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 struct nvmet_ns *ns; 78 u64 host_reads, host_writes, data_units_read, data_units_written; 79 80 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->get_log_page.nsid); 81 if (!ns) { 82 pr_err("Could not find namespace id : %d\n", 83 le32_to_cpu(req->cmd->get_log_page.nsid)); 84 req->error_loc = offsetof(struct nvme_rw_command, nsid); 85 return NVME_SC_INVALID_NS; 86 } 87 88 /* we don't have the right data for file backed ns */ 89 if (!ns->bdev) 90 goto out; 91 92 host_reads = part_stat_read(ns->bdev->bd_part, ios[READ]); 93 data_units_read = DIV_ROUND_UP(part_stat_read(ns->bdev->bd_part, 94 sectors[READ]), 1000); 95 host_writes = part_stat_read(ns->bdev->bd_part, ios[WRITE]); 96 data_units_written = DIV_ROUND_UP(part_stat_read(ns->bdev->bd_part, 97 sectors[WRITE]), 1000); 98 99 put_unaligned_le64(host_reads, &slog->host_reads[0]); 100 put_unaligned_le64(data_units_read, &slog->data_units_read[0]); 101 put_unaligned_le64(host_writes, &slog->host_writes[0]); 102 put_unaligned_le64(data_units_written, &slog->data_units_written[0]); 103 out: 104 nvmet_put_namespace(ns); 105 106 return NVME_SC_SUCCESS; 107 } 108 109 static u16 nvmet_get_smart_log_all(struct nvmet_req *req, 110 struct nvme_smart_log *slog) 111 { 112 u64 host_reads = 0, host_writes = 0; 113 u64 data_units_read = 0, data_units_written = 0; 114 struct nvmet_ns *ns; 115 struct nvmet_ctrl *ctrl; 116 117 ctrl = req->sq->ctrl; 118 119 rcu_read_lock(); 120 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) { 121 /* we don't have the right data for file backed ns */ 122 if (!ns->bdev) 123 continue; 124 host_reads += part_stat_read(ns->bdev->bd_part, ios[READ]); 125 data_units_read += DIV_ROUND_UP( 126 part_stat_read(ns->bdev->bd_part, sectors[READ]), 1000); 127 host_writes += part_stat_read(ns->bdev->bd_part, ios[WRITE]); 128 data_units_written += DIV_ROUND_UP( 129 part_stat_read(ns->bdev->bd_part, sectors[WRITE]), 1000); 130 131 } 132 rcu_read_unlock(); 133 134 put_unaligned_le64(host_reads, &slog->host_reads[0]); 135 put_unaligned_le64(data_units_read, &slog->data_units_read[0]); 136 put_unaligned_le64(host_writes, &slog->host_writes[0]); 137 put_unaligned_le64(data_units_written, &slog->data_units_written[0]); 138 139 return NVME_SC_SUCCESS; 140 } 141 142 static void nvmet_execute_get_log_page_smart(struct nvmet_req *req) 143 { 144 struct nvme_smart_log *log; 145 u16 status = NVME_SC_INTERNAL; 146 unsigned long flags; 147 148 if (req->transfer_len != sizeof(*log)) 149 goto out; 150 151 log = kzalloc(sizeof(*log), GFP_KERNEL); 152 if (!log) 153 goto out; 154 155 if (req->cmd->get_log_page.nsid == cpu_to_le32(NVME_NSID_ALL)) 156 status = nvmet_get_smart_log_all(req, log); 157 else 158 status = nvmet_get_smart_log_nsid(req, log); 159 if (status) 160 goto out_free_log; 161 162 spin_lock_irqsave(&req->sq->ctrl->error_lock, flags); 163 put_unaligned_le64(req->sq->ctrl->err_counter, 164 &log->num_err_log_entries); 165 spin_unlock_irqrestore(&req->sq->ctrl->error_lock, flags); 166 167 status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log)); 168 out_free_log: 169 kfree(log); 170 out: 171 nvmet_req_complete(req, status); 172 } 173 174 static void nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req *req) 175 { 176 u16 status = NVME_SC_INTERNAL; 177 struct nvme_effects_log *log; 178 179 log = kzalloc(sizeof(*log), GFP_KERNEL); 180 if (!log) 181 goto out; 182 183 log->acs[nvme_admin_get_log_page] = cpu_to_le32(1 << 0); 184 log->acs[nvme_admin_identify] = cpu_to_le32(1 << 0); 185 log->acs[nvme_admin_abort_cmd] = cpu_to_le32(1 << 0); 186 log->acs[nvme_admin_set_features] = cpu_to_le32(1 << 0); 187 log->acs[nvme_admin_get_features] = cpu_to_le32(1 << 0); 188 log->acs[nvme_admin_async_event] = cpu_to_le32(1 << 0); 189 log->acs[nvme_admin_keep_alive] = cpu_to_le32(1 << 0); 190 191 log->iocs[nvme_cmd_read] = cpu_to_le32(1 << 0); 192 log->iocs[nvme_cmd_write] = cpu_to_le32(1 << 0); 193 log->iocs[nvme_cmd_flush] = cpu_to_le32(1 << 0); 194 log->iocs[nvme_cmd_dsm] = cpu_to_le32(1 << 0); 195 log->iocs[nvme_cmd_write_zeroes] = cpu_to_le32(1 << 0); 196 197 status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log)); 198 199 kfree(log); 200 out: 201 nvmet_req_complete(req, status); 202 } 203 204 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req) 205 { 206 struct nvmet_ctrl *ctrl = req->sq->ctrl; 207 u16 status = NVME_SC_INTERNAL; 208 size_t len; 209 210 if (req->transfer_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32)) 211 goto out; 212 213 mutex_lock(&ctrl->lock); 214 if (ctrl->nr_changed_ns == U32_MAX) 215 len = sizeof(__le32); 216 else 217 len = ctrl->nr_changed_ns * sizeof(__le32); 218 status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len); 219 if (!status) 220 status = nvmet_zero_sgl(req, len, req->transfer_len - len); 221 ctrl->nr_changed_ns = 0; 222 nvmet_clear_aen_bit(req, NVME_AEN_BIT_NS_ATTR); 223 mutex_unlock(&ctrl->lock); 224 out: 225 nvmet_req_complete(req, status); 226 } 227 228 static u32 nvmet_format_ana_group(struct nvmet_req *req, u32 grpid, 229 struct nvme_ana_group_desc *desc) 230 { 231 struct nvmet_ctrl *ctrl = req->sq->ctrl; 232 struct nvmet_ns *ns; 233 u32 count = 0; 234 235 if (!(req->cmd->get_log_page.lsp & NVME_ANA_LOG_RGO)) { 236 rcu_read_lock(); 237 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) 238 if (ns->anagrpid == grpid) 239 desc->nsids[count++] = cpu_to_le32(ns->nsid); 240 rcu_read_unlock(); 241 } 242 243 desc->grpid = cpu_to_le32(grpid); 244 desc->nnsids = cpu_to_le32(count); 245 desc->chgcnt = cpu_to_le64(nvmet_ana_chgcnt); 246 desc->state = req->port->ana_state[grpid]; 247 memset(desc->rsvd17, 0, sizeof(desc->rsvd17)); 248 return sizeof(struct nvme_ana_group_desc) + count * sizeof(__le32); 249 } 250 251 static void nvmet_execute_get_log_page_ana(struct nvmet_req *req) 252 { 253 struct nvme_ana_rsp_hdr hdr = { 0, }; 254 struct nvme_ana_group_desc *desc; 255 size_t offset = sizeof(struct nvme_ana_rsp_hdr); /* start beyond hdr */ 256 size_t len; 257 u32 grpid; 258 u16 ngrps = 0; 259 u16 status; 260 261 status = NVME_SC_INTERNAL; 262 desc = kmalloc(sizeof(struct nvme_ana_group_desc) + 263 NVMET_MAX_NAMESPACES * sizeof(__le32), GFP_KERNEL); 264 if (!desc) 265 goto out; 266 267 down_read(&nvmet_ana_sem); 268 for (grpid = 1; grpid <= NVMET_MAX_ANAGRPS; grpid++) { 269 if (!nvmet_ana_group_enabled[grpid]) 270 continue; 271 len = nvmet_format_ana_group(req, grpid, desc); 272 status = nvmet_copy_to_sgl(req, offset, desc, len); 273 if (status) 274 break; 275 offset += len; 276 ngrps++; 277 } 278 for ( ; grpid <= NVMET_MAX_ANAGRPS; grpid++) { 279 if (nvmet_ana_group_enabled[grpid]) 280 ngrps++; 281 } 282 283 hdr.chgcnt = cpu_to_le64(nvmet_ana_chgcnt); 284 hdr.ngrps = cpu_to_le16(ngrps); 285 nvmet_clear_aen_bit(req, NVME_AEN_BIT_ANA_CHANGE); 286 up_read(&nvmet_ana_sem); 287 288 kfree(desc); 289 290 /* copy the header last once we know the number of groups */ 291 status = nvmet_copy_to_sgl(req, 0, &hdr, sizeof(hdr)); 292 out: 293 nvmet_req_complete(req, status); 294 } 295 296 static void nvmet_execute_get_log_page(struct nvmet_req *req) 297 { 298 if (!nvmet_check_data_len(req, nvmet_get_log_page_len(req->cmd))) 299 return; 300 301 switch (req->cmd->get_log_page.lid) { 302 case NVME_LOG_ERROR: 303 return nvmet_execute_get_log_page_error(req); 304 case NVME_LOG_SMART: 305 return nvmet_execute_get_log_page_smart(req); 306 case NVME_LOG_FW_SLOT: 307 /* 308 * We only support a single firmware slot which always is 309 * active, so we can zero out the whole firmware slot log and 310 * still claim to fully implement this mandatory log page. 311 */ 312 return nvmet_execute_get_log_page_noop(req); 313 case NVME_LOG_CHANGED_NS: 314 return nvmet_execute_get_log_changed_ns(req); 315 case NVME_LOG_CMD_EFFECTS: 316 return nvmet_execute_get_log_cmd_effects_ns(req); 317 case NVME_LOG_ANA: 318 return nvmet_execute_get_log_page_ana(req); 319 } 320 pr_err("unhandled lid %d on qid %d\n", 321 req->cmd->get_log_page.lid, req->sq->qid); 322 req->error_loc = offsetof(struct nvme_get_log_page_command, lid); 323 nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR); 324 } 325 326 static void nvmet_id_set_model_number(struct nvme_id_ctrl *id, 327 struct nvmet_subsys *subsys) 328 { 329 const char *model = NVMET_DEFAULT_CTRL_MODEL; 330 struct nvmet_subsys_model *subsys_model; 331 332 rcu_read_lock(); 333 subsys_model = rcu_dereference(subsys->model); 334 if (subsys_model) 335 model = subsys_model->number; 336 memcpy_and_pad(id->mn, sizeof(id->mn), model, strlen(model), ' '); 337 rcu_read_unlock(); 338 } 339 340 static void nvmet_execute_identify_ctrl(struct nvmet_req *req) 341 { 342 struct nvmet_ctrl *ctrl = req->sq->ctrl; 343 struct nvme_id_ctrl *id; 344 u16 status = 0; 345 346 id = kzalloc(sizeof(*id), GFP_KERNEL); 347 if (!id) { 348 status = NVME_SC_INTERNAL; 349 goto out; 350 } 351 352 /* XXX: figure out how to assign real vendors IDs. */ 353 id->vid = 0; 354 id->ssvid = 0; 355 356 memset(id->sn, ' ', sizeof(id->sn)); 357 bin2hex(id->sn, &ctrl->subsys->serial, 358 min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2)); 359 nvmet_id_set_model_number(id, ctrl->subsys); 360 memcpy_and_pad(id->fr, sizeof(id->fr), 361 UTS_RELEASE, strlen(UTS_RELEASE), ' '); 362 363 id->rab = 6; 364 365 /* 366 * XXX: figure out how we can assign a IEEE OUI, but until then 367 * the safest is to leave it as zeroes. 368 */ 369 370 /* we support multiple ports, multiples hosts and ANA: */ 371 id->cmic = (1 << 0) | (1 << 1) | (1 << 3); 372 373 /* Limit MDTS according to transport capability */ 374 if (ctrl->ops->get_mdts) 375 id->mdts = ctrl->ops->get_mdts(ctrl); 376 else 377 id->mdts = 0; 378 379 id->cntlid = cpu_to_le16(ctrl->cntlid); 380 id->ver = cpu_to_le32(ctrl->subsys->ver); 381 382 /* XXX: figure out what to do about RTD3R/RTD3 */ 383 id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL); 384 id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT | 385 NVME_CTRL_ATTR_TBKAS); 386 387 id->oacs = 0; 388 389 /* 390 * We don't really have a practical limit on the number of abort 391 * comands. But we don't do anything useful for abort either, so 392 * no point in allowing more abort commands than the spec requires. 393 */ 394 id->acl = 3; 395 396 id->aerl = NVMET_ASYNC_EVENTS - 1; 397 398 /* first slot is read-only, only one slot supported */ 399 id->frmw = (1 << 0) | (1 << 1); 400 id->lpa = (1 << 0) | (1 << 1) | (1 << 2); 401 id->elpe = NVMET_ERROR_LOG_SLOTS - 1; 402 id->npss = 0; 403 404 /* We support keep-alive timeout in granularity of seconds */ 405 id->kas = cpu_to_le16(NVMET_KAS); 406 407 id->sqes = (0x6 << 4) | 0x6; 408 id->cqes = (0x4 << 4) | 0x4; 409 410 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */ 411 id->maxcmd = cpu_to_le16(NVMET_MAX_CMD); 412 413 id->nn = cpu_to_le32(ctrl->subsys->max_nsid); 414 id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES); 415 id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM | 416 NVME_CTRL_ONCS_WRITE_ZEROES); 417 418 /* XXX: don't report vwc if the underlying device is write through */ 419 id->vwc = NVME_CTRL_VWC_PRESENT; 420 421 /* 422 * We can't support atomic writes bigger than a LBA without support 423 * from the backend device. 424 */ 425 id->awun = 0; 426 id->awupf = 0; 427 428 id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */ 429 if (ctrl->ops->has_keyed_sgls) 430 id->sgls |= cpu_to_le32(1 << 2); 431 if (req->port->inline_data_size) 432 id->sgls |= cpu_to_le32(1 << 20); 433 434 strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn)); 435 436 /* Max command capsule size is sqe + single page of in-capsule data */ 437 id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) + 438 req->port->inline_data_size) / 16); 439 /* Max response capsule size is cqe */ 440 id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16); 441 442 id->msdbd = ctrl->ops->msdbd; 443 444 id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4); 445 id->anatt = 10; /* random value */ 446 id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS); 447 id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS); 448 449 /* 450 * Meh, we don't really support any power state. Fake up the same 451 * values that qemu does. 452 */ 453 id->psd[0].max_power = cpu_to_le16(0x9c4); 454 id->psd[0].entry_lat = cpu_to_le32(0x10); 455 id->psd[0].exit_lat = cpu_to_le32(0x4); 456 457 id->nwpc = 1 << 0; /* write protect and no write protect */ 458 459 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 460 461 kfree(id); 462 out: 463 nvmet_req_complete(req, status); 464 } 465 466 static void nvmet_execute_identify_ns(struct nvmet_req *req) 467 { 468 struct nvmet_ns *ns; 469 struct nvme_id_ns *id; 470 u16 status = 0; 471 472 if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) { 473 req->error_loc = offsetof(struct nvme_identify, nsid); 474 status = NVME_SC_INVALID_NS | NVME_SC_DNR; 475 goto out; 476 } 477 478 id = kzalloc(sizeof(*id), GFP_KERNEL); 479 if (!id) { 480 status = NVME_SC_INTERNAL; 481 goto out; 482 } 483 484 /* return an all zeroed buffer if we can't find an active namespace */ 485 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid); 486 if (!ns) 487 goto done; 488 489 /* 490 * nuse = ncap = nsze isn't always true, but we have no way to find 491 * that out from the underlying device. 492 */ 493 id->ncap = id->nsze = cpu_to_le64(ns->size >> ns->blksize_shift); 494 switch (req->port->ana_state[ns->anagrpid]) { 495 case NVME_ANA_INACCESSIBLE: 496 case NVME_ANA_PERSISTENT_LOSS: 497 break; 498 default: 499 id->nuse = id->nsze; 500 break; 501 } 502 503 if (ns->bdev) 504 nvmet_bdev_set_limits(ns->bdev, id); 505 506 /* 507 * We just provide a single LBA format that matches what the 508 * underlying device reports. 509 */ 510 id->nlbaf = 0; 511 id->flbas = 0; 512 513 /* 514 * Our namespace might always be shared. Not just with other 515 * controllers, but also with any other user of the block device. 516 */ 517 id->nmic = (1 << 0); 518 id->anagrpid = cpu_to_le32(ns->anagrpid); 519 520 memcpy(&id->nguid, &ns->nguid, sizeof(id->nguid)); 521 522 id->lbaf[0].ds = ns->blksize_shift; 523 524 if (ns->readonly) 525 id->nsattr |= (1 << 0); 526 nvmet_put_namespace(ns); 527 done: 528 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 529 kfree(id); 530 out: 531 nvmet_req_complete(req, status); 532 } 533 534 static void nvmet_execute_identify_nslist(struct nvmet_req *req) 535 { 536 static const int buf_size = NVME_IDENTIFY_DATA_SIZE; 537 struct nvmet_ctrl *ctrl = req->sq->ctrl; 538 struct nvmet_ns *ns; 539 u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid); 540 __le32 *list; 541 u16 status = 0; 542 int i = 0; 543 544 list = kzalloc(buf_size, GFP_KERNEL); 545 if (!list) { 546 status = NVME_SC_INTERNAL; 547 goto out; 548 } 549 550 rcu_read_lock(); 551 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) { 552 if (ns->nsid <= min_nsid) 553 continue; 554 list[i++] = cpu_to_le32(ns->nsid); 555 if (i == buf_size / sizeof(__le32)) 556 break; 557 } 558 rcu_read_unlock(); 559 560 status = nvmet_copy_to_sgl(req, 0, list, buf_size); 561 562 kfree(list); 563 out: 564 nvmet_req_complete(req, status); 565 } 566 567 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len, 568 void *id, off_t *off) 569 { 570 struct nvme_ns_id_desc desc = { 571 .nidt = type, 572 .nidl = len, 573 }; 574 u16 status; 575 576 status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc)); 577 if (status) 578 return status; 579 *off += sizeof(desc); 580 581 status = nvmet_copy_to_sgl(req, *off, id, len); 582 if (status) 583 return status; 584 *off += len; 585 586 return 0; 587 } 588 589 static void nvmet_execute_identify_desclist(struct nvmet_req *req) 590 { 591 struct nvmet_ns *ns; 592 u16 status = 0; 593 off_t off = 0; 594 595 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid); 596 if (!ns) { 597 req->error_loc = offsetof(struct nvme_identify, nsid); 598 status = NVME_SC_INVALID_NS | NVME_SC_DNR; 599 goto out; 600 } 601 602 if (memchr_inv(&ns->uuid, 0, sizeof(ns->uuid))) { 603 status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID, 604 NVME_NIDT_UUID_LEN, 605 &ns->uuid, &off); 606 if (status) 607 goto out_put_ns; 608 } 609 if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) { 610 status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID, 611 NVME_NIDT_NGUID_LEN, 612 &ns->nguid, &off); 613 if (status) 614 goto out_put_ns; 615 } 616 617 if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off, 618 off) != NVME_IDENTIFY_DATA_SIZE - off) 619 status = NVME_SC_INTERNAL | NVME_SC_DNR; 620 out_put_ns: 621 nvmet_put_namespace(ns); 622 out: 623 nvmet_req_complete(req, status); 624 } 625 626 static void nvmet_execute_identify(struct nvmet_req *req) 627 { 628 if (!nvmet_check_data_len(req, NVME_IDENTIFY_DATA_SIZE)) 629 return; 630 631 switch (req->cmd->identify.cns) { 632 case NVME_ID_CNS_NS: 633 return nvmet_execute_identify_ns(req); 634 case NVME_ID_CNS_CTRL: 635 return nvmet_execute_identify_ctrl(req); 636 case NVME_ID_CNS_NS_ACTIVE_LIST: 637 return nvmet_execute_identify_nslist(req); 638 case NVME_ID_CNS_NS_DESC_LIST: 639 return nvmet_execute_identify_desclist(req); 640 } 641 642 pr_err("unhandled identify cns %d on qid %d\n", 643 req->cmd->identify.cns, req->sq->qid); 644 req->error_loc = offsetof(struct nvme_identify, cns); 645 nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR); 646 } 647 648 /* 649 * A "minimum viable" abort implementation: the command is mandatory in the 650 * spec, but we are not required to do any useful work. We couldn't really 651 * do a useful abort, so don't bother even with waiting for the command 652 * to be exectuted and return immediately telling the command to abort 653 * wasn't found. 654 */ 655 static void nvmet_execute_abort(struct nvmet_req *req) 656 { 657 if (!nvmet_check_data_len(req, 0)) 658 return; 659 nvmet_set_result(req, 1); 660 nvmet_req_complete(req, 0); 661 } 662 663 static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req) 664 { 665 u16 status; 666 667 if (req->ns->file) 668 status = nvmet_file_flush(req); 669 else 670 status = nvmet_bdev_flush(req); 671 672 if (status) 673 pr_err("write protect flush failed nsid: %u\n", req->ns->nsid); 674 return status; 675 } 676 677 static u16 nvmet_set_feat_write_protect(struct nvmet_req *req) 678 { 679 u32 write_protect = le32_to_cpu(req->cmd->common.cdw11); 680 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 681 u16 status = NVME_SC_FEATURE_NOT_CHANGEABLE; 682 683 req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->rw.nsid); 684 if (unlikely(!req->ns)) { 685 req->error_loc = offsetof(struct nvme_common_command, nsid); 686 return status; 687 } 688 689 mutex_lock(&subsys->lock); 690 switch (write_protect) { 691 case NVME_NS_WRITE_PROTECT: 692 req->ns->readonly = true; 693 status = nvmet_write_protect_flush_sync(req); 694 if (status) 695 req->ns->readonly = false; 696 break; 697 case NVME_NS_NO_WRITE_PROTECT: 698 req->ns->readonly = false; 699 status = 0; 700 break; 701 default: 702 break; 703 } 704 705 if (!status) 706 nvmet_ns_changed(subsys, req->ns->nsid); 707 mutex_unlock(&subsys->lock); 708 return status; 709 } 710 711 u16 nvmet_set_feat_kato(struct nvmet_req *req) 712 { 713 u32 val32 = le32_to_cpu(req->cmd->common.cdw11); 714 715 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000); 716 717 nvmet_set_result(req, req->sq->ctrl->kato); 718 719 return 0; 720 } 721 722 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask) 723 { 724 u32 val32 = le32_to_cpu(req->cmd->common.cdw11); 725 726 if (val32 & ~mask) { 727 req->error_loc = offsetof(struct nvme_common_command, cdw11); 728 return NVME_SC_INVALID_FIELD | NVME_SC_DNR; 729 } 730 731 WRITE_ONCE(req->sq->ctrl->aen_enabled, val32); 732 nvmet_set_result(req, val32); 733 734 return 0; 735 } 736 737 static void nvmet_execute_set_features(struct nvmet_req *req) 738 { 739 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 740 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10); 741 u32 cdw11 = le32_to_cpu(req->cmd->common.cdw11); 742 u16 status = 0; 743 u16 nsqr; 744 u16 ncqr; 745 746 if (!nvmet_check_data_len(req, 0)) 747 return; 748 749 switch (cdw10 & 0xff) { 750 case NVME_FEAT_NUM_QUEUES: 751 ncqr = (cdw11 >> 16) & 0xffff; 752 nsqr = cdw11 & 0xffff; 753 if (ncqr == 0xffff || nsqr == 0xffff) { 754 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 755 break; 756 } 757 nvmet_set_result(req, 758 (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16)); 759 break; 760 case NVME_FEAT_KATO: 761 status = nvmet_set_feat_kato(req); 762 break; 763 case NVME_FEAT_ASYNC_EVENT: 764 status = nvmet_set_feat_async_event(req, NVMET_AEN_CFG_ALL); 765 break; 766 case NVME_FEAT_HOST_ID: 767 status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR; 768 break; 769 case NVME_FEAT_WRITE_PROTECT: 770 status = nvmet_set_feat_write_protect(req); 771 break; 772 default: 773 req->error_loc = offsetof(struct nvme_common_command, cdw10); 774 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 775 break; 776 } 777 778 nvmet_req_complete(req, status); 779 } 780 781 static u16 nvmet_get_feat_write_protect(struct nvmet_req *req) 782 { 783 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 784 u32 result; 785 786 req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->common.nsid); 787 if (!req->ns) { 788 req->error_loc = offsetof(struct nvme_common_command, nsid); 789 return NVME_SC_INVALID_NS | NVME_SC_DNR; 790 } 791 mutex_lock(&subsys->lock); 792 if (req->ns->readonly == true) 793 result = NVME_NS_WRITE_PROTECT; 794 else 795 result = NVME_NS_NO_WRITE_PROTECT; 796 nvmet_set_result(req, result); 797 mutex_unlock(&subsys->lock); 798 799 return 0; 800 } 801 802 void nvmet_get_feat_kato(struct nvmet_req *req) 803 { 804 nvmet_set_result(req, req->sq->ctrl->kato * 1000); 805 } 806 807 void nvmet_get_feat_async_event(struct nvmet_req *req) 808 { 809 nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled)); 810 } 811 812 static void nvmet_execute_get_features(struct nvmet_req *req) 813 { 814 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 815 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10); 816 u16 status = 0; 817 818 if (!nvmet_check_data_len(req, nvmet_feat_data_len(req, cdw10))) 819 return; 820 821 switch (cdw10 & 0xff) { 822 /* 823 * These features are mandatory in the spec, but we don't 824 * have a useful way to implement them. We'll eventually 825 * need to come up with some fake values for these. 826 */ 827 #if 0 828 case NVME_FEAT_ARBITRATION: 829 break; 830 case NVME_FEAT_POWER_MGMT: 831 break; 832 case NVME_FEAT_TEMP_THRESH: 833 break; 834 case NVME_FEAT_ERR_RECOVERY: 835 break; 836 case NVME_FEAT_IRQ_COALESCE: 837 break; 838 case NVME_FEAT_IRQ_CONFIG: 839 break; 840 case NVME_FEAT_WRITE_ATOMIC: 841 break; 842 #endif 843 case NVME_FEAT_ASYNC_EVENT: 844 nvmet_get_feat_async_event(req); 845 break; 846 case NVME_FEAT_VOLATILE_WC: 847 nvmet_set_result(req, 1); 848 break; 849 case NVME_FEAT_NUM_QUEUES: 850 nvmet_set_result(req, 851 (subsys->max_qid-1) | ((subsys->max_qid-1) << 16)); 852 break; 853 case NVME_FEAT_KATO: 854 nvmet_get_feat_kato(req); 855 break; 856 case NVME_FEAT_HOST_ID: 857 /* need 128-bit host identifier flag */ 858 if (!(req->cmd->common.cdw11 & cpu_to_le32(1 << 0))) { 859 req->error_loc = 860 offsetof(struct nvme_common_command, cdw11); 861 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 862 break; 863 } 864 865 status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid, 866 sizeof(req->sq->ctrl->hostid)); 867 break; 868 case NVME_FEAT_WRITE_PROTECT: 869 status = nvmet_get_feat_write_protect(req); 870 break; 871 default: 872 req->error_loc = 873 offsetof(struct nvme_common_command, cdw10); 874 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 875 break; 876 } 877 878 nvmet_req_complete(req, status); 879 } 880 881 void nvmet_execute_async_event(struct nvmet_req *req) 882 { 883 struct nvmet_ctrl *ctrl = req->sq->ctrl; 884 885 if (!nvmet_check_data_len(req, 0)) 886 return; 887 888 mutex_lock(&ctrl->lock); 889 if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) { 890 mutex_unlock(&ctrl->lock); 891 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR); 892 return; 893 } 894 ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req; 895 mutex_unlock(&ctrl->lock); 896 897 schedule_work(&ctrl->async_event_work); 898 } 899 900 void nvmet_execute_keep_alive(struct nvmet_req *req) 901 { 902 struct nvmet_ctrl *ctrl = req->sq->ctrl; 903 904 if (!nvmet_check_data_len(req, 0)) 905 return; 906 907 pr_debug("ctrl %d update keep-alive timer for %d secs\n", 908 ctrl->cntlid, ctrl->kato); 909 910 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ); 911 nvmet_req_complete(req, 0); 912 } 913 914 u16 nvmet_parse_admin_cmd(struct nvmet_req *req) 915 { 916 struct nvme_command *cmd = req->cmd; 917 u16 ret; 918 919 if (nvme_is_fabrics(cmd)) 920 return nvmet_parse_fabrics_cmd(req); 921 if (req->sq->ctrl->subsys->type == NVME_NQN_DISC) 922 return nvmet_parse_discovery_cmd(req); 923 924 ret = nvmet_check_ctrl_status(req, cmd); 925 if (unlikely(ret)) 926 return ret; 927 928 switch (cmd->common.opcode) { 929 case nvme_admin_get_log_page: 930 req->execute = nvmet_execute_get_log_page; 931 return 0; 932 case nvme_admin_identify: 933 req->execute = nvmet_execute_identify; 934 return 0; 935 case nvme_admin_abort_cmd: 936 req->execute = nvmet_execute_abort; 937 return 0; 938 case nvme_admin_set_features: 939 req->execute = nvmet_execute_set_features; 940 return 0; 941 case nvme_admin_get_features: 942 req->execute = nvmet_execute_get_features; 943 return 0; 944 case nvme_admin_async_event: 945 req->execute = nvmet_execute_async_event; 946 return 0; 947 case nvme_admin_keep_alive: 948 req->execute = nvmet_execute_keep_alive; 949 return 0; 950 } 951 952 pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode, 953 req->sq->qid); 954 req->error_loc = offsetof(struct nvme_common_command, opcode); 955 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR; 956 } 957