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
 396         /*
 397          * fabrics spec sets a minimum of depth 32 for admin queue,
 398          * so set the queue with this depth always until
 399          * justification otherwise.
 400          */
 401         cmd.connect.sqsize = cpu_to_le16(NVMF_AQ_DEPTH - 1);
 402
 403         /*
 404          * Set keep-alive timeout in seconds granularity (ms * 1000)
 405          * and add a grace period for controller kato enforcement
 406          */
 407         cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
 408                 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
 409
 410         data = kzalloc(sizeof(*data), GFP_KERNEL);
 411         if (!data)
 412                 return -ENOMEM;
 413
 414         uuid_copy(&data->hostid, &ctrl->opts->host->id);
 415         data->cntlid = cpu_to_le16(0xffff);
 416         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 417         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 418
 419         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res,
 420                         data, sizeof(*data), 0, NVME_QID_ANY, 1,
 421                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
 422         if (ret) {
 423                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
 424                                        &cmd, data);
 425                 goto out_free_data;
 426         }
 427
 428         ctrl->cntlid = le16_to_cpu(res.u16);
 429
 430 out_free_data:
 431         kfree(data);
 432         return ret;
 433 }
 434 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
 435
 436 /**
 437  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
 438  *                           API function.
 439  * @ctrl:       Host nvme controller instance used to establish an
 440  *              NVMe I/O queue connection to the already allocated NVMe
 441  *              controller on the target system.
 442  * @qid:        NVMe I/O queue number for the new I/O connection between
 443  *              host and target (note qid == 0 is illegal as this is
 444  *              the Admin queue, per NVMe standard).
 445  *
 446  * This function issues a fabrics-protocol connection
 447  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
 448  * between the host system device and the allocated NVMe controller
 449  * on the target system.
 450  *
 451  * Return:
 452  *      0: success
 453  *      > 0: NVMe error status code
 454  *      < 0: Linux errno error code
 455  */
 456 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
 457 {
 458         struct nvme_command cmd;
 459         struct nvmf_connect_data *data;
 460         union nvme_result res;
 461         int ret;
 462
 463         memset(&cmd, 0, sizeof(cmd));
 464         cmd.connect.opcode = nvme_fabrics_command;
 465         cmd.connect.fctype = nvme_fabrics_type_connect;
 466         cmd.connect.qid = cpu_to_le16(qid);
 467         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
 468
 469         data = kzalloc(sizeof(*data), GFP_KERNEL);
 470         if (!data)
 471                 return -ENOMEM;
 472
 473         uuid_copy(&data->hostid, &ctrl->opts->host->id);
 474         data->cntlid = cpu_to_le16(ctrl->cntlid);
 475         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 476         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 477
 478         ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
 479                         data, sizeof(*data), 0, qid, 1,
 480                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
 481         if (ret) {
 482                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
 483                                        &cmd, data);
 484         }
 485         kfree(data);
 486         return ret;
 487 }
 488 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
 489
 490 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
 491 {
 492         if (ctrl->opts->max_reconnects != -1 &&
 493             ctrl->nr_reconnects < ctrl->opts->max_reconnects)
 494                 return true;
 495
 496         return false;
 497 }
 498 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
 499
 500 /**
 501  * nvmf_register_transport() - NVMe Fabrics Library registration function.
 502  * @ops:        Transport ops instance to be registered to the
 503  *              common fabrics library.
 504  *
 505  * API function that registers the type of specific transport fabric
 506  * being implemented to the common NVMe fabrics library. Part of
 507  * the overall init sequence of starting up a fabrics driver.
 508  */
 509 int nvmf_register_transport(struct nvmf_transport_ops *ops)
 510 {
 511         if (!ops->create_ctrl)
 512                 return -EINVAL;
 513
 514         mutex_lock(&nvmf_transports_mutex);
 515         list_add_tail(&ops->entry, &nvmf_transports);
 516         mutex_unlock(&nvmf_transports_mutex);
 517
 518         return 0;
 519 }
 520 EXPORT_SYMBOL_GPL(nvmf_register_transport);
 521
 522 /**
 523  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
 524  * @ops:        Transport ops instance to be unregistered from the
 525  *              common fabrics library.
 526  *
 527  * Fabrics API function that unregisters the type of specific transport
 528  * fabric being implemented from the common NVMe fabrics library.
 529  * Part of the overall exit sequence of unloading the implemented driver.
 530  */
 531 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
 532 {
 533         mutex_lock(&nvmf_transports_mutex);
 534         list_del(&ops->entry);
 535         mutex_unlock(&nvmf_transports_mutex);
 536 }
 537 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
 538
 539 static struct nvmf_transport_ops *nvmf_lookup_transport(
 540                 struct nvmf_ctrl_options *opts)
 541 {
 542         struct nvmf_transport_ops *ops;
 543
 544         lockdep_assert_held(&nvmf_transports_mutex);
 545
 546         list_for_each_entry(ops, &nvmf_transports, entry) {
 547                 if (strcmp(ops->name, opts->transport) == 0)
 548                         return ops;
 549         }
 550
 551         return NULL;
 552 }
 553
 554 static const match_table_t opt_tokens = {
 555         { NVMF_OPT_TRANSPORT,           "transport=%s"          },
 556         { NVMF_OPT_TRADDR,              "traddr=%s"             },
 557         { NVMF_OPT_TRSVCID,             "trsvcid=%s"            },
 558         { NVMF_OPT_NQN,                 "nqn=%s"                },
 559         { NVMF_OPT_QUEUE_SIZE,          "queue_size=%d"         },
 560         { NVMF_OPT_NR_IO_QUEUES,        "nr_io_queues=%d"       },
 561         { NVMF_OPT_RECONNECT_DELAY,     "reconnect_delay=%d"    },
 562         { NVMF_OPT_CTRL_LOSS_TMO,       "ctrl_loss_tmo=%d"      },
 563         { NVMF_OPT_KATO,                "keep_alive_tmo=%d"     },
 564         { NVMF_OPT_HOSTNQN,             "hostnqn=%s"            },
 565         { NVMF_OPT_HOST_TRADDR,         "host_traddr=%s"        },
 566         { NVMF_OPT_HOST_ID,             "hostid=%s"             },
 567         { NVMF_OPT_ERR,                 NULL                    }
 568 };
 569
 570 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 571                 const char *buf)
 572 {
 573         substring_t args[MAX_OPT_ARGS];
 574         char *options, *o, *p;
 575         int token, ret = 0;
 576         size_t nqnlen  = 0;
 577         int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
 578         uuid_t hostid;
 579
 580         /* Set defaults */
 581         opts->queue_size = NVMF_DEF_QUEUE_SIZE;
 582         opts->nr_io_queues = num_online_cpus();
 583         opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
 584
 585         options = o = kstrdup(buf, GFP_KERNEL);
 586         if (!options)
 587                 return -ENOMEM;
 588
 589         uuid_gen(&hostid);
 590
 591         while ((p = strsep(&o, ",\n")) != NULL) {
 592                 if (!*p)
 593                         continue;
 594
 595                 token = match_token(p, opt_tokens, args);
 596                 opts->mask |= token;
 597                 switch (token) {
 598                 case NVMF_OPT_TRANSPORT:
 599                         p = match_strdup(args);
 600                         if (!p) {
 601                                 ret = -ENOMEM;
 602                                 goto out;
 603                         }
 604                         opts->transport = p;
 605                         break;
 606                 case NVMF_OPT_NQN:
 607                         p = match_strdup(args);
 608                         if (!p) {
 609                                 ret = -ENOMEM;
 610                                 goto out;
 611                         }
 612                         opts->subsysnqn = p;
 613                         nqnlen = strlen(opts->subsysnqn);
 614                         if (nqnlen >= NVMF_NQN_SIZE) {
 615                                 pr_err("%s needs to be < %d bytes\n",
 616                                         opts->subsysnqn, NVMF_NQN_SIZE);
 617                                 ret = -EINVAL;
 618                                 goto out;
 619                         }
 620                         opts->discovery_nqn =
 621                                 !(strcmp(opts->subsysnqn,
 622                                          NVME_DISC_SUBSYS_NAME));
 623                         if (opts->discovery_nqn)
 624                                 opts->nr_io_queues = 0;
 625                         break;
 626                 case NVMF_OPT_TRADDR:
 627                         p = match_strdup(args);
 628                         if (!p) {
 629                                 ret = -ENOMEM;
 630                                 goto out;
 631                         }
 632                         opts->traddr = p;
 633                         break;
 634                 case NVMF_OPT_TRSVCID:
 635                         p = match_strdup(args);
 636                         if (!p) {
 637                                 ret = -ENOMEM;
 638                                 goto out;
 639                         }
 640                         opts->trsvcid = p;
 641                         break;
 642                 case NVMF_OPT_QUEUE_SIZE:
 643                         if (match_int(args, &token)) {
 644                                 ret = -EINVAL;
 645                                 goto out;
 646                         }
 647                         if (token < NVMF_MIN_QUEUE_SIZE ||
 648                             token > NVMF_MAX_QUEUE_SIZE) {
 649                                 pr_err("Invalid queue_size %d\n", token);
 650                                 ret = -EINVAL;
 651                                 goto out;
 652                         }
 653                         opts->queue_size = token;
 654                         break;
 655                 case NVMF_OPT_NR_IO_QUEUES:
 656                         if (match_int(args, &token)) {
 657                                 ret = -EINVAL;
 658                                 goto out;
 659                         }
 660                         if (token <= 0) {
 661                                 pr_err("Invalid number of IOQs %d\n", token);
 662                                 ret = -EINVAL;
 663                                 goto out;
 664                         }
 665                         opts->nr_io_queues = min_t(unsigned int,
 666                                         num_online_cpus(), token);
 667                         break;
 668                 case NVMF_OPT_KATO:
 669                         if (match_int(args, &token)) {
 670                                 ret = -EINVAL;
 671                                 goto out;
 672                         }
 673
 674                         if (opts->discovery_nqn) {
 675                                 pr_err("Discovery controllers cannot accept keep_alive_tmo != 0\n");
 676                                 ret = -EINVAL;
 677                                 goto out;
 678                         }
 679
 680                         if (token < 0) {
 681                                 pr_err("Invalid keep_alive_tmo %d\n", token);
 682                                 ret = -EINVAL;
 683                                 goto out;
 684                         } else if (token == 0) {
 685                                 /* Allowed for debug */
 686                                 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
 687                         }
 688                         opts->kato = token;
 689                         break;
 690                 case NVMF_OPT_CTRL_LOSS_TMO:
 691                         if (match_int(args, &token)) {
 692                                 ret = -EINVAL;
 693                                 goto out;
 694                         }
 695
 696                         if (token < 0)
 697                                 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
 698                         ctrl_loss_tmo = token;
 699                         break;
 700                 case NVMF_OPT_HOSTNQN:
 701                         if (opts->host) {
 702                                 pr_err("hostnqn already user-assigned: %s\n",
 703                                        opts->host->nqn);
 704                                 ret = -EADDRINUSE;
 705                                 goto out;
 706                         }
 707                         p = match_strdup(args);
 708                         if (!p) {
 709                                 ret = -ENOMEM;
 710                                 goto out;
 711                         }
 712                         nqnlen = strlen(p);
 713                         if (nqnlen >= NVMF_NQN_SIZE) {
 714                                 pr_err("%s needs to be < %d bytes\n",
 715                                         p, NVMF_NQN_SIZE);
 716                                 kfree(p);
 717                                 ret = -EINVAL;
 718                                 goto out;
 719                         }
 720                         opts->host = nvmf_host_add(p);
 721                         kfree(p);
 722                         if (!opts->host) {
 723                                 ret = -ENOMEM;
 724                                 goto out;
 725                         }
 726                         break;
 727                 case NVMF_OPT_RECONNECT_DELAY:
 728                         if (match_int(args, &token)) {
 729                                 ret = -EINVAL;
 730                                 goto out;
 731                         }
 732                         if (token <= 0) {
 733                                 pr_err("Invalid reconnect_delay %d\n", token);
 734                                 ret = -EINVAL;
 735                                 goto out;
 736                         }
 737                         opts->reconnect_delay = token;
 738                         break;
 739                 case NVMF_OPT_HOST_TRADDR:
 740                         p = match_strdup(args);
 741                         if (!p) {
 742                                 ret = -ENOMEM;
 743                                 goto out;
 744                         }
 745                         opts->host_traddr = p;
 746                         break;
 747                 case NVMF_OPT_HOST_ID:
 748                         p = match_strdup(args);
 749                         if (!p) {
 750                                 ret = -ENOMEM;
 751                                 goto out;
 752                         }
 753                         if (uuid_parse(p, &hostid)) {
 754                                 ret = -EINVAL;
 755                                 goto out;
 756                         }
 757                         break;
 758                 default:
 759                         pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
 760                                 p);
 761                         ret = -EINVAL;
 762                         goto out;
 763                 }
 764         }
 765
 766         if (ctrl_loss_tmo < 0)
 767                 opts->max_reconnects = -1;
 768         else
 769                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
 770                                                 opts->reconnect_delay);
 771
 772         if (!opts->host) {
 773                 kref_get(&nvmf_default_host->ref);
 774                 opts->host = nvmf_default_host;
 775         }
 776
 777         uuid_copy(&opts->host->id, &hostid);
 778
 779 out:
 780         if (!opts->discovery_nqn && !opts->kato)
 781                 opts->kato = NVME_DEFAULT_KATO;
 782         kfree(options);
 783         return ret;
 784 }
 785
 786 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
 787                 unsigned int required_opts)
 788 {
 789         if ((opts->mask & required_opts) != required_opts) {
 790                 int i;
 791
 792                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 793                         if ((opt_tokens[i].token & required_opts) &&
 794                             !(opt_tokens[i].token & opts->mask)) {
 795                                 pr_warn("missing parameter '%s'\n",
 796                                         opt_tokens[i].pattern);
 797                         }
 798                 }
 799
 800                 return -EINVAL;
 801         }
 802
 803         return 0;
 804 }
 805
 806 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
 807                 unsigned int allowed_opts)
 808 {
 809         if (opts->mask & ~allowed_opts) {
 810                 int i;
 811
 812                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 813                         if (opt_tokens[i].token & ~allowed_opts) {
 814                                 pr_warn("invalid parameter '%s'\n",
 815                                         opt_tokens[i].pattern);
 816                         }
 817                 }
 818
 819                 return -EINVAL;
 820         }
 821
 822         return 0;
 823 }
 824
 825 void nvmf_free_options(struct nvmf_ctrl_options *opts)
 826 {
 827         nvmf_host_put(opts->host);
 828         kfree(opts->transport);
 829         kfree(opts->traddr);
 830         kfree(opts->trsvcid);
 831         kfree(opts->subsysnqn);
 832         kfree(opts->host_traddr);
 833         kfree(opts);
 834 }
 835 EXPORT_SYMBOL_GPL(nvmf_free_options);
 836
 837 #define NVMF_REQUIRED_OPTS      (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
 838 #define NVMF_ALLOWED_OPTS       (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
 839                                  NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
 840                                  NVMF_OPT_HOST_ID)
 841
 842 static struct nvme_ctrl *
 843 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
 844 {
 845         struct nvmf_ctrl_options *opts;
 846         struct nvmf_transport_ops *ops;
 847         struct nvme_ctrl *ctrl;
 848         int ret;
 849
 850         opts = kzalloc(sizeof(*opts), GFP_KERNEL);
 851         if (!opts)
 852                 return ERR_PTR(-ENOMEM);
 853
 854         ret = nvmf_parse_options(opts, buf);
 855         if (ret)
 856                 goto out_free_opts;
 857
 858         /*
 859          * Check the generic options first as we need a valid transport for
 860          * the lookup below.  Then clear the generic flags so that transport
 861          * drivers don't have to care about them.
 862          */
 863         ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
 864         if (ret)
 865                 goto out_free_opts;
 866         opts->mask &= ~NVMF_REQUIRED_OPTS;
 867
 868         mutex_lock(&nvmf_transports_mutex);
 869         ops = nvmf_lookup_transport(opts);
 870         if (!ops) {
 871                 pr_info("no handler found for transport %s.\n",
 872                         opts->transport);
 873                 ret = -EINVAL;
 874                 goto out_unlock;
 875         }
 876
 877         ret = nvmf_check_required_opts(opts, ops->required_opts);
 878         if (ret)
 879                 goto out_unlock;
 880         ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
 881                                 ops->allowed_opts | ops->required_opts);
 882         if (ret)
 883                 goto out_unlock;
 884
 885         ctrl = ops->create_ctrl(dev, opts);
 886         if (IS_ERR(ctrl)) {
 887                 ret = PTR_ERR(ctrl);
 888                 goto out_unlock;
 889         }
 890
 891         mutex_unlock(&nvmf_transports_mutex);
 892         return ctrl;
 893
 894 out_unlock:
 895         mutex_unlock(&nvmf_transports_mutex);
 896 out_free_opts:
 897         nvmf_free_options(opts);
 898         return ERR_PTR(ret);
 899 }
 900
 901 static struct class *nvmf_class;
 902 static struct device *nvmf_device;
 903 static DEFINE_MUTEX(nvmf_dev_mutex);
 904
 905 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
 906                 size_t count, loff_t *pos)
 907 {
 908         struct seq_file *seq_file = file->private_data;
 909         struct nvme_ctrl *ctrl;
 910         const char *buf;
 911         int ret = 0;
 912
 913         if (count > PAGE_SIZE)
 914                 return -ENOMEM;
 915
 916         buf = memdup_user_nul(ubuf, count);
 917         if (IS_ERR(buf))
 918                 return PTR_ERR(buf);
 919
 920         mutex_lock(&nvmf_dev_mutex);
 921         if (seq_file->private) {
 922                 ret = -EINVAL;
 923                 goto out_unlock;
 924         }
 925
 926         ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
 927         if (IS_ERR(ctrl)) {
 928                 ret = PTR_ERR(ctrl);
 929                 goto out_unlock;
 930         }
 931
 932         seq_file->private = ctrl;
 933
 934 out_unlock:
 935         mutex_unlock(&nvmf_dev_mutex);
 936         kfree(buf);
 937         return ret ? ret : count;
 938 }
 939
 940 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
 941 {
 942         struct nvme_ctrl *ctrl;
 943         int ret = 0;
 944
 945         mutex_lock(&nvmf_dev_mutex);
 946         ctrl = seq_file->private;
 947         if (!ctrl) {
 948                 ret = -EINVAL;
 949                 goto out_unlock;
 950         }
 951
 952         seq_printf(seq_file, "instance=%d,cntlid=%d\n",
 953                         ctrl->instance, ctrl->cntlid);
 954
 955 out_unlock:
 956         mutex_unlock(&nvmf_dev_mutex);
 957         return ret;
 958 }
 959
 960 static int nvmf_dev_open(struct inode *inode, struct file *file)
 961 {
 962         /*
 963          * The miscdevice code initializes file->private_data, but doesn't
 964          * make use of it later.
 965          */
 966         file->private_data = NULL;
 967         return single_open(file, nvmf_dev_show, NULL);
 968 }
 969
 970 static int nvmf_dev_release(struct inode *inode, struct file *file)
 971 {
 972         struct seq_file *seq_file = file->private_data;
 973         struct nvme_ctrl *ctrl = seq_file->private;
 974
 975         if (ctrl)
 976                 nvme_put_ctrl(ctrl);
 977         return single_release(inode, file);
 978 }
 979
 980 static const struct file_operations nvmf_dev_fops = {
 981         .owner          = THIS_MODULE,
 982         .write          = nvmf_dev_write,
 983         .read           = seq_read,
 984         .open           = nvmf_dev_open,
 985         .release        = nvmf_dev_release,
 986 };
 987
 988 static struct miscdevice nvmf_misc = {
 989         .minor          = MISC_DYNAMIC_MINOR,
 990         .name           = "nvme-fabrics",
 991         .fops           = &nvmf_dev_fops,
 992 };
 993
 994 static int __init nvmf_init(void)
 995 {
 996         int ret;
 997
 998         nvmf_default_host = nvmf_host_default();
 999         if (!nvmf_default_host)
1000                 return -ENOMEM;
1001
1002         nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1003         if (IS_ERR(nvmf_class)) {
1004                 pr_err("couldn't register class nvme-fabrics\n");
1005                 ret = PTR_ERR(nvmf_class);
1006                 goto out_free_host;
1007         }
1008
1009         nvmf_device =
1010                 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1011         if (IS_ERR(nvmf_device)) {
1012                 pr_err("couldn't create nvme-fabris device!\n");
1013                 ret = PTR_ERR(nvmf_device);
1014                 goto out_destroy_class;
1015         }
1016
1017         ret = misc_register(&nvmf_misc);
1018         if (ret) {
1019                 pr_err("couldn't register misc device: %d\n", ret);
1020                 goto out_destroy_device;
1021         }
1022
1023         return 0;
1024
1025 out_destroy_device:
1026         device_destroy(nvmf_class, MKDEV(0, 0));
1027 out_destroy_class:
1028         class_destroy(nvmf_class);
1029 out_free_host:
1030         nvmf_host_put(nvmf_default_host);
1031         return ret;
1032 }
1033
1034 static void __exit nvmf_exit(void)
1035 {
1036         misc_deregister(&nvmf_misc);
1037         device_destroy(nvmf_class, MKDEV(0, 0));
1038         class_destroy(nvmf_class);
1039         nvmf_host_put(nvmf_default_host);
1040
1041         BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1042         BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1043         BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1044         BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1045 }
1046
1047 MODULE_LICENSE("GPL v2");
1048
1049 module_init(nvmf_init);
1050 module_exit(nvmf_exit);