drivers/nvme/target/admin-cmd.c

   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
  79         req->ns = nvmet_find_namespace(req->sq->ctrl,
  80                                        req->cmd->get_log_page.nsid);
  81         if (!req->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 (!req->ns->bdev)
  90                 return NVME_SC_SUCCESS;
  91
  92         host_reads = part_stat_read(req->ns->bdev, ios[READ]);
  93         data_units_read =
  94                 DIV_ROUND_UP(part_stat_read(req->ns->bdev, sectors[READ]), 1000);
  95         host_writes = part_stat_read(req->ns->bdev, ios[WRITE]);
  96         data_units_written =
  97                 DIV_ROUND_UP(part_stat_read(req->ns->bdev, 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
 104         return NVME_SC_SUCCESS;
 105 }
 106
 107 static u16 nvmet_get_smart_log_all(struct nvmet_req *req,
 108                 struct nvme_smart_log *slog)
 109 {
 110         u64 host_reads = 0, host_writes = 0;
 111         u64 data_units_read = 0, data_units_written = 0;
 112         struct nvmet_ns *ns;
 113         struct nvmet_ctrl *ctrl;
 114         unsigned long idx;
 115
 116         ctrl = req->sq->ctrl;
 117         xa_for_each(&ctrl->subsys->namespaces, idx, ns) {
 118                 /* we don't have the right data for file backed ns */
 119                 if (!ns->bdev)
 120                         continue;
 121                 host_reads += part_stat_read(ns->bdev, ios[READ]);
 122                 data_units_read += DIV_ROUND_UP(
 123                         part_stat_read(ns->bdev, sectors[READ]), 1000);
 124                 host_writes += part_stat_read(ns->bdev, ios[WRITE]);
 125                 data_units_written += DIV_ROUND_UP(
 126                         part_stat_read(ns->bdev, sectors[WRITE]), 1000);
 127         }
 128
 129         put_unaligned_le64(host_reads, &slog->host_reads[0]);
 130         put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
 131         put_unaligned_le64(host_writes, &slog->host_writes[0]);
 132         put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
 133
 134         return NVME_SC_SUCCESS;
 135 }
 136
 137 static void nvmet_execute_get_log_page_smart(struct nvmet_req *req)
 138 {
 139         struct nvme_smart_log *log;
 140         u16 status = NVME_SC_INTERNAL;
 141         unsigned long flags;
 142
 143         if (req->transfer_len != sizeof(*log))
 144                 goto out;
 145
 146         log = kzalloc(sizeof(*log), GFP_KERNEL);
 147         if (!log)
 148                 goto out;
 149
 150         if (req->cmd->get_log_page.nsid == cpu_to_le32(NVME_NSID_ALL))
 151                 status = nvmet_get_smart_log_all(req, log);
 152         else
 153                 status = nvmet_get_smart_log_nsid(req, log);
 154         if (status)
 155                 goto out_free_log;
 156
 157         spin_lock_irqsave(&req->sq->ctrl->error_lock, flags);
 158         put_unaligned_le64(req->sq->ctrl->err_counter,
 159                         &log->num_err_log_entries);
 160         spin_unlock_irqrestore(&req->sq->ctrl->error_lock, flags);
 161
 162         status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
 163 out_free_log:
 164         kfree(log);
 165 out:
 166         nvmet_req_complete(req, status);
 167 }
 168
 169 static void nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req *req)
 170 {
 171         u16 status = NVME_SC_INTERNAL;
 172         struct nvme_effects_log *log;
 173
 174         log = kzalloc(sizeof(*log), GFP_KERNEL);
 175         if (!log)
 176                 goto out;
 177
 178         log->acs[nvme_admin_get_log_page]       = cpu_to_le32(1 << 0);
 179         log->acs[nvme_admin_identify]           = cpu_to_le32(1 << 0);
 180         log->acs[nvme_admin_abort_cmd]          = cpu_to_le32(1 << 0);
 181         log->acs[nvme_admin_set_features]       = cpu_to_le32(1 << 0);
 182         log->acs[nvme_admin_get_features]       = cpu_to_le32(1 << 0);
 183         log->acs[nvme_admin_async_event]        = cpu_to_le32(1 << 0);
 184         log->acs[nvme_admin_keep_alive]         = cpu_to_le32(1 << 0);
 185
 186         log->iocs[nvme_cmd_read]                = cpu_to_le32(1 << 0);
 187         log->iocs[nvme_cmd_write]               = cpu_to_le32(1 << 0);
 188         log->iocs[nvme_cmd_flush]               = cpu_to_le32(1 << 0);
 189         log->iocs[nvme_cmd_dsm]                 = cpu_to_le32(1 << 0);
 190         log->iocs[nvme_cmd_write_zeroes]        = cpu_to_le32(1 << 0);
 191
 192         status = nvmet_copy_to_sgl(req, 0, log, sizeof(*log));
 193
 194         kfree(log);
 195 out:
 196         nvmet_req_complete(req, status);
 197 }
 198
 199 static void nvmet_execute_get_log_changed_ns(struct nvmet_req *req)
 200 {
 201         struct nvmet_ctrl *ctrl = req->sq->ctrl;
 202         u16 status = NVME_SC_INTERNAL;
 203         size_t len;
 204
 205         if (req->transfer_len != NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32))
 206                 goto out;
 207
 208         mutex_lock(&ctrl->lock);
 209         if (ctrl->nr_changed_ns == U32_MAX)
 210                 len = sizeof(__le32);
 211         else
 212                 len = ctrl->nr_changed_ns * sizeof(__le32);
 213         status = nvmet_copy_to_sgl(req, 0, ctrl->changed_ns_list, len);
 214         if (!status)
 215                 status = nvmet_zero_sgl(req, len, req->transfer_len - len);
 216         ctrl->nr_changed_ns = 0;
 217         nvmet_clear_aen_bit(req, NVME_AEN_BIT_NS_ATTR);
 218         mutex_unlock(&ctrl->lock);
 219 out:
 220         nvmet_req_complete(req, status);
 221 }
 222
 223 static u32 nvmet_format_ana_group(struct nvmet_req *req, u32 grpid,
 224                 struct nvme_ana_group_desc *desc)
 225 {
 226         struct nvmet_ctrl *ctrl = req->sq->ctrl;
 227         struct nvmet_ns *ns;
 228         unsigned long idx;
 229         u32 count = 0;
 230
 231         if (!(req->cmd->get_log_page.lsp & NVME_ANA_LOG_RGO)) {
 232                 xa_for_each(&ctrl->subsys->namespaces, idx, ns)
 233                         if (ns->anagrpid == grpid)
 234                                 desc->nsids[count++] = cpu_to_le32(ns->nsid);
 235         }
 236
 237         desc->grpid = cpu_to_le32(grpid);
 238         desc->nnsids = cpu_to_le32(count);
 239         desc->chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
 240         desc->state = req->port->ana_state[grpid];
 241         memset(desc->rsvd17, 0, sizeof(desc->rsvd17));
 242         return sizeof(struct nvme_ana_group_desc) + count * sizeof(__le32);
 243 }
 244
 245 static void nvmet_execute_get_log_page_ana(struct nvmet_req *req)
 246 {
 247         struct nvme_ana_rsp_hdr hdr = { 0, };
 248         struct nvme_ana_group_desc *desc;
 249         size_t offset = sizeof(struct nvme_ana_rsp_hdr); /* start beyond hdr */
 250         size_t len;
 251         u32 grpid;
 252         u16 ngrps = 0;
 253         u16 status;
 254
 255         status = NVME_SC_INTERNAL;
 256         desc = kmalloc(sizeof(struct nvme_ana_group_desc) +
 257                         NVMET_MAX_NAMESPACES * sizeof(__le32), GFP_KERNEL);
 258         if (!desc)
 259                 goto out;
 260
 261         down_read(&nvmet_ana_sem);
 262         for (grpid = 1; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
 263                 if (!nvmet_ana_group_enabled[grpid])
 264                         continue;
 265                 len = nvmet_format_ana_group(req, grpid, desc);
 266                 status = nvmet_copy_to_sgl(req, offset, desc, len);
 267                 if (status)
 268                         break;
 269                 offset += len;
 270                 ngrps++;
 271         }
 272         for ( ; grpid <= NVMET_MAX_ANAGRPS; grpid++) {
 273                 if (nvmet_ana_group_enabled[grpid])
 274                         ngrps++;
 275         }
 276
 277         hdr.chgcnt = cpu_to_le64(nvmet_ana_chgcnt);
 278         hdr.ngrps = cpu_to_le16(ngrps);
 279         nvmet_clear_aen_bit(req, NVME_AEN_BIT_ANA_CHANGE);
 280         up_read(&nvmet_ana_sem);
 281
 282         kfree(desc);
 283
 284         /* copy the header last once we know the number of groups */
 285         status = nvmet_copy_to_sgl(req, 0, &hdr, sizeof(hdr));
 286 out:
 287         nvmet_req_complete(req, status);
 288 }
 289
 290 static void nvmet_execute_get_log_page(struct nvmet_req *req)
 291 {
 292         if (!nvmet_check_transfer_len(req, nvmet_get_log_page_len(req->cmd)))
 293                 return;
 294
 295         switch (req->cmd->get_log_page.lid) {
 296         case NVME_LOG_ERROR:
 297                 return nvmet_execute_get_log_page_error(req);
 298         case NVME_LOG_SMART:
 299                 return nvmet_execute_get_log_page_smart(req);
 300         case NVME_LOG_FW_SLOT:
 301                 /*
 302                  * We only support a single firmware slot which always is
 303                  * active, so we can zero out the whole firmware slot log and
 304                  * still claim to fully implement this mandatory log page.
 305                  */
 306                 return nvmet_execute_get_log_page_noop(req);
 307         case NVME_LOG_CHANGED_NS:
 308                 return nvmet_execute_get_log_changed_ns(req);
 309         case NVME_LOG_CMD_EFFECTS:
 310                 return nvmet_execute_get_log_cmd_effects_ns(req);
 311         case NVME_LOG_ANA:
 312                 return nvmet_execute_get_log_page_ana(req);
 313         }
 314         pr_err("unhandled lid %d on qid %d\n",
 315                req->cmd->get_log_page.lid, req->sq->qid);
 316         req->error_loc = offsetof(struct nvme_get_log_page_command, lid);
 317         nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
 318 }
 319
 320 static void nvmet_id_set_model_number(struct nvme_id_ctrl *id,
 321                                       struct nvmet_subsys *subsys)
 322 {
 323         const char *model = NVMET_DEFAULT_CTRL_MODEL;
 324         struct nvmet_subsys_model *subsys_model;
 325
 326         rcu_read_lock();
 327         subsys_model = rcu_dereference(subsys->model);
 328         if (subsys_model)
 329                 model = subsys_model->number;
 330         memcpy_and_pad(id->mn, sizeof(id->mn), model, strlen(model), ' ');
 331         rcu_read_unlock();
 332 }
 333
 334 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
 335 {
 336         struct nvmet_ctrl *ctrl = req->sq->ctrl;
 337         struct nvme_id_ctrl *id;
 338         u32 cmd_capsule_size;
 339         u16 status = 0;
 340
 341         id = kzalloc(sizeof(*id), GFP_KERNEL);
 342         if (!id) {
 343                 status = NVME_SC_INTERNAL;
 344                 goto out;
 345         }
 346
 347         /* XXX: figure out how to assign real vendors IDs. */
 348         id->vid = 0;
 349         id->ssvid = 0;
 350
 351         memset(id->sn, ' ', sizeof(id->sn));
 352         bin2hex(id->sn, &ctrl->subsys->serial,
 353                 min(sizeof(ctrl->subsys->serial), sizeof(id->sn) / 2));
 354         nvmet_id_set_model_number(id, ctrl->subsys);
 355         memcpy_and_pad(id->fr, sizeof(id->fr),
 356                        UTS_RELEASE, strlen(UTS_RELEASE), ' ');
 357
 358         id->rab = 6;
 359
 360         /*
 361          * XXX: figure out how we can assign a IEEE OUI, but until then
 362          * the safest is to leave it as zeroes.
 363          */
 364
 365         /* we support multiple ports, multiples hosts and ANA: */
 366         id->cmic = (1 << 0) | (1 << 1) | (1 << 3);
 367
 368         /* Limit MDTS according to transport capability */
 369         if (ctrl->ops->get_mdts)
 370                 id->mdts = ctrl->ops->get_mdts(ctrl);
 371         else
 372                 id->mdts = 0;
 373
 374         id->cntlid = cpu_to_le16(ctrl->cntlid);
 375         id->ver = cpu_to_le32(ctrl->subsys->ver);
 376
 377         /* XXX: figure out what to do about RTD3R/RTD3 */
 378         id->oaes = cpu_to_le32(NVMET_AEN_CFG_OPTIONAL);
 379         id->ctratt = cpu_to_le32(NVME_CTRL_ATTR_HID_128_BIT |
 380                 NVME_CTRL_ATTR_TBKAS);
 381
 382         id->oacs = 0;
 383
 384         /*
 385          * We don't really have a practical limit on the number of abort
 386          * comands.  But we don't do anything useful for abort either, so
 387          * no point in allowing more abort commands than the spec requires.
 388          */
 389         id->acl = 3;
 390
 391         id->aerl = NVMET_ASYNC_EVENTS - 1;
 392
 393         /* first slot is read-only, only one slot supported */
 394         id->frmw = (1 << 0) | (1 << 1);
 395         id->lpa = (1 << 0) | (1 << 1) | (1 << 2);
 396         id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
 397         id->npss = 0;
 398
 399         /* We support keep-alive timeout in granularity of seconds */
 400         id->kas = cpu_to_le16(NVMET_KAS);
 401
 402         id->sqes = (0x6 << 4) | 0x6;
 403         id->cqes = (0x4 << 4) | 0x4;
 404
 405         /* no enforcement soft-limit for maxcmd - pick arbitrary high value */
 406         id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
 407
 408         id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
 409         id->mnan = cpu_to_le32(NVMET_MAX_NAMESPACES);
 410         id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM |
 411                         NVME_CTRL_ONCS_WRITE_ZEROES);
 412
 413         /* XXX: don't report vwc if the underlying device is write through */
 414         id->vwc = NVME_CTRL_VWC_PRESENT;
 415
 416         /*
 417          * We can't support atomic writes bigger than a LBA without support
 418          * from the backend device.
 419          */
 420         id->awun = 0;
 421         id->awupf = 0;
 422
 423         id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */
 424         if (ctrl->ops->flags & NVMF_KEYED_SGLS)
 425                 id->sgls |= cpu_to_le32(1 << 2);
 426         if (req->port->inline_data_size)
 427                 id->sgls |= cpu_to_le32(1 << 20);
 428
 429         strlcpy(id->subnqn, ctrl->subsys->subsysnqn, sizeof(id->subnqn));
 430
 431         /*
 432          * Max command capsule size is sqe + in-capsule data size.
 433          * Disable in-capsule data for Metadata capable controllers.
 434          */
 435         cmd_capsule_size = sizeof(struct nvme_command);
 436         if (!ctrl->pi_support)
 437                 cmd_capsule_size += req->port->inline_data_size;
 438         id->ioccsz = cpu_to_le32(cmd_capsule_size / 16);
 439
 440         /* Max response capsule size is cqe */
 441         id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
 442
 443         id->msdbd = ctrl->ops->msdbd;
 444
 445         id->anacap = (1 << 0) | (1 << 1) | (1 << 2) | (1 << 3) | (1 << 4);
 446         id->anatt = 10; /* random value */
 447         id->anagrpmax = cpu_to_le32(NVMET_MAX_ANAGRPS);
 448         id->nanagrpid = cpu_to_le32(NVMET_MAX_ANAGRPS);
 449
 450         /*
 451          * Meh, we don't really support any power state.  Fake up the same
 452          * values that qemu does.
 453          */
 454         id->psd[0].max_power = cpu_to_le16(0x9c4);
 455         id->psd[0].entry_lat = cpu_to_le32(0x10);
 456         id->psd[0].exit_lat = cpu_to_le32(0x4);
 457
 458         id->nwpc = 1 << 0; /* write protect and no write protect */
 459
 460         status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
 461
 462         kfree(id);
 463 out:
 464         nvmet_req_complete(req, status);
 465 }
 466
 467 static void nvmet_execute_identify_ns(struct nvmet_req *req)
 468 {
 469         struct nvmet_ctrl *ctrl = req->sq->ctrl;
 470         struct nvmet_ns *ns;
 471         struct nvme_id_ns *id;
 472         u16 status = 0;
 473
 474         if (le32_to_cpu(req->cmd->identify.nsid) == NVME_NSID_ALL) {
 475                 req->error_loc = offsetof(struct nvme_identify, nsid);
 476                 status = NVME_SC_INVALID_NS | NVME_SC_DNR;
 477                 goto out;
 478         }
 479
 480         id = kzalloc(sizeof(*id), GFP_KERNEL);
 481         if (!id) {
 482                 status = NVME_SC_INTERNAL;
 483                 goto out;
 484         }
 485
 486         /* return an all zeroed buffer if we can't find an active namespace */
 487         ns = nvmet_find_namespace(ctrl, req->cmd->identify.nsid);
 488         if (!ns) {
 489                 status = NVME_SC_INVALID_NS;
 490                 goto done;
 491         }
 492
 493         nvmet_ns_revalidate(ns);
 494
 495         /*
 496          * nuse = ncap = nsze isn't always true, but we have no way to find
 497          * that out from the underlying device.
 498          */
 499         id->ncap = id->nsze = cpu_to_le64(ns->size >> ns->blksize_shift);
 500         switch (req->port->ana_state[ns->anagrpid]) {
 501         case NVME_ANA_INACCESSIBLE:
 502         case NVME_ANA_PERSISTENT_LOSS:
 503                 break;
 504         default:
 505                 id->nuse = id->nsze;
 506                 break;
 507         }
 508
 509         if (ns->bdev)
 510                 nvmet_bdev_set_limits(ns->bdev, id);
 511
 512         /*
 513          * We just provide a single LBA format that matches what the
 514          * underlying device reports.
 515          */
 516         id->nlbaf = 0;
 517         id->flbas = 0;
 518
 519         /*
 520          * Our namespace might always be shared.  Not just with other
 521          * controllers, but also with any other user of the block device.
 522          */
 523         id->nmic = (1 << 0);
 524         id->anagrpid = cpu_to_le32(ns->anagrpid);
 525
 526         memcpy(&id->nguid, &ns->nguid, sizeof(id->nguid));
 527
 528         id->lbaf[0].ds = ns->blksize_shift;
 529
 530         if (ctrl->pi_support && nvmet_ns_has_pi(ns)) {
 531                 id->dpc = NVME_NS_DPC_PI_FIRST | NVME_NS_DPC_PI_LAST |
 532                           NVME_NS_DPC_PI_TYPE1 | NVME_NS_DPC_PI_TYPE2 |
 533                           NVME_NS_DPC_PI_TYPE3;
 534                 id->mc = NVME_MC_EXTENDED_LBA;
 535                 id->dps = ns->pi_type;
 536                 id->flbas = NVME_NS_FLBAS_META_EXT;
 537                 id->lbaf[0].ms = cpu_to_le16(ns->metadata_size);
 538         }
 539
 540         if (ns->readonly)
 541                 id->nsattr |= (1 << 0);
 542         nvmet_put_namespace(ns);
 543 done:
 544         if (!status)
 545                 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
 546
 547         kfree(id);
 548 out:
 549         nvmet_req_complete(req, status);
 550 }
 551
 552 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
 553 {
 554         static const int buf_size = NVME_IDENTIFY_DATA_SIZE;
 555         struct nvmet_ctrl *ctrl = req->sq->ctrl;
 556         struct nvmet_ns *ns;
 557         unsigned long idx;
 558         u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
 559         __le32 *list;
 560         u16 status = 0;
 561         int i = 0;
 562
 563         list = kzalloc(buf_size, GFP_KERNEL);
 564         if (!list) {
 565                 status = NVME_SC_INTERNAL;
 566                 goto out;
 567         }
 568
 569         xa_for_each(&ctrl->subsys->namespaces, idx, ns) {
 570                 if (ns->nsid <= min_nsid)
 571                         continue;
 572                 list[i++] = cpu_to_le32(ns->nsid);
 573                 if (i == buf_size / sizeof(__le32))
 574                         break;
 575         }
 576
 577         status = nvmet_copy_to_sgl(req, 0, list, buf_size);
 578
 579         kfree(list);
 580 out:
 581         nvmet_req_complete(req, status);
 582 }
 583
 584 static u16 nvmet_copy_ns_identifier(struct nvmet_req *req, u8 type, u8 len,
 585                                     void *id, off_t *off)
 586 {
 587         struct nvme_ns_id_desc desc = {
 588                 .nidt = type,
 589                 .nidl = len,
 590         };
 591         u16 status;
 592
 593         status = nvmet_copy_to_sgl(req, *off, &desc, sizeof(desc));
 594         if (status)
 595                 return status;
 596         *off += sizeof(desc);
 597
 598         status = nvmet_copy_to_sgl(req, *off, id, len);
 599         if (status)
 600                 return status;
 601         *off += len;
 602
 603         return 0;
 604 }
 605
 606 static void nvmet_execute_identify_desclist(struct nvmet_req *req)
 607 {
 608         u16 status = 0;
 609         off_t off = 0;
 610
 611         req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
 612         if (!req->ns) {
 613                 req->error_loc = offsetof(struct nvme_identify, nsid);
 614                 status = NVME_SC_INVALID_NS | NVME_SC_DNR;
 615                 goto out;
 616         }
 617
 618         if (memchr_inv(&req->ns->uuid, 0, sizeof(req->ns->uuid))) {
 619                 status = nvmet_copy_ns_identifier(req, NVME_NIDT_UUID,
 620                                                   NVME_NIDT_UUID_LEN,
 621                                                   &req->ns->uuid, &off);
 622                 if (status)
 623                         goto out;
 624         }
 625         if (memchr_inv(req->ns->nguid, 0, sizeof(req->ns->nguid))) {
 626                 status = nvmet_copy_ns_identifier(req, NVME_NIDT_NGUID,
 627                                                   NVME_NIDT_NGUID_LEN,
 628                                                   &req->ns->nguid, &off);
 629                 if (status)
 630                         goto out;
 631         }
 632
 633         if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
 634                         off) != NVME_IDENTIFY_DATA_SIZE - off)
 635                 status = NVME_SC_INTERNAL | NVME_SC_DNR;
 636
 637 out:
 638         nvmet_req_complete(req, status);
 639 }
 640
 641 static void nvmet_execute_identify(struct nvmet_req *req)
 642 {
 643         if (!nvmet_check_transfer_len(req, NVME_IDENTIFY_DATA_SIZE))
 644                 return;
 645
 646         switch (req->cmd->identify.cns) {
 647         case NVME_ID_CNS_NS:
 648                 return nvmet_execute_identify_ns(req);
 649         case NVME_ID_CNS_CTRL:
 650                 return nvmet_execute_identify_ctrl(req);
 651         case NVME_ID_CNS_NS_ACTIVE_LIST:
 652                 return nvmet_execute_identify_nslist(req);
 653         case NVME_ID_CNS_NS_DESC_LIST:
 654                 return nvmet_execute_identify_desclist(req);
 655         }
 656
 657         pr_err("unhandled identify cns %d on qid %d\n",
 658                req->cmd->identify.cns, req->sq->qid);
 659         req->error_loc = offsetof(struct nvme_identify, cns);
 660         nvmet_req_complete(req, NVME_SC_INVALID_FIELD | NVME_SC_DNR);
 661 }
 662
 663 /*
 664  * A "minimum viable" abort implementation: the command is mandatory in the
 665  * spec, but we are not required to do any useful work.  We couldn't really
 666  * do a useful abort, so don't bother even with waiting for the command
 667  * to be exectuted and return immediately telling the command to abort
 668  * wasn't found.
 669  */
 670 static void nvmet_execute_abort(struct nvmet_req *req)
 671 {
 672         if (!nvmet_check_transfer_len(req, 0))
 673                 return;
 674         nvmet_set_result(req, 1);
 675         nvmet_req_complete(req, 0);
 676 }
 677
 678 static u16 nvmet_write_protect_flush_sync(struct nvmet_req *req)
 679 {
 680         u16 status;
 681
 682         if (req->ns->file)
 683                 status = nvmet_file_flush(req);
 684         else
 685                 status = nvmet_bdev_flush(req);
 686
 687         if (status)
 688                 pr_err("write protect flush failed nsid: %u\n", req->ns->nsid);
 689         return status;
 690 }
 691
 692 static u16 nvmet_set_feat_write_protect(struct nvmet_req *req)
 693 {
 694         u32 write_protect = le32_to_cpu(req->cmd->common.cdw11);
 695         struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
 696         u16 status = NVME_SC_FEATURE_NOT_CHANGEABLE;
 697
 698         req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->rw.nsid);
 699         if (unlikely(!req->ns)) {
 700                 req->error_loc = offsetof(struct nvme_common_command, nsid);
 701                 return status;
 702         }
 703
 704         mutex_lock(&subsys->lock);
 705         switch (write_protect) {
 706         case NVME_NS_WRITE_PROTECT:
 707                 req->ns->readonly = true;
 708                 status = nvmet_write_protect_flush_sync(req);
 709                 if (status)
 710                         req->ns->readonly = false;
 711                 break;
 712         case NVME_NS_NO_WRITE_PROTECT:
 713                 req->ns->readonly = false;
 714                 status = 0;
 715                 break;
 716         default:
 717                 break;
 718         }
 719
 720         if (!status)
 721                 nvmet_ns_changed(subsys, req->ns->nsid);
 722         mutex_unlock(&subsys->lock);
 723         return status;
 724 }
 725
 726 u16 nvmet_set_feat_kato(struct nvmet_req *req)
 727 {
 728         u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
 729
 730         nvmet_stop_keep_alive_timer(req->sq->ctrl);
 731         req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
 732         nvmet_start_keep_alive_timer(req->sq->ctrl);
 733
 734         nvmet_set_result(req, req->sq->ctrl->kato);
 735
 736         return 0;
 737 }
 738
 739 u16 nvmet_set_feat_async_event(struct nvmet_req *req, u32 mask)
 740 {
 741         u32 val32 = le32_to_cpu(req->cmd->common.cdw11);
 742
 743         if (val32 & ~mask) {
 744                 req->error_loc = offsetof(struct nvme_common_command, cdw11);
 745                 return NVME_SC_INVALID_FIELD | NVME_SC_DNR;
 746         }
 747
 748         WRITE_ONCE(req->sq->ctrl->aen_enabled, val32);
 749         nvmet_set_result(req, val32);
 750
 751         return 0;
 752 }
 753
 754 void nvmet_execute_set_features(struct nvmet_req *req)
 755 {
 756         struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
 757         u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
 758         u32 cdw11 = le32_to_cpu(req->cmd->common.cdw11);
 759         u16 status = 0;
 760         u16 nsqr;
 761         u16 ncqr;
 762
 763         if (!nvmet_check_transfer_len(req, 0))
 764                 return;
 765
 766         switch (cdw10 & 0xff) {
 767         case NVME_FEAT_NUM_QUEUES:
 768                 ncqr = (cdw11 >> 16) & 0xffff;
 769                 nsqr = cdw11 & 0xffff;
 770                 if (ncqr == 0xffff || nsqr == 0xffff) {
 771                         status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
 772                         break;
 773                 }
 774                 nvmet_set_result(req,
 775                         (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
 776                 break;
 777         case NVME_FEAT_KATO:
 778                 status = nvmet_set_feat_kato(req);
 779                 break;
 780         case NVME_FEAT_ASYNC_EVENT:
 781                 status = nvmet_set_feat_async_event(req, NVMET_AEN_CFG_ALL);
 782                 break;
 783         case NVME_FEAT_HOST_ID:
 784                 status = NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
 785                 break;
 786         case NVME_FEAT_WRITE_PROTECT:
 787                 status = nvmet_set_feat_write_protect(req);
 788                 break;
 789         default:
 790                 req->error_loc = offsetof(struct nvme_common_command, cdw10);
 791                 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
 792                 break;
 793         }
 794
 795         nvmet_req_complete(req, status);
 796 }
 797
 798 static u16 nvmet_get_feat_write_protect(struct nvmet_req *req)
 799 {
 800         struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
 801         u32 result;
 802
 803         req->ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->common.nsid);
 804         if (!req->ns)  {
 805                 req->error_loc = offsetof(struct nvme_common_command, nsid);
 806                 return NVME_SC_INVALID_NS | NVME_SC_DNR;
 807         }
 808         mutex_lock(&subsys->lock);
 809         if (req->ns->readonly == true)
 810                 result = NVME_NS_WRITE_PROTECT;
 811         else
 812                 result = NVME_NS_NO_WRITE_PROTECT;
 813         nvmet_set_result(req, result);
 814         mutex_unlock(&subsys->lock);
 815
 816         return 0;
 817 }
 818
 819 void nvmet_get_feat_kato(struct nvmet_req *req)
 820 {
 821         nvmet_set_result(req, req->sq->ctrl->kato * 1000);
 822 }
 823
 824 void nvmet_get_feat_async_event(struct nvmet_req *req)
 825 {
 826         nvmet_set_result(req, READ_ONCE(req->sq->ctrl->aen_enabled));
 827 }
 828
 829 void nvmet_execute_get_features(struct nvmet_req *req)
 830 {
 831         struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
 832         u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10);
 833         u16 status = 0;
 834
 835         if (!nvmet_check_transfer_len(req, nvmet_feat_data_len(req, cdw10)))
 836                 return;
 837
 838         switch (cdw10 & 0xff) {
 839         /*
 840          * These features are mandatory in the spec, but we don't
 841          * have a useful way to implement them.  We'll eventually
 842          * need to come up with some fake values for these.
 843          */
 844 #if 0
 845         case NVME_FEAT_ARBITRATION:
 846                 break;
 847         case NVME_FEAT_POWER_MGMT:
 848                 break;
 849         case NVME_FEAT_TEMP_THRESH:
 850                 break;
 851         case NVME_FEAT_ERR_RECOVERY:
 852                 break;
 853         case NVME_FEAT_IRQ_COALESCE:
 854                 break;
 855         case NVME_FEAT_IRQ_CONFIG:
 856                 break;
 857         case NVME_FEAT_WRITE_ATOMIC:
 858                 break;
 859 #endif
 860         case NVME_FEAT_ASYNC_EVENT:
 861                 nvmet_get_feat_async_event(req);
 862                 break;
 863         case NVME_FEAT_VOLATILE_WC:
 864                 nvmet_set_result(req, 1);
 865                 break;
 866         case NVME_FEAT_NUM_QUEUES:
 867                 nvmet_set_result(req,
 868                         (subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
 869                 break;
 870         case NVME_FEAT_KATO:
 871                 nvmet_get_feat_kato(req);
 872                 break;
 873         case NVME_FEAT_HOST_ID:
 874                 /* need 128-bit host identifier flag */
 875                 if (!(req->cmd->common.cdw11 & cpu_to_le32(1 << 0))) {
 876                         req->error_loc =
 877                                 offsetof(struct nvme_common_command, cdw11);
 878                         status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
 879                         break;
 880                 }
 881
 882                 status = nvmet_copy_to_sgl(req, 0, &req->sq->ctrl->hostid,
 883                                 sizeof(req->sq->ctrl->hostid));
 884                 break;
 885         case NVME_FEAT_WRITE_PROTECT:
 886                 status = nvmet_get_feat_write_protect(req);
 887                 break;
 888         default:
 889                 req->error_loc =
 890                         offsetof(struct nvme_common_command, cdw10);
 891                 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
 892                 break;
 893         }
 894
 895         nvmet_req_complete(req, status);
 896 }
 897
 898 void nvmet_execute_async_event(struct nvmet_req *req)
 899 {
 900         struct nvmet_ctrl *ctrl = req->sq->ctrl;
 901
 902         if (!nvmet_check_transfer_len(req, 0))
 903                 return;
 904
 905         mutex_lock(&ctrl->lock);
 906         if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
 907                 mutex_unlock(&ctrl->lock);
 908                 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
 909                 return;
 910         }
 911         ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
 912         mutex_unlock(&ctrl->lock);
 913
 914         schedule_work(&ctrl->async_event_work);
 915 }
 916
 917 void nvmet_execute_keep_alive(struct nvmet_req *req)
 918 {
 919         struct nvmet_ctrl *ctrl = req->sq->ctrl;
 920
 921         if (!nvmet_check_transfer_len(req, 0))
 922                 return;
 923
 924         pr_debug("ctrl %d update keep-alive timer for %d secs\n",
 925                 ctrl->cntlid, ctrl->kato);
 926
 927         mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
 928         nvmet_req_complete(req, 0);
 929 }
 930
 931 u16 nvmet_parse_admin_cmd(struct nvmet_req *req)
 932 {
 933         struct nvme_command *cmd = req->cmd;
 934         u16 ret;
 935
 936         if (nvme_is_fabrics(cmd))
 937                 return nvmet_parse_fabrics_cmd(req);
 938         if (req->sq->ctrl->subsys->type == NVME_NQN_DISC)
 939                 return nvmet_parse_discovery_cmd(req);
 940
 941         ret = nvmet_check_ctrl_status(req, cmd);
 942         if (unlikely(ret))
 943                 return ret;
 944
 945         if (nvmet_req_passthru_ctrl(req))
 946                 return nvmet_parse_passthru_admin_cmd(req);
 947
 948         switch (cmd->common.opcode) {
 949         case nvme_admin_get_log_page:
 950                 req->execute = nvmet_execute_get_log_page;
 951                 return 0;
 952         case nvme_admin_identify:
 953                 req->execute = nvmet_execute_identify;
 954                 return 0;
 955         case nvme_admin_abort_cmd:
 956                 req->execute = nvmet_execute_abort;
 957                 return 0;
 958         case nvme_admin_set_features:
 959                 req->execute = nvmet_execute_set_features;
 960                 return 0;
 961         case nvme_admin_get_features:
 962                 req->execute = nvmet_execute_get_features;
 963                 return 0;
 964         case nvme_admin_async_event:
 965                 req->execute = nvmet_execute_async_event;
 966                 return 0;
 967         case nvme_admin_keep_alive:
 968                 req->execute = nvmet_execute_keep_alive;
 969                 return 0;
 970         }
 971
 972         pr_err("unhandled cmd %d on qid %d\n", cmd->common.opcode,
 973                req->sq->qid);
 974         req->error_loc = offsetof(struct nvme_common_command, opcode);
 975         return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
 976 }