct spdk_bdev *bdev = unmap_ctx->bdev; 671 struct spdk_io_channel *ch = unmap_ctx->ch; 672 673 nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, unmap_ctx); 674 } 675 676 static int 677 nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 678 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 679 struct nvmf_bdev_ctrlr_unmap *unmap_ctx) 680 { 681 uint16_t nr, i; 682 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 683 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 684 uint64_t max_discard_size = req->qpair->ctrlr->subsys->max_discard_size_kib; 685 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 686 struct spdk_iov_xfer ix; 687 uint64_t lba; 688 uint32_t lba_count; 689 int rc; 690 691 nr = cmd->cdw10_bits.dsm.nr + 1; 692 if (nr * sizeof(struct spdk_nvme_dsm_range) > req->length) { 693 SPDK_ERRLOG("Dataset Management number of ranges > SGL length\n"); 694 response->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 695 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 696 } 697 698 if (unmap_ctx == NULL) { 699 unmap_ctx = calloc(1, sizeof(*unmap_ctx)); 700 if (!unmap_ctx) { 701 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 702 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 703 } 704 705 unmap_ctx->req = req; 706 unmap_ctx->desc = desc; 707 unmap_ctx->ch = ch; 708 unmap_ctx->bdev = bdev; 709 710 response->status.sct = SPDK_NVME_SCT_GENERIC; 711 response->status.sc = SPDK_NVME_SC_SUCCESS; 712 } else { 713 unmap_ctx->count--; /* dequeued */ 714 } 715 716 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 717 718 for (i = unmap_ctx->range_index; i < nr; i++) { 719 struct spdk_nvme_dsm_range dsm_range = { 0 }; 720 721 spdk_iov_xfer_to_buf(&ix, &dsm_range, sizeof(dsm_range)); 722 723 lba = dsm_range.starting_lba; 724 lba_count = dsm_range.length; 725 if (max_discard_size > 0 && lba_count > (max_discard_size << 10) / block_size) { 726 SPDK_ERRLOG("invalid unmap size %" PRIu32 " blocks, should not exceed %" PRIu64 " blocks\n", 727 lba_count, max_discard_size << 1); 728 response->status.sct = SPDK_NVME_SCT_GENERIC; 729 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 730 break; 731 } 732 733 unmap_ctx->count++; 734 735 rc = spdk_bdev_unmap_blocks(desc, ch, lba, lba_count, 736 nvmf_bdev_ctrlr_unmap_cpl, unmap_ctx); 737 if (rc) { 738 if (rc == -ENOMEM) { 739 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_bdev_ctrlr_unmap_resubmit, unmap_ctx); 740 /* Unmap was not yet submitted to bdev */ 741 /* unmap_ctx->count will be decremented when the request is dequeued */ 742 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 743 } 744 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 745 unmap_ctx->count--; 746 /* We can't return here - we may have to wait for any other 747 * unmaps already sent to complete */ 748 break; 749 } 750 unmap_ctx->range_index++; 751 } 752 753 if (unmap_ctx->count == 0) { 754 free(unmap_ctx); 755 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 756 } 757 758 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 759 } 760 761 int 762 nvmf_bdev_ctrlr_dsm_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 763 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 764 { 765 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 766 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 767 768 if (cmd->cdw11_bits.dsm.ad) { 769 return nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, NULL); 770 } 771 772 response->status.sct = SPDK_NVME_SCT_GENERIC; 773 response->status.sc = SPDK_NVME_SC_SUCCESS; 774 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 775 } 776 777 int 778 nvmf_bdev_ctrlr_copy_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 779 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 780 { 781 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 782 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 783 uint64_t sdlba = ((uint64_t)cmd->cdw11 << 32) + cmd->cdw10; 784 struct spdk_nvme_scc_source_range range = { 0 }; 785 struct spdk_iov_xfer ix; 786 int rc; 787 788 SPDK_DEBUGLOG(nvmf, "Copy command: SDLBA %lu, NR %u, desc format %u, PRINFOR %u, " 789 "DTYPE %u, STCW %u, PRINFOW %u, FUA %u, LR %u\n", 790 sdlba, 791 cmd->cdw12_bits.copy.nr, 792 cmd->cdw12_bits.copy.df, 793 cmd->cdw12_bits.copy.prinfor, 794 cmd->cdw12_bits.copy.dtype, 795 cmd->cdw12_bits.copy.stcw, 796 cmd->cdw12_bits.copy.prinfow, 797 cmd->cdw12_bits.copy.fua, 798 cmd->cdw12_bits.copy.lr); 799 800 if (spdk_unlikely(req->length != (cmd->cdw12_bits.copy.nr + 1) * 801 sizeof(struct spdk_nvme_scc_source_range))) { 802 response->status.sct = SPDK_NVME_SCT_GENERIC; 803 response->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 804 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 805 } 806 807 /* 808 * We support only one source range, and rely on this with the xfer 809 * below. 810 */ 811 if (cmd->cdw12_bits.copy.nr > 0) { 812 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 813 response->status.sc = SPDK_NVME_SC_CMD_SIZE_LIMIT_SIZE_EXCEEDED; 814 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 815 } 816 817 if (cmd->cdw12_bits.copy.df != 0) { 818 response->status.sct = SPDK_NVME_SCT_GENERIC; 819 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 820 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 821 } 822 823 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 824 spdk_iov_xfer_to_buf(&ix, &range, sizeof(range)); 825 826 rc = spdk_bdev_copy_blocks(desc, ch, sdlba, range.slba, range.nlb + 1, 827 nvmf_bdev_ctrlr_complete_cmd, req); 828 if (spdk_unlikely(rc)) { 829 if (rc == -ENOMEM) { 830 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 831 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 832 } 833 834 response->status.sct = SPDK_NVME_SCT_GENERIC; 835 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 836 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 837 } 838 839 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 840 } 841 842 int 843 nvmf_bdev_ctrlr_nvme_passthru_io(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 844 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 845 { 846 int rc; 847 848 rc = spdk_bdev_nvme_iov_passthru_md(desc, ch, &req->cmd->nvme_cmd, req->iov, req->iovcnt, 849 req->length, NULL, 0, nvmf_bdev_ctrlr_complete_cmd, req); 850 851 if (spdk_unlikely(rc)) { 852 if (rc == -ENOMEM) { 853 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 854 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 855 } 856 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 857 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 858 req->rsp->nvme_cpl.status.dnr = 1; 859 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 860 } 861 862 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 863 } 864 865 int 866 spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 867 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 868 spdk_nvmf_nvme_passthru_cmd_cb cb_fn) 869 { 870 int rc; 871 872 if (spdk_unlikely(req->iovcnt > 1)) { 873 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 874 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 875 req->rsp->nvme_cpl.status.dnr = 1; 876 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 877 } 878 879 req->cmd_cb_fn = cb_fn; 880 881 rc = spdk_bdev_nvme_admin_passthru(desc, ch, &req->cmd->nvme_cmd, req->iov[0].iov_base, req->length, 882 nvmf_bdev_ctrlr_complete_admin_cmd, req); 883 if (spdk_unlikely(rc)) { 884 if (rc == -ENOMEM) { 885 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 886 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 887 } 888 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 889 if (rc == -ENOTSUP) { 890 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 891 } else { 892 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 893 } 894 895 req->rsp->nvme_cpl.status.dnr = 1; 896 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 897 } 898 899 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 900 } 901 902 static void 903 nvmf_bdev_ctrlr_complete_abort_cmd(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg) 904 { 905 struct spdk_nvmf_request *req = cb_arg; 906 907 if (success) { 908 req->rsp->nvme_cpl.cdw0 &= ~1U; 909 } 910 911 spdk_nvmf_request_complete(req); 912 spdk_bdev_free_io(bdev_io); 913 } 914 915 int 916 spdk_nvmf_bdev_ctrlr_abort_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 917 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 918 struct spdk_nvmf_request *req_to_abort) 919 { 920 int rc; 921 922 assert((req->rsp->nvme_cpl.cdw0 & 1U) != 0); 923 924 rc = spdk_bdev_abort(desc, ch, req_to_abort, nvmf_bdev_ctrlr_complete_abort_cmd, req); 925 if (spdk_likely(rc == 0)) { 926 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 927 } else if (rc == -ENOMEM) { 928 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 929 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 930 } else { 931 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 932 } 933 } 934 935 bool 936 nvmf_bdev_ctrlr_get_dif_ctx(struct spdk_bdev_desc *desc, struct spdk_nvme_cmd *cmd, 937 struct spdk_dif_ctx *dif_ctx) 938 { 939 uint32_t init_ref_tag, dif_check_flags = 0; 940 int rc; 941 struct spdk_dif_ctx_init_ext_opts dif_opts; 942 943 if (spdk_bdev_desc_get_md_size(desc) == 0) { 944 return false; 945 } 946 947 /* Initial Reference Tag is the lower 32 bits of the start LBA. */ 948 init_ref_tag = (uint32_t)from_le64(&cmd->cdw10); 949 950 if (spdk_bdev_desc_is_dif_check_enabled(desc, SPDK_DIF_CHECK_TYPE_REFTAG)) { 951 dif_check_flags |= SPDK_DIF_FLAGS_REFTAG_CHECK; 952 } 953 954 if (spdk_bdev_desc_is_dif_check_enabled(desc, SPDK_DIF_CHECK_TYPE_GUARD)) { 955 dif_check_flags |= SPDK_DIF_FLAGS_GUARD_CHECK; 956 } 957 958 dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format); 959 dif_opts.dif_pi_format = SPDK_DIF_PI_FORMAT_16; 960 rc = spdk_dif_ctx_init(dif_ctx, 961 spdk_bdev_desc_get_block_size(desc), 962 spdk_bdev_desc_get_md_size(desc), 963 spdk_bdev_desc_is_md_interleaved(desc), 964 spdk_bdev_desc_is_dif_head_of_md(desc), 965 spdk_bdev_desc_get_dif_type(desc), 966 dif_check_flags, 967 init_ref_tag, 0, 0, 0, 0, &dif_opts); 968 969 return (rc == 0) ? true : false; 970 } 971 972 static void 973 nvmf_bdev_ctrlr_zcopy_start_complete(struct spdk_bdev_io *bdev_io, bool success, 974 void *cb_arg) 975 { 976 struct spdk_nvmf_request *req = cb_arg; 977 struct iovec *iov; 978 int iovcnt = 0; 979 980 if (spdk_unlikely(!success)) { 981 int sc = 0, sct = 0; 982 uint32_t cdw0 = 0; 983 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 984 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 985 986 response->cdw0 = cdw0; 987 response->status.sc = sc; 988 response->status.sct = sct; 989 990 spdk_bdev_free_io(bdev_io); 991 spdk_nvmf_request_complete(req); 992 return; 993 } 994 995 spdk_bdev_io_get_iovec(bdev_io, &iov, &iovcnt); 996 997 assert(iovcnt <= NVMF_REQ_MAX_BUFFERS); 998 assert(iovcnt > 0); 999 1000 req->iovcnt = iovcnt; 1001 1002 assert(req->iov == iov); 1003 1004 req->zcopy_bdev_io = bdev_io; /* Preserve the bdev_io for the end zcopy */ 1005 1006 spdk_nvmf_request_complete(req); 1007 /* Don't free the bdev_io here as it is needed for the END ZCOPY */ 1008 } 1009 1010 int 1011 nvmf_bdev_ctrlr_zcopy_start(struct spdk_bdev *bdev, 1012 struct spdk_bdev_desc *desc, 1013 struct spdk_io_channel *ch, 1014 struct spdk_nvmf_request *req) 1015 { 1016 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1017 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 1018 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 1019 uint64_t start_lba; 1020 uint64_t num_blocks; 1021 int rc; 1022 1023 nvmf_bdev_ctrlr_get_rw_params(&req->cmd->nvme_cmd, &start_lba, &num_blocks); 1024 1025 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 1026 SPDK_ERRLOG("end of media\n"); 1027 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1028 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 1029 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1030 } 1031 1032 if (spdk_unlikely(num_blocks * block_size > req->length)) { 1033 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 1034 num_blocks, block_size, req->length); 1035 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1036 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 1037 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1038 } 1039 1040 bool populate = (req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_READ) ? true : false; 1041 1042 rc = spdk_bdev_zcopy_start(desc, ch, req->iov, req->iovcnt, start_lba, 1043 num_blocks, populate, nvmf_bdev_ctrlr_zcopy_start_complete, req); 1044 if (spdk_unlikely(rc != 0)) { 1045 if (rc == -ENOMEM) { 1046 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 1047 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 1048 } 1049 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1050 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 1051 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1052 } 1053 1054 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 1055 } 1056 1057 static void 1058 nvmf_bdev_ctrlr_zcopy_end_complete(struct spdk_bdev_io *bdev_io, bool success, 1059 void *cb_arg) 1060 { 1061 struct spdk_nvmf_request *req = cb_arg; 1062 1063 if (spdk_unlikely(!success)) { 1064 int sc = 0, sct = 0; 1065 uint32_t cdw0 = 0; 1066 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1067 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 1068 1069 response->cdw0 = cdw0; 1070 response->status.sc = sc; 1071 response->status.sct = sct; 1072 } 1073 1074 spdk_bdev_free_io(bdev_io); 1075 req->zcopy_bdev_io = NULL; 1076 spdk_nvmf_request_complete(req); 1077 } 1078 1079 void 1080 nvmf_bdev_ctrlr_zcopy_end(struct spdk_nvmf_request *req, bool commit) 1081 { 1082 int rc __attribute__((unused)); 1083 1084 rc = spdk_bdev_zcopy_end(req->zcopy_bdev_io, commit, nvmf_bdev_ctrlr_zcopy_end_complete, req); 1085 1086 /* The only way spdk_bdev_zcopy_end() can fail is if we pass a bdev_io type that isn't ZCOPY */ 1087 assert(rc == 0); 1088 } 1089