drivers/nvme/host/fabrics.c

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