drivers/nvme/host/fabrics.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * NVMe over Fabrics common host code.
   4  * Copyright (c) 2015-2016 HGST, a Western Digital Company.
   5  */
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7 #include <linux/init.h>
   8 #include <linux/miscdevice.h>
   9 #include <linux/module.h>
  10 #include <linux/mutex.h>
  11 #include <linux/parser.h>
  12 #include <linux/seq_file.h>
  13 #include "nvme.h"
  14 #include "fabrics.h"
  15
  16 static LIST_HEAD(nvmf_transports);
  17 static DECLARE_RWSEM(nvmf_transports_rwsem);
  18
  19 static LIST_HEAD(nvmf_hosts);
  20 static DEFINE_MUTEX(nvmf_hosts_mutex);
  21
  22 static struct nvmf_host *nvmf_default_host;
  23
  24 static struct nvmf_host *__nvmf_host_find(const char *hostnqn)
  25 {
  26         struct nvmf_host *host;
  27
  28         list_for_each_entry(host, &nvmf_hosts, list) {
  29                 if (!strcmp(host->nqn, hostnqn))
  30                         return host;
  31         }
  32
  33         return NULL;
  34 }
  35
  36 static struct nvmf_host *nvmf_host_add(const char *hostnqn)
  37 {
  38         struct nvmf_host *host;
  39
  40         mutex_lock(&nvmf_hosts_mutex);
  41         host = __nvmf_host_find(hostnqn);
  42         if (host) {
  43                 kref_get(&host->ref);
  44                 goto out_unlock;
  45         }
  46
  47         host = kmalloc(sizeof(*host), GFP_KERNEL);
  48         if (!host)
  49                 goto out_unlock;
  50
  51         kref_init(&host->ref);
  52         strlcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
  53
  54         list_add_tail(&host->list, &nvmf_hosts);
  55 out_unlock:
  56         mutex_unlock(&nvmf_hosts_mutex);
  57         return host;
  58 }
  59
  60 static struct nvmf_host *nvmf_host_default(void)
  61 {
  62         struct nvmf_host *host;
  63
  64         host = kmalloc(sizeof(*host), GFP_KERNEL);
  65         if (!host)
  66                 return NULL;
  67
  68         kref_init(&host->ref);
  69         uuid_gen(&host->id);
  70         snprintf(host->nqn, NVMF_NQN_SIZE,
  71                 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id);
  72
  73         mutex_lock(&nvmf_hosts_mutex);
  74         list_add_tail(&host->list, &nvmf_hosts);
  75         mutex_unlock(&nvmf_hosts_mutex);
  76
  77         return host;
  78 }
  79
  80 static void nvmf_host_destroy(struct kref *ref)
  81 {
  82         struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
  83
  84         mutex_lock(&nvmf_hosts_mutex);
  85         list_del(&host->list);
  86         mutex_unlock(&nvmf_hosts_mutex);
  87
  88         kfree(host);
  89 }
  90
  91 static void nvmf_host_put(struct nvmf_host *host)
  92 {
  93         if (host)
  94                 kref_put(&host->ref, nvmf_host_destroy);
  95 }
  96
  97 /**
  98  * nvmf_get_address() -  Get address/port
  99  * @ctrl:       Host NVMe controller instance which we got the address
 100  * @buf:        OUTPUT parameter that will contain the address/port
 101  * @size:       buffer size
 102  */
 103 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
 104 {
 105         int len = 0;
 106
 107         if (ctrl->opts->mask & NVMF_OPT_TRADDR)
 108                 len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
 109         if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
 110                 len += scnprintf(buf + len, size - len, "%strsvcid=%s",
 111                                 (len) ? "," : "", ctrl->opts->trsvcid);
 112         if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
 113                 len += scnprintf(buf + len, size - len, "%shost_traddr=%s",
 114                                 (len) ? "," : "", ctrl->opts->host_traddr);
 115         len += scnprintf(buf + len, size - len, "\n");
 116
 117         return len;
 118 }
 119 EXPORT_SYMBOL_GPL(nvmf_get_address);
 120
 121 /**
 122  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
 123  * @ctrl:       Host NVMe controller instance maintaining the admin
 124  *              queue used to submit the property read command to
 125  *              the allocated NVMe controller resource on the target system.
 126  * @off:        Starting offset value of the targeted property
 127  *              register (see the fabrics section of the NVMe standard).
 128  * @val:        OUTPUT parameter that will contain the value of
 129  *              the property after a successful read.
 130  *
 131  * Used by the host system to retrieve a 32-bit capsule property value
 132  * from an NVMe controller on the target system.
 133  *
 134  * ("Capsule property" is an "PCIe register concept" applied to the
 135  * NVMe fabrics space.)
 136  *
 137  * Return:
 138  *      0: successful read
 139  *      > 0: NVMe error status code
 140  *      < 0: Linux errno error code
 141  */
 142 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
 143 {
 144         struct nvme_command cmd;
 145         union nvme_result res;
 146         int ret;
 147
 148         memset(&cmd, 0, sizeof(cmd));
 149         cmd.prop_get.opcode = nvme_fabrics_command;
 150         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
 151         cmd.prop_get.offset = cpu_to_le32(off);
 152
 153         ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
 154                         NVME_QID_ANY, 0, 0, false);
 155
 156         if (ret >= 0)
 157                 *val = le64_to_cpu(res.u64);
 158         if (unlikely(ret != 0))
 159                 dev_err(ctrl->device,
 160                         "Property Get error: %d, offset %#x\n",
 161                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
 162
 163         return ret;
 164 }
 165 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
 166
 167 /**
 168  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
 169  * @ctrl:       Host NVMe controller instance maintaining the admin
 170  *              queue used to submit the property read command to
 171  *              the allocated controller resource on the target system.
 172  * @off:        Starting offset value of the targeted property
 173  *              register (see the fabrics section of the NVMe standard).
 174  * @val:        OUTPUT parameter that will contain the value of
 175  *              the property after a successful read.
 176  *
 177  * Used by the host system to retrieve a 64-bit capsule property value
 178  * from an NVMe controller on the target system.
 179  *
 180  * ("Capsule property" is an "PCIe register concept" applied to the
 181  * NVMe fabrics space.)
 182  *
 183  * Return:
 184  *      0: successful read
 185  *      > 0: NVMe error status code
 186  *      < 0: Linux errno error code
 187  */
 188 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
 189 {
 190         struct nvme_command cmd;
 191         union nvme_result res;
 192         int ret;
 193
 194         memset(&cmd, 0, sizeof(cmd));
 195         cmd.prop_get.opcode = nvme_fabrics_command;
 196         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
 197         cmd.prop_get.attrib = 1;
 198         cmd.prop_get.offset = cpu_to_le32(off);
 199
 200         ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 0,
 201                         NVME_QID_ANY, 0, 0, false);
 202
 203         if (ret >= 0)
 204                 *val = le64_to_cpu(res.u64);
 205         if (unlikely(ret != 0))
 206                 dev_err(ctrl->device,
 207                         "Property Get error: %d, offset %#x\n",
 208                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
 209         return ret;
 210 }
 211 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
 212
 213 /**
 214  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
 215  * @ctrl:       Host NVMe controller instance maintaining the admin
 216  *              queue used to submit the property read command to
 217  *              the allocated NVMe controller resource on the target system.
 218  * @off:        Starting offset value of the targeted property
 219  *              register (see the fabrics section of the NVMe standard).
 220  * @val:        Input parameter that contains the value to be
 221  *              written to the property.
 222  *
 223  * Used by the NVMe host system to write a 32-bit capsule property value
 224  * to an NVMe controller on the target system.
 225  *
 226  * ("Capsule property" is an "PCIe register concept" applied to the
 227  * NVMe fabrics space.)
 228  *
 229  * Return:
 230  *      0: successful write
 231  *      > 0: NVMe error status code
 232  *      < 0: Linux errno error code
 233  */
 234 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
 235 {
 236         struct nvme_command cmd;
 237         int ret;
 238
 239         memset(&cmd, 0, sizeof(cmd));
 240         cmd.prop_set.opcode = nvme_fabrics_command;
 241         cmd.prop_set.fctype = nvme_fabrics_type_property_set;
 242         cmd.prop_set.attrib = 0;
 243         cmd.prop_set.offset = cpu_to_le32(off);
 244         cmd.prop_set.value = cpu_to_le64(val);
 245
 246         ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0, 0,
 247                         NVME_QID_ANY, 0, 0, false);
 248         if (unlikely(ret))
 249                 dev_err(ctrl->device,
 250                         "Property Set error: %d, offset %#x\n",
 251                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
 252         return ret;
 253 }
 254 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
 255
 256 /**
 257  * nvmf_log_connect_error() - Error-parsing-diagnostic print
 258  * out function for connect() errors.
 259  *
 260  * @ctrl: the specific /dev/nvmeX device that had the error.
 261  *
 262  * @errval: Error code to be decoded in a more human-friendly
 263  *          printout.
 264  *
 265  * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
 266  *
 267  * @cmd: This is the SQE portion of a submission capsule.
 268  *
 269  * @data: This is the "Data" portion of a submission capsule.
 270  */
 271 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
 272                 int errval, int offset, struct nvme_command *cmd,
 273                 struct nvmf_connect_data *data)
 274 {
 275         int err_sctype = errval & (~NVME_SC_DNR);
 276
 277         switch (err_sctype) {
 278
 279         case (NVME_SC_CONNECT_INVALID_PARAM):
 280                 if (offset >> 16) {
 281                         char *inv_data = "Connect Invalid Data Parameter";
 282
 283                         switch (offset & 0xffff) {
 284                         case (offsetof(struct nvmf_connect_data, cntlid)):
 285                                 dev_err(ctrl->device,
 286                                         "%s, cntlid: %d\n",
 287                                         inv_data, data->cntlid);
 288                                 break;
 289                         case (offsetof(struct nvmf_connect_data, hostnqn)):
 290                                 dev_err(ctrl->device,
 291                                         "%s, hostnqn \"%s\"\n",
 292                                         inv_data, data->hostnqn);
 293                                 break;
 294                         case (offsetof(struct nvmf_connect_data, subsysnqn)):
 295                                 dev_err(ctrl->device,
 296                                         "%s, subsysnqn \"%s\"\n",
 297                                         inv_data, data->subsysnqn);
 298                                 break;
 299                         default:
 300                                 dev_err(ctrl->device,
 301                                         "%s, starting byte offset: %d\n",
 302                                        inv_data, offset & 0xffff);
 303                                 break;
 304                         }
 305                 } else {
 306                         char *inv_sqe = "Connect Invalid SQE Parameter";
 307
 308                         switch (offset) {
 309                         case (offsetof(struct nvmf_connect_command, qid)):
 310                                 dev_err(ctrl->device,
 311                                        "%s, qid %d\n",
 312                                         inv_sqe, cmd->connect.qid);
 313                                 break;
 314                         default:
 315                                 dev_err(ctrl->device,
 316                                         "%s, starting byte offset: %d\n",
 317                                         inv_sqe, offset);
 318                         }
 319                 }
 320                 break;
 321
 322         case NVME_SC_CONNECT_INVALID_HOST:
 323                 dev_err(ctrl->device,
 324                         "Connect for subsystem %s is not allowed, hostnqn: %s\n",
 325                         data->subsysnqn, data->hostnqn);
 326                 break;
 327
 328         case NVME_SC_CONNECT_CTRL_BUSY:
 329                 dev_err(ctrl->device,
 330                         "Connect command failed: controller is busy or not available\n");
 331                 break;
 332
 333         case NVME_SC_CONNECT_FORMAT:
 334                 dev_err(ctrl->device,
 335                         "Connect incompatible format: %d",
 336                         cmd->connect.recfmt);
 337                 break;
 338
 339         default:
 340                 dev_err(ctrl->device,
 341                         "Connect command failed, error wo/DNR bit: %d\n",
 342                         err_sctype);
 343                 break;
 344         } /* switch (err_sctype) */
 345 }
 346
 347 /**
 348  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
 349  *                              API function.
 350  * @ctrl:       Host nvme controller instance used to request
 351  *              a new NVMe controller allocation on the target
 352  *              system and  establish an NVMe Admin connection to
 353  *              that controller.
 354  *
 355  * This function enables an NVMe host device to request a new allocation of
 356  * an NVMe controller resource on a target system as well establish a
 357  * fabrics-protocol connection of the NVMe Admin queue between the
 358  * host system device and the allocated NVMe controller on the
 359  * target system via a NVMe Fabrics "Connect" command.
 360  *
 361  * Return:
 362  *      0: success
 363  *      > 0: NVMe error status code
 364  *      < 0: Linux errno error code
 365  *
 366  */
 367 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
 368 {
 369         struct nvme_command cmd;
 370         union nvme_result res;
 371         struct nvmf_connect_data *data;
 372         int ret;
 373
 374         memset(&cmd, 0, sizeof(cmd));
 375         cmd.connect.opcode = nvme_fabrics_command;
 376         cmd.connect.fctype = nvme_fabrics_type_connect;
 377         cmd.connect.qid = 0;
 378         cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
 379
 380         /*
 381          * Set keep-alive timeout in seconds granularity (ms * 1000)
 382          * and add a grace period for controller kato enforcement
 383          */
 384         cmd.connect.kato = ctrl->kato ?
 385                 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000) : 0;
 386
 387         if (ctrl->opts->disable_sqflow)
 388                 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
 389
 390         data = kzalloc(sizeof(*data), GFP_KERNEL);
 391         if (!data)
 392                 return -ENOMEM;
 393
 394         uuid_copy(&data->hostid, &ctrl->opts->host->id);
 395         data->cntlid = cpu_to_le16(0xffff);
 396         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 397         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 398
 399         ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
 400                         data, sizeof(*data), 0, NVME_QID_ANY, 1,
 401                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, false);
 402         if (ret) {
 403                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
 404                                        &cmd, data);
 405                 goto out_free_data;
 406         }
 407
 408         ctrl->cntlid = le16_to_cpu(res.u16);
 409
 410 out_free_data:
 411         kfree(data);
 412         return ret;
 413 }
 414 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
 415
 416 /**
 417  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
 418  *                           API function.
 419  * @ctrl:       Host nvme controller instance used to establish an
 420  *              NVMe I/O queue connection to the already allocated NVMe
 421  *              controller on the target system.
 422  * @qid:        NVMe I/O queue number for the new I/O connection between
 423  *              host and target (note qid == 0 is illegal as this is
 424  *              the Admin queue, per NVMe standard).
 425  * @poll:       Whether or not to poll for the completion of the connect cmd.
 426  *
 427  * This function issues a fabrics-protocol connection
 428  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
 429  * between the host system device and the allocated NVMe controller
 430  * on the target system.
 431  *
 432  * Return:
 433  *      0: success
 434  *      > 0: NVMe error status code
 435  *      < 0: Linux errno error code
 436  */
 437 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid, bool poll)
 438 {
 439         struct nvme_command cmd;
 440         struct nvmf_connect_data *data;
 441         union nvme_result res;
 442         int ret;
 443
 444         memset(&cmd, 0, sizeof(cmd));
 445         cmd.connect.opcode = nvme_fabrics_command;
 446         cmd.connect.fctype = nvme_fabrics_type_connect;
 447         cmd.connect.qid = cpu_to_le16(qid);
 448         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
 449
 450         if (ctrl->opts->disable_sqflow)
 451                 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
 452
 453         data = kzalloc(sizeof(*data), GFP_KERNEL);
 454         if (!data)
 455                 return -ENOMEM;
 456
 457         uuid_copy(&data->hostid, &ctrl->opts->host->id);
 458         data->cntlid = cpu_to_le16(ctrl->cntlid);
 459         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 460         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 461
 462         ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
 463                         data, sizeof(*data), 0, qid, 1,
 464                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT, poll);
 465         if (ret) {
 466                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
 467                                        &cmd, data);
 468         }
 469         kfree(data);
 470         return ret;
 471 }
 472 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
 473
 474 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
 475 {
 476         if (ctrl->opts->max_reconnects == -1 ||
 477             ctrl->nr_reconnects < ctrl->opts->max_reconnects)
 478                 return true;
 479
 480         return false;
 481 }
 482 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
 483
 484 /**
 485  * nvmf_register_transport() - NVMe Fabrics Library registration function.
 486  * @ops:        Transport ops instance to be registered to the
 487  *              common fabrics library.
 488  *
 489  * API function that registers the type of specific transport fabric
 490  * being implemented to the common NVMe fabrics library. Part of
 491  * the overall init sequence of starting up a fabrics driver.
 492  */
 493 int nvmf_register_transport(struct nvmf_transport_ops *ops)
 494 {
 495         if (!ops->create_ctrl)
 496                 return -EINVAL;
 497
 498         down_write(&nvmf_transports_rwsem);
 499         list_add_tail(&ops->entry, &nvmf_transports);
 500         up_write(&nvmf_transports_rwsem);
 501
 502         return 0;
 503 }
 504 EXPORT_SYMBOL_GPL(nvmf_register_transport);
 505
 506 /**
 507  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
 508  * @ops:        Transport ops instance to be unregistered from the
 509  *              common fabrics library.
 510  *
 511  * Fabrics API function that unregisters the type of specific transport
 512  * fabric being implemented from the common NVMe fabrics library.
 513  * Part of the overall exit sequence of unloading the implemented driver.
 514  */
 515 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
 516 {
 517         down_write(&nvmf_transports_rwsem);
 518         list_del(&ops->entry);
 519         up_write(&nvmf_transports_rwsem);
 520 }
 521 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
 522
 523 static struct nvmf_transport_ops *nvmf_lookup_transport(
 524                 struct nvmf_ctrl_options *opts)
 525 {
 526         struct nvmf_transport_ops *ops;
 527
 528         lockdep_assert_held(&nvmf_transports_rwsem);
 529
 530         list_for_each_entry(ops, &nvmf_transports, entry) {
 531                 if (strcmp(ops->name, opts->transport) == 0)
 532                         return ops;
 533         }
 534
 535         return NULL;
 536 }
 537
 538 /*
 539  * For something we're not in a state to send to the device the default action
 540  * is to busy it and retry it after the controller state is recovered.  However,
 541  * if the controller is deleting or if anything is marked for failfast or
 542  * nvme multipath it is immediately failed.
 543  *
 544  * Note: commands used to initialize the controller will be marked for failfast.
 545  * Note: nvme cli/ioctl commands are marked for failfast.
 546  */
 547 blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
 548                 struct request *rq)
 549 {
 550         if (ctrl->state != NVME_CTRL_DELETING &&
 551             ctrl->state != NVME_CTRL_DEAD &&
 552             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 553                 return BLK_STS_RESOURCE;
 554
 555         nvme_req(rq)->status = NVME_SC_HOST_PATH_ERROR;
 556         blk_mq_start_request(rq);
 557         nvme_complete_rq(rq);
 558         return BLK_STS_OK;
 559 }
 560 EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command);
 561
 562 bool __nvmf_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 563                 bool queue_live)
 564 {
 565         struct nvme_request *req = nvme_req(rq);
 566
 567         /*
 568          * If we are in some state of setup or teardown only allow
 569          * internally generated commands.
 570          */
 571         if (!blk_rq_is_passthrough(rq) || (req->flags & NVME_REQ_USERCMD))
 572                 return false;
 573
 574         /*
 575          * Only allow commands on a live queue, except for the connect command,
 576          * which is require to set the queue live in the appropinquate states.
 577          */
 578         switch (ctrl->state) {
 579         case NVME_CTRL_NEW:
 580         case NVME_CTRL_CONNECTING:
 581                 if (nvme_is_fabrics(req->cmd) &&
 582                     req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
 583                         return true;
 584                 break;
 585         default:
 586                 break;
 587         case NVME_CTRL_DEAD:
 588                 return false;
 589         }
 590
 591         return queue_live;
 592 }
 593 EXPORT_SYMBOL_GPL(__nvmf_check_ready);
 594
 595 static const match_table_t opt_tokens = {
 596         { NVMF_OPT_TRANSPORT,           "transport=%s"          },
 597         { NVMF_OPT_TRADDR,              "traddr=%s"             },
 598         { NVMF_OPT_TRSVCID,             "trsvcid=%s"            },
 599         { NVMF_OPT_NQN,                 "nqn=%s"                },
 600         { NVMF_OPT_QUEUE_SIZE,          "queue_size=%d"         },
 601         { NVMF_OPT_NR_IO_QUEUES,        "nr_io_queues=%d"       },
 602         { NVMF_OPT_RECONNECT_DELAY,     "reconnect_delay=%d"    },
 603         { NVMF_OPT_CTRL_LOSS_TMO,       "ctrl_loss_tmo=%d"      },
 604         { NVMF_OPT_KATO,                "keep_alive_tmo=%d"     },
 605         { NVMF_OPT_HOSTNQN,             "hostnqn=%s"            },
 606         { NVMF_OPT_HOST_TRADDR,         "host_traddr=%s"        },
 607         { NVMF_OPT_HOST_ID,             "hostid=%s"             },
 608         { NVMF_OPT_DUP_CONNECT,         "duplicate_connect"     },
 609         { NVMF_OPT_DISABLE_SQFLOW,      "disable_sqflow"        },
 610         { NVMF_OPT_HDR_DIGEST,          "hdr_digest"            },
 611         { NVMF_OPT_DATA_DIGEST,         "data_digest"           },
 612         { NVMF_OPT_NR_WRITE_QUEUES,     "nr_write_queues=%d"    },
 613         { NVMF_OPT_NR_POLL_QUEUES,      "nr_poll_queues=%d"     },
 614         { NVMF_OPT_TOS,                 "tos=%d"                },
 615         { NVMF_OPT_ERR,                 NULL                    }
 616 };
 617
 618 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 619                 const char *buf)
 620 {
 621         substring_t args[MAX_OPT_ARGS];
 622         char *options, *o, *p;
 623         int token, ret = 0;
 624         size_t nqnlen  = 0;
 625         int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
 626         uuid_t hostid;
 627
 628         /* Set defaults */
 629         opts->queue_size = NVMF_DEF_QUEUE_SIZE;
 630         opts->nr_io_queues = num_online_cpus();
 631         opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
 632         opts->kato = NVME_DEFAULT_KATO;
 633         opts->duplicate_connect = false;
 634         opts->hdr_digest = false;
 635         opts->data_digest = false;
 636         opts->tos = -1; /* < 0 == use transport default */
 637
 638         options = o = kstrdup(buf, GFP_KERNEL);
 639         if (!options)
 640                 return -ENOMEM;
 641
 642         uuid_gen(&hostid);
 643
 644         while ((p = strsep(&o, ",\n")) != NULL) {
 645                 if (!*p)
 646                         continue;
 647
 648                 token = match_token(p, opt_tokens, args);
 649                 opts->mask |= token;
 650                 switch (token) {
 651                 case NVMF_OPT_TRANSPORT:
 652                         p = match_strdup(args);
 653                         if (!p) {
 654                                 ret = -ENOMEM;
 655                                 goto out;
 656                         }
 657                         kfree(opts->transport);
 658                         opts->transport = p;
 659                         break;
 660                 case NVMF_OPT_NQN:
 661                         p = match_strdup(args);
 662                         if (!p) {
 663                                 ret = -ENOMEM;
 664                                 goto out;
 665                         }
 666                         kfree(opts->subsysnqn);
 667                         opts->subsysnqn = p;
 668                         nqnlen = strlen(opts->subsysnqn);
 669                         if (nqnlen >= NVMF_NQN_SIZE) {
 670                                 pr_err("%s needs to be < %d bytes\n",
 671                                         opts->subsysnqn, NVMF_NQN_SIZE);
 672                                 ret = -EINVAL;
 673                                 goto out;
 674                         }
 675                         opts->discovery_nqn =
 676                                 !(strcmp(opts->subsysnqn,
 677                                          NVME_DISC_SUBSYS_NAME));
 678                         break;
 679                 case NVMF_OPT_TRADDR:
 680                         p = match_strdup(args);
 681                         if (!p) {
 682                                 ret = -ENOMEM;
 683                                 goto out;
 684                         }
 685                         kfree(opts->traddr);
 686                         opts->traddr = p;
 687                         break;
 688                 case NVMF_OPT_TRSVCID:
 689                         p = match_strdup(args);
 690                         if (!p) {
 691                                 ret = -ENOMEM;
 692                                 goto out;
 693                         }
 694                         kfree(opts->trsvcid);
 695                         opts->trsvcid = p;
 696                         break;
 697                 case NVMF_OPT_QUEUE_SIZE:
 698                         if (match_int(args, &token)) {
 699                                 ret = -EINVAL;
 700                                 goto out;
 701                         }
 702                         if (token < NVMF_MIN_QUEUE_SIZE ||
 703                             token > NVMF_MAX_QUEUE_SIZE) {
 704                                 pr_err("Invalid queue_size %d\n", token);
 705                                 ret = -EINVAL;
 706                                 goto out;
 707                         }
 708                         opts->queue_size = token;
 709                         break;
 710                 case NVMF_OPT_NR_IO_QUEUES:
 711                         if (match_int(args, &token)) {
 712                                 ret = -EINVAL;
 713                                 goto out;
 714                         }
 715                         if (token <= 0) {
 716                                 pr_err("Invalid number of IOQs %d\n", token);
 717                                 ret = -EINVAL;
 718                                 goto out;
 719                         }
 720                         if (opts->discovery_nqn) {
 721                                 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
 722                                 break;
 723                         }
 724
 725                         opts->nr_io_queues = min_t(unsigned int,
 726                                         num_online_cpus(), token);
 727                         break;
 728                 case NVMF_OPT_KATO:
 729                         if (match_int(args, &token)) {
 730                                 ret = -EINVAL;
 731                                 goto out;
 732                         }
 733
 734                         if (token < 0) {
 735                                 pr_err("Invalid keep_alive_tmo %d\n", token);
 736                                 ret = -EINVAL;
 737                                 goto out;
 738                         } else if (token == 0 && !opts->discovery_nqn) {
 739                                 /* Allowed for debug */
 740                                 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
 741                         }
 742                         opts->kato = token;
 743                         break;
 744                 case NVMF_OPT_CTRL_LOSS_TMO:
 745                         if (match_int(args, &token)) {
 746                                 ret = -EINVAL;
 747                                 goto out;
 748                         }
 749
 750                         if (token < 0)
 751                                 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
 752                         ctrl_loss_tmo = token;
 753                         break;
 754                 case NVMF_OPT_HOSTNQN:
 755                         if (opts->host) {
 756                                 pr_err("hostnqn already user-assigned: %s\n",
 757                                        opts->host->nqn);
 758                                 ret = -EADDRINUSE;
 759                                 goto out;
 760                         }
 761                         p = match_strdup(args);
 762                         if (!p) {
 763                                 ret = -ENOMEM;
 764                                 goto out;
 765                         }
 766                         nqnlen = strlen(p);
 767                         if (nqnlen >= NVMF_NQN_SIZE) {
 768                                 pr_err("%s needs to be < %d bytes\n",
 769                                         p, NVMF_NQN_SIZE);
 770                                 kfree(p);
 771                                 ret = -EINVAL;
 772                                 goto out;
 773                         }
 774                         nvmf_host_put(opts->host);
 775                         opts->host = nvmf_host_add(p);
 776                         kfree(p);
 777                         if (!opts->host) {
 778                                 ret = -ENOMEM;
 779                                 goto out;
 780                         }
 781                         break;
 782                 case NVMF_OPT_RECONNECT_DELAY:
 783                         if (match_int(args, &token)) {
 784                                 ret = -EINVAL;
 785                                 goto out;
 786                         }
 787                         if (token <= 0) {
 788                                 pr_err("Invalid reconnect_delay %d\n", token);
 789                                 ret = -EINVAL;
 790                                 goto out;
 791                         }
 792                         opts->reconnect_delay = token;
 793                         break;
 794                 case NVMF_OPT_HOST_TRADDR:
 795                         p = match_strdup(args);
 796                         if (!p) {
 797                                 ret = -ENOMEM;
 798                                 goto out;
 799                         }
 800                         kfree(opts->host_traddr);
 801                         opts->host_traddr = p;
 802                         break;
 803                 case NVMF_OPT_HOST_ID:
 804                         p = match_strdup(args);
 805                         if (!p) {
 806                                 ret = -ENOMEM;
 807                                 goto out;
 808                         }
 809                         ret = uuid_parse(p, &hostid);
 810                         if (ret) {
 811                                 pr_err("Invalid hostid %s\n", p);
 812                                 ret = -EINVAL;
 813                                 kfree(p);
 814                                 goto out;
 815                         }
 816                         kfree(p);
 817                         break;
 818                 case NVMF_OPT_DUP_CONNECT:
 819                         opts->duplicate_connect = true;
 820                         break;
 821                 case NVMF_OPT_DISABLE_SQFLOW:
 822                         opts->disable_sqflow = true;
 823                         break;
 824                 case NVMF_OPT_HDR_DIGEST:
 825                         opts->hdr_digest = true;
 826                         break;
 827                 case NVMF_OPT_DATA_DIGEST:
 828                         opts->data_digest = true;
 829                         break;
 830                 case NVMF_OPT_NR_WRITE_QUEUES:
 831                         if (match_int(args, &token)) {
 832                                 ret = -EINVAL;
 833                                 goto out;
 834                         }
 835                         if (token <= 0) {
 836                                 pr_err("Invalid nr_write_queues %d\n", token);
 837                                 ret = -EINVAL;
 838                                 goto out;
 839                         }
 840                         opts->nr_write_queues = token;
 841                         break;
 842                 case NVMF_OPT_NR_POLL_QUEUES:
 843                         if (match_int(args, &token)) {
 844                                 ret = -EINVAL;
 845                                 goto out;
 846                         }
 847                         if (token <= 0) {
 848                                 pr_err("Invalid nr_poll_queues %d\n", token);
 849                                 ret = -EINVAL;
 850                                 goto out;
 851                         }
 852                         opts->nr_poll_queues = token;
 853                         break;
 854                 case NVMF_OPT_TOS:
 855                         if (match_int(args, &token)) {
 856                                 ret = -EINVAL;
 857                                 goto out;
 858                         }
 859                         if (token < 0) {
 860                                 pr_err("Invalid type of service %d\n", token);
 861                                 ret = -EINVAL;
 862                                 goto out;
 863                         }
 864                         if (token > 255) {
 865                                 pr_warn("Clamping type of service to 255\n");
 866                                 token = 255;
 867                         }
 868                         opts->tos = token;
 869                         break;
 870                 default:
 871                         pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
 872                                 p);
 873                         ret = -EINVAL;
 874                         goto out;
 875                 }
 876         }
 877
 878         if (opts->discovery_nqn) {
 879                 opts->nr_io_queues = 0;
 880                 opts->nr_write_queues = 0;
 881                 opts->nr_poll_queues = 0;
 882                 opts->duplicate_connect = true;
 883         }
 884         if (ctrl_loss_tmo < 0)
 885                 opts->max_reconnects = -1;
 886         else
 887                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
 888                                                 opts->reconnect_delay);
 889
 890         if (!opts->host) {
 891                 kref_get(&nvmf_default_host->ref);
 892                 opts->host = nvmf_default_host;
 893         }
 894
 895         uuid_copy(&opts->host->id, &hostid);
 896
 897 out:
 898         kfree(options);
 899         return ret;
 900 }
 901
 902 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
 903                 unsigned int required_opts)
 904 {
 905         if ((opts->mask & required_opts) != required_opts) {
 906                 int i;
 907
 908                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 909                         if ((opt_tokens[i].token & required_opts) &&
 910                             !(opt_tokens[i].token & opts->mask)) {
 911                                 pr_warn("missing parameter '%s'\n",
 912                                         opt_tokens[i].pattern);
 913                         }
 914                 }
 915
 916                 return -EINVAL;
 917         }
 918
 919         return 0;
 920 }
 921
 922 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
 923                 struct nvmf_ctrl_options *opts)
 924 {
 925         if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
 926             strcmp(opts->traddr, ctrl->opts->traddr) ||
 927             strcmp(opts->trsvcid, ctrl->opts->trsvcid))
 928                 return false;
 929
 930         /*
 931          * Checking the local address is rough. In most cases, none is specified
 932          * and the host port is selected by the stack.
 933          *
 934          * Assume no match if:
 935          * -  local address is specified and address is not the same
 936          * -  local address is not specified but remote is, or vice versa
 937          *    (admin using specific host_traddr when it matters).
 938          */
 939         if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
 940             (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
 941                 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
 942                         return false;
 943         } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
 944                    (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
 945                 return false;
 946         }
 947
 948         return true;
 949 }
 950 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
 951
 952 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
 953                 unsigned int allowed_opts)
 954 {
 955         if (opts->mask & ~allowed_opts) {
 956                 int i;
 957
 958                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 959                         if ((opt_tokens[i].token & opts->mask) &&
 960                             (opt_tokens[i].token & ~allowed_opts)) {
 961                                 pr_warn("invalid parameter '%s'\n",
 962                                         opt_tokens[i].pattern);
 963                         }
 964                 }
 965
 966                 return -EINVAL;
 967         }
 968
 969         return 0;
 970 }
 971
 972 void nvmf_free_options(struct nvmf_ctrl_options *opts)
 973 {
 974         nvmf_host_put(opts->host);
 975         kfree(opts->transport);
 976         kfree(opts->traddr);
 977         kfree(opts->trsvcid);
 978         kfree(opts->subsysnqn);
 979         kfree(opts->host_traddr);
 980         kfree(opts);
 981 }
 982 EXPORT_SYMBOL_GPL(nvmf_free_options);
 983
 984 #define NVMF_REQUIRED_OPTS      (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
 985 #define NVMF_ALLOWED_OPTS       (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
 986                                  NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
 987                                  NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
 988                                  NVMF_OPT_DISABLE_SQFLOW)
 989
 990 static struct nvme_ctrl *
 991 nvmf_create_ctrl(struct device *dev, const char *buf)
 992 {
 993         struct nvmf_ctrl_options *opts;
 994         struct nvmf_transport_ops *ops;
 995         struct nvme_ctrl *ctrl;
 996         int ret;
 997
 998         opts = kzalloc(sizeof(*opts), GFP_KERNEL);
 999         if (!opts)
1000                 return ERR_PTR(-ENOMEM);
1001
1002         ret = nvmf_parse_options(opts, buf);
1003         if (ret)
1004                 goto out_free_opts;
1005
1006
1007         request_module("nvme-%s", opts->transport);
1008
1009         /*
1010          * Check the generic options first as we need a valid transport for
1011          * the lookup below.  Then clear the generic flags so that transport
1012          * drivers don't have to care about them.
1013          */
1014         ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1015         if (ret)
1016                 goto out_free_opts;
1017         opts->mask &= ~NVMF_REQUIRED_OPTS;
1018
1019         down_read(&nvmf_transports_rwsem);
1020         ops = nvmf_lookup_transport(opts);
1021         if (!ops) {
1022                 pr_info("no handler found for transport %s.\n",
1023                         opts->transport);
1024                 ret = -EINVAL;
1025                 goto out_unlock;
1026         }
1027
1028         if (!try_module_get(ops->module)) {
1029                 ret = -EBUSY;
1030                 goto out_unlock;
1031         }
1032         up_read(&nvmf_transports_rwsem);
1033
1034         ret = nvmf_check_required_opts(opts, ops->required_opts);
1035         if (ret)
1036                 goto out_module_put;
1037         ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1038                                 ops->allowed_opts | ops->required_opts);
1039         if (ret)
1040                 goto out_module_put;
1041
1042         ctrl = ops->create_ctrl(dev, opts);
1043         if (IS_ERR(ctrl)) {
1044                 ret = PTR_ERR(ctrl);
1045                 goto out_module_put;
1046         }
1047
1048         module_put(ops->module);
1049         return ctrl;
1050
1051 out_module_put:
1052         module_put(ops->module);
1053         goto out_free_opts;
1054 out_unlock:
1055         up_read(&nvmf_transports_rwsem);
1056 out_free_opts:
1057         nvmf_free_options(opts);
1058         return ERR_PTR(ret);
1059 }
1060
1061 static struct class *nvmf_class;
1062 static struct device *nvmf_device;
1063 static DEFINE_MUTEX(nvmf_dev_mutex);
1064
1065 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1066                 size_t count, loff_t *pos)
1067 {
1068         struct seq_file *seq_file = file->private_data;
1069         struct nvme_ctrl *ctrl;
1070         const char *buf;
1071         int ret = 0;
1072
1073         if (count > PAGE_SIZE)
1074                 return -ENOMEM;
1075
1076         buf = memdup_user_nul(ubuf, count);
1077         if (IS_ERR(buf))
1078                 return PTR_ERR(buf);
1079
1080         mutex_lock(&nvmf_dev_mutex);
1081         if (seq_file->private) {
1082                 ret = -EINVAL;
1083                 goto out_unlock;
1084         }
1085
1086         ctrl = nvmf_create_ctrl(nvmf_device, buf);
1087         if (IS_ERR(ctrl)) {
1088                 ret = PTR_ERR(ctrl);
1089                 goto out_unlock;
1090         }
1091
1092         seq_file->private = ctrl;
1093
1094 out_unlock:
1095         mutex_unlock(&nvmf_dev_mutex);
1096         kfree(buf);
1097         return ret ? ret : count;
1098 }
1099
1100 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1101 {
1102         struct nvme_ctrl *ctrl;
1103         int ret = 0;
1104
1105         mutex_lock(&nvmf_dev_mutex);
1106         ctrl = seq_file->private;
1107         if (!ctrl) {
1108                 ret = -EINVAL;
1109                 goto out_unlock;
1110         }
1111
1112         seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1113                         ctrl->instance, ctrl->cntlid);
1114
1115 out_unlock:
1116         mutex_unlock(&nvmf_dev_mutex);
1117         return ret;
1118 }
1119
1120 static int nvmf_dev_open(struct inode *inode, struct file *file)
1121 {
1122         /*
1123          * The miscdevice code initializes file->private_data, but doesn't
1124          * make use of it later.
1125          */
1126         file->private_data = NULL;
1127         return single_open(file, nvmf_dev_show, NULL);
1128 }
1129
1130 static int nvmf_dev_release(struct inode *inode, struct file *file)
1131 {
1132         struct seq_file *seq_file = file->private_data;
1133         struct nvme_ctrl *ctrl = seq_file->private;
1134
1135         if (ctrl)
1136                 nvme_put_ctrl(ctrl);
1137         return single_release(inode, file);
1138 }
1139
1140 static const struct file_operations nvmf_dev_fops = {
1141         .owner          = THIS_MODULE,
1142         .write          = nvmf_dev_write,
1143         .read           = seq_read,
1144         .open           = nvmf_dev_open,
1145         .release        = nvmf_dev_release,
1146 };
1147
1148 static struct miscdevice nvmf_misc = {
1149         .minor          = MISC_DYNAMIC_MINOR,
1150         .name           = "nvme-fabrics",
1151         .fops           = &nvmf_dev_fops,
1152 };
1153
1154 static int __init nvmf_init(void)
1155 {
1156         int ret;
1157
1158         nvmf_default_host = nvmf_host_default();
1159         if (!nvmf_default_host)
1160                 return -ENOMEM;
1161
1162         nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1163         if (IS_ERR(nvmf_class)) {
1164                 pr_err("couldn't register class nvme-fabrics\n");
1165                 ret = PTR_ERR(nvmf_class);
1166                 goto out_free_host;
1167         }
1168
1169         nvmf_device =
1170                 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1171         if (IS_ERR(nvmf_device)) {
1172                 pr_err("couldn't create nvme-fabris device!\n");
1173                 ret = PTR_ERR(nvmf_device);
1174                 goto out_destroy_class;
1175         }
1176
1177         ret = misc_register(&nvmf_misc);
1178         if (ret) {
1179                 pr_err("couldn't register misc device: %d\n", ret);
1180                 goto out_destroy_device;
1181         }
1182
1183         return 0;
1184
1185 out_destroy_device:
1186         device_destroy(nvmf_class, MKDEV(0, 0));
1187 out_destroy_class:
1188         class_destroy(nvmf_class);
1189 out_free_host:
1190         nvmf_host_put(nvmf_default_host);
1191         return ret;
1192 }
1193
1194 static void __exit nvmf_exit(void)
1195 {
1196         misc_deregister(&nvmf_misc);
1197         device_destroy(nvmf_class, MKDEV(0, 0));
1198         class_destroy(nvmf_class);
1199         nvmf_host_put(nvmf_default_host);
1200
1201         BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1202         BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1203         BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1204         BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1205         BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1206 }
1207
1208 MODULE_LICENSE("GPL v2");
1209
1210 module_init(nvmf_init);
1211 module_exit(nvmf_exit);