1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * NVMe admin command implementation. 4 * Copyright (c) 2015-2016 HGST, a Western Digital Company. 5 */ 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 #include <linux/module.h> 8 #include <linux/rculist.h> 9 #include <linux/part_stat.h> 10 11 #include <generated/utsrelease.h> 12 #include <asm/unaligned.h> 13 #include "nvmet.h" 14 15 u32 nvmet_get_log_page_len(struct nvme_command *cmd) 16 { 17 u32 len = le16_to_cpu(cmd->get_log_page.numdu); 18 19 len <<= 16; 20 len += le16_to_cpu(cmd->get_log_page.numdl); 21 /* NUMD is a 0's based value */ 22 len += 1; 23 len *= sizeof(u32); 24 25 return len; 26 } 27 28 static u32 nvmet_feat_data_len(struct nvmet_req *req, u32 cdw10) 29 { 30 switch (cdw10 & 0xff) { 31 case NVME_FEAT_HOST_ID: 32 return sizeof(req->sq->ctrl->hostid); 33 default: 34 return 0; 35 } 36 } 37 38 u64 nvmet_get_log_page_offset(struct nvme_command *cmd) 39 { 40 return le64_to_cpu(cmd->get_log_page.lpo); 41 } 42 43 static void nvmet_execute_get_log_page_noop(struct nvmet_req *req) 44 { 45 nvmet_req_complete(req, nvmet_zero_sgl(req, 0, req->transfer_len)); 46 } 47 48 static void nvmet_execute_get_log_page_error(struct nvmet_req *req) 49 { 50 struct nvmet_ctrl *ctrl = req->sq->ctrl; 51 unsigned long flags; 52 off_t offset = 0; 53 u64 slot; 54 u64 i; 55 56 spin_lock_irqsave(&ctrl->error_lock, flags); 57 slot = ctrl->err_counter % NVMET_ERROR_LOG_SLOTS; 58 59 for (i = 0; i < NVMET_ERROR_LOG_SLOTS; i++) { 60 if (nvmet_copy_to_sgl(req, offset, &ctrl->slots[slot], 61 sizeof(struct nvme_error_slot))) 62 break; 63 64 if (slot == 0) 65 slot = NVMET_ERROR_LOG_SLOTS - 1; 66 else 67 slot--; 68 offset += sizeof(struct nvme_error_slot); 69 } 70 spin_unlock_irqrestore(&ctrl->error_lock, flags); 71 nvmet_req_complete(req, 0); 72 } 73 74 static u16 nvmet_get_smart_log_nsid(struct nvmet_req *req, 75 struct nvme_smart_log *slog) 76 { 77 u64 host_reads, host_writes, data_units_read, data_units_written; 78 u16 status; 79 80 status = nvmet_req_find_ns(req); 81 if (status) 82 return status; 83 84 /* we don't have the right data for file backed ns */ 85 if (!req->ns->bdev) 86 return NVME_SC_SUCCESS; 87 88 host_reads = part_stat_read(req->ns->bdev, ios[READ]); 89 data_units_read = 90 DIV_ROUND_UP(part_stat_read(req->ns->bdev, sectors[READ]), 1000); 91 host_writes = part_stat_read(req->ns->bdev, ios[WRITE]); 92 data_units_written = 93 DIV_ROUND_UP(part_stat_read(req->ns->bdev, sectors[WRITE]), 1000); 94 95 put_unaligned_le64(host_reads, &slog->host_reads[0]); 96 put_unaligned_le64(data_units_read, &slog->data_units_read[0]); 97 put_unaligned_le64(host_writes, &slog->host_writes[0]); 98 put_unaligned_le64(data_units_written, &slog->data_units_written[0]); 99 100 return NVME_SC_SUCCESS; 101 } 102 103 static u16 nvmet_get_smart_log_all(struct nvmet_req *req, 104 struct nvme_smart_log *slog) 105 { 106 u64 host_reads = 0, host_writes = 0; 107 u64 data_units_read = 0, data_units_written = 0; 108 struct nvmet_ns *ns; 109 struct nvmet_ctrl *ctrl; 110 unsigned long idx; 111 112 ctrl = req->sq->ctrl; 113 xa_for_each(&ctrl->subsys->namespaces, idx, ns) { 114 /* we don't have the right data for file backed ns */ 115 if (!ns->bdev) 116 continue; 117 host_reads += part_stat_read(ns->bdev, ios[READ]); 118 data_units_read += DIV_ROUND_UP( 119 part_stat_read(ns->bdev, sectors[READ]), 1000); 120 host_writes += part_stat_read(ns->bdev, ios[WRITE]); 121 data_units_written += DIV_ROUND_UP( 122 part_stat_read(ns->bdev, sectors[WRITE]), 1000); 123 } 124 125 put_unaligned_le64(host_reads, &slog->host_reads[0]); 126 put_unaligned_le64(data_units_read, &slog->data_units_read[0]); 127 put_unaligned_le64(host_writes, &slog->host_writes[0]); 128 put_unaligned_le64(data_units_written, &slog->data_units_written[0]); 129 130 return NVME_SC_SUCCESS; 131 } 132 133 static void nvmet_execute_get_log_page_smart(struct nvmet_req *req) 134 { 135 struct nvme_smart_log *log; 136 u16 status = NVME_SC_INTERNAL; 137 unsigned long flags; 138 139 if (req->transfer_len != sizeof(*log)) 140 goto out; 141 142 log = kzalloc(sizeof(*log), GFP_KERNEL); 143 if (!log) 144 goto out; 145 146 if (req->cmd->get_log_page.nsid == cpu_to_le32(NVME_NSID_ALL)) 147 status = nvmet_get_smart_log_all(req, log); 148 else 149 status = nvmet_get_smart_log_nsid(req, log); 150 if (status) 151 goto out_free_log; 152 153 spin_lock_irqsave(&req->sq->ctrl->error_lock, flags); 154 put_unaligned_le64(req->sq->ctrl->err_counter, 155 &log->num_err_log_entries); 156 spin_unlock_irqrestore(&req->sq->ctrl->error_lock, flags); 157 158 status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log)); 159 out_free_log: 160 kfree(log); 161 out: 162 nvmet_req_complete(req, status); 163 } 164 165 static void nvmet_get_cmd_effects_nvm(struct nvme_effects_log *log) 166 { 167 log->acs[nvme_admin_get_log_page] = cpu_to_le32(1 << 0); 168 log->acs[nvme_admin_identify] = cpu_to_le32(1 << 0); 169 log->acs[nvme_admin_abort_cmd] = cpu_to_le32(1 << 0); 170 log->acs[nvme_admin_set_features] = cpu_to_le32(1 << 0); 171 log->acs[nvme_admin_get_features] = cpu_to_le32(1 << 0); 172 log->acs[nvme_admin_async_event] = cpu_to_le32(1 << 0); 173 log->acs[nvme_admin_keep_alive] = cpu_to_le32(1 << 0); 174 175 log->iocs[nvme_cmd_read] = cpu_to_le32(1 << 0); 176 log->iocs[nvme_cmd_write] = cpu_to_le32(1 << 0); 177 log->iocs[nvme_cmd_flush] = cpu_to_le32(1 << 0); 178 log->iocs[nvme_cmd_dsm] = cpu_to_le32(1 << 0); 179 log->iocs[nvme_cmd_write_zeroes] = cpu_to_le32(1 << 0); 180 } 181 182 static void nvmet_get_cmd_effects_zns(struct nvme_effects_log *log) 183 { 184 log->iocs[nvme_cmd_zone_append] = cpu_to_le32(1 << 0); 185 log->iocs[nvme_cmd_zone_mgmt_send] = cpu_to_le32(1 << 0); 186 log->iocs[nvme_cmd_zone_mgmt_recv] = cpu_to_le32(1 << 0); 187 } 188 189 static void nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req *req) 190 { 191 struct nvme_effects_log *log; 192 u16 status = NVME_SC_SUCCESS; 193 194 log = kzalloc(sizeof(*log), GFP_KERNEL); 195 if (!log) { 196 status = NVME_SC_INTERNAL; 197 goto out; 198 } 199 200 switch (req->cmd->get_log_page.csi) { 201 case NVME_CSI_NVM: 202 nvmet_get_cmd_effects_nvm(log); 203 break; 204 case NVME_CSI_ZNS: 205 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) { 206 status = NVME_SC_INVALID_IO_CMD_SET; 207 goto free; 208 } 209 nvmet_get_cmd_effects_nvm(log); 210 nvmet_get_cmd_effects_zns(log); 211 break; 212 default: 213 status = NVME_SC_INVALID_LOG_PAGE; 214 goto free; 215 } 216 217 status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log)); 218 free: 219 kfree(log); 220 out: 221 nvmet_req_complete(req, status); 222 } 223 224 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req) 225 { 226 struct nvmet_ctrl *ctrl = req->sq->ctrl; 227 u16 status = NVME_SC_INTERNAL; 228 size_t len; 229 230 if (req->transfer_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32)) 231 goto out; 232 233 mutex_lock(&ctrl->lock); 234 if (ctrl->nr_changed_ns == U32_MAX) 235 len = sizeof(__le32); 236 else 237 len = ctrl->nr_changed_ns * sizeof(__le32); 238 status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len); 239 if (!status) 240 status = nvmet_zero_sgl(req, len, req->transfer_len - len); 241 ctrl->nr_changed_ns = 0; 242 nvmet_clear_aen_bit(req, NVME_AEN_BIT_NS_ATTR); 243 mutex_unlock(&ctrl->lock); 244 out: 245 nvmet_req_complete(req, status); 246 } 247 248 static u32 nvmet_format_ana_group(struct nvmet_req *req, u32 grpid, 249 struct nvme_ana_group_desc *desc) 250 { 251 struct nvmet_ctrl *ctrl = req->sq->ctrl; 252 struct nvmet_ns *ns; 253 unsigned long idx; 254 u32 count = 0; 255 256 if (!(req->cmd->get_log_page.lsp & NVME_ANA_LOG_RGO)) { 257 xa_for_each(&ctrl->subsys->namespaces, idx, ns) 258 if (ns->anagrpid == grpid) 259 desc->nsids[count++] = cpu_to_le32(ns->nsid); 260 } 261 262 desc->grpid = cpu_to_le32(grpid); 263 desc->nnsids = cpu_to_le32(count); 264 desc->chgcnt = cpu_to_le64(nvmet_ana_chgcnt); 265 desc->state = req->port->ana_state[grpid]; 266 memset(desc->rsvd17, 0, sizeof(desc->rsvd17)); 267 return struct_size(desc, nsids, count); 268 } 269 270 static void nvmet_execute_get_log_page_ana(struct nvmet_req *req) 271 { 272 struct nvme_ana_rsp_hdr hdr = { 0, }; 273 struct nvme_ana_group_desc *desc; 274 size_t offset = sizeof(struct nvme_ana_rsp_hdr); /* start beyond hdr */ 275 size_t len; 276 u32 grpid; 277 u16 ngrps = 0; 278 u16 status; 279 280 status = NVME_SC_INTERNAL; 281 desc = kmalloc(struct_size(desc, nsids, NVMET_MAX_NAMESPACES), 282 GFP_KERNEL); 283 if (!desc) 284 goto out; 285 286 down_read(&nvmet_ana_sem); 287 for (grpid = 1; grpid <= NVMET_MAX_ANAGRPS; grpid++) { 288 if (!nvmet_ana_group_enabled[grpid]) 289 continue; 290 len = nvmet_format_ana_group(req, grpid, desc); 291 status = nvmet_copy_to_sgl(req, offset, desc, len); 292 if (status) 293 break; 294 offset += len; 295 ngrps++; 296 } 297 for ( ; grpid <= NVMET_MAX_ANAGRPS; grpid++) { 298 if (nvmet_ana_group_enabled[grpid]) 299 ngrps++; 300 } 301 302 hdr.chgcnt = cpu_to_le64(nvmet_ana_chgcnt); 303 hdr.ngrps = cpu_to_le16(ngrps); 304 nvmet_clear_aen_bit(req, NVME_AEN_BIT_ANA_CHANGE); 305 up_read(&nvmet_ana_sem); 306 307 kfree(desc); 308 309 /* copy the header last once we know the number of groups */ 310 status = nvmet_copy_to_sgl(req, 0, &hdr, sizeof(hdr)); 311 out: 312 nvmet_req_complete(req, status); 313 } 314 315 static void nvmet_execute_get_log_page(struct nvmet_req *req) 316 { 317 if (!nvmet_check_transfer_len(req, nvmet_get_log_page_len(req->cmd))) 318 return; 319 320 switch (req->cmd->get_log_page.lid) { 321 case NVME_LOG_ERROR: 322 return nvmet_execute_get_log_page_error(req); 323 case NVME_LOG_SMART: 324 return nvmet_execute_get_log_page_smart(req); 325 case NVME_LOG_FW_SLOT: 326 /* 327 * We only support a single firmware slot which always is 328 * active, so we can zero out the whole firmware slot log and 329 * still claim to fully implement this mandatory log page. 330 */ 331 return nvmet_execute_get_log_page_noop(req); 332 case NVME_LOG_CHANGED_NS: 333 return nvmet_execute_get_log_changed_ns(req); 334 case NVME_LOG_CMD_EFFECTS: 335 return nvmet_execute_get_log_cmd_effects_ns(req); 336 case NVME_LOG_ANA: 337 return nvmet_execute_get_log_page_ana(req); 338 } 339 pr_debug("unhandled lid %d on qid %d\n", 340 req->cmd->get_log_page.lid, req->sq->qid); 341 req->error_loc = offsetof(struct nvme_get_log_page_command, lid); 342 nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR); 343 } 344 345 static void nvmet_execute_identify_ctrl(struct nvmet_req *req) 346 { 347 struct nvmet_ctrl *ctrl = req->sq->ctrl; 348 struct nvmet_subsys *subsys = ctrl->subsys; 349 struct nvme_id_ctrl *id; 350 u32 cmd_capsule_size; 351 u16 status = 0; 352 353 if (!subsys->subsys_discovered) { 354 mutex_lock(&subsys->lock); 355 subsys->subsys_discovered = true; 356 mutex_unlock(&subsys->lock); 357 } 358 359 id = kzalloc(sizeof(*id), GFP_KERNEL); 360 if (!id) { 361 status = NVME_SC_INTERNAL; 362 goto out; 363 } 364 365 /* XXX: figure out how to assign real vendors IDs. */ 366 id->vid = 0; 367 id->ssvid = 0; 368 369 memcpy(id->sn, ctrl->subsys->serial, NVMET_SN_MAX_SIZE); 370 memcpy_and_pad(id->mn, sizeof(id->mn), subsys->model_number, 371 strlen(subsys->model_number), ' '); 372 memcpy_and_pad(id->fr, sizeof(id->fr), 373 UTS_RELEASE, strlen(UTS_RELEASE), ' '); 374 375 id->rab = 6; 376 377 if (nvmet_is_disc_subsys(ctrl->subsys)) 378 id->cntrltype = NVME_CTRL_DISC; 379 else 380 id->cntrltype = NVME_CTRL_IO; 381 382 /* 383 * XXX: figure out how we can assign a IEEE OUI, but until then 384 * the safest is to leave it as zeroes. 385 */ 386 387 /* we support multiple ports, multiples hosts and ANA: */ 388 id->cmic = NVME_CTRL_CMIC_MULTI_PORT | NVME_CTRL_CMIC_MULTI_CTRL | 389 NVME_CTRL_CMIC_ANA; 390 391 /* Limit MDTS according to transport capability */ 392 if (ctrl->ops->get_mdts) 393 id->mdts = ctrl->ops->get_mdts(ctrl); 394 else 395 id->mdts = 0; 396 397 id->cntlid = cpu_to_le16(ctrl->cntlid); 398 id->ver = cpu_to_le32(ctrl->subsys->ver); 399 400 /* XXX: figure out what to do about RTD3R/RTD3 */ 401 id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL); 402 id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT | 403 NVME_CTRL_ATTR_TBKAS); 404 405 id->oacs = 0; 406 407 /* 408 * We don't really have a practical limit on the number of abort 409 * comands. But we don't do anything useful for abort either, so 410 * no point in allowing more abort commands than the spec requires. 411 */ 412 id->acl = 3; 413 414 id->aerl = NVMET_ASYNC_EVENTS - 1; 415 416 /* first slot is read-only, only one slot supported */ 417 id->frmw = (1 << 0) | (1 << 1); 418 id->lpa = (1 << 0) | (1 << 1) | (1 << 2); 419 id->elpe = NVMET_ERROR_LOG_SLOTS - 1; 420 id->npss = 0; 421 422 /* We support keep-alive timeout in granularity of seconds */ 423 id->kas = cpu_to_le16(NVMET_KAS); 424 425 id->sqes = (0x6 << 4) | 0x6; 426 id->cqes = (0x4 << 4) | 0x4; 427 428 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */ 429 id->maxcmd = cpu_to_le16(NVMET_MAX_CMD); 430 431 id->nn = cpu_to_le32(NVMET_MAX_NAMESPACES); 432 id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES); 433 id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM | 434 NVME_CTRL_ONCS_WRITE_ZEROES); 435 436 /* XXX: don't report vwc if the underlying device is write through */ 437 id->vwc = NVME_CTRL_VWC_PRESENT; 438 439 /* 440 * We can't support atomic writes bigger than a LBA without support 441 * from the backend device. 442 */ 443 id->awun = 0; 444 id->awupf = 0; 445 446 id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */ 447 if (ctrl->ops->flags & NVMF_KEYED_SGLS) 448 id->sgls |= cpu_to_le32(1 << 2); 449 if (req->port->inline_data_size) 450 id->sgls |= cpu_to_le32(1 << 20); 451 452 strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn)); 453 454 /* 455 * Max command capsule size is sqe + in-capsule data size. 456 * Disable in-capsule data for Metadata capable controllers. 457 */ 458 cmd_capsule_size = sizeof(struct nvme_command); 459 if (!ctrl->pi_support) 460 cmd_capsule_size += req->port->inline_data_size; 461 id->ioccsz = cpu_to_le32(cmd_capsule_size / 16); 462 463 /* Max response capsule size is cqe */ 464 id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16); 465 466 id->msdbd = ctrl->ops->msdbd; 467 468 id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4); 469 id->anatt = 10; /* random value */ 470 id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS); 471 id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS); 472 473 /* 474 * Meh, we don't really support any power state. Fake up the same 475 * values that qemu does. 476 */ 477 id->psd[0].max_power = cpu_to_le16(0x9c4); 478 id->psd[0].entry_lat = cpu_to_le32(0x10); 479 id->psd[0].exit_lat = cpu_to_le32(0x4); 480 481 id->nwpc = 1 << 0; /* write protect and no write protect */ 482 483 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 484 485 kfree(id); 486 out: 487 nvmet_req_complete(req, status); 488 } 489 490 static void nvmet_execute_identify_ns(struct nvmet_req *req) 491 { 492 struct nvme_id_ns *id; 493 u16 status; 494 495 if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) { 496 req->error_loc = offsetof(struct nvme_identify, nsid); 497 status = NVME_SC_INVALID_NS | NVME_SC_DNR; 498 goto out; 499 } 500 501 id = kzalloc(sizeof(*id), GFP_KERNEL); 502 if (!id) { 503 status = NVME_SC_INTERNAL; 504 goto out; 505 } 506 507 /* return an all zeroed buffer if we can't find an active namespace */ 508 status = nvmet_req_find_ns(req); 509 if (status) { 510 status = 0; 511 goto done; 512 } 513 514 nvmet_ns_revalidate(req->ns); 515 516 /* 517 * nuse = ncap = nsze isn't always true, but we have no way to find 518 * that out from the underlying device. 519 */ 520 id->ncap = id->nsze = 521 cpu_to_le64(req->ns->size >> req->ns->blksize_shift); 522 switch (req->port->ana_state[req->ns->anagrpid]) { 523 case NVME_ANA_INACCESSIBLE: 524 case NVME_ANA_PERSISTENT_LOSS: 525 break; 526 default: 527 id->nuse = id->nsze; 528 break; 529 } 530 531 if (req->ns->bdev) 532 nvmet_bdev_set_limits(req->ns->bdev, id); 533 534 /* 535 * We just provide a single LBA format that matches what the 536 * underlying device reports. 537 */ 538 id->nlbaf = 0; 539 id->flbas = 0; 540 541 /* 542 * Our namespace might always be shared. Not just with other 543 * controllers, but also with any other user of the block device. 544 */ 545 id->nmic = NVME_NS_NMIC_SHARED; 546 id->anagrpid = cpu_to_le32(req->ns->anagrpid); 547 548 memcpy(&id->nguid, &req->ns->nguid, sizeof(id->nguid)); 549 550 id->lbaf[0].ds = req->ns->blksize_shift; 551 552 if (req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns)) { 553 id->dpc = NVME_NS_DPC_PI_FIRST | NVME_NS_DPC_PI_LAST | 554 NVME_NS_DPC_PI_TYPE1 | NVME_NS_DPC_PI_TYPE2 | 555 NVME_NS_DPC_PI_TYPE3; 556 id->mc = NVME_MC_EXTENDED_LBA; 557 id->dps = req->ns->pi_type; 558 id->flbas = NVME_NS_FLBAS_META_EXT; 559 id->lbaf[0].ms = cpu_to_le16(req->ns->metadata_size); 560 } 561 562 if (req->ns->readonly) 563 id->nsattr |= (1 << 0); 564 done: 565 if (!status) 566 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 567 568 kfree(id); 569 out: 570 nvmet_req_complete(req, status); 571 } 572 573 static void nvmet_execute_identify_nslist(struct nvmet_req *req) 574 { 575 static const int buf_size = NVME_IDENTIFY_DATA_SIZE; 576 struct nvmet_ctrl *ctrl = req->sq->ctrl; 577 struct nvmet_ns *ns; 578 unsigned long idx; 579 u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid); 580 __le32 *list; 581 u16 status = 0; 582 int i = 0; 583 584 list = kzalloc(buf_size, GFP_KERNEL); 585 if (!list) { 586 status = NVME_SC_INTERNAL; 587 goto out; 588 } 589 590 xa_for_each(&ctrl->subsys->namespaces, idx, ns) { 591 if (ns->nsid <= min_nsid) 592 continue; 593 list[i++] = cpu_to_le32(ns->nsid); 594 if (i == buf_size / sizeof(__le32)) 595 break; 596 } 597 598 status = nvmet_copy_to_sgl(req, 0, list, buf_size); 599 600 kfree(list); 601 out: 602 nvmet_req_complete(req, status); 603 } 604 605 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len, 606 void *id, off_t *off) 607 { 608 struct nvme_ns_id_desc desc = { 609 .nidt = type, 610 .nidl = len, 611 }; 612 u16 status; 613 614 status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc)); 615 if (status) 616 return status; 617 *off += sizeof(desc); 618 619 status = nvmet_copy_to_sgl(req, *off, id, len); 620 if (status) 621 return status; 622 *off += len; 623 624 return 0; 625 } 626 627 static void nvmet_execute_identify_desclist(struct nvmet_req *req) 628 { 629 off_t off = 0; 630 u16 status; 631 632 status = nvmet_req_find_ns(req); 633 if (status) 634 goto out; 635 636 if (memchr_inv(&req->ns->uuid, 0, sizeof(req->ns->uuid))) { 637 status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID, 638 NVME_NIDT_UUID_LEN, 639 &req->ns->uuid, &off); 640 if (status) 641 goto out; 642 } 643 if (memchr_inv(req->ns->nguid, 0, sizeof(req->ns->nguid))) { 644 status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID, 645 NVME_NIDT_NGUID_LEN, 646 &req->ns->nguid, &off); 647 if (status) 648 goto out; 649 } 650 651 status = nvmet_copy_ns_identifier(req, NVME_NIDT_CSI, 652 NVME_NIDT_CSI_LEN, 653 &req->ns->csi, &off); 654 if (status) 655 goto out; 656 657 if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off, 658 off) != NVME_IDENTIFY_DATA_SIZE - off) 659 status = NVME_SC_INTERNAL | NVME_SC_DNR; 660 661 out: 662 nvmet_req_complete(req, status); 663 } 664 665 static bool nvmet_handle_identify_desclist(struct nvmet_req *req) 666 { 667 switch (req->cmd->identify.csi) { 668 case NVME_CSI_NVM: 669 nvmet_execute_identify_desclist(req); 670 return true; 671 case NVME_CSI_ZNS: 672 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) { 673 nvmet_execute_identify_desclist(req); 674 return true; 675 } 676 return false; 677 default: 678 return false; 679 } 680 } 681 682 static void nvmet_execute_identify(struct nvmet_req *req) 683 { 684 if (!nvmet_check_transfer_len(req, NVME_IDENTIFY_DATA_SIZE)) 685 return; 686 687 switch (req->cmd->identify.cns) { 688 case NVME_ID_CNS_NS: 689 switch (req->cmd->identify.csi) { 690 case NVME_CSI_NVM: 691 return nvmet_execute_identify_ns(req); 692 default: 693 break; 694 } 695 break; 696 case NVME_ID_CNS_CS_NS: 697 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) { 698 switch (req->cmd->identify.csi) { 699 case NVME_CSI_ZNS: 700 return nvmet_execute_identify_cns_cs_ns(req); 701 default: 702 break; 703 } 704 } 705 break; 706 case NVME_ID_CNS_CTRL: 707 switch (req->cmd->identify.csi) { 708 case NVME_CSI_NVM: 709 return nvmet_execute_identify_ctrl(req); 710 } 711 break; 712 case NVME_ID_CNS_CS_CTRL: 713 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) { 714 switch (req->cmd->identify.csi) { 715 case NVME_CSI_ZNS: 716 return nvmet_execute_identify_cns_cs_ctrl(req); 717 default: 718 break; 719 } 720 } 721 break; 722 case NVME_ID_CNS_NS_ACTIVE_LIST: 723 switch (req->cmd->identify.csi) { 724 case NVME_CSI_NVM: 725 return nvmet_execute_identify_nslist(req); 726 default: 727 break; 728 } 729 break; 730 case NVME_ID_CNS_NS_DESC_LIST: 731 if (nvmet_handle_identify_desclist(req) == true) 732 return; 733 break; 734 } 735 736 nvmet_req_cns_error_complete(req); 737 } 738 739 /* 740 * A "minimum viable" abort implementation: the command is mandatory in the 741 * spec, but we are not required to do any useful work. We couldn't really 742 * do a useful abort, so don't bother even with waiting for the command 743 * to be exectuted and return immediately telling the command to abort 744 * wasn't found. 745 */ 746 static void nvmet_execute_abort(struct nvmet_req *req) 747 { 748 if (!nvmet_check_transfer_len(req, 0)) 749 return; 750 nvmet_set_result(req, 1); 751 nvmet_req_complete(req, 0); 752 } 753 754 static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req) 755 { 756 u16 status; 757 758 if (req->ns->file) 759 status = nvmet_file_flush(req); 760 else 761 status = nvmet_bdev_flush(req); 762 763 if (status) 764 pr_err("write protect flush failed nsid: %u\n", req->ns->nsid); 765 return status; 766 } 767 768 static u16 nvmet_set_feat_write_protect(struct nvmet_req *req) 769 { 770 u32 write_protect = le32_to_cpu(req->cmd->common.cdw11); 771 struct nvmet_subsys *subsys = nvmet_req_subsys(req); 772 u16 status; 773 774 status = nvmet_req_find_ns(req); 775 if (status) 776 return status; 777 778 mutex_lock(&subsys->lock); 779 switch (write_protect) { 780 case NVME_NS_WRITE_PROTECT: 781 req->ns->readonly = true; 782 status = nvmet_write_protect_flush_sync(req); 783 if (status) 784 req->ns->readonly = false; 785 break; 786 case NVME_NS_NO_WRITE_PROTECT: 787 req->ns->readonly = false; 788 status = 0; 789 break; 790 default: 791 break; 792 } 793 794 if (!status) 795 nvmet_ns_changed(subsys, req->ns->nsid); 796 mutex_unlock(&subsys->lock); 797 return status; 798 } 799 800 u16 nvmet_set_feat_kato(struct nvmet_req *req) 801 { 802 u32 val32 = le32_to_cpu(req->cmd->common.cdw11); 803 804 nvmet_stop_keep_alive_timer(req->sq->ctrl); 805 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000); 806 nvmet_start_keep_alive_timer(req->sq->ctrl); 807 808 nvmet_set_result(req, req->sq->ctrl->kato); 809 810 return 0; 811 } 812 813 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask) 814 { 815 u32 val32 = le32_to_cpu(req->cmd->common.cdw11); 816 817 if (val32 & ~mask) { 818 req->error_loc = offsetof(struct nvme_common_command, cdw11); 819 return NVME_SC_INVALID_FIELD | NVME_SC_DNR; 820 } 821 822 WRITE_ONCE(req->sq->ctrl->aen_enabled, val32); 823 nvmet_set_result(req, val32); 824 825 return 0; 826 } 827 828 void nvmet_execute_set_features(struct nvmet_req *req) 829 { 830 struct nvmet_subsys *subsys = nvmet_req_subsys(req); 831 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10); 832 u32 cdw11 = le32_to_cpu(req->cmd->common.cdw11); 833 u16 status = 0; 834 u16 nsqr; 835 u16 ncqr; 836 837 if (!nvmet_check_transfer_len(req, 0)) 838 return; 839 840 switch (cdw10 & 0xff) { 841 case NVME_FEAT_NUM_QUEUES: 842 ncqr = (cdw11 >> 16) & 0xffff; 843 nsqr = cdw11 & 0xffff; 844 if (ncqr == 0xffff || nsqr == 0xffff) { 845 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 846 break; 847 } 848 nvmet_set_result(req, 849 (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16)); 850 break; 851 case NVME_FEAT_KATO: 852 status = nvmet_set_feat_kato(req); 853 break; 854 case NVME_FEAT_ASYNC_EVENT: 855 status = nvmet_set_feat_async_event(req, NVMET_AEN_CFG_ALL); 856 break; 857 case NVME_FEAT_HOST_ID: 858 status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR; 859 break; 860 case NVME_FEAT_WRITE_PROTECT: 861 status = nvmet_set_feat_write_protect(req); 862 break; 863 default: 864 req->error_loc = offsetof(struct nvme_common_command, cdw10); 865 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 866 break; 867 } 868 869 nvmet_req_complete(req, status); 870 } 871 872 static u16 nvmet_get_feat_write_protect(struct nvmet_req *req) 873 { 874 struct nvmet_subsys *subsys = nvmet_req_subsys(req); 875 u32 result; 876 877 result = nvmet_req_find_ns(req); 878 if (result) 879 return result; 880 881 mutex_lock(&subsys->lock); 882 if (req->ns->readonly == true) 883 result = NVME_NS_WRITE_PROTECT; 884 else 885 result = NVME_NS_NO_WRITE_PROTECT; 886 nvmet_set_result(req, result); 887 mutex_unlock(&subsys->lock); 888 889 return 0; 890 } 891 892 void nvmet_get_feat_kato(struct nvmet_req *req) 893 { 894 nvmet_set_result(req, req->sq->ctrl->kato * 1000); 895 } 896 897 void nvmet_get_feat_async_event(struct nvmet_req *req) 898 { 899 nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled)); 900 } 901 902 void nvmet_execute_get_features(struct nvmet_req *req) 903 { 904 struct nvmet_subsys *subsys = nvmet_req_subsys(req); 905 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10); 906 u16 status = 0; 907 908 if (!nvmet_check_transfer_len(req, nvmet_feat_data_len(req, cdw10))) 909 return; 910 911 switch (cdw10 & 0xff) { 912 /* 913 * These features are mandatory in the spec, but we don't 914 * have a useful way to implement them. We'll eventually 915 * need to come up with some fake values for these. 916 */ 917 #if 0 918 case NVME_FEAT_ARBITRATION: 919 break; 920 case NVME_FEAT_POWER_MGMT: 921 break; 922 case NVME_FEAT_TEMP_THRESH: 923 break; 924 case NVME_FEAT_ERR_RECOVERY: 925 break; 926 case NVME_FEAT_IRQ_COALESCE: 927 break; 928 case NVME_FEAT_IRQ_CONFIG: 929 break; 930 case NVME_FEAT_WRITE_ATOMIC: 931 break; 932 #endif 933 case NVME_FEAT_ASYNC_EVENT: 934 nvmet_get_feat_async_event(req); 935 break; 936 case NVME_FEAT_VOLATILE_WC: 937 nvmet_set_result(req, 1); 938 break; 939 case NVME_FEAT_NUM_QUEUES: 940 nvmet_set_result(req, 941 (subsys->max_qid-1) | ((subsys->max_qid-1) << 16)); 942 break; 943 case NVME_FEAT_KATO: 944 nvmet_get_feat_kato(req); 945 break; 946 case NVME_FEAT_HOST_ID: 947 /* need 128-bit host identifier flag */ 948 if (!(req->cmd->common.cdw11 & cpu_to_le32(1 << 0))) { 949 req->error_loc = 950 offsetof(struct nvme_common_command, cdw11); 951 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 952 break; 953 } 954 955 status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid, 956 sizeof(req->sq->ctrl->hostid)); 957 break; 958 case NVME_FEAT_WRITE_PROTECT: 959 status = nvmet_get_feat_write_protect(req); 960 break; 961 default: 962 req->error_loc = 963 offsetof(struct nvme_common_command, cdw10); 964 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 965 break; 966 } 967 968 nvmet_req_complete(req, status); 969 } 970 971 void nvmet_execute_async_event(struct nvmet_req *req) 972 { 973 struct nvmet_ctrl *ctrl = req->sq->ctrl; 974 975 if (!nvmet_check_transfer_len(req, 0)) 976 return; 977 978 mutex_lock(&ctrl->lock); 979 if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) { 980 mutex_unlock(&ctrl->lock); 981 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR); 982 return; 983 } 984 ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req; 985 mutex_unlock(&ctrl->lock); 986 987 schedule_work(&ctrl->async_event_work); 988 } 989 990 void nvmet_execute_keep_alive(struct nvmet_req *req) 991 { 992 struct nvmet_ctrl *ctrl = req->sq->ctrl; 993 u16 status = 0; 994 995 if (!nvmet_check_transfer_len(req, 0)) 996 return; 997 998 if (!ctrl->kato) { 999 status = NVME_SC_KA_TIMEOUT_INVALID; 1000 goto out; 1001 } 1002 1003 pr_debug("ctrl %d update keep-alive timer for %d secs\n", 1004 ctrl->cntlid, ctrl->kato); 1005 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ); 1006 out: 1007 nvmet_req_complete(req, status); 1008 } 1009 1010 u16 nvmet_parse_admin_cmd(struct nvmet_req *req) 1011 { 1012 struct nvme_command *cmd = req->cmd; 1013 u16 ret; 1014 1015 if (nvme_is_fabrics(cmd)) 1016 return nvmet_parse_fabrics_cmd(req); 1017 if (nvmet_is_disc_subsys(nvmet_req_subsys(req))) 1018 return nvmet_parse_discovery_cmd(req); 1019 1020 ret = nvmet_check_ctrl_status(req); 1021 if (unlikely(ret)) 1022 return ret; 1023 1024 if (nvmet_is_passthru_req(req)) 1025 return nvmet_parse_passthru_admin_cmd(req); 1026 1027 switch (cmd->common.opcode) { 1028 case nvme_admin_get_log_page: 1029 req->execute = nvmet_execute_get_log_page; 1030 return 0; 1031 case nvme_admin_identify: 1032 req->execute = nvmet_execute_identify; 1033 return 0; 1034 case nvme_admin_abort_cmd: 1035 req->execute = nvmet_execute_abort; 1036 return 0; 1037 case nvme_admin_set_features: 1038 req->execute = nvmet_execute_set_features; 1039 return 0; 1040 case nvme_admin_get_features: 1041 req->execute = nvmet_execute_get_features; 1042 return 0; 1043 case nvme_admin_async_event: 1044 req->execute = nvmet_execute_async_event; 1045 return 0; 1046 case nvme_admin_keep_alive: 1047 req->execute = nvmet_execute_keep_alive; 1048 return 0; 1049 default: 1050 return nvmet_report_invalid_opcode(req); 1051 } 1052 } 1053