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