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); 145 146 switch (spdk_bdev_get_dif_type(bdev)) { 147 case SPDK_DIF_TYPE1: 148 nsdata->dpc.pit1 = 1; 149 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE1; 150 break; 151 case SPDK_DIF_TYPE2: 152 nsdata->dpc.pit2 = 1; 153 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE2; 154 break; 155 case SPDK_DIF_TYPE3: 156 nsdata->dpc.pit3 = 1; 157 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE3; 158 break; 159 default: 160 SPDK_DEBUGLOG(nvmf, "Protection Disabled\n"); 161 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_DISABLE; 162 break; 163 } 164 } 165 } else { 166 nsdata->lbaf[0].ms = 0; 167 nsdata->lbaf[0].lbads = spdk_u32log2(spdk_bdev_get_data_block_size(bdev)); 168 } 169 170 phys_blocklen = spdk_bdev_get_physical_block_size(bdev); 171 assert(phys_blocklen > 0); 172 /* Linux driver uses min(nawupf, npwg) to set physical_block_size */ 173 nsdata->nsfeat.optperf = 1; 174 nsdata->nsfeat.ns_atomic_write_unit = 1; 175 nsdata->npwg = (phys_blocklen >> nsdata->lbaf[0].lbads) - 1; 176 nsdata->nawupf = nsdata->npwg; 177 nsdata->npwa = nsdata->npwg; 178 nsdata->npdg = nsdata->npwg; 179 nsdata->npda = nsdata->npwg; 180 181 if (spdk_bdev_get_write_unit_size(bdev) == 1) { 182 nsdata->noiob = spdk_bdev_get_optimal_io_boundary(bdev); 183 } 184 nsdata->nmic.can_share = 1; 185 if (nvmf_ns_is_ptpl_capable(ns)) { 186 nsdata->nsrescap.rescap.persist = 1; 187 } 188 nsdata->nsrescap.rescap.write_exclusive = 1; 189 nsdata->nsrescap.rescap.exclusive_access = 1; 190 nsdata->nsrescap.rescap.write_exclusive_reg_only = 1; 191 nsdata->nsrescap.rescap.exclusive_access_reg_only = 1; 192 nsdata->nsrescap.rescap.write_exclusive_all_reg = 1; 193 nsdata->nsrescap.rescap.exclusive_access_all_reg = 1; 194 nsdata->nsrescap.rescap.ignore_existing_key = 1; 195 196 SPDK_STATIC_ASSERT(sizeof(nsdata->nguid) == sizeof(ns->opts.nguid), "size mismatch"); 197 memcpy(nsdata->nguid, ns->opts.nguid, sizeof(nsdata->nguid)); 198 199 SPDK_STATIC_ASSERT(sizeof(nsdata->eui64) == sizeof(ns->opts.eui64), "size mismatch"); 200 memcpy(&nsdata->eui64, ns->opts.eui64, sizeof(nsdata->eui64)); 201 202 /* For now we support just one source range for copy command */ 203 nsdata->msrc = 0; 204 205 max_copy = spdk_bdev_get_max_copy(bdev); 206 if (max_copy == 0 || max_copy > UINT16_MAX) { 207 /* Zero means copy size is unlimited */ 208 nsdata->mcl = UINT16_MAX; 209 nsdata->mssrl = UINT16_MAX; 210 } else { 211 nsdata->mcl = max_copy; 212 nsdata->mssrl = max_copy; 213 } 214 } 215 216 void 217 nvmf_bdev_ctrlr_identify_iocs_nvm(struct spdk_nvmf_ns *ns, 218 struct spdk_nvme_nvm_ns_data *nsdata_nvm) 219 { 220 struct spdk_bdev_desc *desc = ns->desc; 221 uint8_t _16bpists; 222 uint32_t sts, pif; 223 224 if (spdk_bdev_desc_get_dif_type(desc) == SPDK_DIF_DISABLE) { 225 return; 226 } 227 228 pif = spdk_bdev_desc_get_dif_pi_format(desc); 229 230 /* 231 * 16BPISTS shall be 1 for 32/64b Guard PI. 232 * STCRS shall be 1 if 16BPISTS is 1. 233 * 16 is the minimum value of STS for 32b Guard PI. 234 */ 235 switch (pif) { 236 case SPDK_DIF_PI_FORMAT_16: 237 _16bpists = 0; 238 sts = 0; 239 break; 240 case SPDK_DIF_PI_FORMAT_32: 241 _16bpists = 1; 242 sts = 16; 243 break; 244 case SPDK_DIF_PI_FORMAT_64: 245 _16bpists = 1; 246 sts = 0; 247 break; 248 default: 249 SPDK_WARNLOG("PI format %u is not supported\n", pif); 250 return; 251 } 252 253 /* For 16b Guard PI, Storage Tag is not available because we set STS to 0. 254 * In this case, we do not have to set 16BPISTM to 1. For simplicity, 255 * set 16BPISTM to 0 and set LBSTM to all zeroes. 256 * 257 * We will revisit here when we find any OS uses Storage Tag. 258 */ 259 nsdata_nvm->lbstm = 0; 260 nsdata_nvm->pic._16bpistm = 0; 261 262 nsdata_nvm->pic._16bpists = _16bpists; 263 nsdata_nvm->pic.stcrs = 0; 264 nsdata_nvm->elbaf[0].sts = sts; 265 nsdata_nvm->elbaf[0].pif = pif; 266 } 267 268 static void 269 nvmf_bdev_ctrlr_get_rw_params(const struct spdk_nvme_cmd *cmd, uint64_t *start_lba, 270 uint64_t *num_blocks) 271 { 272 /* SLBA: CDW10 and CDW11 */ 273 *start_lba = from_le64(&cmd->cdw10); 274 275 /* NLB: CDW12 bits 15:00, 0's based */ 276 *num_blocks = (from_le32(&cmd->cdw12) & 0xFFFFu) + 1; 277 } 278 279 static void 280 nvmf_bdev_ctrlr_get_rw_ext_params(const struct spdk_nvme_cmd *cmd, 281 struct spdk_bdev_ext_io_opts *opts) 282 { 283 /* Get CDW12 values */ 284 opts->nvme_cdw12.raw = from_le32(&cmd->cdw12); 285 286 /* Get CDW13 values */ 287 opts->nvme_cdw13.raw = from_le32(&cmd->cdw13); 288 289 /* Bdev layer checks PRACT in CDW12 because it is NVMe specific, but 290 * it does not check DIF check flags in CDW because DIF is not NVMe 291 * specific. Hence, copy DIF check flags from CDW12 to dif_check_flags_exclude_mask. 292 */ 293 opts->dif_check_flags_exclude_mask = (~opts->nvme_cdw12.raw) & SPDK_NVME_IO_FLAGS_PRCHK_MASK; 294 } 295 296 static bool 297 nvmf_bdev_ctrlr_lba_in_range(uint64_t bdev_num_blocks, uint64_t io_start_lba, 298 uint64_t io_num_blocks) 299 { 300 if (io_start_lba + io_num_blocks > bdev_num_blocks || 301 io_start_lba + io_num_blocks < io_start_lba) { 302 return false; 303 } 304 305 return true; 306 } 307 308 static void 309 nvmf_ctrlr_process_io_cmd_resubmit(void *arg) 310 { 311 struct spdk_nvmf_request *req = arg; 312 int rc; 313 314 rc = nvmf_ctrlr_process_io_cmd(req); 315 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 316 spdk_nvmf_request_complete(req); 317 } 318 } 319 320 static void 321 nvmf_ctrlr_process_admin_cmd_resubmit(void *arg) 322 { 323 struct spdk_nvmf_request *req = arg; 324 int rc; 325 326 rc = nvmf_ctrlr_process_admin_cmd(req); 327 if (rc == SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE) { 328 spdk_nvmf_request_complete(req); 329 } 330 } 331 332 static void 333 nvmf_bdev_ctrl_queue_io(struct spdk_nvmf_request *req, struct spdk_bdev *bdev, 334 struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn, void *cb_arg) 335 { 336 int rc; 337 338 req->bdev_io_wait.bdev = bdev; 339 req->bdev_io_wait.cb_fn = cb_fn; 340 req->bdev_io_wait.cb_arg = cb_arg; 341 342 rc = spdk_bdev_queue_io_wait(bdev, ch, &req->bdev_io_wait); 343 if (rc != 0) { 344 assert(false); 345 } 346 req->qpair->group->stat.pending_bdev_io++; 347 } 348 349 bool 350 nvmf_bdev_zcopy_enabled(struct spdk_bdev *bdev) 351 { 352 return spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY); 353 } 354 355 int 356 nvmf_bdev_ctrlr_read_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 357 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 358 { 359 struct spdk_bdev_ext_io_opts opts = { 360 .size = SPDK_SIZEOF(&opts, accel_sequence), 361 .memory_domain = req->memory_domain, 362 .memory_domain_ctx = req->memory_domain_ctx, 363 .accel_sequence = req->accel_sequence, 364 }; 365 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 366 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 367 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 368 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 369 uint64_t start_lba; 370 uint64_t num_blocks; 371 int rc; 372 373 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 374 nvmf_bdev_ctrlr_get_rw_ext_params(cmd, &opts); 375 376 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 377 SPDK_ERRLOG("end of media\n"); 378 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 379 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 380 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 381 } 382 383 if (spdk_unlikely(num_blocks * block_size > req->length)) { 384 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 385 num_blocks, block_size, req->length); 386 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 387 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 388 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 389 } 390 391 assert(!spdk_nvmf_request_using_zcopy(req)); 392 393 rc = spdk_bdev_readv_blocks_ext(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 394 nvmf_bdev_ctrlr_complete_cmd, req, &opts); 395 if (spdk_unlikely(rc)) { 396 if (rc == -ENOMEM) { 397 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 398 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 399 } 400 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 401 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 402 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 403 } 404 405 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 406 } 407 408 int 409 nvmf_bdev_ctrlr_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 410 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 411 { 412 struct spdk_bdev_ext_io_opts opts = { 413 .size = SPDK_SIZEOF(&opts, nvme_cdw13), 414 .memory_domain = req->memory_domain, 415 .memory_domain_ctx = req->memory_domain_ctx, 416 .accel_sequence = req->accel_sequence, 417 }; 418 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 419 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 420 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 421 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 422 uint64_t start_lba; 423 uint64_t num_blocks; 424 int rc; 425 426 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 427 nvmf_bdev_ctrlr_get_rw_ext_params(cmd, &opts); 428 429 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 430 SPDK_ERRLOG("end of media\n"); 431 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 432 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 433 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 434 } 435 436 if (spdk_unlikely(num_blocks * block_size > req->length)) { 437 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 438 num_blocks, block_size, req->length); 439 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 440 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 441 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 442 } 443 444 assert(!spdk_nvmf_request_using_zcopy(req)); 445 446 rc = spdk_bdev_writev_blocks_ext(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 447 nvmf_bdev_ctrlr_complete_cmd, req, &opts); 448 if (spdk_unlikely(rc)) { 449 if (rc == -ENOMEM) { 450 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 451 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 452 } 453 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 454 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 455 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 456 } 457 458 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 459 } 460 461 int 462 nvmf_bdev_ctrlr_compare_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 463 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 464 { 465 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 466 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 467 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 468 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 469 uint64_t start_lba; 470 uint64_t num_blocks; 471 int rc; 472 473 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 474 475 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 476 SPDK_ERRLOG("end of media\n"); 477 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 478 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 479 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 480 } 481 482 if (spdk_unlikely(num_blocks * block_size > req->length)) { 483 SPDK_ERRLOG("Compare NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 484 num_blocks, block_size, req->length); 485 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 486 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 487 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 488 } 489 490 rc = spdk_bdev_comparev_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 491 nvmf_bdev_ctrlr_complete_cmd, req); 492 if (spdk_unlikely(rc)) { 493 if (rc == -ENOMEM) { 494 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 495 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 496 } 497 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 498 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 499 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 500 } 501 502 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 503 } 504 505 int 506 nvmf_bdev_ctrlr_compare_and_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 507 struct spdk_io_channel *ch, struct spdk_nvmf_request *cmp_req, struct spdk_nvmf_request *write_req) 508 { 509 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 510 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 511 struct spdk_nvme_cmd *cmp_cmd = &cmp_req->cmd->nvme_cmd; 512 struct spdk_nvme_cmd *write_cmd = &write_req->cmd->nvme_cmd; 513 struct spdk_nvme_cpl *rsp = &write_req->rsp->nvme_cpl; 514 uint64_t write_start_lba, cmp_start_lba; 515 uint64_t write_num_blocks, cmp_num_blocks; 516 int rc; 517 518 nvmf_bdev_ctrlr_get_rw_params(cmp_cmd, &cmp_start_lba, &cmp_num_blocks); 519 nvmf_bdev_ctrlr_get_rw_params(write_cmd, &write_start_lba, &write_num_blocks); 520 521 if (spdk_unlikely(write_start_lba != cmp_start_lba || write_num_blocks != cmp_num_blocks)) { 522 SPDK_ERRLOG("Fused command start lba / num blocks mismatch\n"); 523 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 524 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 525 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 526 } 527 528 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, write_start_lba, 529 write_num_blocks))) { 530 SPDK_ERRLOG("end of media\n"); 531 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 532 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 533 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 534 } 535 536 if (spdk_unlikely(write_num_blocks * block_size > write_req->length)) { 537 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 538 write_num_blocks, block_size, write_req->length); 539 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 540 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 541 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 542 } 543 544 rc = spdk_bdev_comparev_and_writev_blocks(desc, ch, cmp_req->iov, cmp_req->iovcnt, write_req->iov, 545 write_req->iovcnt, write_start_lba, write_num_blocks, nvmf_bdev_ctrlr_complete_cmd, write_req); 546 if (spdk_unlikely(rc)) { 547 if (rc == -ENOMEM) { 548 nvmf_bdev_ctrl_queue_io(cmp_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, cmp_req); 549 nvmf_bdev_ctrl_queue_io(write_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, write_req); 550 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 551 } 552 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 553 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 554 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 555 } 556 557 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 558 } 559 560 int 561 nvmf_bdev_ctrlr_write_zeroes_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 562 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 563 { 564 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 565 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 566 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 567 uint64_t max_write_zeroes_size = req->qpair->ctrlr->subsys->max_write_zeroes_size_kib; 568 uint64_t start_lba; 569 uint64_t num_blocks; 570 int rc; 571 572 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 573 if (spdk_unlikely(max_write_zeroes_size > 0 && 574 num_blocks > (max_write_zeroes_size << 10) / spdk_bdev_desc_get_block_size(desc))) { 575 SPDK_ERRLOG("invalid write zeroes size, should not exceed %" PRIu64 "Kib\n", max_write_zeroes_size); 576 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 577 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 578 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 579 } 580 581 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 582 SPDK_ERRLOG("end of media\n"); 583 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 584 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 585 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 586 } 587 588 if (spdk_unlikely(cmd->cdw12_bits.write_zeroes.deac)) { 589 SPDK_ERRLOG("Write Zeroes Deallocate is not supported\n"); 590 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 591 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 592 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 593 } 594 595 rc = spdk_bdev_write_zeroes_blocks(desc, ch, start_lba, num_blocks, 596 nvmf_bdev_ctrlr_complete_cmd, req); 597 if (spdk_unlikely(rc)) { 598 if (rc == -ENOMEM) { 599 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 600 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 601 } 602 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 603 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 604 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 605 } 606 607 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 608 } 609 610 int 611 nvmf_bdev_ctrlr_flush_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 612 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 613 { 614 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 615 int rc; 616 617 /* As for NVMeoF controller, SPDK always set volatile write 618 * cache bit to 1, return success for those block devices 619 * which can't support FLUSH command. 620 */ 621 if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_FLUSH)) { 622 response->status.sct = SPDK_NVME_SCT_GENERIC; 623 response->status.sc = SPDK_NVME_SC_SUCCESS; 624 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 625 } 626 627 rc = spdk_bdev_flush_blocks(desc, ch, 0, spdk_bdev_get_num_blocks(bdev), 628 nvmf_bdev_ctrlr_complete_cmd, req); 629 if (spdk_unlikely(rc)) { 630 if (rc == -ENOMEM) { 631 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 632 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 633 } 634 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 635 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 636 } 637 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 638 } 639 640 struct nvmf_bdev_ctrlr_unmap { 641 struct spdk_nvmf_request *req; 642 uint32_t count; 643 struct spdk_bdev_desc *desc; 644 struct spdk_bdev *bdev; 645 struct spdk_io_channel *ch; 646 uint32_t range_index; 647 }; 648 649 static void 650 nvmf_bdev_ctrlr_unmap_cpl(struct spdk_bdev_io *bdev_io, bool success, 651 void *cb_arg) 652 { 653 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = cb_arg; 654 struct spdk_nvmf_request *req = unmap_ctx->req; 655 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 656 int sc, sct; 657 uint32_t cdw0; 658 659 unmap_ctx->count--; 660 661 if (response->status.sct == SPDK_NVME_SCT_GENERIC && 662 response->status.sc == SPDK_NVME_SC_SUCCESS) { 663 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 664 response->cdw0 = cdw0; 665 response->status.sc = sc; 666 response->status.sct = sct; 667 } 668 669 if (unmap_ctx->count == 0) { 670 spdk_nvmf_request_complete(req); 671 free(unmap_ctx); 672 } 673 spdk_bdev_free_io(bdev_io); 674 } 675 676 static int nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 677 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 678 struct nvmf_bdev_ctrlr_unmap *unmap_ctx); 679 static void 680 nvmf_bdev_ctrlr_unmap_resubmit(void *arg) 681 { 682 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = arg; 683 struct spdk_nvmf_request *req = unmap_ctx->req; 684 struct spdk_bdev_desc *desc = unmap_ctx->desc; 685 struct spdk_bdev *bdev = unmap_ctx->bdev; 686 struct spdk_io_channel *ch = unmap_ctx->ch; 687 688 nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, unmap_ctx); 689 } 690 691 static int 692 nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 693 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 694 struct nvmf_bdev_ctrlr_unmap *unmap_ctx) 695 { 696 uint16_t nr, i; 697 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 698 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 699 uint64_t max_discard_size = req->qpair->ctrlr->subsys->max_discard_size_kib; 700 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 701 struct spdk_iov_xfer ix; 702 uint64_t lba; 703 uint32_t lba_count; 704 int rc; 705 706 nr = cmd->cdw10_bits.dsm.nr + 1; 707 if (nr * sizeof(struct spdk_nvme_dsm_range) > req->length) { 708 SPDK_ERRLOG("Dataset Management number of ranges > SGL length\n"); 709 response->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 710 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 711 } 712 713 if (unmap_ctx == NULL) { 714 unmap_ctx = calloc(1, sizeof(*unmap_ctx)); 715 if (!unmap_ctx) { 716 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 717 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 718 } 719 720 unmap_ctx->req = req; 721 unmap_ctx->desc = desc; 722 unmap_ctx->ch = ch; 723 unmap_ctx->bdev = bdev; 724 725 response->status.sct = SPDK_NVME_SCT_GENERIC; 726 response->status.sc = SPDK_NVME_SC_SUCCESS; 727 } else { 728 unmap_ctx->count--; /* dequeued */ 729 } 730 731 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 732 733 for (i = unmap_ctx->range_index; i < nr; i++) { 734 struct spdk_nvme_dsm_range dsm_range = { 0 }; 735 736 spdk_iov_xfer_to_buf(&ix, &dsm_range, sizeof(dsm_range)); 737 738 lba = dsm_range.starting_lba; 739 lba_count = dsm_range.length; 740 if (max_discard_size > 0 && lba_count > (max_discard_size << 10) / block_size) { 741 SPDK_ERRLOG("invalid unmap size %" PRIu32 " blocks, should not exceed %" PRIu64 " blocks\n", 742 lba_count, max_discard_size << 1); 743 response->status.sct = SPDK_NVME_SCT_GENERIC; 744 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 745 break; 746 } 747 748 unmap_ctx->count++; 749 750 rc = spdk_bdev_unmap_blocks(desc, ch, lba, lba_count, 751 nvmf_bdev_ctrlr_unmap_cpl, unmap_ctx); 752 if (rc) { 753 if (rc == -ENOMEM) { 754 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_bdev_ctrlr_unmap_resubmit, unmap_ctx); 755 /* Unmap was not yet submitted to bdev */ 756 /* unmap_ctx->count will be decremented when the request is dequeued */ 757 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 758 } 759 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 760 unmap_ctx->count--; 761 /* We can't return here - we may have to wait for any other 762 * unmaps already sent to complete */ 763 break; 764 } 765 unmap_ctx->range_index++; 766 } 767 768 if (unmap_ctx->count == 0) { 769 free(unmap_ctx); 770 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 771 } 772 773 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 774 } 775 776 int 777 nvmf_bdev_ctrlr_dsm_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 778 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 779 { 780 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 781 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 782 783 if (cmd->cdw11_bits.dsm.ad) { 784 return nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, NULL); 785 } 786 787 response->status.sct = SPDK_NVME_SCT_GENERIC; 788 response->status.sc = SPDK_NVME_SC_SUCCESS; 789 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 790 } 791 792 int 793 nvmf_bdev_ctrlr_copy_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 794 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 795 { 796 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 797 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 798 uint64_t sdlba = ((uint64_t)cmd->cdw11 << 32) + cmd->cdw10; 799 struct spdk_nvme_scc_source_range range = { 0 }; 800 struct spdk_iov_xfer ix; 801 int rc; 802 803 SPDK_DEBUGLOG(nvmf, "Copy command: SDLBA %lu, NR %u, desc format %u, PRINFOR %u, " 804 "DTYPE %u, STCW %u, PRINFOW %u, FUA %u, LR %u\n", 805 sdlba, 806 cmd->cdw12_bits.copy.nr, 807 cmd->cdw12_bits.copy.df, 808 cmd->cdw12_bits.copy.prinfor, 809 cmd->cdw12_bits.copy.dtype, 810 cmd->cdw12_bits.copy.stcw, 811 cmd->cdw12_bits.copy.prinfow, 812 cmd->cdw12_bits.copy.fua, 813 cmd->cdw12_bits.copy.lr); 814 815 if (spdk_unlikely(req->length != (cmd->cdw12_bits.copy.nr + 1) * 816 sizeof(struct spdk_nvme_scc_source_range))) { 817 response->status.sct = SPDK_NVME_SCT_GENERIC; 818 response->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 819 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 820 } 821 822 /* 823 * We support only one source range, and rely on this with the xfer 824 * below. 825 */ 826 if (cmd->cdw12_bits.copy.nr > 0) { 827 response->status.sct = SPDK_NVME_SCT_COMMAND_SPECIFIC; 828 response->status.sc = SPDK_NVME_SC_CMD_SIZE_LIMIT_SIZE_EXCEEDED; 829 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 830 } 831 832 if (cmd->cdw12_bits.copy.df != 0) { 833 response->status.sct = SPDK_NVME_SCT_GENERIC; 834 response->status.sc = SPDK_NVME_SC_INVALID_FIELD; 835 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 836 } 837 838 spdk_iov_xfer_init(&ix, req->iov, req->iovcnt); 839 spdk_iov_xfer_to_buf(&ix, &range, sizeof(range)); 840 841 rc = spdk_bdev_copy_blocks(desc, ch, sdlba, range.slba, range.nlb + 1, 842 nvmf_bdev_ctrlr_complete_cmd, req); 843 if (spdk_unlikely(rc)) { 844 if (rc == -ENOMEM) { 845 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 846 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 847 } 848 849 response->status.sct = SPDK_NVME_SCT_GENERIC; 850 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 851 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 852 } 853 854 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 855 } 856 857 int 858 nvmf_bdev_ctrlr_nvme_passthru_io(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 859 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 860 { 861 int rc; 862 863 rc = spdk_bdev_nvme_iov_passthru_md(desc, ch, &req->cmd->nvme_cmd, req->iov, req->iovcnt, 864 req->length, NULL, 0, nvmf_bdev_ctrlr_complete_cmd, req); 865 866 if (spdk_unlikely(rc)) { 867 if (rc == -ENOMEM) { 868 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 869 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 870 } 871 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 872 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 873 req->rsp->nvme_cpl.status.dnr = 1; 874 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 875 } 876 877 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 878 } 879 880 int 881 spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 882 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 883 spdk_nvmf_nvme_passthru_cmd_cb cb_fn) 884 { 885 int rc; 886 887 if (spdk_unlikely(req->iovcnt > 1)) { 888 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 889 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 890 req->rsp->nvme_cpl.status.dnr = 1; 891 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 892 } 893 894 req->cmd_cb_fn = cb_fn; 895 896 rc = spdk_bdev_nvme_admin_passthru(desc, ch, &req->cmd->nvme_cmd, req->iov[0].iov_base, req->length, 897 nvmf_bdev_ctrlr_complete_admin_cmd, req); 898 if (spdk_unlikely(rc)) { 899 if (rc == -ENOMEM) { 900 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 901 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 902 } 903 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 904 if (rc == -ENOTSUP) { 905 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 906 } else { 907 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 908 } 909 910 req->rsp->nvme_cpl.status.dnr = 1; 911 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 912 } 913 914 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 915 } 916 917 static void 918 nvmf_bdev_ctrlr_complete_abort_cmd(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg) 919 { 920 struct spdk_nvmf_request *req = cb_arg; 921 922 if (success) { 923 req->rsp->nvme_cpl.cdw0 &= ~1U; 924 } 925 926 spdk_nvmf_request_complete(req); 927 spdk_bdev_free_io(bdev_io); 928 } 929 930 int 931 spdk_nvmf_bdev_ctrlr_abort_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 932 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 933 struct spdk_nvmf_request *req_to_abort) 934 { 935 int rc; 936 937 assert((req->rsp->nvme_cpl.cdw0 & 1U) != 0); 938 939 rc = spdk_bdev_abort(desc, ch, req_to_abort, nvmf_bdev_ctrlr_complete_abort_cmd, req); 940 if (spdk_likely(rc == 0)) { 941 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 942 } else if (rc == -ENOMEM) { 943 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 944 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 945 } else { 946 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 947 } 948 } 949 950 bool 951 nvmf_bdev_ctrlr_get_dif_ctx(struct spdk_bdev_desc *desc, struct spdk_nvme_cmd *cmd, 952 struct spdk_dif_ctx *dif_ctx) 953 { 954 uint32_t init_ref_tag, dif_check_flags = 0; 955 int rc; 956 struct spdk_dif_ctx_init_ext_opts dif_opts; 957 958 if (spdk_bdev_desc_get_md_size(desc) == 0) { 959 return false; 960 } 961 962 /* Initial Reference Tag is the lower 32 bits of the start LBA. */ 963 init_ref_tag = (uint32_t)from_le64(&cmd->cdw10); 964 965 if (spdk_bdev_desc_is_dif_check_enabled(desc, SPDK_DIF_CHECK_TYPE_REFTAG)) { 966 dif_check_flags |= SPDK_DIF_FLAGS_REFTAG_CHECK; 967 } 968 969 if (spdk_bdev_desc_is_dif_check_enabled(desc, SPDK_DIF_CHECK_TYPE_GUARD)) { 970 dif_check_flags |= SPDK_DIF_FLAGS_GUARD_CHECK; 971 } 972 973 dif_opts.size = SPDK_SIZEOF(&dif_opts, dif_pi_format); 974 dif_opts.dif_pi_format = SPDK_DIF_PI_FORMAT_16; 975 rc = spdk_dif_ctx_init(dif_ctx, 976 spdk_bdev_desc_get_block_size(desc), 977 spdk_bdev_desc_get_md_size(desc), 978 spdk_bdev_desc_is_md_interleaved(desc), 979 spdk_bdev_desc_is_dif_head_of_md(desc), 980 spdk_bdev_desc_get_dif_type(desc), 981 dif_check_flags, 982 init_ref_tag, 0, 0, 0, 0, &dif_opts); 983 984 return (rc == 0) ? true : false; 985 } 986 987 static void 988 nvmf_bdev_ctrlr_zcopy_start_complete(struct spdk_bdev_io *bdev_io, bool success, 989 void *cb_arg) 990 { 991 struct spdk_nvmf_request *req = cb_arg; 992 struct iovec *iov; 993 int iovcnt = 0; 994 995 if (spdk_unlikely(!success)) { 996 int sc = 0, sct = 0; 997 uint32_t cdw0 = 0; 998 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 999 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 1000 1001 response->cdw0 = cdw0; 1002 response->status.sc = sc; 1003 response->status.sct = sct; 1004 1005 spdk_bdev_free_io(bdev_io); 1006 spdk_nvmf_request_complete(req); 1007 return; 1008 } 1009 1010 spdk_bdev_io_get_iovec(bdev_io, &iov, &iovcnt); 1011 1012 assert(iovcnt <= NVMF_REQ_MAX_BUFFERS); 1013 assert(iovcnt > 0); 1014 1015 req->iovcnt = iovcnt; 1016 1017 assert(req->iov == iov); 1018 1019 req->zcopy_bdev_io = bdev_io; /* Preserve the bdev_io for the end zcopy */ 1020 1021 spdk_nvmf_request_complete(req); 1022 /* Don't free the bdev_io here as it is needed for the END ZCOPY */ 1023 } 1024 1025 int 1026 nvmf_bdev_ctrlr_zcopy_start(struct spdk_bdev *bdev, 1027 struct spdk_bdev_desc *desc, 1028 struct spdk_io_channel *ch, 1029 struct spdk_nvmf_request *req) 1030 { 1031 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 1032 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 1033 uint32_t block_size = spdk_bdev_desc_get_block_size(desc); 1034 uint64_t start_lba; 1035 uint64_t num_blocks; 1036 int rc; 1037 1038 nvmf_bdev_ctrlr_get_rw_params(&req->cmd->nvme_cmd, &start_lba, &num_blocks); 1039 1040 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 1041 SPDK_ERRLOG("end of media\n"); 1042 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1043 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 1044 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1045 } 1046 1047 if (spdk_unlikely(num_blocks * block_size > req->length)) { 1048 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 1049 num_blocks, block_size, req->length); 1050 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1051 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 1052 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1053 } 1054 1055 bool populate = (req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_READ) ? true : false; 1056 1057 rc = spdk_bdev_zcopy_start(desc, ch, req->iov, req->iovcnt, start_lba, 1058 num_blocks, populate, nvmf_bdev_ctrlr_zcopy_start_complete, req); 1059 if (spdk_unlikely(rc != 0)) { 1060 if (rc == -ENOMEM) { 1061 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 1062 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 1063 } 1064 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 1065 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 1066 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 1067 } 1068 1069 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 1070 } 1071 1072 static void 1073 nvmf_bdev_ctrlr_zcopy_end_complete(struct spdk_bdev_io *bdev_io, bool success, 1074 void *cb_arg) 1075 { 1076 struct spdk_nvmf_request *req = cb_arg; 1077 1078 if (spdk_unlikely(!success)) { 1079 int sc = 0, sct = 0; 1080 uint32_t cdw0 = 0; 1081 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 1082 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 1083 1084 response->cdw0 = cdw0; 1085 response->status.sc = sc; 1086 response->status.sct = sct; 1087 } 1088 1089 spdk_bdev_free_io(bdev_io); 1090 req->zcopy_bdev_io = NULL; 1091 spdk_nvmf_request_complete(req); 1092 } 1093 1094 void 1095 nvmf_bdev_ctrlr_zcopy_end(struct spdk_nvmf_request *req, bool commit) 1096 { 1097 int rc __attribute__((unused)); 1098 1099 rc = spdk_bdev_zcopy_end(req->zcopy_bdev_io, commit, nvmf_bdev_ctrlr_zcopy_end_complete, req); 1100 1101 /* The only way spdk_bdev_zcopy_end() can fail is if we pass a bdev_io type that isn't ZCOPY */ 1102 assert(rc == 0); 1103 } 1104