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                 goto out_unlock;
  52
  53         host = kmalloc(sizeof(*host), GFP_KERNEL);
  54         if (!host)
  55                 goto out_unlock;
  56
  57         kref_init(&host->ref);
  58         memcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
  59         uuid_le_gen(&host->id);
  60
  61         list_add_tail(&host->list, &nvmf_hosts);
  62 out_unlock:
  63         mutex_unlock(&nvmf_hosts_mutex);
  64         return host;
  65 }
  66
  67 static struct nvmf_host *nvmf_host_default(void)
  68 {
  69         struct nvmf_host *host;
  70
  71         host = kmalloc(sizeof(*host), GFP_KERNEL);
  72         if (!host)
  73                 return NULL;
  74
  75         kref_init(&host->ref);
  76         uuid_le_gen(&host->id);
  77         snprintf(host->nqn, NVMF_NQN_SIZE,
  78                 "nqn.2014-08.org.nvmexpress:NVMf:uuid:%pUl", &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         kfree(host);
  92 }
  93
  94 static void nvmf_host_put(struct nvmf_host *host)
  95 {
  96         if (host)
  97                 kref_put(&host->ref, nvmf_host_destroy);
  98 }
  99
 100 /**
 101  * nvmf_get_address() -  Get address/port
 102  * @ctrl:       Host NVMe controller instance which we got the address
 103  * @buf:        OUTPUT parameter that will contain the address/port
 104  * @size:       buffer size
 105  */
 106 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
 107 {
 108         return snprintf(buf, size, "traddr=%s,trsvcid=%s\n",
 109                         ctrl->opts->traddr, ctrl->opts->trsvcid);
 110 }
 111 EXPORT_SYMBOL_GPL(nvmf_get_address);
 112
 113 /**
 114  * nvmf_get_subsysnqn() - Get subsystem NQN
 115  * @ctrl:       Host NVMe controller instance which we got the NQN
 116  */
 117 const char *nvmf_get_subsysnqn(struct nvme_ctrl *ctrl)
 118 {
 119         return ctrl->opts->subsysnqn;
 120 }
 121 EXPORT_SYMBOL_GPL(nvmf_get_subsysnqn);
 122
 123 /**
 124  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
 125  * @ctrl:       Host NVMe controller instance maintaining the admin
 126  *              queue used to submit the property read command to
 127  *              the allocated NVMe controller resource on the target system.
 128  * @off:        Starting offset value of the targeted property
 129  *              register (see the fabrics section of the NVMe standard).
 130  * @val:        OUTPUT parameter that will contain the value of
 131  *              the property after a successful read.
 132  *
 133  * Used by the host system to retrieve a 32-bit capsule property value
 134  * from an NVMe controller on the target system.
 135  *
 136  * ("Capsule property" is an "PCIe register concept" applied to the
 137  * NVMe fabrics space.)
 138  *
 139  * Return:
 140  *      0: successful read
 141  *      > 0: NVMe error status code
 142  *      < 0: Linux errno error code
 143  */
 144 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
 145 {
 146         struct nvme_command cmd;
 147         struct nvme_completion cqe;
 148         int ret;
 149
 150         memset(&cmd, 0, sizeof(cmd));
 151         cmd.prop_get.opcode = nvme_fabrics_command;
 152         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
 153         cmd.prop_get.offset = cpu_to_le32(off);
 154
 155         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe, NULL, 0, 0,
 156                         NVME_QID_ANY, 0, 0);
 157
 158         if (ret >= 0)
 159                 *val = le64_to_cpu(cqe.result64);
 160         if (unlikely(ret != 0))
 161                 dev_err(ctrl->device,
 162                         "Property Get error: %d, offset %#x\n",
 163                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
 164
 165         return ret;
 166 }
 167 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
 168
 169 /**
 170  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
 171  * @ctrl:       Host NVMe controller instance maintaining the admin
 172  *              queue used to submit the property read command to
 173  *              the allocated controller resource on the target system.
 174  * @off:        Starting offset value of the targeted property
 175  *              register (see the fabrics section of the NVMe standard).
 176  * @val:        OUTPUT parameter that will contain the value of
 177  *              the property after a successful read.
 178  *
 179  * Used by the host system to retrieve a 64-bit capsule property value
 180  * from an NVMe controller on the target system.
 181  *
 182  * ("Capsule property" is an "PCIe register concept" applied to the
 183  * NVMe fabrics space.)
 184  *
 185  * Return:
 186  *      0: successful read
 187  *      > 0: NVMe error status code
 188  *      < 0: Linux errno error code
 189  */
 190 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
 191 {
 192         struct nvme_command cmd;
 193         struct nvme_completion cqe;
 194         int ret;
 195
 196         memset(&cmd, 0, sizeof(cmd));
 197         cmd.prop_get.opcode = nvme_fabrics_command;
 198         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
 199         cmd.prop_get.attrib = 1;
 200         cmd.prop_get.offset = cpu_to_le32(off);
 201
 202         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe, NULL, 0, 0,
 203                         NVME_QID_ANY, 0, 0);
 204
 205         if (ret >= 0)
 206                 *val = le64_to_cpu(cqe.result64);
 207         if (unlikely(ret != 0))
 208                 dev_err(ctrl->device,
 209                         "Property Get error: %d, offset %#x\n",
 210                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
 211         return ret;
 212 }
 213 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
 214
 215 /**
 216  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
 217  * @ctrl:       Host NVMe controller instance maintaining the admin
 218  *              queue used to submit the property read command to
 219  *              the allocated NVMe controller resource on the target system.
 220  * @off:        Starting offset value of the targeted property
 221  *              register (see the fabrics section of the NVMe standard).
 222  * @val:        Input parameter that contains the value to be
 223  *              written to the property.
 224  *
 225  * Used by the NVMe host system to write a 32-bit capsule property value
 226  * to an NVMe controller on the target system.
 227  *
 228  * ("Capsule property" is an "PCIe register concept" applied to the
 229  * NVMe fabrics space.)
 230  *
 231  * Return:
 232  *      0: successful write
 233  *      > 0: NVMe error status code
 234  *      < 0: Linux errno error code
 235  */
 236 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
 237 {
 238         struct nvme_command cmd;
 239         int ret;
 240
 241         memset(&cmd, 0, sizeof(cmd));
 242         cmd.prop_set.opcode = nvme_fabrics_command;
 243         cmd.prop_set.fctype = nvme_fabrics_type_property_set;
 244         cmd.prop_set.attrib = 0;
 245         cmd.prop_set.offset = cpu_to_le32(off);
 246         cmd.prop_set.value = cpu_to_le64(val);
 247
 248         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
 249                         NVME_QID_ANY, 0, 0);
 250         if (unlikely(ret))
 251                 dev_err(ctrl->device,
 252                         "Property Set error: %d, offset %#x\n",
 253                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
 254         return ret;
 255 }
 256 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
 257
 258 /**
 259  * nvmf_log_connect_error() - Error-parsing-diagnostic print
 260  * out function for connect() errors.
 261  *
 262  * @ctrl: the specific /dev/nvmeX device that had the error.
 263  *
 264  * @errval: Error code to be decoded in a more human-friendly
 265  *          printout.
 266  *
 267  * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
 268  *
 269  * @cmd: This is the SQE portion of a submission capsule.
 270  *
 271  * @data: This is the "Data" portion of a submission capsule.
 272  */
 273 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
 274                 int errval, int offset, struct nvme_command *cmd,
 275                 struct nvmf_connect_data *data)
 276 {
 277         int err_sctype = errval & (~NVME_SC_DNR);
 278
 279         switch (err_sctype) {
 280
 281         case (NVME_SC_CONNECT_INVALID_PARAM):
 282                 if (offset >> 16) {
 283                         char *inv_data = "Connect Invalid Data Parameter";
 284
 285                         switch (offset & 0xffff) {
 286                         case (offsetof(struct nvmf_connect_data, cntlid)):
 287                                 dev_err(ctrl->device,
 288                                         "%s, cntlid: %d\n",
 289                                         inv_data, data->cntlid);
 290                                 break;
 291                         case (offsetof(struct nvmf_connect_data, hostnqn)):
 292                                 dev_err(ctrl->device,
 293                                         "%s, hostnqn \"%s\"\n",
 294                                         inv_data, data->hostnqn);
 295                                 break;
 296                         case (offsetof(struct nvmf_connect_data, subsysnqn)):
 297                                 dev_err(ctrl->device,
 298                                         "%s, subsysnqn \"%s\"\n",
 299                                         inv_data, data->subsysnqn);
 300                                 break;
 301                         default:
 302                                 dev_err(ctrl->device,
 303                                         "%s, starting byte offset: %d\n",
 304                                        inv_data, offset & 0xffff);
 305                                 break;
 306                         }
 307                 } else {
 308                         char *inv_sqe = "Connect Invalid SQE Parameter";
 309
 310                         switch (offset) {
 311                         case (offsetof(struct nvmf_connect_command, qid)):
 312                                 dev_err(ctrl->device,
 313                                        "%s, qid %d\n",
 314                                         inv_sqe, cmd->connect.qid);
 315                                 break;
 316                         default:
 317                                 dev_err(ctrl->device,
 318                                         "%s, starting byte offset: %d\n",
 319                                         inv_sqe, offset);
 320                         }
 321                 }
 322                 break;
 323         default:
 324                 dev_err(ctrl->device,
 325                         "Connect command failed, error wo/DNR bit: %d\n",
 326                         err_sctype);
 327                 break;
 328         } /* switch (err_sctype) */
 329 }
 330
 331 /**
 332  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
 333  *                              API function.
 334  * @ctrl:       Host nvme controller instance used to request
 335  *              a new NVMe controller allocation on the target
 336  *              system and  establish an NVMe Admin connection to
 337  *              that controller.
 338  *
 339  * This function enables an NVMe host device to request a new allocation of
 340  * an NVMe controller resource on a target system as well establish a
 341  * fabrics-protocol connection of the NVMe Admin queue between the
 342  * host system device and the allocated NVMe controller on the
 343  * target system via a NVMe Fabrics "Connect" command.
 344  *
 345  * Return:
 346  *      0: success
 347  *      > 0: NVMe error status code
 348  *      < 0: Linux errno error code
 349  *
 350  */
 351 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
 352 {
 353         struct nvme_command cmd;
 354         struct nvme_completion cqe;
 355         struct nvmf_connect_data *data;
 356         int ret;
 357
 358         memset(&cmd, 0, sizeof(cmd));
 359         cmd.connect.opcode = nvme_fabrics_command;
 360         cmd.connect.fctype = nvme_fabrics_type_connect;
 361         cmd.connect.qid = 0;
 362         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
 363         /*
 364          * Set keep-alive timeout in seconds granularity (ms * 1000)
 365          * and add a grace period for controller kato enforcement
 366          */
 367         cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
 368                 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
 369
 370         data = kzalloc(sizeof(*data), GFP_KERNEL);
 371         if (!data)
 372                 return -ENOMEM;
 373
 374         memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_le));
 375         data->cntlid = cpu_to_le16(0xffff);
 376         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 377         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 378
 379         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe,
 380                         data, sizeof(*data), 0, NVME_QID_ANY, 1,
 381                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
 382         if (ret) {
 383                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(cqe.result),
 384                                        &cmd, data);
 385                 goto out_free_data;
 386         }
 387
 388         ctrl->cntlid = le16_to_cpu(cqe.result16);
 389
 390 out_free_data:
 391         kfree(data);
 392         return ret;
 393 }
 394 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
 395
 396 /**
 397  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
 398  *                           API function.
 399  * @ctrl:       Host nvme controller instance used to establish an
 400  *              NVMe I/O queue connection to the already allocated NVMe
 401  *              controller on the target system.
 402  * @qid:        NVMe I/O queue number for the new I/O connection between
 403  *              host and target (note qid == 0 is illegal as this is
 404  *              the Admin queue, per NVMe standard).
 405  *
 406  * This function issues a fabrics-protocol connection
 407  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
 408  * between the host system device and the allocated NVMe controller
 409  * on the target system.
 410  *
 411  * Return:
 412  *      0: success
 413  *      > 0: NVMe error status code
 414  *      < 0: Linux errno error code
 415  */
 416 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
 417 {
 418         struct nvme_command cmd;
 419         struct nvmf_connect_data *data;
 420         struct nvme_completion cqe;
 421         int ret;
 422
 423         memset(&cmd, 0, sizeof(cmd));
 424         cmd.connect.opcode = nvme_fabrics_command;
 425         cmd.connect.fctype = nvme_fabrics_type_connect;
 426         cmd.connect.qid = cpu_to_le16(qid);
 427         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
 428
 429         data = kzalloc(sizeof(*data), GFP_KERNEL);
 430         if (!data)
 431                 return -ENOMEM;
 432
 433         memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_le));
 434         data->cntlid = cpu_to_le16(ctrl->cntlid);
 435         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 436         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 437
 438         ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &cqe,
 439                         data, sizeof(*data), 0, qid, 1,
 440                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
 441         if (ret) {
 442                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(cqe.result),
 443                                        &cmd, data);
 444         }
 445         kfree(data);
 446         return ret;
 447 }
 448 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
 449
 450 /**
 451  * nvmf_register_transport() - NVMe Fabrics Library registration function.
 452  * @ops:        Transport ops instance to be registered to the
 453  *              common fabrics library.
 454  *
 455  * API function that registers the type of specific transport fabric
 456  * being implemented to the common NVMe fabrics library. Part of
 457  * the overall init sequence of starting up a fabrics driver.
 458  */
 459 void nvmf_register_transport(struct nvmf_transport_ops *ops)
 460 {
 461         mutex_lock(&nvmf_transports_mutex);
 462         list_add_tail(&ops->entry, &nvmf_transports);
 463         mutex_unlock(&nvmf_transports_mutex);
 464 }
 465 EXPORT_SYMBOL_GPL(nvmf_register_transport);
 466
 467 /**
 468  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
 469  * @ops:        Transport ops instance to be unregistered from the
 470  *              common fabrics library.
 471  *
 472  * Fabrics API function that unregisters the type of specific transport
 473  * fabric being implemented from the common NVMe fabrics library.
 474  * Part of the overall exit sequence of unloading the implemented driver.
 475  */
 476 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
 477 {
 478         mutex_lock(&nvmf_transports_mutex);
 479         list_del(&ops->entry);
 480         mutex_unlock(&nvmf_transports_mutex);
 481 }
 482 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
 483
 484 static struct nvmf_transport_ops *nvmf_lookup_transport(
 485                 struct nvmf_ctrl_options *opts)
 486 {
 487         struct nvmf_transport_ops *ops;
 488
 489         lockdep_assert_held(&nvmf_transports_mutex);
 490
 491         list_for_each_entry(ops, &nvmf_transports, entry) {
 492                 if (strcmp(ops->name, opts->transport) == 0)
 493                         return ops;
 494         }
 495
 496         return NULL;
 497 }
 498
 499 static const match_table_t opt_tokens = {
 500         { NVMF_OPT_TRANSPORT,           "transport=%s"          },
 501         { NVMF_OPT_TRADDR,              "traddr=%s"             },
 502         { NVMF_OPT_TRSVCID,             "trsvcid=%s"            },
 503         { NVMF_OPT_NQN,                 "nqn=%s"                },
 504         { NVMF_OPT_QUEUE_SIZE,          "queue_size=%d"         },
 505         { NVMF_OPT_NR_IO_QUEUES,        "nr_io_queues=%d"       },
 506         { NVMF_OPT_TL_RETRY_COUNT,      "tl_retry_count=%d"     },
 507         { NVMF_OPT_RECONNECT_DELAY,     "reconnect_delay=%d"    },
 508         { NVMF_OPT_KATO,                "keep_alive_tmo=%d"     },
 509         { NVMF_OPT_HOSTNQN,             "hostnqn=%s"            },
 510         { NVMF_OPT_ERR,                 NULL                    }
 511 };
 512
 513 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 514                 const char *buf)
 515 {
 516         substring_t args[MAX_OPT_ARGS];
 517         char *options, *o, *p;
 518         int token, ret = 0;
 519         size_t nqnlen  = 0;
 520
 521         /* Set defaults */
 522         opts->queue_size = NVMF_DEF_QUEUE_SIZE;
 523         opts->nr_io_queues = num_online_cpus();
 524         opts->tl_retry_count = 2;
 525         opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
 526
 527         options = o = kstrdup(buf, GFP_KERNEL);
 528         if (!options)
 529                 return -ENOMEM;
 530
 531         while ((p = strsep(&o, ",\n")) != NULL) {
 532                 if (!*p)
 533                         continue;
 534
 535                 token = match_token(p, opt_tokens, args);
 536                 opts->mask |= token;
 537                 switch (token) {
 538                 case NVMF_OPT_TRANSPORT:
 539                         p = match_strdup(args);
 540                         if (!p) {
 541                                 ret = -ENOMEM;
 542                                 goto out;
 543                         }
 544                         opts->transport = p;
 545                         break;
 546                 case NVMF_OPT_NQN:
 547                         p = match_strdup(args);
 548                         if (!p) {
 549                                 ret = -ENOMEM;
 550                                 goto out;
 551                         }
 552                         opts->subsysnqn = p;
 553                         nqnlen = strlen(opts->subsysnqn);
 554                         if (nqnlen >= NVMF_NQN_SIZE) {
 555                                 pr_err("%s needs to be < %d bytes\n",
 556                                 opts->subsysnqn, NVMF_NQN_SIZE);
 557                                 ret = -EINVAL;
 558                                 goto out;
 559                         }
 560                         opts->discovery_nqn =
 561                                 !(strcmp(opts->subsysnqn,
 562                                          NVME_DISC_SUBSYS_NAME));
 563                         if (opts->discovery_nqn)
 564                                 opts->nr_io_queues = 0;
 565                         break;
 566                 case NVMF_OPT_TRADDR:
 567                         p = match_strdup(args);
 568                         if (!p) {
 569                                 ret = -ENOMEM;
 570                                 goto out;
 571                         }
 572                         opts->traddr = p;
 573                         break;
 574                 case NVMF_OPT_TRSVCID:
 575                         p = match_strdup(args);
 576                         if (!p) {
 577                                 ret = -ENOMEM;
 578                                 goto out;
 579                         }
 580                         opts->trsvcid = p;
 581                         break;
 582                 case NVMF_OPT_QUEUE_SIZE:
 583                         if (match_int(args, &token)) {
 584                                 ret = -EINVAL;
 585                                 goto out;
 586                         }
 587                         if (token < NVMF_MIN_QUEUE_SIZE ||
 588                             token > NVMF_MAX_QUEUE_SIZE) {
 589                                 pr_err("Invalid queue_size %d\n", token);
 590                                 ret = -EINVAL;
 591                                 goto out;
 592                         }
 593                         opts->queue_size = token;
 594                         break;
 595                 case NVMF_OPT_NR_IO_QUEUES:
 596                         if (match_int(args, &token)) {
 597                                 ret = -EINVAL;
 598                                 goto out;
 599                         }
 600                         if (token <= 0) {
 601                                 pr_err("Invalid number of IOQs %d\n", token);
 602                                 ret = -EINVAL;
 603                                 goto out;
 604                         }
 605                         opts->nr_io_queues = min_t(unsigned int,
 606                                         num_online_cpus(), token);
 607                         break;
 608                 case NVMF_OPT_TL_RETRY_COUNT:
 609                         if (match_int(args, &token)) {
 610                                 ret = -EINVAL;
 611                                 goto out;
 612                         }
 613                         if (token < 0) {
 614                                 pr_err("Invalid tl_retry_count %d\n", token);
 615                                 ret = -EINVAL;
 616                                 goto out;
 617                         }
 618                         opts->tl_retry_count = token;
 619                         break;
 620                 case NVMF_OPT_KATO:
 621                         if (match_int(args, &token)) {
 622                                 ret = -EINVAL;
 623                                 goto out;
 624                         }
 625
 626                         if (opts->discovery_nqn) {
 627                                 pr_err("Discovery controllers cannot accept keep_alive_tmo != 0\n");
 628                                 ret = -EINVAL;
 629                                 goto out;
 630                         }
 631
 632                         if (token < 0) {
 633                                 pr_err("Invalid keep_alive_tmo %d\n", token);
 634                                 ret = -EINVAL;
 635                                 goto out;
 636                         } else if (token == 0) {
 637                                 /* Allowed for debug */
 638                                 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
 639                         }
 640                         opts->kato = token;
 641                         break;
 642                 case NVMF_OPT_HOSTNQN:
 643                         if (opts->host) {
 644                                 pr_err("hostnqn already user-assigned: %s\n",
 645                                        opts->host->nqn);
 646                                 ret = -EADDRINUSE;
 647                                 goto out;
 648                         }
 649                         p = match_strdup(args);
 650                         if (!p) {
 651                                 ret = -ENOMEM;
 652                                 goto out;
 653                         }
 654                         nqnlen = strlen(p);
 655                         if (nqnlen >= NVMF_NQN_SIZE) {
 656                                 pr_err("%s needs to be < %d bytes\n",
 657                                         p, NVMF_NQN_SIZE);
 658                                 ret = -EINVAL;
 659                                 goto out;
 660                         }
 661                         opts->host = nvmf_host_add(p);
 662                         if (!opts->host) {
 663                                 ret = -ENOMEM;
 664                                 goto out;
 665                         }
 666                         break;
 667                 case NVMF_OPT_RECONNECT_DELAY:
 668                         if (match_int(args, &token)) {
 669                                 ret = -EINVAL;
 670                                 goto out;
 671                         }
 672                         if (token <= 0) {
 673                                 pr_err("Invalid reconnect_delay %d\n", token);
 674                                 ret = -EINVAL;
 675                                 goto out;
 676                         }
 677                         opts->reconnect_delay = token;
 678                         break;
 679                 default:
 680                         pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
 681                                 p);
 682                         ret = -EINVAL;
 683                         goto out;
 684                 }
 685         }
 686
 687         if (!opts->host) {
 688                 kref_get(&nvmf_default_host->ref);
 689                 opts->host = nvmf_default_host;
 690         }
 691
 692 out:
 693         if (!opts->discovery_nqn && !opts->kato)
 694                 opts->kato = NVME_DEFAULT_KATO;
 695         kfree(options);
 696         return ret;
 697 }
 698
 699 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
 700                 unsigned int required_opts)
 701 {
 702         if ((opts->mask & required_opts) != required_opts) {
 703                 int i;
 704
 705                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 706                         if ((opt_tokens[i].token & required_opts) &&
 707                             !(opt_tokens[i].token & opts->mask)) {
 708                                 pr_warn("missing parameter '%s'\n",
 709                                         opt_tokens[i].pattern);
 710                         }
 711                 }
 712
 713                 return -EINVAL;
 714         }
 715
 716         return 0;
 717 }
 718
 719 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
 720                 unsigned int allowed_opts)
 721 {
 722         if (opts->mask & ~allowed_opts) {
 723                 int i;
 724
 725                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 726                         if (opt_tokens[i].token & ~allowed_opts) {
 727                                 pr_warn("invalid parameter '%s'\n",
 728                                         opt_tokens[i].pattern);
 729                         }
 730                 }
 731
 732                 return -EINVAL;
 733         }
 734
 735         return 0;
 736 }
 737
 738 void nvmf_free_options(struct nvmf_ctrl_options *opts)
 739 {
 740         nvmf_host_put(opts->host);
 741         kfree(opts->transport);
 742         kfree(opts->traddr);
 743         kfree(opts->trsvcid);
 744         kfree(opts->subsysnqn);
 745         kfree(opts);
 746 }
 747 EXPORT_SYMBOL_GPL(nvmf_free_options);
 748
 749 #define NVMF_REQUIRED_OPTS      (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
 750 #define NVMF_ALLOWED_OPTS       (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
 751                                  NVMF_OPT_KATO | NVMF_OPT_HOSTNQN)
 752
 753 static struct nvme_ctrl *
 754 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
 755 {
 756         struct nvmf_ctrl_options *opts;
 757         struct nvmf_transport_ops *ops;
 758         struct nvme_ctrl *ctrl;
 759         int ret;
 760
 761         opts = kzalloc(sizeof(*opts), GFP_KERNEL);
 762         if (!opts)
 763                 return ERR_PTR(-ENOMEM);
 764
 765         ret = nvmf_parse_options(opts, buf);
 766         if (ret)
 767                 goto out_free_opts;
 768
 769         /*
 770          * Check the generic options first as we need a valid transport for
 771          * the lookup below.  Then clear the generic flags so that transport
 772          * drivers don't have to care about them.
 773          */
 774         ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
 775         if (ret)
 776                 goto out_free_opts;
 777         opts->mask &= ~NVMF_REQUIRED_OPTS;
 778
 779         mutex_lock(&nvmf_transports_mutex);
 780         ops = nvmf_lookup_transport(opts);
 781         if (!ops) {
 782                 pr_info("no handler found for transport %s.\n",
 783                         opts->transport);
 784                 ret = -EINVAL;
 785                 goto out_unlock;
 786         }
 787
 788         ret = nvmf_check_required_opts(opts, ops->required_opts);
 789         if (ret)
 790                 goto out_unlock;
 791         ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
 792                                 ops->allowed_opts | ops->required_opts);
 793         if (ret)
 794                 goto out_unlock;
 795
 796         ctrl = ops->create_ctrl(dev, opts);
 797         if (IS_ERR(ctrl)) {
 798                 ret = PTR_ERR(ctrl);
 799                 goto out_unlock;
 800         }
 801
 802         mutex_unlock(&nvmf_transports_mutex);
 803         return ctrl;
 804
 805 out_unlock:
 806         mutex_unlock(&nvmf_transports_mutex);
 807 out_free_opts:
 808         nvmf_host_put(opts->host);
 809         kfree(opts);
 810         return ERR_PTR(ret);
 811 }
 812
 813 static struct class *nvmf_class;
 814 static struct device *nvmf_device;
 815 static DEFINE_MUTEX(nvmf_dev_mutex);
 816
 817 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
 818                 size_t count, loff_t *pos)
 819 {
 820         struct seq_file *seq_file = file->private_data;
 821         struct nvme_ctrl *ctrl;
 822         const char *buf;
 823         int ret = 0;
 824
 825         if (count > PAGE_SIZE)
 826                 return -ENOMEM;
 827
 828         buf = memdup_user_nul(ubuf, count);
 829         if (IS_ERR(buf))
 830                 return PTR_ERR(buf);
 831
 832         mutex_lock(&nvmf_dev_mutex);
 833         if (seq_file->private) {
 834                 ret = -EINVAL;
 835                 goto out_unlock;
 836         }
 837
 838         ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
 839         if (IS_ERR(ctrl)) {
 840                 ret = PTR_ERR(ctrl);
 841                 goto out_unlock;
 842         }
 843
 844         seq_file->private = ctrl;
 845
 846 out_unlock:
 847         mutex_unlock(&nvmf_dev_mutex);
 848         kfree(buf);
 849         return ret ? ret : count;
 850 }
 851
 852 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
 853 {
 854         struct nvme_ctrl *ctrl;
 855         int ret = 0;
 856
 857         mutex_lock(&nvmf_dev_mutex);
 858         ctrl = seq_file->private;
 859         if (!ctrl) {
 860                 ret = -EINVAL;
 861                 goto out_unlock;
 862         }
 863
 864         seq_printf(seq_file, "instance=%d,cntlid=%d\n",
 865                         ctrl->instance, ctrl->cntlid);
 866
 867 out_unlock:
 868         mutex_unlock(&nvmf_dev_mutex);
 869         return ret;
 870 }
 871
 872 static int nvmf_dev_open(struct inode *inode, struct file *file)
 873 {
 874         /*
 875          * The miscdevice code initializes file->private_data, but doesn't
 876          * make use of it later.
 877          */
 878         file->private_data = NULL;
 879         return single_open(file, nvmf_dev_show, NULL);
 880 }
 881
 882 static int nvmf_dev_release(struct inode *inode, struct file *file)
 883 {
 884         struct seq_file *seq_file = file->private_data;
 885         struct nvme_ctrl *ctrl = seq_file->private;
 886
 887         if (ctrl)
 888                 nvme_put_ctrl(ctrl);
 889         return single_release(inode, file);
 890 }
 891
 892 static const struct file_operations nvmf_dev_fops = {
 893         .owner          = THIS_MODULE,
 894         .write          = nvmf_dev_write,
 895         .read           = seq_read,
 896         .open           = nvmf_dev_open,
 897         .release        = nvmf_dev_release,
 898 };
 899
 900 static struct miscdevice nvmf_misc = {
 901         .minor          = MISC_DYNAMIC_MINOR,
 902         .name           = "nvme-fabrics",
 903         .fops           = &nvmf_dev_fops,
 904 };
 905
 906 static int __init nvmf_init(void)
 907 {
 908         int ret;
 909
 910         nvmf_default_host = nvmf_host_default();
 911         if (!nvmf_default_host)
 912                 return -ENOMEM;
 913
 914         nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
 915         if (IS_ERR(nvmf_class)) {
 916                 pr_err("couldn't register class nvme-fabrics\n");
 917                 ret = PTR_ERR(nvmf_class);
 918                 goto out_free_host;
 919         }
 920
 921         nvmf_device =
 922                 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
 923         if (IS_ERR(nvmf_device)) {
 924                 pr_err("couldn't create nvme-fabris device!\n");
 925                 ret = PTR_ERR(nvmf_device);
 926                 goto out_destroy_class;
 927         }
 928
 929         ret = misc_register(&nvmf_misc);
 930         if (ret) {
 931                 pr_err("couldn't register misc device: %d\n", ret);
 932                 goto out_destroy_device;
 933         }
 934
 935         return 0;
 936
 937 out_destroy_device:
 938         device_destroy(nvmf_class, MKDEV(0, 0));
 939 out_destroy_class:
 940         class_destroy(nvmf_class);
 941 out_free_host:
 942         nvmf_host_put(nvmf_default_host);
 943         return ret;
 944 }
 945
 946 static void __exit nvmf_exit(void)
 947 {
 948         misc_deregister(&nvmf_misc);
 949         device_destroy(nvmf_class, MKDEV(0, 0));
 950         class_destroy(nvmf_class);
 951         nvmf_host_put(nvmf_default_host);
 952
 953         BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
 954         BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
 955         BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
 956         BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
 957 }
 958
 959 MODULE_LICENSE("GPL v2");
 960
 961 module_init(nvmf_init);
 962 module_exit(nvmf_exit);