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