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