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(0xFFFFFFFF)) 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 copy_and_pad(char *dst, int dst_len, const char *src, int src_len) 172 { 173 int len = min(src_len, dst_len); 174 175 memcpy(dst, src, len); 176 if (dst_len > len) 177 memset(dst + len, ' ', dst_len - len); 178 } 179 180 static void nvmet_execute_identify_ctrl(struct nvmet_req *req) 181 { 182 struct nvmet_ctrl *ctrl = req->sq->ctrl; 183 struct nvme_id_ctrl *id; 184 u16 status = 0; 185 const char model[] = "Linux"; 186 187 id = kzalloc(sizeof(*id), GFP_KERNEL); 188 if (!id) { 189 status = NVME_SC_INTERNAL; 190 goto out; 191 } 192 193 /* XXX: figure out how to assign real vendors IDs. */ 194 id->vid = 0; 195 id->ssvid = 0; 196 197 bin2hex(id->sn, &ctrl->subsys->serial, 198 min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2)); 199 copy_and_pad(id->mn, sizeof(id->mn), model, sizeof(model) - 1); 200 copy_and_pad(id->fr, sizeof(id->fr), UTS_RELEASE, strlen(UTS_RELEASE)); 201 202 id->rab = 6; 203 204 /* 205 * XXX: figure out how we can assign a IEEE OUI, but until then 206 * the safest is to leave it as zeroes. 207 */ 208 209 /* we support multiple ports and multiples hosts: */ 210 id->cmic = (1 << 0) | (1 << 1); 211 212 /* no limit on data transfer sizes for now */ 213 id->mdts = 0; 214 id->cntlid = cpu_to_le16(ctrl->cntlid); 215 id->ver = cpu_to_le32(ctrl->subsys->ver); 216 217 /* XXX: figure out what to do about RTD3R/RTD3 */ 218 id->oaes = cpu_to_le32(1 << 8); 219 id->ctratt = cpu_to_le32(1 << 0); 220 221 id->oacs = 0; 222 223 /* 224 * We don't really have a practical limit on the number of abort 225 * comands. But we don't do anything useful for abort either, so 226 * no point in allowing more abort commands than the spec requires. 227 */ 228 id->acl = 3; 229 230 id->aerl = NVMET_ASYNC_EVENTS - 1; 231 232 /* first slot is read-only, only one slot supported */ 233 id->frmw = (1 << 0) | (1 << 1); 234 id->lpa = (1 << 0) | (1 << 2); 235 id->elpe = NVMET_ERROR_LOG_SLOTS - 1; 236 id->npss = 0; 237 238 /* We support keep-alive timeout in granularity of seconds */ 239 id->kas = cpu_to_le16(NVMET_KAS); 240 241 id->sqes = (0x6 << 4) | 0x6; 242 id->cqes = (0x4 << 4) | 0x4; 243 244 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */ 245 id->maxcmd = cpu_to_le16(NVMET_MAX_CMD); 246 247 id->nn = cpu_to_le32(ctrl->subsys->max_nsid); 248 id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM | 249 NVME_CTRL_ONCS_WRITE_ZEROES); 250 251 /* XXX: don't report vwc if the underlying device is write through */ 252 id->vwc = NVME_CTRL_VWC_PRESENT; 253 254 /* 255 * We can't support atomic writes bigger than a LBA without support 256 * from the backend device. 257 */ 258 id->awun = 0; 259 id->awupf = 0; 260 261 id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */ 262 if (ctrl->ops->has_keyed_sgls) 263 id->sgls |= cpu_to_le32(1 << 2); 264 if (ctrl->ops->sqe_inline_size) 265 id->sgls |= cpu_to_le32(1 << 20); 266 267 strcpy(id->subnqn, ctrl->subsys->subsysnqn); 268 269 /* Max command capsule size is sqe + single page of in-capsule data */ 270 id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) + 271 ctrl->ops->sqe_inline_size) / 16); 272 /* Max response capsule size is cqe */ 273 id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16); 274 275 id->msdbd = ctrl->ops->msdbd; 276 277 /* 278 * Meh, we don't really support any power state. Fake up the same 279 * values that qemu does. 280 */ 281 id->psd[0].max_power = cpu_to_le16(0x9c4); 282 id->psd[0].entry_lat = cpu_to_le32(0x10); 283 id->psd[0].exit_lat = cpu_to_le32(0x4); 284 285 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 286 287 kfree(id); 288 out: 289 nvmet_req_complete(req, status); 290 } 291 292 static void nvmet_execute_identify_ns(struct nvmet_req *req) 293 { 294 struct nvmet_ns *ns; 295 struct nvme_id_ns *id; 296 u16 status = 0; 297 298 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid); 299 if (!ns) { 300 status = NVME_SC_INVALID_NS | NVME_SC_DNR; 301 goto out; 302 } 303 304 id = kzalloc(sizeof(*id), GFP_KERNEL); 305 if (!id) { 306 status = NVME_SC_INTERNAL; 307 goto out_put_ns; 308 } 309 310 /* 311 * nuse = ncap = nsze isn't aways true, but we have no way to find 312 * that out from the underlying device. 313 */ 314 id->ncap = id->nuse = id->nsze = 315 cpu_to_le64(ns->size >> ns->blksize_shift); 316 317 /* 318 * We just provide a single LBA format that matches what the 319 * underlying device reports. 320 */ 321 id->nlbaf = 0; 322 id->flbas = 0; 323 324 /* 325 * Our namespace might always be shared. Not just with other 326 * controllers, but also with any other user of the block device. 327 */ 328 id->nmic = (1 << 0); 329 330 memcpy(&id->nguid, &ns->nguid, sizeof(uuid_le)); 331 332 id->lbaf[0].ds = ns->blksize_shift; 333 334 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 335 336 kfree(id); 337 out_put_ns: 338 nvmet_put_namespace(ns); 339 out: 340 nvmet_req_complete(req, status); 341 } 342 343 static void nvmet_execute_identify_nslist(struct nvmet_req *req) 344 { 345 static const int buf_size = NVME_IDENTIFY_DATA_SIZE; 346 struct nvmet_ctrl *ctrl = req->sq->ctrl; 347 struct nvmet_ns *ns; 348 u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid); 349 __le32 *list; 350 u16 status = 0; 351 int i = 0; 352 353 list = kzalloc(buf_size, GFP_KERNEL); 354 if (!list) { 355 status = NVME_SC_INTERNAL; 356 goto out; 357 } 358 359 rcu_read_lock(); 360 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) { 361 if (ns->nsid <= min_nsid) 362 continue; 363 list[i++] = cpu_to_le32(ns->nsid); 364 if (i == buf_size / sizeof(__le32)) 365 break; 366 } 367 rcu_read_unlock(); 368 369 status = nvmet_copy_to_sgl(req, 0, list, buf_size); 370 371 kfree(list); 372 out: 373 nvmet_req_complete(req, status); 374 } 375 376 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len, 377 void *id, off_t *off) 378 { 379 struct nvme_ns_id_desc desc = { 380 .nidt = type, 381 .nidl = len, 382 }; 383 u16 status; 384 385 status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc)); 386 if (status) 387 return status; 388 *off += sizeof(desc); 389 390 status = nvmet_copy_to_sgl(req, *off, id, len); 391 if (status) 392 return status; 393 *off += len; 394 395 return 0; 396 } 397 398 static void nvmet_execute_identify_desclist(struct nvmet_req *req) 399 { 400 struct nvmet_ns *ns; 401 u16 status = 0; 402 off_t off = 0; 403 404 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid); 405 if (!ns) { 406 status = NVME_SC_INVALID_NS | NVME_SC_DNR; 407 goto out; 408 } 409 410 if (memchr_inv(&ns->uuid, 0, sizeof(ns->uuid))) { 411 status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID, 412 NVME_NIDT_UUID_LEN, 413 &ns->uuid, &off); 414 if (status) 415 goto out_put_ns; 416 } 417 if (memchr_inv(ns->nguid, 0, sizeof(ns->nguid))) { 418 status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID, 419 NVME_NIDT_NGUID_LEN, 420 &ns->nguid, &off); 421 if (status) 422 goto out_put_ns; 423 } 424 425 if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off, 426 off) != NVME_IDENTIFY_DATA_SIZE - off) 427 status = NVME_SC_INTERNAL | NVME_SC_DNR; 428 out_put_ns: 429 nvmet_put_namespace(ns); 430 out: 431 nvmet_req_complete(req, status); 432 } 433 434 /* 435 * A "mimimum viable" abort implementation: the command is mandatory in the 436 * spec, but we are not required to do any useful work. We couldn't really 437 * do a useful abort, so don't bother even with waiting for the command 438 * to be exectuted and return immediately telling the command to abort 439 * wasn't found. 440 */ 441 static void nvmet_execute_abort(struct nvmet_req *req) 442 { 443 nvmet_set_result(req, 1); 444 nvmet_req_complete(req, 0); 445 } 446 447 static void nvmet_execute_set_features(struct nvmet_req *req) 448 { 449 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 450 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]); 451 u32 val32; 452 u16 status = 0; 453 454 switch (cdw10 & 0xf) { 455 case NVME_FEAT_NUM_QUEUES: 456 nvmet_set_result(req, 457 (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16)); 458 break; 459 case NVME_FEAT_KATO: 460 val32 = le32_to_cpu(req->cmd->common.cdw10[1]); 461 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000); 462 nvmet_set_result(req, req->sq->ctrl->kato); 463 break; 464 default: 465 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 466 break; 467 } 468 469 nvmet_req_complete(req, status); 470 } 471 472 static void nvmet_execute_get_features(struct nvmet_req *req) 473 { 474 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 475 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]); 476 u16 status = 0; 477 478 switch (cdw10 & 0xf) { 479 /* 480 * These features are mandatory in the spec, but we don't 481 * have a useful way to implement them. We'll eventually 482 * need to come up with some fake values for these. 483 */ 484 #if 0 485 case NVME_FEAT_ARBITRATION: 486 break; 487 case NVME_FEAT_POWER_MGMT: 488 break; 489 case NVME_FEAT_TEMP_THRESH: 490 break; 491 case NVME_FEAT_ERR_RECOVERY: 492 break; 493 case NVME_FEAT_IRQ_COALESCE: 494 break; 495 case NVME_FEAT_IRQ_CONFIG: 496 break; 497 case NVME_FEAT_WRITE_ATOMIC: 498 break; 499 case NVME_FEAT_ASYNC_EVENT: 500 break; 501 #endif 502 case NVME_FEAT_VOLATILE_WC: 503 nvmet_set_result(req, 1); 504 break; 505 case NVME_FEAT_NUM_QUEUES: 506 nvmet_set_result(req, 507 (subsys->max_qid-1) | ((subsys->max_qid-1) << 16)); 508 break; 509 case NVME_FEAT_KATO: 510 nvmet_set_result(req, req->sq->ctrl->kato * 1000); 511 break; 512 default: 513 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 514 break; 515 } 516 517 nvmet_req_complete(req, status); 518 } 519 520 static void nvmet_execute_async_event(struct nvmet_req *req) 521 { 522 struct nvmet_ctrl *ctrl = req->sq->ctrl; 523 524 mutex_lock(&ctrl->lock); 525 if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) { 526 mutex_unlock(&ctrl->lock); 527 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR); 528 return; 529 } 530 ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req; 531 mutex_unlock(&ctrl->lock); 532 533 schedule_work(&ctrl->async_event_work); 534 } 535 536 static void nvmet_execute_keep_alive(struct nvmet_req *req) 537 { 538 struct nvmet_ctrl *ctrl = req->sq->ctrl; 539 540 pr_debug("ctrl %d update keep-alive timer for %d secs\n", 541 ctrl->cntlid, ctrl->kato); 542 543 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ); 544 nvmet_req_complete(req, 0); 545 } 546 547 u16 nvmet_parse_admin_cmd(struct nvmet_req *req) 548 { 549 struct nvme_command *cmd = req->cmd; 550 u16 ret; 551 552 req->ns = NULL; 553 554 ret = nvmet_check_ctrl_status(req, cmd); 555 if (unlikely(ret)) 556 return ret; 557 558 switch (cmd->common.opcode) { 559 case nvme_admin_get_log_page: 560 req->data_len = nvmet_get_log_page_len(cmd); 561 562 switch (cmd->get_log_page.lid) { 563 case NVME_LOG_ERROR: 564 case NVME_LOG_SMART: 565 case NVME_LOG_FW_SLOT: 566 req->execute = nvmet_execute_get_log_page; 567 return 0; 568 } 569 break; 570 case nvme_admin_identify: 571 req->data_len = NVME_IDENTIFY_DATA_SIZE; 572 switch (cmd->identify.cns) { 573 case NVME_ID_CNS_NS: 574 req->execute = nvmet_execute_identify_ns; 575 return 0; 576 case NVME_ID_CNS_CTRL: 577 req->execute = nvmet_execute_identify_ctrl; 578 return 0; 579 case NVME_ID_CNS_NS_ACTIVE_LIST: 580 req->execute = nvmet_execute_identify_nslist; 581 return 0; 582 case NVME_ID_CNS_NS_DESC_LIST: 583 req->execute = nvmet_execute_identify_desclist; 584 return 0; 585 } 586 break; 587 case nvme_admin_abort_cmd: 588 req->execute = nvmet_execute_abort; 589 req->data_len = 0; 590 return 0; 591 case nvme_admin_set_features: 592 req->execute = nvmet_execute_set_features; 593 req->data_len = 0; 594 return 0; 595 case nvme_admin_get_features: 596 req->execute = nvmet_execute_get_features; 597 req->data_len = 0; 598 return 0; 599 case nvme_admin_async_event: 600 req->execute = nvmet_execute_async_event; 601 req->data_len = 0; 602 return 0; 603 case nvme_admin_keep_alive: 604 req->execute = nvmet_execute_keep_alive; 605 req->data_len = 0; 606 return 0; 607 } 608 609 pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode, 610 req->sq->qid); 611 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR; 612 } 613