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 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; 176 177 id = kzalloc(sizeof(*id), GFP_KERNEL); 178 if (!id) { 179 status = NVME_SC_INTERNAL; 180 goto out; 181 } 182 183 /* XXX: figure out how to assign real vendors IDs. */ 184 id->vid = 0; 185 id->ssvid = 0; 186 187 memset(id->sn, ' ', sizeof(id->sn)); 188 snprintf(id->sn, sizeof(id->sn), "%llx", ctrl->serial); 189 190 memset(id->mn, ' ', sizeof(id->mn)); 191 strncpy((char *)id->mn, "Linux", sizeof(id->mn)); 192 193 memset(id->fr, ' ', sizeof(id->fr)); 194 strncpy((char *)id->fr, UTS_RELEASE, sizeof(id->fr)); 195 196 id->rab = 6; 197 198 /* 199 * XXX: figure out how we can assign a IEEE OUI, but until then 200 * the safest is to leave it as zeroes. 201 */ 202 203 /* we support multiple ports and multiples hosts: */ 204 id->cmic = (1 << 0) | (1 << 1); 205 206 /* no limit on data transfer sizes for now */ 207 id->mdts = 0; 208 id->cntlid = cpu_to_le16(ctrl->cntlid); 209 id->ver = cpu_to_le32(ctrl->subsys->ver); 210 211 /* XXX: figure out what to do about RTD3R/RTD3 */ 212 id->oaes = cpu_to_le32(1 << 8); 213 id->ctratt = cpu_to_le32(1 << 0); 214 215 id->oacs = 0; 216 217 /* 218 * We don't really have a practical limit on the number of abort 219 * comands. But we don't do anything useful for abort either, so 220 * no point in allowing more abort commands than the spec requires. 221 */ 222 id->acl = 3; 223 224 id->aerl = NVMET_ASYNC_EVENTS - 1; 225 226 /* first slot is read-only, only one slot supported */ 227 id->frmw = (1 << 0) | (1 << 1); 228 id->lpa = (1 << 0) | (1 << 2); 229 id->elpe = NVMET_ERROR_LOG_SLOTS - 1; 230 id->npss = 0; 231 232 /* We support keep-alive timeout in granularity of seconds */ 233 id->kas = cpu_to_le16(NVMET_KAS); 234 235 id->sqes = (0x6 << 4) | 0x6; 236 id->cqes = (0x4 << 4) | 0x4; 237 238 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */ 239 id->maxcmd = cpu_to_le16(NVMET_MAX_CMD); 240 241 id->nn = cpu_to_le32(ctrl->subsys->max_nsid); 242 id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM | 243 NVME_CTRL_ONCS_WRITE_ZEROES); 244 245 /* XXX: don't report vwc if the underlying device is write through */ 246 id->vwc = NVME_CTRL_VWC_PRESENT; 247 248 /* 249 * We can't support atomic writes bigger than a LBA without support 250 * from the backend device. 251 */ 252 id->awun = 0; 253 id->awupf = 0; 254 255 id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */ 256 if (ctrl->ops->has_keyed_sgls) 257 id->sgls |= cpu_to_le32(1 << 2); 258 if (ctrl->ops->sqe_inline_size) 259 id->sgls |= cpu_to_le32(1 << 20); 260 261 strcpy(id->subnqn, ctrl->subsys->subsysnqn); 262 263 /* Max command capsule size is sqe + single page of in-capsule data */ 264 id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) + 265 ctrl->ops->sqe_inline_size) / 16); 266 /* Max response capsule size is cqe */ 267 id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16); 268 269 id->msdbd = ctrl->ops->msdbd; 270 271 /* 272 * Meh, we don't really support any power state. Fake up the same 273 * values that qemu does. 274 */ 275 id->psd[0].max_power = cpu_to_le16(0x9c4); 276 id->psd[0].entry_lat = cpu_to_le32(0x10); 277 id->psd[0].exit_lat = cpu_to_le32(0x4); 278 279 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 280 281 kfree(id); 282 out: 283 nvmet_req_complete(req, status); 284 } 285 286 static void nvmet_execute_identify_ns(struct nvmet_req *req) 287 { 288 struct nvmet_ns *ns; 289 struct nvme_id_ns *id; 290 u16 status = 0; 291 292 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid); 293 if (!ns) { 294 status = NVME_SC_INVALID_NS | NVME_SC_DNR; 295 goto out; 296 } 297 298 id = kzalloc(sizeof(*id), GFP_KERNEL); 299 if (!id) { 300 status = NVME_SC_INTERNAL; 301 goto out_put_ns; 302 } 303 304 /* 305 * nuse = ncap = nsze isn't aways true, but we have no way to find 306 * that out from the underlying device. 307 */ 308 id->ncap = id->nuse = id->nsze = 309 cpu_to_le64(ns->size >> ns->blksize_shift); 310 311 /* 312 * We just provide a single LBA format that matches what the 313 * underlying device reports. 314 */ 315 id->nlbaf = 0; 316 id->flbas = 0; 317 318 /* 319 * Our namespace might always be shared. Not just with other 320 * controllers, but also with any other user of the block device. 321 */ 322 id->nmic = (1 << 0); 323 324 memcpy(&id->nguid, &ns->nguid, sizeof(uuid_le)); 325 326 id->lbaf[0].ds = ns->blksize_shift; 327 328 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id)); 329 330 kfree(id); 331 out_put_ns: 332 nvmet_put_namespace(ns); 333 out: 334 nvmet_req_complete(req, status); 335 } 336 337 static void nvmet_execute_identify_nslist(struct nvmet_req *req) 338 { 339 static const int buf_size = 4096; 340 struct nvmet_ctrl *ctrl = req->sq->ctrl; 341 struct nvmet_ns *ns; 342 u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid); 343 __le32 *list; 344 u16 status = 0; 345 int i = 0; 346 347 list = kzalloc(buf_size, GFP_KERNEL); 348 if (!list) { 349 status = NVME_SC_INTERNAL; 350 goto out; 351 } 352 353 rcu_read_lock(); 354 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) { 355 if (ns->nsid <= min_nsid) 356 continue; 357 list[i++] = cpu_to_le32(ns->nsid); 358 if (i == buf_size / sizeof(__le32)) 359 break; 360 } 361 rcu_read_unlock(); 362 363 status = nvmet_copy_to_sgl(req, 0, list, buf_size); 364 365 kfree(list); 366 out: 367 nvmet_req_complete(req, status); 368 } 369 370 /* 371 * A "mimimum viable" abort implementation: the command is mandatory in the 372 * spec, but we are not required to do any useful work. We couldn't really 373 * do a useful abort, so don't bother even with waiting for the command 374 * to be exectuted and return immediately telling the command to abort 375 * wasn't found. 376 */ 377 static void nvmet_execute_abort(struct nvmet_req *req) 378 { 379 nvmet_set_result(req, 1); 380 nvmet_req_complete(req, 0); 381 } 382 383 static void nvmet_execute_set_features(struct nvmet_req *req) 384 { 385 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 386 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]); 387 u32 val32; 388 u16 status = 0; 389 390 switch (cdw10 & 0xf) { 391 case NVME_FEAT_NUM_QUEUES: 392 nvmet_set_result(req, 393 (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16)); 394 break; 395 case NVME_FEAT_KATO: 396 val32 = le32_to_cpu(req->cmd->common.cdw10[1]); 397 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000); 398 nvmet_set_result(req, req->sq->ctrl->kato); 399 break; 400 default: 401 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 402 break; 403 } 404 405 nvmet_req_complete(req, status); 406 } 407 408 static void nvmet_execute_get_features(struct nvmet_req *req) 409 { 410 struct nvmet_subsys *subsys = req->sq->ctrl->subsys; 411 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]); 412 u16 status = 0; 413 414 switch (cdw10 & 0xf) { 415 /* 416 * These features are mandatory in the spec, but we don't 417 * have a useful way to implement them. We'll eventually 418 * need to come up with some fake values for these. 419 */ 420 #if 0 421 case NVME_FEAT_ARBITRATION: 422 break; 423 case NVME_FEAT_POWER_MGMT: 424 break; 425 case NVME_FEAT_TEMP_THRESH: 426 break; 427 case NVME_FEAT_ERR_RECOVERY: 428 break; 429 case NVME_FEAT_IRQ_COALESCE: 430 break; 431 case NVME_FEAT_IRQ_CONFIG: 432 break; 433 case NVME_FEAT_WRITE_ATOMIC: 434 break; 435 case NVME_FEAT_ASYNC_EVENT: 436 break; 437 #endif 438 case NVME_FEAT_VOLATILE_WC: 439 nvmet_set_result(req, 1); 440 break; 441 case NVME_FEAT_NUM_QUEUES: 442 nvmet_set_result(req, 443 (subsys->max_qid-1) | ((subsys->max_qid-1) << 16)); 444 break; 445 case NVME_FEAT_KATO: 446 nvmet_set_result(req, req->sq->ctrl->kato * 1000); 447 break; 448 default: 449 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; 450 break; 451 } 452 453 nvmet_req_complete(req, status); 454 } 455 456 static void nvmet_execute_async_event(struct nvmet_req *req) 457 { 458 struct nvmet_ctrl *ctrl = req->sq->ctrl; 459 460 mutex_lock(&ctrl->lock); 461 if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) { 462 mutex_unlock(&ctrl->lock); 463 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR); 464 return; 465 } 466 ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req; 467 mutex_unlock(&ctrl->lock); 468 469 schedule_work(&ctrl->async_event_work); 470 } 471 472 static void nvmet_execute_keep_alive(struct nvmet_req *req) 473 { 474 struct nvmet_ctrl *ctrl = req->sq->ctrl; 475 476 pr_debug("ctrl %d update keep-alive timer for %d secs\n", 477 ctrl->cntlid, ctrl->kato); 478 479 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ); 480 nvmet_req_complete(req, 0); 481 } 482 483 u16 nvmet_parse_admin_cmd(struct nvmet_req *req) 484 { 485 struct nvme_command *cmd = req->cmd; 486 u16 ret; 487 488 req->ns = NULL; 489 490 ret = nvmet_check_ctrl_status(req, cmd); 491 if (unlikely(ret)) 492 return ret; 493 494 switch (cmd->common.opcode) { 495 case nvme_admin_get_log_page: 496 req->data_len = nvmet_get_log_page_len(cmd); 497 498 switch (cmd->get_log_page.lid) { 499 case NVME_LOG_ERROR: 500 case NVME_LOG_SMART: 501 case NVME_LOG_FW_SLOT: 502 req->execute = nvmet_execute_get_log_page; 503 return 0; 504 } 505 break; 506 case nvme_admin_identify: 507 req->data_len = 4096; 508 switch (cmd->identify.cns) { 509 case NVME_ID_CNS_NS: 510 req->execute = nvmet_execute_identify_ns; 511 return 0; 512 case NVME_ID_CNS_CTRL: 513 req->execute = nvmet_execute_identify_ctrl; 514 return 0; 515 case NVME_ID_CNS_NS_ACTIVE_LIST: 516 req->execute = nvmet_execute_identify_nslist; 517 return 0; 518 } 519 break; 520 case nvme_admin_abort_cmd: 521 req->execute = nvmet_execute_abort; 522 req->data_len = 0; 523 return 0; 524 case nvme_admin_set_features: 525 req->execute = nvmet_execute_set_features; 526 req->data_len = 0; 527 return 0; 528 case nvme_admin_get_features: 529 req->execute = nvmet_execute_get_features; 530 req->data_len = 0; 531 return 0; 532 case nvme_admin_async_event: 533 req->execute = nvmet_execute_async_event; 534 req->data_len = 0; 535 return 0; 536 case nvme_admin_keep_alive: 537 req->execute = nvmet_execute_keep_alive; 538 req->data_len = 0; 539 return 0; 540 } 541 542 pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode, 543 req->sq->qid); 544 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR; 545 } 546