1 /* 2 * NVMe admin command implementation. 3 * Copyright (c) 2015-2016 HGST, a Western Digital Company. 4 * 5 * This program is free software; you can redistribute it and/or modify it 6 * under the terms and conditions of the GNU General Public License, 7 * version 2, as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 12 * more details. 13 */ 14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 15 #include <linux/module.h> 16 #include <linux/rculist.h> 17 18 #include <generated/utsrelease.h> 19 #include <asm/unaligned.h> 20 #include "nvmet.h" 21 22 u32 nvmet_get_log_page_len(struct nvme_command *cmd) 23 { 24 u32 len = le16_to_cpu(cmd->get_log_page.numdu); 25 26 len <<= 16; 27 len += le16_to_cpu(cmd->get_log_page.numdl); 28 /* NUMD is a 0's based value */ 29 len += 1; 30 len *= sizeof(u32); 31 32 return len; 33 } 34 35 static u16 nvmet_get_smart_log_nsid(struct nvmet_req *req, 36 struct nvme_smart_log *slog) 37 { 38 u16 status; 39 struct nvmet_ns *ns; 40 u64 host_reads, host_writes, data_units_read, data_units_written; 41 42 status = NVME_SC_SUCCESS; 43 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->get_log_page.nsid); 44 if (!ns) { 45 status = NVME_SC_INVALID_NS; 46 pr_err("nvmet : Could not find namespace id : %d\n", 47 le32_to_cpu(req->cmd->get_log_page.nsid)); 48 goto out; 49 } 50 51 host_reads = part_stat_read(ns->bdev->bd_part, ios[READ]); 52 data_units_read = part_stat_read(ns->bdev->bd_part, sectors[READ]); 53 host_writes = part_stat_read(ns->bdev->bd_part, ios[WRITE]); 54 data_units_written = part_stat_read(ns->bdev->bd_part, sectors[WRITE]); 55 56 put_unaligned_le64(host_reads, &slog->host_reads[0]); 57 put_unaligned_le64(data_units_read, &slog->data_units_read[0]); 58 put_unaligned_le64(host_writes, &slog->host_writes[0]); 59 put_unaligned_le64(data_units_written, &slog->data_units_written[0]); 60 nvmet_put_namespace(ns); 61 out: 62 return status; 63 } 64 65 static u16 nvmet_get_smart_log_all(struct nvmet_req *req, 66 struct nvme_smart_log *slog) 67 { 68 u16 status; 69 u64 host_reads = 0, host_writes = 0; 70 u64 data_units_read = 0, data_units_written = 0; 71 struct nvmet_ns *ns; 72 struct nvmet_ctrl *ctrl; 73 74 status = NVME_SC_SUCCESS; 75 ctrl = req->sq->ctrl; 76 77 rcu_read_lock(); 78 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) { 79 host_reads += part_stat_read(ns->bdev->bd_part, ios[READ]); 80 data_units_read += 81 part_stat_read(ns->bdev->bd_part, sectors[READ]); 82 host_writes += part_stat_read(ns->bdev->bd_part, ios[WRITE]); 83 data_units_written += 84 part_stat_read(ns->bdev->bd_part, sectors[WRITE]); 85 86 } 87 rcu_read_unlock(); 88 89 put_unaligned_le64(host_reads, &slog->host_reads[0]); 90 put_unaligned_le64(data_units_read, &slog->data_units_read[0]); 91 put_unaligned_le64(host_writes, &slog->host_writes[0]); 92 put_unaligned_le64(data_units_written, &slog->data_units_written[0]); 93 94 return status; 95 } 96 97 static u16 nvmet_get_smart_log(struct nvmet_req *req, 98 struct nvme_smart_log *slog) 99 { 100 u16 status; 101 102 WARN_ON(req == NULL || slog == NULL); 103 if (req->cmd->get_log_page.nsid == cpu_to_le32(NVME_NSID_ALL)) 104 status = nvmet_get_smart_log_all(req, slog); 105 else 106 status = nvmet_get_smart_log_nsid(req, slog); 107 return status; 108 } 109 110 static void nvmet_execute_get_log_page(struct nvmet_req *req) 111 { 112 struct nvme_smart_log *smart_log; 113 size_t data_len = nvmet_get_log_page_len(req->cmd); 114 void *buf; 115 u16 status = 0; 116 117 buf = kzalloc(data_len, GFP_KERNEL); 118 if (!buf) { 119 status = NVME_SC_INTERNAL; 120 goto out; 121 } 122 123 switch (req->cmd->get_log_page.lid) { 124 case NVME_LOG_ERROR: 125 /* 126 * We currently never set the More bit in the status field, 127 * so all error log entries are invalid and can be zeroed out. 128 * This is called a minum viable implementation (TM) of this 129 * mandatory log page. 130 */ 131 break; 132 case NVME_LOG_SMART: 133 /* 134 * XXX: fill out actual smart log 135 * 136 * We might have a hard time coming up with useful values for 137 * many of the fields, and even when we have useful data 138 * available (e.g. units or commands read/written) those aren't 139 * persistent over power loss. 140 */ 141 if (data_len != sizeof(*smart_log)) { 142 status = NVME_SC_INTERNAL; 143 goto err; 144 } 145 smart_log = buf; 146 status = nvmet_get_smart_log(req, smart_log); 147 if (status) { 148 memset(buf, '\0', data_len); 149 goto err; 150 } 151 break; 152 case NVME_LOG_FW_SLOT: 153 /* 154 * We only support a single firmware slot which always is 155 * active, so we can zero out the whole firmware slot log and 156 * still claim to fully implement this mandatory log page. 157 */ 158 break; 159 default: 160 BUG(); 161 } 162 163 status = nvmet_copy_to_sgl(req, 0, buf, data_len); 164 165 err: 166 kfree(buf); 167 out: 168 nvmet_req_complete(req, status); 169 } 170 171 static void nvmet_execute_identify_ctrl(struct nvmet_req *req) 172 { 173 struct nvmet_ctrl *ctrl = req->sq->ctrl; 174 struct nvme_id_ctrl *id; 175 u16 status = 0; 176 const char model[] = "Linux"; 177 178 id = kzalloc(sizeof(*id), GFP_KERNEL); 179 if (!id) { 180 status = NVME_SC_INTERNAL; 181 goto out; 182 } 183 184 /* XXX: figure out how to assign real vendors IDs. */ 185 id->vid = 0; 186 id->ssvid = 0; 187 188 bin2hex(id->sn, &ctrl->subsys->serial, 189 min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2)); 190 memcpy_and_pad(id->mn, sizeof(id->mn), model, sizeof(model) - 1, ' '); 191 memcpy_and_pad(id->fr, sizeof(id->fr), 192 UTS_RELEASE, strlen(UTS_RELEASE), ' '); 193 194 id->rab = 6; 195 196 /* 197 * XXX: figure out how we can assign a IEEE OUI, but until then 198 * the safest is to leave it as zeroes. 199 */ 200 201 /* we support multiple ports and multiples hosts: */ 202 id->cmic = (1 << 0) | (1 << 1); 203 204 /* no limit on data transfer sizes for now */ 205 id->mdts = 0; 206 id->cntlid = cpu_to_le16(ctrl->cntlid); 207 id->ver = cpu_to_le32(ctrl->subsys->ver); 208 209 /* XXX: figure out what to do about RTD3R/RTD3 */ 210 id->oaes = cpu_to_le32(1 << 8); 211 id->ctratt = cpu_to_le32(1 << 0); 212 213 id->oacs = 0; 214 215 /* 216 * We don't really have a practical limit on the number of abort 217 * comands. But we don't do anything useful for abort either, so 218 * no point in allowing more abort commands than the spec requires. 219 */ 220 id->acl = 3; 221 222 id->aerl = NVMET_ASYNC_EVENTS - 1; 223 224 /* first slot is read-only, only one slot supported */ 225 id->frmw = (1 << 0) | (1 << 1); 226 id->lpa = (1 << 0) | (1 << 2); 227 id->elpe = NVMET_ERROR_LOG_SLOTS - 1; 228 id->npss = 0; 229 230 /* We support keep-alive timeout in granularity of seconds */ 231 id->kas = cpu_to_le16(NVMET_KAS); 232 233 id->sqes = (0x6 << 4) | 0x6; 234 id->cqes = (0x4 << 4) | 0x4; 235 236 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */ 237 id->maxcmd = cpu_to_le16(NVMET_MAX_CMD); 238 239 id->nn = cpu_to_le32(ctrl->subsys->max_nsid); 240 id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM | 241 NVME_CTRL_ONCS_WRITE_ZEROES); 242 243 /* XXX: don't report vwc if the underlying device is write through */ 244 id->vwc = NVME_CTRL_VWC_PRESENT; 245 246 /* 247 * We can't support atomic writes bigger than a LBA without support 248 * from the backend device. 249 */ 250 id->awun = 0; 251 id->awupf = 0; 252 253 id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */ 254 if (ctrl->ops->has_keyed_sgls) 255 id->sgls |= cpu_to_le32(1 << 2); 256 if (ctrl->ops->sqe_inline_size) 257 id->sgls |= cpu_to_le32(1 << 20); 258 259 strcpy(id->subnqn, ctrl->subsys->subsysnqn); 260 261 /* Max command capsule size is sqe + single page of in-capsule data */ 262 id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) + 263 ctrl->ops->sqe_inline_size) / 16); 264 /* Max response capsule size is cqe */ 265 id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16); 266 267 id->msdbd = ctrl->ops->msdbd; 268 269 /* 270 * Meh, we don't really support any power state. Fake up the same 271 * values that qemu does. 272 */ 273 id->psd[0].max_power = cpu_to_le16(0x9c4); 274 id->psd[0].entry_lat = cpu_to_le32(0x10); 275 id->psd[0].exit_lat = cpu_to_le32(0x4); 276 277 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 278 279 kfree(id); 280 out: 281 nvmet_req_complete(req, status); 282 } 283 284 static void nvmet_execute_identify_ns(struct nvmet_req *req) 285 { 286 struct nvmet_ns *ns; 287 struct nvme_id_ns *id; 288 u16 status = 0; 289 290 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid); 291 if (!ns) { 292 status = NVME_SC_INVALID_NS | NVME_SC_DNR; 293 goto out; 294 } 295 296 id = kzalloc(sizeof(*id), GFP_KERNEL); 297 if (!id) { 298 status = NVME_SC_INTERNAL; 299 goto out_put_ns; 300 } 301 302 /* 303 * nuse = ncap = nsze isn't aways true, but we have no way to find 304 * that out from the underlying device. 305 */ 306 id->ncap = id->nuse = id->nsze = 307 cpu_to_le64(ns->size >> ns->blksize_shift); 308 309 /* 310 * We just provide a single LBA format that matches what the 311 * underlying device reports. 312 */ 313 id->nlbaf = 0; 314 id->flbas = 0; 315 316 /* 317 * Our namespace might always be shared. Not just with other 318 * controllers, but also with any other user of the block device. 319 */ 320 id->nmic = (1 << 0); 321 322 memcpy(&id->nguid, &ns->nguid, sizeof(uuid_le)); 323 324 id->lbaf[0].ds = ns->blksize_shift; 325 326 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 327 328 kfree(id); 329 out_put_ns: 330 nvmet_put_namespace(ns); 331 out: 332 nvmet_req_complete(req, status); 333 } 334 335 static void nvmet_execute_identify_nslist(struct nvmet_req *req) 336 { 337 static const int buf_size = NVME_IDENTIFY_DATA_SIZE; 338 struct nvmet_ctrl *ctrl = req->sq->ctrl; 339 struct nvmet_ns *ns; 340 u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid); 341 __le32 *list; 342 u16 status = 0; 343 int i = 0; 344 345 list = kzalloc(buf_size, GFP_KERNEL); 346 if (!list) { 347 status = NVME_SC_INTERNAL; 348 goto out; 349 } 350 351 rcu_read_lock(); 352 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) { 353 if (ns->nsid <= min_nsid) 354 continue; 355 list[i++] = cpu_to_le32(ns->nsid); 356 if (i == buf_size / sizeof(__le32)) 357 break; 358 } 359 rcu_read_unlock(); 360 361 status = nvmet_copy_to_sgl(req, 0, list, buf_size); 362 363 kfree(list); 364 out: 365 nvmet_req_complete(req, status); 366 } 367 368 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len, 369 void *id, off_t *off) 370 { 371 struct nvme_ns_id_desc desc = { 372 .nidt = type, 373 .nidl = len, 374 }; 375 u16 status; 376 377 status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc)); 378 if (status) 379 return status; 380 *off += sizeof(desc); 381 382 status = nvmet_copy_to_sgl(req, *off, id, len); 383 if (status) 384 return status; 385 *off += len; 386 387 return 0; 388 } 389 390 static void nvmet_execute_identify_desclist(struct nvmet_req *req) 391 { 392 struct nvmet_ns *ns; 393 u16 status = 0; 394 off_t off = 0; 395 396 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid); 397 if (!ns) { 398 status = NVME_SC_INVALID_NS | NVME_SC_DNR; 399 goto out; 400 } 401 402 if (memchr_inv(&ns->uuid, 0, sizeof(ns->uuid))) { 403 status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID, 404 NVME_NIDT_UUID_LEN, 405 &ns->uuid, &off); 406 if (status) 407 goto out_put_ns; 408 } 409 if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) { 410 status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID, 411 NVME_NIDT_NGUID_LEN, 412 &ns->nguid, &off); 413 if (status) 414 goto out_put_ns; 415 } 416 417 if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off, 418 off) != NVME_IDENTIFY_DATA_SIZE - off) 419 status = NVME_SC_INTERNAL | NVME_SC_DNR; 420 out_put_ns: 421 nvmet_put_namespace(ns); 422 out: 423 nvmet_req_complete(req, status); 424 } 425 426 /* 427 * A "mimimum viable" abort implementation: the command is mandatory in the 428 * spec, but we are not required to do any useful work. We couldn't really 429 * do a useful abort, so don't bother even with waiting for the command 430 * to be exectuted and return immediately telling the command to abort 431 * wasn't found. 432 */ 433 static void nvmet_execute_abort(struct nvmet_req *req) 434 { 435 nvmet_set_result(req, 1); 436 nvmet_req_complete(req, 0); 437 } 438 439 static void nvmet_execute_set_features(struct nvmet_req *req) 440 { 441 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 442 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]); 443 u32 val32; 444 u16 status = 0; 445 446 switch (cdw10 & 0xff) { 447 case NVME_FEAT_NUM_QUEUES: 448 nvmet_set_result(req, 449 (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16)); 450 break; 451 case NVME_FEAT_KATO: 452 val32 = le32_to_cpu(req->cmd->common.cdw10[1]); 453 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000); 454 nvmet_set_result(req, req->sq->ctrl->kato); 455 break; 456 case NVME_FEAT_HOST_ID: 457 status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR; 458 break; 459 default: 460 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 461 break; 462 } 463 464 nvmet_req_complete(req, status); 465 } 466 467 static void nvmet_execute_get_features(struct nvmet_req *req) 468 { 469 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 470 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]); 471 u16 status = 0; 472 473 switch (cdw10 & 0xff) { 474 /* 475 * These features are mandatory in the spec, but we don't 476 * have a useful way to implement them. We'll eventually 477 * need to come up with some fake values for these. 478 */ 479 #if 0 480 case NVME_FEAT_ARBITRATION: 481 break; 482 case NVME_FEAT_POWER_MGMT: 483 break; 484 case NVME_FEAT_TEMP_THRESH: 485 break; 486 case NVME_FEAT_ERR_RECOVERY: 487 break; 488 case NVME_FEAT_IRQ_COALESCE: 489 break; 490 case NVME_FEAT_IRQ_CONFIG: 491 break; 492 case NVME_FEAT_WRITE_ATOMIC: 493 break; 494 case NVME_FEAT_ASYNC_EVENT: 495 break; 496 #endif 497 case NVME_FEAT_VOLATILE_WC: 498 nvmet_set_result(req, 1); 499 break; 500 case NVME_FEAT_NUM_QUEUES: 501 nvmet_set_result(req, 502 (subsys->max_qid-1) | ((subsys->max_qid-1) << 16)); 503 break; 504 case NVME_FEAT_KATO: 505 nvmet_set_result(req, req->sq->ctrl->kato * 1000); 506 break; 507 case NVME_FEAT_HOST_ID: 508 /* need 128-bit host identifier flag */ 509 if (!(req->cmd->common.cdw10[1] & cpu_to_le32(1 << 0))) { 510 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 511 break; 512 } 513 514 status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid, 515 sizeof(req->sq->ctrl->hostid)); 516 break; 517 default: 518 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 519 break; 520 } 521 522 nvmet_req_complete(req, status); 523 } 524 525 static void nvmet_execute_async_event(struct nvmet_req *req) 526 { 527 struct nvmet_ctrl *ctrl = req->sq->ctrl; 528 529 mutex_lock(&ctrl->lock); 530 if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) { 531 mutex_unlock(&ctrl->lock); 532 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR); 533 return; 534 } 535 ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req; 536 mutex_unlock(&ctrl->lock); 537 538 schedule_work(&ctrl->async_event_work); 539 } 540 541 static void nvmet_execute_keep_alive(struct nvmet_req *req) 542 { 543 struct nvmet_ctrl *ctrl = req->sq->ctrl; 544 545 pr_debug("ctrl %d update keep-alive timer for %d secs\n", 546 ctrl->cntlid, ctrl->kato); 547 548 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ); 549 nvmet_req_complete(req, 0); 550 } 551 552 u16 nvmet_parse_admin_cmd(struct nvmet_req *req) 553 { 554 struct nvme_command *cmd = req->cmd; 555 u16 ret; 556 557 req->ns = NULL; 558 559 ret = nvmet_check_ctrl_status(req, cmd); 560 if (unlikely(ret)) 561 return ret; 562 563 switch (cmd->common.opcode) { 564 case nvme_admin_get_log_page: 565 req->data_len = nvmet_get_log_page_len(cmd); 566 567 switch (cmd->get_log_page.lid) { 568 case NVME_LOG_ERROR: 569 case NVME_LOG_SMART: 570 case NVME_LOG_FW_SLOT: 571 req->execute = nvmet_execute_get_log_page; 572 return 0; 573 } 574 break; 575 case nvme_admin_identify: 576 req->data_len = NVME_IDENTIFY_DATA_SIZE; 577 switch (cmd->identify.cns) { 578 case NVME_ID_CNS_NS: 579 req->execute = nvmet_execute_identify_ns; 580 return 0; 581 case NVME_ID_CNS_CTRL: 582 req->execute = nvmet_execute_identify_ctrl; 583 return 0; 584 case NVME_ID_CNS_NS_ACTIVE_LIST: 585 req->execute = nvmet_execute_identify_nslist; 586 return 0; 587 case NVME_ID_CNS_NS_DESC_LIST: 588 req->execute = nvmet_execute_identify_desclist; 589 return 0; 590 } 591 break; 592 case nvme_admin_abort_cmd: 593 req->execute = nvmet_execute_abort; 594 req->data_len = 0; 595 return 0; 596 case nvme_admin_set_features: 597 req->execute = nvmet_execute_set_features; 598 req->data_len = 0; 599 return 0; 600 case nvme_admin_get_features: 601 req->execute = nvmet_execute_get_features; 602 req->data_len = 0; 603 return 0; 604 case nvme_admin_async_event: 605 req->execute = nvmet_execute_async_event; 606 req->data_len = 0; 607 return 0; 608 case nvme_admin_keep_alive: 609 req->execute = nvmet_execute_keep_alive; 610 req->data_len = 0; 611 return 0; 612 } 613 614 pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode, 615 req->sq->qid); 616 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR; 617 } 618