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, sc = 0, 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 *first_response = &first_req->rsp->nvme_cpl; 103 104 spdk_bdev_io_get_nvme_fused_status(bdev_io, &cdw0, &first_sct, &first_sc, &sct, &sc); 105 first_response->cdw0 = cdw0; 106 first_response->status.sc = first_sc; 107 first_response->status.sct = first_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, &sct, &sc); 114 } 115 116 response->cdw0 = cdw0; 117 response->status.sc = sc; 118 response->status.sct = 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 nsdata->npwa = nsdata->npwg; 194 nsdata->npdg = nsdata->npwg; 195 nsdata->npda = nsdata->npwg; 196 197 nsdata->noiob = spdk_bdev_get_optimal_io_boundary(bdev); 198 nsdata->nmic.can_share = 1; 199 if (ns->ptpl_file != NULL) { 200 nsdata->nsrescap.rescap.persist = 1; 201 } 202 nsdata->nsrescap.rescap.write_exclusive = 1; 203 nsdata->nsrescap.rescap.exclusive_access = 1; 204 nsdata->nsrescap.rescap.write_exclusive_reg_only = 1; 205 nsdata->nsrescap.rescap.exclusive_access_reg_only = 1; 206 nsdata->nsrescap.rescap.write_exclusive_all_reg = 1; 207 nsdata->nsrescap.rescap.exclusive_access_all_reg = 1; 208 nsdata->nsrescap.rescap.ignore_existing_key = 1; 209 210 SPDK_STATIC_ASSERT(sizeof(nsdata->nguid) == sizeof(ns->opts.nguid), "size mismatch"); 211 memcpy(nsdata->nguid, ns->opts.nguid, sizeof(nsdata->nguid)); 212 213 SPDK_STATIC_ASSERT(sizeof(nsdata->eui64) == sizeof(ns->opts.eui64), "size mismatch"); 214 memcpy(&nsdata->eui64, ns->opts.eui64, sizeof(nsdata->eui64)); 215 } 216 217 static void 218 nvmf_bdev_ctrlr_get_rw_params(const struct spdk_nvme_cmd *cmd, uint64_t *start_lba, 219 uint64_t *num_blocks) 220 { 221 /* SLBA: CDW10 and CDW11 */ 222 *start_lba = from_le64(&cmd->cdw10); 223 224 /* NLB: CDW12 bits 15:00, 0's based */ 225 *num_blocks = (from_le32(&cmd->cdw12) & 0xFFFFu) + 1; 226 } 227 228 static bool 229 nvmf_bdev_ctrlr_lba_in_range(uint64_t bdev_num_blocks, uint64_t io_start_lba, 230 uint64_t io_num_blocks) 231 { 232 if (io_start_lba + io_num_blocks > bdev_num_blocks || 233 io_start_lba + io_num_blocks < io_start_lba) { 234 return false; 235 } 236 237 return true; 238 } 239 240 static void 241 nvmf_ctrlr_process_io_cmd_resubmit(void *arg) 242 { 243 struct spdk_nvmf_request *req = arg; 244 245 nvmf_ctrlr_process_io_cmd(req); 246 } 247 248 static void 249 nvmf_ctrlr_process_admin_cmd_resubmit(void *arg) 250 { 251 struct spdk_nvmf_request *req = arg; 252 253 nvmf_ctrlr_process_admin_cmd(req); 254 } 255 256 static void 257 nvmf_bdev_ctrl_queue_io(struct spdk_nvmf_request *req, struct spdk_bdev *bdev, 258 struct spdk_io_channel *ch, spdk_bdev_io_wait_cb cb_fn, void *cb_arg) 259 { 260 int rc; 261 262 req->bdev_io_wait.bdev = bdev; 263 req->bdev_io_wait.cb_fn = cb_fn; 264 req->bdev_io_wait.cb_arg = cb_arg; 265 266 rc = spdk_bdev_queue_io_wait(bdev, ch, &req->bdev_io_wait); 267 if (rc != 0) { 268 assert(false); 269 } 270 req->qpair->group->stat.pending_bdev_io++; 271 } 272 273 bool 274 nvmf_bdev_zcopy_enabled(struct spdk_bdev *bdev) 275 { 276 return spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_ZCOPY); 277 } 278 279 int 280 nvmf_bdev_ctrlr_read_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 281 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 282 { 283 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 284 uint32_t block_size = spdk_bdev_get_block_size(bdev); 285 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 286 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 287 uint64_t start_lba; 288 uint64_t num_blocks; 289 int rc; 290 291 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 292 293 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 294 SPDK_ERRLOG("end of media\n"); 295 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 296 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 297 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 298 } 299 300 if (spdk_unlikely(num_blocks * block_size > req->length)) { 301 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 302 num_blocks, block_size, req->length); 303 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 304 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 305 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 306 } 307 308 if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) { 309 /* Return here after checking the lba etc */ 310 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 311 } 312 313 assert(!spdk_nvmf_using_zcopy(req->zcopy_phase)); 314 315 rc = spdk_bdev_readv_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 316 nvmf_bdev_ctrlr_complete_cmd, req); 317 if (spdk_unlikely(rc)) { 318 if (rc == -ENOMEM) { 319 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 320 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 321 } 322 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 323 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 324 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 325 } 326 327 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 328 } 329 330 int 331 nvmf_bdev_ctrlr_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 332 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 333 { 334 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 335 uint32_t block_size = spdk_bdev_get_block_size(bdev); 336 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 337 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 338 uint64_t start_lba; 339 uint64_t num_blocks; 340 int rc; 341 342 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 343 344 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 345 SPDK_ERRLOG("end of media\n"); 346 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 347 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 348 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 349 } 350 351 if (spdk_unlikely(num_blocks * block_size > req->length)) { 352 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 353 num_blocks, block_size, req->length); 354 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 355 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 356 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 357 } 358 359 if (req->zcopy_phase == NVMF_ZCOPY_PHASE_EXECUTE) { 360 /* Return here after checking the lba etc */ 361 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 362 } 363 364 assert(!spdk_nvmf_using_zcopy(req->zcopy_phase)); 365 366 rc = spdk_bdev_writev_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_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 383 struct spdk_io_channel *ch, struct spdk_nvmf_request *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 *cmd = &req->cmd->nvme_cmd; 388 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 389 uint64_t start_lba; 390 uint64_t num_blocks; 391 int rc; 392 393 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 394 395 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 396 SPDK_ERRLOG("end of media\n"); 397 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 398 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 399 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 400 } 401 402 if (spdk_unlikely(num_blocks * block_size > req->length)) { 403 SPDK_ERRLOG("Compare NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 404 num_blocks, block_size, req->length); 405 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 406 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 407 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 408 } 409 410 rc = spdk_bdev_comparev_blocks(desc, ch, req->iov, req->iovcnt, start_lba, num_blocks, 411 nvmf_bdev_ctrlr_complete_cmd, req); 412 if (spdk_unlikely(rc)) { 413 if (rc == -ENOMEM) { 414 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 415 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 416 } 417 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 418 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 419 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 420 } 421 422 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 423 } 424 425 int 426 nvmf_bdev_ctrlr_compare_and_write_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 427 struct spdk_io_channel *ch, struct spdk_nvmf_request *cmp_req, struct spdk_nvmf_request *write_req) 428 { 429 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 430 uint32_t block_size = spdk_bdev_get_block_size(bdev); 431 struct spdk_nvme_cmd *cmp_cmd = &cmp_req->cmd->nvme_cmd; 432 struct spdk_nvme_cmd *write_cmd = &write_req->cmd->nvme_cmd; 433 struct spdk_nvme_cpl *rsp = &write_req->rsp->nvme_cpl; 434 uint64_t write_start_lba, cmp_start_lba; 435 uint64_t write_num_blocks, cmp_num_blocks; 436 int rc; 437 438 nvmf_bdev_ctrlr_get_rw_params(cmp_cmd, &cmp_start_lba, &cmp_num_blocks); 439 nvmf_bdev_ctrlr_get_rw_params(write_cmd, &write_start_lba, &write_num_blocks); 440 441 if (spdk_unlikely(write_start_lba != cmp_start_lba || write_num_blocks != cmp_num_blocks)) { 442 SPDK_ERRLOG("Fused command start lba / num blocks mismatch\n"); 443 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 444 rsp->status.sc = SPDK_NVME_SC_INVALID_FIELD; 445 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 446 } 447 448 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, write_start_lba, 449 write_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 if (spdk_unlikely(write_num_blocks * block_size > write_req->length)) { 457 SPDK_ERRLOG("Write NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 458 write_num_blocks, block_size, write_req->length); 459 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 460 rsp->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 461 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 462 } 463 464 rc = spdk_bdev_comparev_and_writev_blocks(desc, ch, cmp_req->iov, cmp_req->iovcnt, write_req->iov, 465 write_req->iovcnt, write_start_lba, write_num_blocks, nvmf_bdev_ctrlr_complete_cmd, write_req); 466 if (spdk_unlikely(rc)) { 467 if (rc == -ENOMEM) { 468 nvmf_bdev_ctrl_queue_io(cmp_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, cmp_req); 469 nvmf_bdev_ctrl_queue_io(write_req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, write_req); 470 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 471 } 472 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 473 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 474 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 475 } 476 477 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 478 } 479 480 int 481 nvmf_bdev_ctrlr_write_zeroes_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 482 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 483 { 484 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 485 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 486 struct spdk_nvme_cpl *rsp = &req->rsp->nvme_cpl; 487 uint64_t start_lba; 488 uint64_t num_blocks; 489 int rc; 490 491 nvmf_bdev_ctrlr_get_rw_params(cmd, &start_lba, &num_blocks); 492 493 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 494 SPDK_ERRLOG("end of media\n"); 495 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 496 rsp->status.sc = SPDK_NVME_SC_LBA_OUT_OF_RANGE; 497 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 498 } 499 500 rc = spdk_bdev_write_zeroes_blocks(desc, ch, start_lba, num_blocks, 501 nvmf_bdev_ctrlr_complete_cmd, req); 502 if (spdk_unlikely(rc)) { 503 if (rc == -ENOMEM) { 504 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 505 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 506 } 507 rsp->status.sct = SPDK_NVME_SCT_GENERIC; 508 rsp->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 509 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 510 } 511 512 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 513 } 514 515 int 516 nvmf_bdev_ctrlr_flush_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 517 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 518 { 519 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 520 int rc; 521 522 /* As for NVMeoF controller, SPDK always set volatile write 523 * cache bit to 1, return success for those block devices 524 * which can't support FLUSH command. 525 */ 526 if (!spdk_bdev_io_type_supported(bdev, SPDK_BDEV_IO_TYPE_FLUSH)) { 527 response->status.sct = SPDK_NVME_SCT_GENERIC; 528 response->status.sc = SPDK_NVME_SC_SUCCESS; 529 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 530 } 531 532 rc = spdk_bdev_flush_blocks(desc, ch, 0, spdk_bdev_get_num_blocks(bdev), 533 nvmf_bdev_ctrlr_complete_cmd, req); 534 if (spdk_unlikely(rc)) { 535 if (rc == -ENOMEM) { 536 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 537 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 538 } 539 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 540 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 541 } 542 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 543 } 544 545 struct nvmf_bdev_ctrlr_unmap { 546 struct spdk_nvmf_request *req; 547 uint32_t count; 548 struct spdk_bdev_desc *desc; 549 struct spdk_bdev *bdev; 550 struct spdk_io_channel *ch; 551 uint32_t range_index; 552 }; 553 554 static void 555 nvmf_bdev_ctrlr_unmap_cpl(struct spdk_bdev_io *bdev_io, bool success, 556 void *cb_arg) 557 { 558 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = cb_arg; 559 struct spdk_nvmf_request *req = unmap_ctx->req; 560 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 561 int sc, sct; 562 uint32_t cdw0; 563 564 unmap_ctx->count--; 565 566 if (response->status.sct == SPDK_NVME_SCT_GENERIC && 567 response->status.sc == SPDK_NVME_SC_SUCCESS) { 568 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 569 response->cdw0 = cdw0; 570 response->status.sc = sc; 571 response->status.sct = sct; 572 } 573 574 if (unmap_ctx->count == 0) { 575 spdk_nvmf_request_complete(req); 576 free(unmap_ctx); 577 } 578 spdk_bdev_free_io(bdev_io); 579 } 580 581 static int 582 nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 583 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 584 struct nvmf_bdev_ctrlr_unmap *unmap_ctx); 585 static void 586 nvmf_bdev_ctrlr_unmap_resubmit(void *arg) 587 { 588 struct nvmf_bdev_ctrlr_unmap *unmap_ctx = arg; 589 struct spdk_nvmf_request *req = unmap_ctx->req; 590 struct spdk_bdev_desc *desc = unmap_ctx->desc; 591 struct spdk_bdev *bdev = unmap_ctx->bdev; 592 struct spdk_io_channel *ch = unmap_ctx->ch; 593 594 nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, unmap_ctx); 595 } 596 597 static int 598 nvmf_bdev_ctrlr_unmap(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 599 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 600 struct nvmf_bdev_ctrlr_unmap *unmap_ctx) 601 { 602 uint16_t nr, i; 603 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 604 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 605 struct spdk_nvme_dsm_range *dsm_range; 606 uint64_t lba; 607 uint32_t lba_count; 608 int rc; 609 610 nr = cmd->cdw10_bits.dsm.nr + 1; 611 if (nr * sizeof(struct spdk_nvme_dsm_range) > req->length) { 612 SPDK_ERRLOG("Dataset Management number of ranges > SGL length\n"); 613 response->status.sc = SPDK_NVME_SC_DATA_SGL_LENGTH_INVALID; 614 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 615 } 616 617 if (unmap_ctx == NULL) { 618 unmap_ctx = calloc(1, sizeof(*unmap_ctx)); 619 if (!unmap_ctx) { 620 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 621 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 622 } 623 624 unmap_ctx->req = req; 625 unmap_ctx->desc = desc; 626 unmap_ctx->ch = ch; 627 unmap_ctx->bdev = bdev; 628 629 response->status.sct = SPDK_NVME_SCT_GENERIC; 630 response->status.sc = SPDK_NVME_SC_SUCCESS; 631 } else { 632 unmap_ctx->count--; /* dequeued */ 633 } 634 635 dsm_range = (struct spdk_nvme_dsm_range *)req->data; 636 for (i = unmap_ctx->range_index; i < nr; i++) { 637 lba = dsm_range[i].starting_lba; 638 lba_count = dsm_range[i].length; 639 640 unmap_ctx->count++; 641 642 rc = spdk_bdev_unmap_blocks(desc, ch, lba, lba_count, 643 nvmf_bdev_ctrlr_unmap_cpl, unmap_ctx); 644 if (rc) { 645 if (rc == -ENOMEM) { 646 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_bdev_ctrlr_unmap_resubmit, unmap_ctx); 647 /* Unmap was not yet submitted to bdev */ 648 /* unmap_ctx->count will be decremented when the request is dequeued */ 649 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 650 } 651 response->status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 652 unmap_ctx->count--; 653 /* We can't return here - we may have to wait for any other 654 * unmaps already sent to complete */ 655 break; 656 } 657 unmap_ctx->range_index++; 658 } 659 660 if (unmap_ctx->count == 0) { 661 free(unmap_ctx); 662 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 663 } 664 665 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 666 } 667 668 int 669 nvmf_bdev_ctrlr_dsm_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 670 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 671 { 672 struct spdk_nvme_cmd *cmd = &req->cmd->nvme_cmd; 673 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 674 675 if (cmd->cdw11_bits.dsm.ad) { 676 return nvmf_bdev_ctrlr_unmap(bdev, desc, ch, req, NULL); 677 } 678 679 response->status.sct = SPDK_NVME_SCT_GENERIC; 680 response->status.sc = SPDK_NVME_SC_SUCCESS; 681 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 682 } 683 684 int 685 nvmf_bdev_ctrlr_nvme_passthru_io(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 686 struct spdk_io_channel *ch, struct spdk_nvmf_request *req) 687 { 688 int rc; 689 690 rc = spdk_bdev_nvme_io_passthru(desc, ch, &req->cmd->nvme_cmd, req->data, req->length, 691 nvmf_bdev_ctrlr_complete_cmd, req); 692 if (spdk_unlikely(rc)) { 693 if (rc == -ENOMEM) { 694 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_io_cmd_resubmit, req); 695 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 696 } 697 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 698 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INVALID_OPCODE; 699 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 700 } 701 702 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 703 } 704 705 int 706 spdk_nvmf_bdev_ctrlr_nvme_passthru_admin(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 707 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 708 spdk_nvmf_nvme_passthru_cmd_cb cb_fn) 709 { 710 int rc; 711 712 req->cmd_cb_fn = cb_fn; 713 714 rc = spdk_bdev_nvme_admin_passthru(desc, ch, &req->cmd->nvme_cmd, req->data, req->length, 715 nvmf_bdev_ctrlr_complete_admin_cmd, req); 716 if (spdk_unlikely(rc)) { 717 if (rc == -ENOMEM) { 718 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 719 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 720 } 721 req->rsp->nvme_cpl.status.sct = SPDK_NVME_SCT_GENERIC; 722 req->rsp->nvme_cpl.status.sc = SPDK_NVME_SC_INTERNAL_DEVICE_ERROR; 723 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 724 } 725 726 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 727 } 728 729 static void 730 nvmf_bdev_ctrlr_complete_abort_cmd(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg) 731 { 732 struct spdk_nvmf_request *req = cb_arg; 733 734 if (success) { 735 req->rsp->nvme_cpl.cdw0 &= ~1U; 736 } 737 738 spdk_nvmf_request_complete(req); 739 spdk_bdev_free_io(bdev_io); 740 } 741 742 int 743 spdk_nvmf_bdev_ctrlr_abort_cmd(struct spdk_bdev *bdev, struct spdk_bdev_desc *desc, 744 struct spdk_io_channel *ch, struct spdk_nvmf_request *req, 745 struct spdk_nvmf_request *req_to_abort) 746 { 747 int rc; 748 749 assert((req->rsp->nvme_cpl.cdw0 & 1U) != 0); 750 751 rc = spdk_bdev_abort(desc, ch, req_to_abort, nvmf_bdev_ctrlr_complete_abort_cmd, req); 752 if (spdk_likely(rc == 0)) { 753 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 754 } else if (rc == -ENOMEM) { 755 nvmf_bdev_ctrl_queue_io(req, bdev, ch, nvmf_ctrlr_process_admin_cmd_resubmit, req); 756 return SPDK_NVMF_REQUEST_EXEC_STATUS_ASYNCHRONOUS; 757 } else { 758 return SPDK_NVMF_REQUEST_EXEC_STATUS_COMPLETE; 759 } 760 } 761 762 bool 763 nvmf_bdev_ctrlr_get_dif_ctx(struct spdk_bdev *bdev, struct spdk_nvme_cmd *cmd, 764 struct spdk_dif_ctx *dif_ctx) 765 { 766 uint32_t init_ref_tag, dif_check_flags = 0; 767 int rc; 768 769 if (spdk_bdev_get_md_size(bdev) == 0) { 770 return false; 771 } 772 773 /* Initial Reference Tag is the lower 32 bits of the start LBA. */ 774 init_ref_tag = (uint32_t)from_le64(&cmd->cdw10); 775 776 if (spdk_bdev_is_dif_check_enabled(bdev, SPDK_DIF_CHECK_TYPE_REFTAG)) { 777 dif_check_flags |= SPDK_DIF_FLAGS_REFTAG_CHECK; 778 } 779 780 if (spdk_bdev_is_dif_check_enabled(bdev, SPDK_DIF_CHECK_TYPE_GUARD)) { 781 dif_check_flags |= SPDK_DIF_FLAGS_GUARD_CHECK; 782 } 783 784 rc = spdk_dif_ctx_init(dif_ctx, 785 spdk_bdev_get_block_size(bdev), 786 spdk_bdev_get_md_size(bdev), 787 spdk_bdev_is_md_interleaved(bdev), 788 spdk_bdev_is_dif_head_of_md(bdev), 789 spdk_bdev_get_dif_type(bdev), 790 dif_check_flags, 791 init_ref_tag, 0, 0, 0, 0); 792 793 return (rc == 0) ? true : false; 794 } 795 796 static void 797 nvmf_bdev_ctrlr_start_zcopy_complete(struct spdk_bdev_io *bdev_io, bool success, 798 void *cb_arg) 799 { 800 struct spdk_nvmf_request *req = cb_arg; 801 struct iovec *iov; 802 int iovcnt; 803 804 if (spdk_unlikely(!success)) { 805 int sc = 0, sct = 0; 806 uint32_t cdw0 = 0; 807 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 808 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 809 810 response->cdw0 = cdw0; 811 response->status.sc = sc; 812 response->status.sct = sct; 813 814 spdk_bdev_free_io(bdev_io); 815 spdk_nvmf_request_complete(req); 816 return; 817 } 818 819 spdk_bdev_io_get_iovec(bdev_io, &iov, &iovcnt); 820 821 assert(iovcnt <= NVMF_REQ_MAX_BUFFERS); 822 assert(iovcnt > 0); 823 824 req->iovcnt = iovcnt; 825 826 assert(req->iov == iov); 827 828 /* backward compatible */ 829 req->data = req->iov[0].iov_base; 830 831 req->zcopy_bdev_io = bdev_io; /* Preserve the bdev_io for the end zcopy */ 832 833 spdk_nvmf_request_complete(req); 834 /* Don't free the bdev_io here as it is needed for the END ZCOPY */ 835 } 836 837 int 838 nvmf_bdev_ctrlr_start_zcopy(struct spdk_bdev *bdev, 839 struct spdk_bdev_desc *desc, 840 struct spdk_io_channel *ch, 841 struct spdk_nvmf_request *req) 842 { 843 uint64_t bdev_num_blocks = spdk_bdev_get_num_blocks(bdev); 844 uint32_t block_size = spdk_bdev_get_block_size(bdev); 845 uint64_t start_lba; 846 uint64_t num_blocks; 847 848 nvmf_bdev_ctrlr_get_rw_params(&req->cmd->nvme_cmd, &start_lba, &num_blocks); 849 850 if (spdk_unlikely(!nvmf_bdev_ctrlr_lba_in_range(bdev_num_blocks, start_lba, num_blocks))) { 851 SPDK_ERRLOG("end of media\n"); 852 return -ENXIO; 853 } 854 855 if (spdk_unlikely(num_blocks * block_size > req->length)) { 856 SPDK_ERRLOG("Read NLB %" PRIu64 " * block size %" PRIu32 " > SGL length %" PRIu32 "\n", 857 num_blocks, block_size, req->length); 858 return -ENXIO; 859 } 860 861 bool populate = (req->cmd->nvme_cmd.opc == SPDK_NVME_OPC_READ) ? true : false; 862 863 return spdk_bdev_zcopy_start(desc, ch, req->iov, req->iovcnt, start_lba, 864 num_blocks, populate, nvmf_bdev_ctrlr_start_zcopy_complete, req); 865 } 866 867 static void 868 nvmf_bdev_ctrlr_end_zcopy_complete(struct spdk_bdev_io *bdev_io, bool success, 869 void *cb_arg) 870 { 871 struct spdk_nvmf_request *req = cb_arg; 872 873 if (spdk_unlikely(!success)) { 874 int sc = 0, sct = 0; 875 uint32_t cdw0 = 0; 876 struct spdk_nvme_cpl *response = &req->rsp->nvme_cpl; 877 spdk_bdev_io_get_nvme_status(bdev_io, &cdw0, &sct, &sc); 878 879 response->cdw0 = cdw0; 880 response->status.sc = sc; 881 response->status.sct = sct; 882 } 883 884 spdk_bdev_free_io(bdev_io); 885 req->zcopy_bdev_io = NULL; 886 spdk_nvmf_request_complete(req); 887 } 888 889 int 890 nvmf_bdev_ctrlr_end_zcopy(struct spdk_nvmf_request *req, bool commit) 891 { 892 return spdk_bdev_zcopy_end(req->zcopy_bdev_io, commit, nvmf_bdev_ctrlr_end_zcopy_complete, req); 893 } 894