1 /*- 2 * BSD LICENSE 3 * 4 * Copyright (c) Intel Corporation. All rights reserved. 5 * Copyright (c) 2019 Mellanox Technologies LTD. All rights reserved. 6 * Copyright (c) 2021 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 12 * * Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * * Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in 16 * the documentation and/or other materials provided with the 17 * distribution. 18 * * Neither the name of Intel Corporation nor the names of its 19 * contributors may be used to endorse or promote products derived 20 * from this software without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 26 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 #include "spdk/stdinc.h" 36 37 #include "nvmf_internal.h" 38 39 #include "spdk/bdev.h" 40 #include "spdk/endian.h" 41 #include "spdk/thread.h" 42 #include "spdk/likely.h" 43 #include "spdk/nvme.h" 44 #include "spdk/nvmf_cmd.h" 45 #include "spdk/nvmf_spec.h" 46 #include "spdk/trace.h" 47 #include "spdk/scsi_spec.h" 48 #include "spdk/string.h" 49 #include "spdk/util.h" 50 51 #include "spdk/log.h" 52 53 static bool 54 nvmf_subsystem_bdev_io_type_supported(struct spdk_nvmf_subsystem *subsystem, 55 enum spdk_bdev_io_type io_type) 56 { 57 struct spdk_nvmf_ns *ns; 58 59 for (ns = spdk_nvmf_subsystem_get_first_ns(subsystem); ns != NULL; 60 ns = spdk_nvmf_subsystem_get_next_ns(subsystem, ns)) { 61 if (ns->bdev == NULL) { 62 continue; 63 } 64 65 if (!spdk_bdev_io_type_supported(ns->bdev, io_type)) { 66 SPDK_DEBUGLOG(nvmf, 67 "Subsystem %s namespace %u (%s) does not support io_type %d\n", 68 spdk_nvmf_subsystem_get_nqn(subsystem), 69 ns->opts.nsid, spdk_bdev_get_name(ns->bdev), (int)io_type); 70 return false; 71 } 72 } 73 74 SPDK_DEBUGLOG(nvmf, "All devices in Subsystem %s support io_type %d\n", 75 spdk_nvmf_subsystem_get_nqn(subsystem), (int)io_type); 76 return true; 77 } 78 79 bool 80 nvmf_ctrlr_dsm_supported(struct spdk_nvmf_ctrlr *ctrlr) 81 { 82 return nvmf_subsystem_bdev_io_type_supported(ctrlr->subsys, SPDK_BDEV_IO_TYPE_UNMAP); 83 } 84 85 bool 86 nvmf_ctrlr_write_zeroes_supported(struct spdk_nvmf_ctrlr *ctrlr) 87 { 88 return nvmf_subsystem_bdev_io_type_supported(ctrlr->subsys, SPDK_BDEV_IO_TYPE_WRITE_ZEROES); 89 } 90 91 static void 92 nvmf_bdev_ctrlr_complete_cmd(struct spdk_bdev_io *bdev_io, bool success, 93 void *cb_arg) 94 { 95 struct spdk_nvmf_request *req = cb_arg; 96 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 97 int first_sc = 0, first_sct = 0, sc = 0, sct = 0; 98 uint32_t cdw0 = 0; 99 struct spdk_nvmf_request *first_req = req->first_fused_req; 100 101 if (spdk_unlikely(first_req != NULL)) { 102 /* fused commands - get status for both operations */ 103 struct spdk_nvme_cpl *first_response = &first_req->rsp->nvme_cpl; 104 105 spdk_bdev_io_get_nvme_fused_status(bdev_io, &cdw0, &first_sct, &first_sc, &sct, &sc); 106 first_response->cdw0 = cdw0; 107 first_response->status.sc = first_sc; 108 first_response->status.sct = first_sct; 109 110 /* first request should be completed */ 111 spdk_nvmf_request_complete(first_req); 112 req->first_fused_req = NULL; 113 } else { 114 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 115 } 116 117 response->cdw0 = cdw0; 118 response->status.sc = sc; 119 response->status.sct = sct; 120 121 spdk_nvmf_request_complete(req); 122 spdk_bdev_free_io(bdev_io); 123 } 124 125 static void 126 nvmf_bdev_ctrlr_complete_admin_cmd(struct spdk_bdev_io *bdev_io, bool success, 127 void *cb_arg) 128 { 129 struct spdk_nvmf_request *req = cb_arg; 130 131 if (req->cmd_cb_fn) { 132 req->cmd_cb_fn(req); 133 } 134 135 nvmf_bdev_ctrlr_complete_cmd(bdev_io, success, req); 136 } 137 138 void 139 nvmf_bdev_ctrlr_identify_ns(struct spdk_nvmf_ns *ns, struct spdk_nvme_ns_data *nsdata, 140 bool dif_insert_or_strip) 141 { 142 struct spdk_bdev *bdev = ns->bdev; 143 uint64_t num_blocks; 144 uint32_t phys_blocklen; 145 146 num_blocks = spdk_bdev_get_num_blocks(bdev); 147 148 nsdata->nsze = num_blocks; 149 nsdata->ncap = num_blocks; 150 nsdata->nuse = num_blocks; 151 nsdata->nlbaf = 0; 152 nsdata->flbas.format = 0; 153 nsdata->nacwu = spdk_bdev_get_acwu(bdev); 154 if (!dif_insert_or_strip) { 155 nsdata->lbaf[0].ms = spdk_bdev_get_md_size(bdev); 156 nsdata->lbaf[0].lbads = spdk_u32log2(spdk_bdev_get_block_size(bdev)); 157 if (nsdata->lbaf[0].ms != 0) { 158 nsdata->flbas.extended = 1; 159 nsdata->mc.extended = 1; 160 nsdata->mc.pointer = 0; 161 nsdata->dps.md_start = spdk_bdev_is_dif_head_of_md(bdev); 162 163 switch (spdk_bdev_get_dif_type(bdev)) { 164 case SPDK_DIF_TYPE1: 165 nsdata->dpc.pit1 = 1; 166 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE1; 167 break; 168 case SPDK_DIF_TYPE2: 169 nsdata->dpc.pit2 = 1; 170 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE2; 171 break; 172 case SPDK_DIF_TYPE3: 173 nsdata->dpc.pit3 = 1; 174 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_TYPE3; 175 break; 176 default: 177 SPDK_DEBUGLOG(nvmf, "Protection Disabled\n"); 178 nsdata->dps.pit = SPDK_NVME_FMT_NVM_PROTECTION_DISABLE; 179 break; 180 } 181 } 182 } else { 183 nsdata->lbaf[0].ms = 0; 184 nsdata->lbaf[0].lbads = spdk_u32log2(spdk_bdev_get_data_block_size(bdev)); 185 } 186 187 phys_blocklen = spdk_bdev_get_physical_block_size(bdev); 188 assert(phys_blocklen > 0); 189 /* Linux driver uses min(nawupf, npwg) to set physical_block_size */ 190 nsdata->nsfeat.optperf = 1; 191 nsdata->nsfeat.ns_atomic_write_unit = 1; 192 nsdata->npwg = (phys_blocklen >> nsdata->lbaf[0].lbads) - 1; 193 nsdata->nawupf = nsdata->npwg; 194 nsdata->npwa = nsdata->npwg; 195 nsdata->npdg = nsdata->npwg; 196 nsdata->npda = nsdata->npwg; 197 198 nsdata->noiob = spdk_bdev_get_optimal_io_boundary(bdev); 199 nsdata->nmic.can_share = 1; 200 if (ns->ptpl_file != NULL) { 201 nsdata->nsrescap.rescap.persist = 1; 202 } 203 nsdata->nsrescap.rescap.write_exclusive = 1; 204 nsdata->nsrescap.rescap.exclusive_access = 1; 205 nsdata->nsrescap.rescap.write_exclusive_reg_only = 1; 206 nsdata->nsrescap.rescap.exclusive_access_reg_only = 1; 207 nsdata->nsrescap.rescap.write_exclusive_all_reg = 1; 208 nsdata->nsrescap.rescap.exclusive_access_all_reg = 1; 209 nsdata->nsrescap.rescap.ignore_existing_key = 1; 210 211 SPDK_STATIC_ASSERT(sizeof(nsdata->nguid) == sizeof(ns->opts.nguid), "size mismatch"); 212 memcpy(nsdata->nguid, ns->opts.nguid, sizeof(nsdata->nguid)); 213 214 SPDK_STATIC_ASSERT(sizeof(nsdata->eui64) == sizeof(ns->opts.eui64), "size mismatch"); 215 memcpy(&nsdata->eui64, ns->opts.eui64, sizeof(nsdata->eui64)); 216 } 217 218 static void 219 nvmf_bdev_ctrlr_get_rw_params(const struct spdk_nvme_cmd *cmd, uint64_t *start_lba, 220 uint64_t *num_blocks) 221 { 222 /* SLBA: CDW10 and CDW11 */ 223 *start_lba = from_le64(&cmd->cdw10); 224 225 /* NLB: CDW12 bits 15:00, 0's based */ 226 *num_blocks = (from_le32(&cmd->cdw12) & 0xFFFFu) + 1; 227 } 228 229 static bool 230 nvmf_bdev_ctrlr_lba_in_range(uint64_t bdev_num_blocks, uint64_t io_start_lba, 231 uint64_t io_num_blocks) 232 { 233 if (io_start_lba + io_num_blocks > bdev_num_blocks || 234 io_start_lba + io_num_blocks < io_start_lba) { 235 return false; 236 } 237 238 return true; 239 } 240 241 static void 242 nvmf_ctrlr_process_io_cmd_resubmit(void *arg) 243 { 244 struct spdk_nvmf_request *req = arg; 245 246 nvmf_ctrlr_process_io_cmd(req); 247 } 248 249 static void 250 nvmf_ctrlr_process_admin_cmd_resubmit(void *arg) 251 { 252 struct spdk_nvmf_request *req = arg; 253 254 nvmf_ctrlr_process_admin_cmd(req); 255 } 256 257 static void 258 nvmf_bdev_ctrl_queue_io(struct spdk_nvmf_request *req, struct spdk_bdev *bdev, 259 struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn, void *cb_arg) 260 { 261 int rc; 262 263 req->bdev_io_wait.bdev = bdev; 264 req->bdev_io_wait.cb_fn = cb_fn; 265 req->bdev_io_wait.cb_arg = cb_arg; 266 267 rc = spdk_bdev_queue_io_wait(bdev, ch, &req->bdev_io_wait); 268 if (rc != 0) { 269 assert(false); 270 } 271 req->qpair->group->stat.pending_bdev_io++; 272 } 273 274 bool 275 nvmf_bdev_zcopy_enabled(struct spdk_bdev *bdev) 276 { 277 return spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY); 278 } 279 280 int 281 nvmf_bdev_ctrlr_read_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 282 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 283 { 284 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 285 uint32_t block_size = spdk_bdev_get_block_size(bdev); 286 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 287 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 288 uint64_t start_lba; 289 uint64_t num_blocks; 290 int rc; 291 292 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 293 294 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 295 SPDK_ERRLOG("end of media\n"); 296 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 297 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 298 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 299 } 300 301 if (spdk_unlikely(num_blocks * block_size > req->length)) { 302 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 303 num_blocks, block_size, req->length); 304 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 305 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 306 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 307 } 308 309 if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) { 310 /* Return here after checking the lba etc */ 311 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 312 } 313 314 assert(!spdk_nvmf_using_zcopy(req->zcopy_phase)); 315 316 rc = spdk_bdev_readv_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 317 nvmf_bdev_ctrlr_complete_cmd, req); 318 if (spdk_unlikely(rc)) { 319 if (rc == -ENOMEM) { 320 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 321 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 322 } 323 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 324 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 325 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 326 } 327 328 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 329 } 330 331 int 332 nvmf_bdev_ctrlr_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 333 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 334 { 335 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 336 uint32_t block_size = spdk_bdev_get_block_size(bdev); 337 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 338 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 339 uint64_t start_lba; 340 uint64_t num_blocks; 341 int rc; 342 343 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 344 345 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 346 SPDK_ERRLOG("end of media\n"); 347 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 348 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 349 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 350 } 351 352 if (spdk_unlikely(num_blocks * block_size > req->length)) { 353 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 354 num_blocks, block_size, req->length); 355 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 356 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 357 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 358 } 359 360 if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) { 361 /* Return here after checking the lba etc */ 362 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 363 } 364 365 assert(!spdk_nvmf_using_zcopy(req->zcopy_phase)); 366 367 rc = spdk_bdev_writev_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 368 nvmf_bdev_ctrlr_complete_cmd, req); 369 if (spdk_unlikely(rc)) { 370 if (rc == -ENOMEM) { 371 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 372 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 373 } 374 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 375 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 376 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 377 } 378 379 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 380 } 381 382 int 383 nvmf_bdev_ctrlr_compare_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 384 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 385 { 386 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 387 uint32_t block_size = spdk_bdev_get_block_size(bdev); 388 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 389 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 390 uint64_t start_lba; 391 uint64_t num_blocks; 392 int rc; 393 394 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 395 396 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 397 SPDK_ERRLOG("end of media\n"); 398 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 399 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 400 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 401 } 402 403 if (spdk_unlikely(num_blocks * block_size > req->length)) { 404 SPDK_ERRLOG("Compare NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 405 num_blocks, block_size, req->length); 406 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 407 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 408 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 409 } 410 411 rc = spdk_bdev_comparev_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 412 nvmf_bdev_ctrlr_complete_cmd, req); 413 if (spdk_unlikely(rc)) { 414 if (rc == -ENOMEM) { 415 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 416 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 417 } 418 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 419 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 420 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 421 } 422 423 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 424 } 425 426 int 427 nvmf_bdev_ctrlr_compare_and_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 428 struct spdk_io_channel *ch, struct spdk_nvmf_request *cmp_req, struct spdk_nvmf_request *write_req) 429 { 430 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 431 uint32_t block_size = spdk_bdev_get_block_size(bdev); 432 struct spdk_nvme_cmd *cmp_cmd = &cmp_req->cmd->nvme_cmd; 433 struct spdk_nvme_cmd *write_cmd = &write_req->cmd->nvme_cmd; 434 struct spdk_nvme_cpl *rsp = &write_req->rsp->nvme_cpl; 435 uint64_t write_start_lba, cmp_start_lba; 436 uint64_t write_num_blocks, cmp_num_blocks; 437 int rc; 438 439 nvmf_bdev_ctrlr_get_rw_params(cmp_cmd, &cmp_start_lba, &cmp_num_blocks); 440 nvmf_bdev_ctrlr_get_rw_params(write_cmd, &write_start_lba, &write_num_blocks); 441 442 if (spdk_unlikely(write_start_lba != cmp_start_lba || write_num_blocks != cmp_num_blocks)) { 443 SPDK_ERRLOG("Fused command start lba / num blocks mismatch\n"); 444 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 445 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 446 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 447 } 448 449 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, write_start_lba, 450 write_num_blocks))) { 451 SPDK_ERRLOG("end of media\n"); 452 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 453 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 454 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 455 } 456 457 if (spdk_unlikely(write_num_blocks * block_size > write_req->length)) { 458 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 459 write_num_blocks, block_size, write_req->length); 460 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 461 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 462 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 463 } 464 465 rc = spdk_bdev_comparev_and_writev_blocks(desc, ch, cmp_req->iov, cmp_req->iovcnt, write_req->iov, 466 write_req->iovcnt, write_start_lba, write_num_blocks, nvmf_bdev_ctrlr_complete_cmd, write_req); 467 if (spdk_unlikely(rc)) { 468 if (rc == -ENOMEM) { 469 nvmf_bdev_ctrl_queue_io(cmp_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, cmp_req); 470 nvmf_bdev_ctrl_queue_io(write_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, write_req); 471 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 472 } 473 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 474 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 475 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 476 } 477 478 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 479 } 480 481 int 482 nvmf_bdev_ctrlr_write_zeroes_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 483 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 484 { 485 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 486 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 487 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 488 uint64_t start_lba; 489 uint64_t num_blocks; 490 int rc; 491 492 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 493 494 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 495 SPDK_ERRLOG("end of media\n"); 496 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 497 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 498 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 499 } 500 501 rc = spdk_bdev_write_zeroes_blocks(desc, ch, start_lba, num_blocks, 502 nvmf_bdev_ctrlr_complete_cmd, req); 503 if (spdk_unlikely(rc)) { 504 if (rc == -ENOMEM) { 505 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 506 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 507 } 508 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 509 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 510 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 511 } 512 513 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 514 } 515 516 int 517 nvmf_bdev_ctrlr_flush_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 518 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 519 { 520 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 521 int rc; 522 523 /* As for NVMeoF controller, SPDK always set volatile write 524 * cache bit to 1, return success for those block devices 525 * which can't support FLUSH command. 526 */ 527 if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_FLUSH)) { 528 response->status.sct = SPDK_NVME_SCT_GENERIC; 529 response->status.sc = SPDK_NVME_SC_SUCCESS; 530 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 531 } 532 533 rc = spdk_bdev_flush_blocks(desc, ch, 0, spdk_bdev_get_num_blocks(bdev), 534 nvmf_bdev_ctrlr_complete_cmd, req); 535 if (spdk_unlikely(rc)) { 536 if (rc == -ENOMEM) { 537 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 538 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 539 } 540 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 541 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 542 } 543 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 544 } 545 546 struct nvmf_bdev_ctrlr_unmap { 547 struct spdk_nvmf_request *req; 548 uint32_t count; 549 struct spdk_bdev_desc *desc; 550 struct spdk_bdev *bdev; 551 struct spdk_io_channel *ch; 552 uint32_t range_index; 553 }; 554 555 static void 556 nvmf_bdev_ctrlr_unmap_cpl(struct spdk_bdev_io *bdev_io, bool success, 557 void *cb_arg) 558 { 559 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = cb_arg; 560 struct spdk_nvmf_request *req = unmap_ctx->req; 561 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 562 int sc, sct; 563 uint32_t cdw0; 564 565 unmap_ctx->count--; 566 567 if (response->status.sct == SPDK_NVME_SCT_GENERIC && 568 response->status.sc == SPDK_NVME_SC_SUCCESS) { 569 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 570 response->cdw0 = cdw0; 571 response->status.sc = sc; 572 response->status.sct = sct; 573 } 574 575 if (unmap_ctx->count == 0) { 576 spdk_nvmf_request_complete(req); 577 free(unmap_ctx); 578 } 579 spdk_bdev_free_io(bdev_io); 580 } 581 582 static int 583 nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 584 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 585 struct nvmf_bdev_ctrlr_unmap *unmap_ctx); 586 static void 587 nvmf_bdev_ctrlr_unmap_resubmit(void *arg) 588 { 589 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = arg; 590 struct spdk_nvmf_request *req = unmap_ctx->req; 591 struct spdk_bdev_desc *desc = unmap_ctx->desc; 592 struct spdk_bdev *bdev = unmap_ctx->bdev; 593 struct spdk_io_channel *ch = unmap_ctx->ch; 594 595 nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, unmap_ctx); 596 } 597 598 static int 599 nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 600 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 601 struct nvmf_bdev_ctrlr_unmap *unmap_ctx) 602 { 603 uint16_t nr, i; 604 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 605 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 606 struct spdk_nvme_dsm_range *dsm_range; 607 uint64_t lba; 608 uint32_t lba_count; 609 int rc; 610 611 nr = cmd->cdw10_bits.dsm.nr + 1; 612 if (nr * sizeof(struct spdk_nvme_dsm_range) > req->length) { 613 SPDK_ERRLOG("Dataset Management number of ranges > SGL length\n"); 614 response->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 615 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 616 } 617 618 if (unmap_ctx == NULL) { 619 unmap_ctx = calloc(1, sizeof(*unmap_ctx)); 620 if (!unmap_ctx) { 621 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 622 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 623 } 624 625 unmap_ctx->req = req; 626 unmap_ctx->desc = desc; 627 unmap_ctx->ch = ch; 628 unmap_ctx->bdev = bdev; 629 630 response->status.sct = SPDK_NVME_SCT_GENERIC; 631 response->status.sc = SPDK_NVME_SC_SUCCESS; 632 } else { 633 unmap_ctx->count--; /* dequeued */ 634 } 635 636 dsm_range = (struct spdk_nvme_dsm_range *)req->data; 637 for (i = unmap_ctx->range_index; i < nr; i++) { 638 lba = dsm_range[i].starting_lba; 639 lba_count = dsm_range[i].length; 640 641 unmap_ctx->count++; 642 643 rc = spdk_bdev_unmap_blocks(desc, ch, lba, lba_count, 644 nvmf_bdev_ctrlr_unmap_cpl, unmap_ctx); 645 if (rc) { 646 if (rc == -ENOMEM) { 647 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_bdev_ctrlr_unmap_resubmit, unmap_ctx); 648 /* Unmap was not yet submitted to bdev */ 649 /* unmap_ctx->count will be decremented when the request is dequeued */ 650 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 651 } 652 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 653 unmap_ctx->count--; 654 /* We can't return here - we may have to wait for any other 655 * unmaps already sent to complete */ 656 break; 657 } 658 unmap_ctx->range_index++; 659 } 660 661 if (unmap_ctx->count == 0) { 662 free(unmap_ctx); 663 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 664 } 665 666 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 667 } 668 669 int 670 nvmf_bdev_ctrlr_dsm_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 671 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 672 { 673 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 674 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 675 676 if (cmd->cdw11_bits.dsm.ad) { 677 return nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, NULL); 678 } 679 680 response->status.sct = SPDK_NVME_SCT_GENERIC; 681 response->status.sc = SPDK_NVME_SC_SUCCESS; 682 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 683 } 684 685 int 686 nvmf_bdev_ctrlr_nvme_passthru_io(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 687 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 688 { 689 int rc; 690 691 rc = spdk_bdev_nvme_io_passthru(desc, ch, &req->cmd->nvme_cmd, req->data, req->length, 692 nvmf_bdev_ctrlr_complete_cmd, req); 693 if (spdk_unlikely(rc)) { 694 if (rc == -ENOMEM) { 695 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 696 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 697 } 698 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 699 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 700 req->rsp->nvme_cpl.status.dnr = 1; 701 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 702 } 703 704 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 705 } 706 707 int 708 spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 709 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 710 spdk_nvmf_nvme_passthru_cmd_cb cb_fn) 711 { 712 int rc; 713 714 req->cmd_cb_fn = cb_fn; 715 716 rc = spdk_bdev_nvme_admin_passthru(desc, ch, &req->cmd->nvme_cmd, req->data, req->length, 717 nvmf_bdev_ctrlr_complete_admin_cmd, req); 718 if (spdk_unlikely(rc)) { 719 if (rc == -ENOMEM) { 720 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 721 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 722 } 723 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 724 if (rc == -ENOTSUP) { 725 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 726 } else { 727 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 728 } 729 730 req->rsp->nvme_cpl.status.dnr = 1; 731 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 732 } 733 734 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 735 } 736 737 static void 738 nvmf_bdev_ctrlr_complete_abort_cmd(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg) 739 { 740 struct spdk_nvmf_request *req = cb_arg; 741 742 if (success) { 743 req->rsp->nvme_cpl.cdw0 &= ~1U; 744 } 745 746 spdk_nvmf_request_complete(req); 747 spdk_bdev_free_io(bdev_io); 748 } 749 750 int 751 spdk_nvmf_bdev_ctrlr_abort_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 752 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 753 struct spdk_nvmf_request *req_to_abort) 754 { 755 int rc; 756 757 assert((req->rsp->nvme_cpl.cdw0 & 1U) != 0); 758 759 rc = spdk_bdev_abort(desc, ch, req_to_abort, nvmf_bdev_ctrlr_complete_abort_cmd, req); 760 if (spdk_likely(rc == 0)) { 761 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 762 } else if (rc == -ENOMEM) { 763 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 764 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 765 } else { 766 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 767 } 768 } 769 770 bool 771 nvmf_bdev_ctrlr_get_dif_ctx(struct spdk_bdev *bdev, struct spdk_nvme_cmd *cmd, 772 struct spdk_dif_ctx *dif_ctx) 773 { 774 uint32_t init_ref_tag, dif_check_flags = 0; 775 int rc; 776 777 if (spdk_bdev_get_md_size(bdev) == 0) { 778 return false; 779 } 780 781 /* Initial Reference Tag is the lower 32 bits of the start LBA. */ 782 init_ref_tag = (uint32_t)from_le64(&cmd->cdw10); 783 784 if (spdk_bdev_is_dif_check_enabled(bdev, SPDK_DIF_CHECK_TYPE_REFTAG)) { 785 dif_check_flags |= SPDK_DIF_FLAGS_REFTAG_CHECK; 786 } 787 788 if (spdk_bdev_is_dif_check_enabled(bdev, SPDK_DIF_CHECK_TYPE_GUARD)) { 789 dif_check_flags |= SPDK_DIF_FLAGS_GUARD_CHECK; 790 } 791 792 rc = spdk_dif_ctx_init(dif_ctx, 793 spdk_bdev_get_block_size(bdev), 794 spdk_bdev_get_md_size(bdev), 795 spdk_bdev_is_md_interleaved(bdev), 796 spdk_bdev_is_dif_head_of_md(bdev), 797 spdk_bdev_get_dif_type(bdev), 798 dif_check_flags, 799 init_ref_tag, 0, 0, 0, 0); 800 801 return (rc == 0) ? true : false; 802 } 803 804 static void 805 nvmf_bdev_ctrlr_start_zcopy_complete(struct spdk_bdev_io *bdev_io, bool success, 806 void *cb_arg) 807 { 808 struct spdk_nvmf_request *req = cb_arg; 809 struct iovec *iov; 810 int iovcnt; 811 812 if (spdk_unlikely(!success)) { 813 int sc = 0, sct = 0; 814 uint32_t cdw0 = 0; 815 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 816 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 817 818 response->cdw0 = cdw0; 819 response->status.sc = sc; 820 response->status.sct = sct; 821 822 spdk_bdev_free_io(bdev_io); 823 spdk_nvmf_request_complete(req); 824 return; 825 } 826 827 spdk_bdev_io_get_iovec(bdev_io, &iov, &iovcnt); 828 829 assert(iovcnt <= NVMF_REQ_MAX_BUFFERS); 830 assert(iovcnt > 0); 831 832 req->iovcnt = iovcnt; 833 834 assert(req->iov == iov); 835 836 /* backward compatible */ 837 req->data = req->iov[0].iov_base; 838 839 req->zcopy_bdev_io = bdev_io; /* Preserve the bdev_io for the end zcopy */ 840 841 spdk_nvmf_request_complete(req); 842 /* Don't free the bdev_io here as it is needed for the END ZCOPY */ 843 } 844 845 int 846 nvmf_bdev_ctrlr_start_zcopy(struct spdk_bdev *bdev, 847 struct spdk_bdev_desc *desc, 848 struct spdk_io_channel *ch, 849 struct spdk_nvmf_request *req) 850 { 851 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 852 uint32_t block_size = spdk_bdev_get_block_size(bdev); 853 uint64_t start_lba; 854 uint64_t num_blocks; 855 856 nvmf_bdev_ctrlr_get_rw_params(&req->cmd->nvme_cmd, &start_lba, &num_blocks); 857 858 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 859 SPDK_ERRLOG("end of media\n"); 860 return -ENXIO; 861 } 862 863 if (spdk_unlikely(num_blocks * block_size > req->length)) { 864 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 865 num_blocks, block_size, req->length); 866 return -ENXIO; 867 } 868 869 bool populate = (req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_READ) ? true : false; 870 871 return spdk_bdev_zcopy_start(desc, ch, req->iov, req->iovcnt, start_lba, 872 num_blocks, populate, nvmf_bdev_ctrlr_start_zcopy_complete, req); 873 } 874 875 static void 876 nvmf_bdev_ctrlr_end_zcopy_complete(struct spdk_bdev_io *bdev_io, bool success, 877 void *cb_arg) 878 { 879 struct spdk_nvmf_request *req = cb_arg; 880 881 if (spdk_unlikely(!success)) { 882 int sc = 0, sct = 0; 883 uint32_t cdw0 = 0; 884 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 885 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 886 887 response->cdw0 = cdw0; 888 response->status.sc = sc; 889 response->status.sct = sct; 890 } 891 892 spdk_bdev_free_io(bdev_io); 893 req->zcopy_bdev_io = NULL; 894 spdk_nvmf_request_complete(req); 895 } 896 897 int 898 nvmf_bdev_ctrlr_end_zcopy(struct spdk_nvmf_request *req, bool commit) 899 { 900 return spdk_bdev_zcopy_end(req->zcopy_bdev_io, commit, nvmf_bdev_ctrlr_end_zcopy_complete, req); 901 } 902