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