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         default:
 348                 dev_err(ctrl->device,
 349                         "Connect command failed, error wo/DNR bit: %d\n",
 350                         err_sctype);
 351                 break;
 352         } /* switch (err_sctype) */
 353 }
 354
 355 /**
 356  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
 357  *                              API function.
 358  * @ctrl:       Host nvme controller instance used to request
 359  *              a new NVMe controller allocation on the target
 360  *              system and  establish an NVMe Admin connection to
 361  *              that controller.
 362  *
 363  * This function enables an NVMe host device to request a new allocation of
 364  * an NVMe controller resource on a target system as well establish a
 365  * fabrics-protocol connection of the NVMe Admin queue between the
 366  * host system device and the allocated NVMe controller on the
 367  * target system via a NVMe Fabrics "Connect" command.
 368  *
 369  * Return:
 370  *      0: success
 371  *      > 0: NVMe error status code
 372  *      < 0: Linux errno error code
 373  *
 374  */
 375 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
 376 {
 377         struct nvme_command cmd;
 378         union nvme_result res;
 379         struct nvmf_connect_data *data;
 380         int ret;
 381
 382         memset(&cmd, 0, sizeof(cmd));
 383         cmd.connect.opcode = nvme_fabrics_command;
 384         cmd.connect.fctype = nvme_fabrics_type_connect;
 385         cmd.connect.qid = 0;
 386         cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
 387
 388         /*
 389          * Set keep-alive timeout in seconds granularity (ms * 1000)
 390          * and add a grace period for controller kato enforcement
 391          */
 392         cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
 393                 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
 394
 395         if (ctrl->opts->disable_sqflow)
 396                 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
 397
 398         data = kzalloc(sizeof(*data), GFP_KERNEL);
 399         if (!data)
 400                 return -ENOMEM;
 401
 402         uuid_copy(&data->hostid, &ctrl->opts->host->id);
 403         data->cntlid = cpu_to_le16(0xffff);
 404         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 405         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 406
 407         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res,
 408                         data, sizeof(*data), 0, NVME_QID_ANY, 1,
 409                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
 410         if (ret) {
 411                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
 412                                        &cmd, data);
 413                 goto out_free_data;
 414         }
 415
 416         ctrl->cntlid = le16_to_cpu(res.u16);
 417
 418 out_free_data:
 419         kfree(data);
 420         return ret;
 421 }
 422 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
 423
 424 /**
 425  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
 426  *                           API function.
 427  * @ctrl:       Host nvme controller instance used to establish an
 428  *              NVMe I/O queue connection to the already allocated NVMe
 429  *              controller on the target system.
 430  * @qid:        NVMe I/O queue number for the new I/O connection between
 431  *              host and target (note qid == 0 is illegal as this is
 432  *              the Admin queue, per NVMe standard).
 433  *
 434  * This function issues a fabrics-protocol connection
 435  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
 436  * between the host system device and the allocated NVMe controller
 437  * on the target system.
 438  *
 439  * Return:
 440  *      0: success
 441  *      > 0: NVMe error status code
 442  *      < 0: Linux errno error code
 443  */
 444 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
 445 {
 446         struct nvme_command cmd;
 447         struct nvmf_connect_data *data;
 448         union nvme_result res;
 449         int ret;
 450
 451         memset(&cmd, 0, sizeof(cmd));
 452         cmd.connect.opcode = nvme_fabrics_command;
 453         cmd.connect.fctype = nvme_fabrics_type_connect;
 454         cmd.connect.qid = cpu_to_le16(qid);
 455         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
 456
 457         if (ctrl->opts->disable_sqflow)
 458                 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
 459
 460         data = kzalloc(sizeof(*data), GFP_KERNEL);
 461         if (!data)
 462                 return -ENOMEM;
 463
 464         uuid_copy(&data->hostid, &ctrl->opts->host->id);
 465         data->cntlid = cpu_to_le16(ctrl->cntlid);
 466         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 467         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 468
 469         ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
 470                         data, sizeof(*data), 0, qid, 1,
 471                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
 472         if (ret) {
 473                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
 474                                        &cmd, data);
 475         }
 476         kfree(data);
 477         return ret;
 478 }
 479 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
 480
 481 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
 482 {
 483         if (ctrl->opts->max_reconnects == -1 ||
 484             ctrl->nr_reconnects < ctrl->opts->max_reconnects)
 485                 return true;
 486
 487         return false;
 488 }
 489 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
 490
 491 /**
 492  * nvmf_register_transport() - NVMe Fabrics Library registration function.
 493  * @ops:        Transport ops instance to be registered to the
 494  *              common fabrics library.
 495  *
 496  * API function that registers the type of specific transport fabric
 497  * being implemented to the common NVMe fabrics library. Part of
 498  * the overall init sequence of starting up a fabrics driver.
 499  */
 500 int nvmf_register_transport(struct nvmf_transport_ops *ops)
 501 {
 502         if (!ops->create_ctrl)
 503                 return -EINVAL;
 504
 505         down_write(&nvmf_transports_rwsem);
 506         list_add_tail(&ops->entry, &nvmf_transports);
 507         up_write(&nvmf_transports_rwsem);
 508
 509         return 0;
 510 }
 511 EXPORT_SYMBOL_GPL(nvmf_register_transport);
 512
 513 /**
 514  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
 515  * @ops:        Transport ops instance to be unregistered from the
 516  *              common fabrics library.
 517  *
 518  * Fabrics API function that unregisters the type of specific transport
 519  * fabric being implemented from the common NVMe fabrics library.
 520  * Part of the overall exit sequence of unloading the implemented driver.
 521  */
 522 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
 523 {
 524         down_write(&nvmf_transports_rwsem);
 525         list_del(&ops->entry);
 526         up_write(&nvmf_transports_rwsem);
 527 }
 528 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
 529
 530 static struct nvmf_transport_ops *nvmf_lookup_transport(
 531                 struct nvmf_ctrl_options *opts)
 532 {
 533         struct nvmf_transport_ops *ops;
 534
 535         lockdep_assert_held(&nvmf_transports_rwsem);
 536
 537         list_for_each_entry(ops, &nvmf_transports, entry) {
 538                 if (strcmp(ops->name, opts->transport) == 0)
 539                         return ops;
 540         }
 541
 542         return NULL;
 543 }
 544
 545 /*
 546  * For something we're not in a state to send to the device the default action
 547  * is to busy it and retry it after the controller state is recovered.  However,
 548  * if the controller is deleting or if anything is marked for failfast or
 549  * nvme multipath it is immediately failed.
 550  *
 551  * Note: commands used to initialize the controller will be marked for failfast.
 552  * Note: nvme cli/ioctl commands are marked for failfast.
 553  */
 554 blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
 555                 struct request *rq)
 556 {
 557         if (ctrl->state != NVME_CTRL_DELETING &&
 558             ctrl->state != NVME_CTRL_DEAD &&
 559             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 560                 return BLK_STS_RESOURCE;
 561
 562         nvme_req(rq)->status = NVME_SC_HOST_PATH_ERROR;
 563         blk_mq_start_request(rq);
 564         nvme_complete_rq(rq);
 565         return BLK_STS_OK;
 566 }
 567 EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command);
 568
 569 bool __nvmf_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 570                 bool queue_live)
 571 {
 572         struct nvme_request *req = nvme_req(rq);
 573
 574         /*
 575          * If we are in some state of setup or teardown only allow
 576          * internally generated commands.
 577          */
 578         if (!blk_rq_is_passthrough(rq) || (req->flags & NVME_REQ_USERCMD))
 579                 return false;
 580
 581         /*
 582          * Only allow commands on a live queue, except for the connect command,
 583          * which is require to set the queue live in the appropinquate states.
 584          */
 585         switch (ctrl->state) {
 586         case NVME_CTRL_NEW:
 587         case NVME_CTRL_CONNECTING:
 588                 if (req->cmd->common.opcode == nvme_fabrics_command &&
 589                     req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
 590                         return true;
 591                 break;
 592         default:
 593                 break;
 594         case NVME_CTRL_DEAD:
 595                 return false;
 596         }
 597
 598         return queue_live;
 599 }
 600 EXPORT_SYMBOL_GPL(__nvmf_check_ready);
 601
 602 static const match_table_t opt_tokens = {
 603         { NVMF_OPT_TRANSPORT,           "transport=%s"          },
 604         { NVMF_OPT_TRADDR,              "traddr=%s"             },
 605         { NVMF_OPT_TRSVCID,             "trsvcid=%s"            },
 606         { NVMF_OPT_NQN,                 "nqn=%s"                },
 607         { NVMF_OPT_QUEUE_SIZE,          "queue_size=%d"         },
 608         { NVMF_OPT_NR_IO_QUEUES,        "nr_io_queues=%d"       },
 609         { NVMF_OPT_RECONNECT_DELAY,     "reconnect_delay=%d"    },
 610         { NVMF_OPT_CTRL_LOSS_TMO,       "ctrl_loss_tmo=%d"      },
 611         { NVMF_OPT_KATO,                "keep_alive_tmo=%d"     },
 612         { NVMF_OPT_HOSTNQN,             "hostnqn=%s"            },
 613         { NVMF_OPT_HOST_TRADDR,         "host_traddr=%s"        },
 614         { NVMF_OPT_HOST_ID,             "hostid=%s"             },
 615         { NVMF_OPT_DUP_CONNECT,         "duplicate_connect"     },
 616         { NVMF_OPT_DISABLE_SQFLOW,      "disable_sqflow"        },
 617         { NVMF_OPT_HDR_DIGEST,          "hdr_digest"            },
 618         { NVMF_OPT_ERR,                 NULL                    }
 619 };
 620
 621 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 622                 const char *buf)
 623 {
 624         substring_t args[MAX_OPT_ARGS];
 625         char *options, *o, *p;
 626         int token, ret = 0;
 627         size_t nqnlen  = 0;
 628         int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
 629         uuid_t hostid;
 630
 631         /* Set defaults */
 632         opts->queue_size = NVMF_DEF_QUEUE_SIZE;
 633         opts->nr_io_queues = num_online_cpus();
 634         opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
 635         opts->kato = NVME_DEFAULT_KATO;
 636         opts->duplicate_connect = false;
 637         opts->hdr_digest = false;
 638
 639         options = o = kstrdup(buf, GFP_KERNEL);
 640         if (!options)
 641                 return -ENOMEM;
 642
 643         uuid_gen(&hostid);
 644
 645         while ((p = strsep(&o, ",\n")) != NULL) {
 646                 if (!*p)
 647                         continue;
 648
 649                 token = match_token(p, opt_tokens, args);
 650                 opts->mask |= token;
 651                 switch (token) {
 652                 case NVMF_OPT_TRANSPORT:
 653                         p = match_strdup(args);
 654                         if (!p) {
 655                                 ret = -ENOMEM;
 656                                 goto out;
 657                         }
 658                         kfree(opts->transport);
 659                         opts->transport = p;
 660                         break;
 661                 case NVMF_OPT_NQN:
 662                         p = match_strdup(args);
 663                         if (!p) {
 664                                 ret = -ENOMEM;
 665                                 goto out;
 666                         }
 667                         kfree(opts->subsysnqn);
 668                         opts->subsysnqn = p;
 669                         nqnlen = strlen(opts->subsysnqn);
 670                         if (nqnlen >= NVMF_NQN_SIZE) {
 671                                 pr_err("%s needs to be < %d bytes\n",
 672                                         opts->subsysnqn, NVMF_NQN_SIZE);
 673                                 ret = -EINVAL;
 674                                 goto out;
 675                         }
 676                         opts->discovery_nqn =
 677                                 !(strcmp(opts->subsysnqn,
 678                                          NVME_DISC_SUBSYS_NAME));
 679                         break;
 680                 case NVMF_OPT_TRADDR:
 681                         p = match_strdup(args);
 682                         if (!p) {
 683                                 ret = -ENOMEM;
 684                                 goto out;
 685                         }
 686                         kfree(opts->traddr);
 687                         opts->traddr = p;
 688                         break;
 689                 case NVMF_OPT_TRSVCID:
 690                         p = match_strdup(args);
 691                         if (!p) {
 692                                 ret = -ENOMEM;
 693                                 goto out;
 694                         }
 695                         kfree(opts->trsvcid);
 696                         opts->trsvcid = p;
 697                         break;
 698                 case NVMF_OPT_QUEUE_SIZE:
 699                         if (match_int(args, &token)) {
 700                                 ret = -EINVAL;
 701                                 goto out;
 702                         }
 703                         if (token < NVMF_MIN_QUEUE_SIZE ||
 704                             token > NVMF_MAX_QUEUE_SIZE) {
 705                                 pr_err("Invalid queue_size %d\n", token);
 706                                 ret = -EINVAL;
 707                                 goto out;
 708                         }
 709                         opts->queue_size = token;
 710                         break;
 711                 case NVMF_OPT_NR_IO_QUEUES:
 712                         if (match_int(args, &token)) {
 713                                 ret = -EINVAL;
 714                                 goto out;
 715                         }
 716                         if (token <= 0) {
 717                                 pr_err("Invalid number of IOQs %d\n", token);
 718                                 ret = -EINVAL;
 719                                 goto out;
 720                         }
 721                         if (opts->discovery_nqn) {
 722                                 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
 723                                 break;
 724                         }
 725
 726                         opts->nr_io_queues = min_t(unsigned int,
 727                                         num_online_cpus(), token);
 728                         break;
 729                 case NVMF_OPT_KATO:
 730                         if (match_int(args, &token)) {
 731                                 ret = -EINVAL;
 732                                 goto out;
 733                         }
 734
 735                         if (token < 0) {
 736                                 pr_err("Invalid keep_alive_tmo %d\n", token);
 737                                 ret = -EINVAL;
 738                                 goto out;
 739                         } else if (token == 0 && !opts->discovery_nqn) {
 740                                 /* Allowed for debug */
 741                                 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
 742                         }
 743                         opts->kato = token;
 744
 745                         if (opts->discovery_nqn && opts->kato) {
 746                                 pr_err("Discovery controllers cannot accept KATO != 0\n");
 747                                 ret = -EINVAL;
 748                                 goto out;
 749                         }
 750
 751                         break;
 752                 case NVMF_OPT_CTRL_LOSS_TMO:
 753                         if (match_int(args, &token)) {
 754                                 ret = -EINVAL;
 755                                 goto out;
 756                         }
 757
 758                         if (token < 0)
 759                                 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
 760                         ctrl_loss_tmo = token;
 761                         break;
 762                 case NVMF_OPT_HOSTNQN:
 763                         if (opts->host) {
 764                                 pr_err("hostnqn already user-assigned: %s\n",
 765                                        opts->host->nqn);
 766                                 ret = -EADDRINUSE;
 767                                 goto out;
 768                         }
 769                         p = match_strdup(args);
 770                         if (!p) {
 771                                 ret = -ENOMEM;
 772                                 goto out;
 773                         }
 774                         nqnlen = strlen(p);
 775                         if (nqnlen >= NVMF_NQN_SIZE) {
 776                                 pr_err("%s needs to be < %d bytes\n",
 777                                         p, NVMF_NQN_SIZE);
 778                                 kfree(p);
 779                                 ret = -EINVAL;
 780                                 goto out;
 781                         }
 782                         nvmf_host_put(opts->host);
 783                         opts->host = nvmf_host_add(p);
 784                         kfree(p);
 785                         if (!opts->host) {
 786                                 ret = -ENOMEM;
 787                                 goto out;
 788                         }
 789                         break;
 790                 case NVMF_OPT_RECONNECT_DELAY:
 791                         if (match_int(args, &token)) {
 792                                 ret = -EINVAL;
 793                                 goto out;
 794                         }
 795                         if (token <= 0) {
 796                                 pr_err("Invalid reconnect_delay %d\n", token);
 797                                 ret = -EINVAL;
 798                                 goto out;
 799                         }
 800                         opts->reconnect_delay = token;
 801                         break;
 802                 case NVMF_OPT_HOST_TRADDR:
 803                         p = match_strdup(args);
 804                         if (!p) {
 805                                 ret = -ENOMEM;
 806                                 goto out;
 807                         }
 808                         kfree(opts->host_traddr);
 809                         opts->host_traddr = p;
 810                         break;
 811                 case NVMF_OPT_HOST_ID:
 812                         p = match_strdup(args);
 813                         if (!p) {
 814                                 ret = -ENOMEM;
 815                                 goto out;
 816                         }
 817                         ret = uuid_parse(p, &hostid);
 818                         if (ret) {
 819                                 pr_err("Invalid hostid %s\n", p);
 820                                 ret = -EINVAL;
 821                                 kfree(p);
 822                                 goto out;
 823                         }
 824                         kfree(p);
 825                         break;
 826                 case NVMF_OPT_DUP_CONNECT:
 827                         opts->duplicate_connect = true;
 828                         break;
 829                 case NVMF_OPT_DISABLE_SQFLOW:
 830                         opts->disable_sqflow = true;
 831                         break;
 832                 case NVMF_OPT_HDR_DIGEST:
 833                         opts->hdr_digest = true;
 834                         break;
 835                 default:
 836                         pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
 837                                 p);
 838                         ret = -EINVAL;
 839                         goto out;
 840                 }
 841         }
 842
 843         if (opts->discovery_nqn) {
 844                 opts->kato = 0;
 845                 opts->nr_io_queues = 0;
 846                 opts->duplicate_connect = true;
 847         }
 848         if (ctrl_loss_tmo < 0)
 849                 opts->max_reconnects = -1;
 850         else
 851                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
 852                                                 opts->reconnect_delay);
 853
 854         if (!opts->host) {
 855                 kref_get(&nvmf_default_host->ref);
 856                 opts->host = nvmf_default_host;
 857         }
 858
 859         uuid_copy(&opts->host->id, &hostid);
 860
 861 out:
 862         kfree(options);
 863         return ret;
 864 }
 865
 866 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
 867                 unsigned int required_opts)
 868 {
 869         if ((opts->mask & required_opts) != required_opts) {
 870                 int i;
 871
 872                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 873                         if ((opt_tokens[i].token & required_opts) &&
 874                             !(opt_tokens[i].token & opts->mask)) {
 875                                 pr_warn("missing parameter '%s'\n",
 876                                         opt_tokens[i].pattern);
 877                         }
 878                 }
 879
 880                 return -EINVAL;
 881         }
 882
 883         return 0;
 884 }
 885
 886 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
 887                 struct nvmf_ctrl_options *opts)
 888 {
 889         if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
 890             strcmp(opts->traddr, ctrl->opts->traddr) ||
 891             strcmp(opts->trsvcid, ctrl->opts->trsvcid))
 892                 return false;
 893
 894         /*
 895          * Checking the local address is rough. In most cases, none is specified
 896          * and the host port is selected by the stack.
 897          *
 898          * Assume no match if:
 899          * -  local address is specified and address is not the same
 900          * -  local address is not specified but remote is, or vice versa
 901          *    (admin using specific host_traddr when it matters).
 902          */
 903         if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
 904             (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
 905                 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
 906                         return false;
 907         } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
 908                    (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
 909                 return false;
 910         }
 911
 912         return true;
 913 }
 914 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
 915
 916 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
 917                 unsigned int allowed_opts)
 918 {
 919         if (opts->mask & ~allowed_opts) {
 920                 int i;
 921
 922                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 923                         if ((opt_tokens[i].token & opts->mask) &&
 924                             (opt_tokens[i].token & ~allowed_opts)) {
 925                                 pr_warn("invalid parameter '%s'\n",
 926                                         opt_tokens[i].pattern);
 927                         }
 928                 }
 929
 930                 return -EINVAL;
 931         }
 932
 933         return 0;
 934 }
 935
 936 void nvmf_free_options(struct nvmf_ctrl_options *opts)
 937 {
 938         nvmf_host_put(opts->host);
 939         kfree(opts->transport);
 940         kfree(opts->traddr);
 941         kfree(opts->trsvcid);
 942         kfree(opts->subsysnqn);
 943         kfree(opts->host_traddr);
 944         kfree(opts);
 945 }
 946 EXPORT_SYMBOL_GPL(nvmf_free_options);
 947
 948 #define NVMF_REQUIRED_OPTS      (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
 949 #define NVMF_ALLOWED_OPTS       (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
 950                                  NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
 951                                  NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
 952                                  NVMF_OPT_DISABLE_SQFLOW)
 953
 954 static struct nvme_ctrl *
 955 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
 956 {
 957         struct nvmf_ctrl_options *opts;
 958         struct nvmf_transport_ops *ops;
 959         struct nvme_ctrl *ctrl;
 960         int ret;
 961
 962         opts = kzalloc(sizeof(*opts), GFP_KERNEL);
 963         if (!opts)
 964                 return ERR_PTR(-ENOMEM);
 965
 966         ret = nvmf_parse_options(opts, buf);
 967         if (ret)
 968                 goto out_free_opts;
 969
 970
 971         request_module("nvme-%s", opts->transport);
 972
 973         /*
 974          * Check the generic options first as we need a valid transport for
 975          * the lookup below.  Then clear the generic flags so that transport
 976          * drivers don't have to care about them.
 977          */
 978         ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
 979         if (ret)
 980                 goto out_free_opts;
 981         opts->mask &= ~NVMF_REQUIRED_OPTS;
 982
 983         down_read(&nvmf_transports_rwsem);
 984         ops = nvmf_lookup_transport(opts);
 985         if (!ops) {
 986                 pr_info("no handler found for transport %s.\n",
 987                         opts->transport);
 988                 ret = -EINVAL;
 989                 goto out_unlock;
 990         }
 991
 992         if (!try_module_get(ops->module)) {
 993                 ret = -EBUSY;
 994                 goto out_unlock;
 995         }
 996         up_read(&nvmf_transports_rwsem);
 997
 998         ret = nvmf_check_required_opts(opts, ops->required_opts);
 999         if (ret)
1000                 goto out_module_put;
1001         ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1002                                 ops->allowed_opts | ops->required_opts);
1003         if (ret)
1004                 goto out_module_put;
1005
1006         ctrl = ops->create_ctrl(dev, opts);
1007         if (IS_ERR(ctrl)) {
1008                 ret = PTR_ERR(ctrl);
1009                 goto out_module_put;
1010         }
1011
1012         module_put(ops->module);
1013         return ctrl;
1014
1015 out_module_put:
1016         module_put(ops->module);
1017         goto out_free_opts;
1018 out_unlock:
1019         up_read(&nvmf_transports_rwsem);
1020 out_free_opts:
1021         nvmf_free_options(opts);
1022         return ERR_PTR(ret);
1023 }
1024
1025 static struct class *nvmf_class;
1026 static struct device *nvmf_device;
1027 static DEFINE_MUTEX(nvmf_dev_mutex);
1028
1029 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1030                 size_t count, loff_t *pos)
1031 {
1032         struct seq_file *seq_file = file->private_data;
1033         struct nvme_ctrl *ctrl;
1034         const char *buf;
1035         int ret = 0;
1036
1037         if (count > PAGE_SIZE)
1038                 return -ENOMEM;
1039
1040         buf = memdup_user_nul(ubuf, count);
1041         if (IS_ERR(buf))
1042                 return PTR_ERR(buf);
1043
1044         mutex_lock(&nvmf_dev_mutex);
1045         if (seq_file->private) {
1046                 ret = -EINVAL;
1047                 goto out_unlock;
1048         }
1049
1050         ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
1051         if (IS_ERR(ctrl)) {
1052                 ret = PTR_ERR(ctrl);
1053                 goto out_unlock;
1054         }
1055
1056         seq_file->private = ctrl;
1057
1058 out_unlock:
1059         mutex_unlock(&nvmf_dev_mutex);
1060         kfree(buf);
1061         return ret ? ret : count;
1062 }
1063
1064 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1065 {
1066         struct nvme_ctrl *ctrl;
1067         int ret = 0;
1068
1069         mutex_lock(&nvmf_dev_mutex);
1070         ctrl = seq_file->private;
1071         if (!ctrl) {
1072                 ret = -EINVAL;
1073                 goto out_unlock;
1074         }
1075
1076         seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1077                         ctrl->instance, ctrl->cntlid);
1078
1079 out_unlock:
1080         mutex_unlock(&nvmf_dev_mutex);
1081         return ret;
1082 }
1083
1084 static int nvmf_dev_open(struct inode *inode, struct file *file)
1085 {
1086         /*
1087          * The miscdevice code initializes file->private_data, but doesn't
1088          * make use of it later.
1089          */
1090         file->private_data = NULL;
1091         return single_open(file, nvmf_dev_show, NULL);
1092 }
1093
1094 static int nvmf_dev_release(struct inode *inode, struct file *file)
1095 {
1096         struct seq_file *seq_file = file->private_data;
1097         struct nvme_ctrl *ctrl = seq_file->private;
1098
1099         if (ctrl)
1100                 nvme_put_ctrl(ctrl);
1101         return single_release(inode, file);
1102 }
1103
1104 static const struct file_operations nvmf_dev_fops = {
1105         .owner          = THIS_MODULE,
1106         .write          = nvmf_dev_write,
1107         .read           = seq_read,
1108         .open           = nvmf_dev_open,
1109         .release        = nvmf_dev_release,
1110 };
1111
1112 static struct miscdevice nvmf_misc = {
1113         .minor          = MISC_DYNAMIC_MINOR,
1114         .name           = "nvme-fabrics",
1115         .fops           = &nvmf_dev_fops,
1116 };
1117
1118 static int __init nvmf_init(void)
1119 {
1120         int ret;
1121
1122         nvmf_default_host = nvmf_host_default();
1123         if (!nvmf_default_host)
1124                 return -ENOMEM;
1125
1126         nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1127         if (IS_ERR(nvmf_class)) {
1128                 pr_err("couldn't register class nvme-fabrics\n");
1129                 ret = PTR_ERR(nvmf_class);
1130                 goto out_free_host;
1131         }
1132
1133         nvmf_device =
1134                 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1135         if (IS_ERR(nvmf_device)) {
1136                 pr_err("couldn't create nvme-fabris device!\n");
1137                 ret = PTR_ERR(nvmf_device);
1138                 goto out_destroy_class;
1139         }
1140
1141         ret = misc_register(&nvmf_misc);
1142         if (ret) {
1143                 pr_err("couldn't register misc device: %d\n", ret);
1144                 goto out_destroy_device;
1145         }
1146
1147         return 0;
1148
1149 out_destroy_device:
1150         device_destroy(nvmf_class, MKDEV(0, 0));
1151 out_destroy_class:
1152         class_destroy(nvmf_class);
1153 out_free_host:
1154         nvmf_host_put(nvmf_default_host);
1155         return ret;
1156 }
1157
1158 static void __exit nvmf_exit(void)
1159 {
1160         misc_deregister(&nvmf_misc);
1161         device_destroy(nvmf_class, MKDEV(0, 0));
1162         class_destroy(nvmf_class);
1163         nvmf_host_put(nvmf_default_host);
1164
1165         BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1166         BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1167         BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1168         BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1169 }
1170
1171 MODULE_LICENSE("GPL v2");
1172
1173 module_init(nvmf_init);
1174 module_exit(nvmf_exit);