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