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