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 memcpy(host->nqn, hostnqn, NVMF_NQN_SIZE); 61 62 list_add_tail(&host->list, &nvmf_hosts); 63 out_unlock: 64 mutex_unlock(&nvmf_hosts_mutex); 65 return host; 66 } 67 68 static struct nvmf_host *nvmf_host_default(void) 69 { 70 struct nvmf_host *host; 71 72 host = kmalloc(sizeof(*host), GFP_KERNEL); 73 if (!host) 74 return NULL; 75 76 kref_init(&host->ref); 77 snprintf(host->nqn, NVMF_NQN_SIZE, 78 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id); 79 80 mutex_lock(&nvmf_hosts_mutex); 81 list_add_tail(&host->list, &nvmf_hosts); 82 mutex_unlock(&nvmf_hosts_mutex); 83 84 return host; 85 } 86 87 static void nvmf_host_destroy(struct kref *ref) 88 { 89 struct nvmf_host *host = container_of(ref, struct nvmf_host, ref); 90 91 mutex_lock(&nvmf_hosts_mutex); 92 list_del(&host->list); 93 mutex_unlock(&nvmf_hosts_mutex); 94 95 kfree(host); 96 } 97 98 static void nvmf_host_put(struct nvmf_host *host) 99 { 100 if (host) 101 kref_put(&host->ref, nvmf_host_destroy); 102 } 103 104 /** 105 * nvmf_get_address() - Get address/port 106 * @ctrl: Host NVMe controller instance which we got the address 107 * @buf: OUTPUT parameter that will contain the address/port 108 * @size: buffer size 109 */ 110 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size) 111 { 112 int len = 0; 113 114 if (ctrl->opts->mask & NVMF_OPT_TRADDR) 115 len += snprintf(buf, size, "traddr=%s", ctrl->opts->traddr); 116 if (ctrl->opts->mask & NVMF_OPT_TRSVCID) 117 len += snprintf(buf + len, size - len, "%strsvcid=%s", 118 (len) ? "," : "", ctrl->opts->trsvcid); 119 if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR) 120 len += snprintf(buf + len, size - len, "%shost_traddr=%s", 121 (len) ? "," : "", ctrl->opts->host_traddr); 122 len += snprintf(buf + len, size - len, "\n"); 123 124 return len; 125 } 126 EXPORT_SYMBOL_GPL(nvmf_get_address); 127 128 /** 129 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function. 130 * @ctrl: Host NVMe controller instance maintaining the admin 131 * queue used to submit the property read command to 132 * the allocated NVMe controller resource on the target system. 133 * @off: Starting offset value of the targeted property 134 * register (see the fabrics section of the NVMe standard). 135 * @val: OUTPUT parameter that will contain the value of 136 * the property after a successful read. 137 * 138 * Used by the host system to retrieve a 32-bit capsule property value 139 * from an NVMe controller on the target system. 140 * 141 * ("Capsule property" is an "PCIe register concept" applied to the 142 * NVMe fabrics space.) 143 * 144 * Return: 145 * 0: successful read 146 * > 0: NVMe error status code 147 * < 0: Linux errno error code 148 */ 149 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) 150 { 151 struct nvme_command cmd; 152 union nvme_result res; 153 int ret; 154 155 memset(&cmd, 0, sizeof(cmd)); 156 cmd.prop_get.opcode = nvme_fabrics_command; 157 cmd.prop_get.fctype = nvme_fabrics_type_property_get; 158 cmd.prop_get.offset = cpu_to_le32(off); 159 160 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0, 161 NVME_QID_ANY, 0, 0); 162 163 if (ret >= 0) 164 *val = le64_to_cpu(res.u64); 165 if (unlikely(ret != 0)) 166 dev_err(ctrl->device, 167 "Property Get error: %d, offset %#x\n", 168 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 169 170 return ret; 171 } 172 EXPORT_SYMBOL_GPL(nvmf_reg_read32); 173 174 /** 175 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function. 176 * @ctrl: Host NVMe controller instance maintaining the admin 177 * queue used to submit the property read command to 178 * the allocated controller resource on the target system. 179 * @off: Starting offset value of the targeted property 180 * register (see the fabrics section of the NVMe standard). 181 * @val: OUTPUT parameter that will contain the value of 182 * the property after a successful read. 183 * 184 * Used by the host system to retrieve a 64-bit capsule property value 185 * from an NVMe controller on the target system. 186 * 187 * ("Capsule property" is an "PCIe register concept" applied to the 188 * NVMe fabrics space.) 189 * 190 * Return: 191 * 0: successful read 192 * > 0: NVMe error status code 193 * < 0: Linux errno error code 194 */ 195 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) 196 { 197 struct nvme_command cmd; 198 union nvme_result res; 199 int ret; 200 201 memset(&cmd, 0, sizeof(cmd)); 202 cmd.prop_get.opcode = nvme_fabrics_command; 203 cmd.prop_get.fctype = nvme_fabrics_type_property_get; 204 cmd.prop_get.attrib = 1; 205 cmd.prop_get.offset = cpu_to_le32(off); 206 207 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0, 208 NVME_QID_ANY, 0, 0); 209 210 if (ret >= 0) 211 *val = le64_to_cpu(res.u64); 212 if (unlikely(ret != 0)) 213 dev_err(ctrl->device, 214 "Property Get error: %d, offset %#x\n", 215 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 216 return ret; 217 } 218 EXPORT_SYMBOL_GPL(nvmf_reg_read64); 219 220 /** 221 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function. 222 * @ctrl: Host NVMe controller instance maintaining the admin 223 * queue used to submit the property read command to 224 * the allocated NVMe controller resource on the target system. 225 * @off: Starting offset value of the targeted property 226 * register (see the fabrics section of the NVMe standard). 227 * @val: Input parameter that contains the value to be 228 * written to the property. 229 * 230 * Used by the NVMe host system to write a 32-bit capsule property value 231 * to an NVMe controller on the target system. 232 * 233 * ("Capsule property" is an "PCIe register concept" applied to the 234 * NVMe fabrics space.) 235 * 236 * Return: 237 * 0: successful write 238 * > 0: NVMe error status code 239 * < 0: Linux errno error code 240 */ 241 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) 242 { 243 struct nvme_command cmd; 244 int ret; 245 246 memset(&cmd, 0, sizeof(cmd)); 247 cmd.prop_set.opcode = nvme_fabrics_command; 248 cmd.prop_set.fctype = nvme_fabrics_type_property_set; 249 cmd.prop_set.attrib = 0; 250 cmd.prop_set.offset = cpu_to_le32(off); 251 cmd.prop_set.value = cpu_to_le64(val); 252 253 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0, 254 NVME_QID_ANY, 0, 0); 255 if (unlikely(ret)) 256 dev_err(ctrl->device, 257 "Property Set error: %d, offset %#x\n", 258 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 259 return ret; 260 } 261 EXPORT_SYMBOL_GPL(nvmf_reg_write32); 262 263 /** 264 * nvmf_log_connect_error() - Error-parsing-diagnostic print 265 * out function for connect() errors. 266 * 267 * @ctrl: the specific /dev/nvmeX device that had the error. 268 * 269 * @errval: Error code to be decoded in a more human-friendly 270 * printout. 271 * 272 * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM. 273 * 274 * @cmd: This is the SQE portion of a submission capsule. 275 * 276 * @data: This is the "Data" portion of a submission capsule. 277 */ 278 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl, 279 int errval, int offset, struct nvme_command *cmd, 280 struct nvmf_connect_data *data) 281 { 282 int err_sctype = errval & (~NVME_SC_DNR); 283 284 switch (err_sctype) { 285 286 case (NVME_SC_CONNECT_INVALID_PARAM): 287 if (offset >> 16) { 288 char *inv_data = "Connect Invalid Data Parameter"; 289 290 switch (offset & 0xffff) { 291 case (offsetof(struct nvmf_connect_data, cntlid)): 292 dev_err(ctrl->device, 293 "%s, cntlid: %d\n", 294 inv_data, data->cntlid); 295 break; 296 case (offsetof(struct nvmf_connect_data, hostnqn)): 297 dev_err(ctrl->device, 298 "%s, hostnqn \"%s\"\n", 299 inv_data, data->hostnqn); 300 break; 301 case (offsetof(struct nvmf_connect_data, subsysnqn)): 302 dev_err(ctrl->device, 303 "%s, subsysnqn \"%s\"\n", 304 inv_data, data->subsysnqn); 305 break; 306 default: 307 dev_err(ctrl->device, 308 "%s, starting byte offset: %d\n", 309 inv_data, offset & 0xffff); 310 break; 311 } 312 } else { 313 char *inv_sqe = "Connect Invalid SQE Parameter"; 314 315 switch (offset) { 316 case (offsetof(struct nvmf_connect_command, qid)): 317 dev_err(ctrl->device, 318 "%s, qid %d\n", 319 inv_sqe, cmd->connect.qid); 320 break; 321 default: 322 dev_err(ctrl->device, 323 "%s, starting byte offset: %d\n", 324 inv_sqe, offset); 325 } 326 } 327 break; 328 329 case NVME_SC_CONNECT_INVALID_HOST: 330 dev_err(ctrl->device, 331 "Connect for subsystem %s is not allowed, hostnqn: %s\n", 332 data->subsysnqn, data->hostnqn); 333 break; 334 335 case NVME_SC_CONNECT_CTRL_BUSY: 336 dev_err(ctrl->device, 337 "Connect command failed: controller is busy or not available\n"); 338 break; 339 340 case NVME_SC_CONNECT_FORMAT: 341 dev_err(ctrl->device, 342 "Connect incompatible format: %d", 343 cmd->connect.recfmt); 344 break; 345 346 default: 347 dev_err(ctrl->device, 348 "Connect command failed, error wo/DNR bit: %d\n", 349 err_sctype); 350 break; 351 } /* switch (err_sctype) */ 352 } 353 354 /** 355 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect" 356 * API function. 357 * @ctrl: Host nvme controller instance used to request 358 * a new NVMe controller allocation on the target 359 * system and establish an NVMe Admin connection to 360 * that controller. 361 * 362 * This function enables an NVMe host device to request a new allocation of 363 * an NVMe controller resource on a target system as well establish a 364 * fabrics-protocol connection of the NVMe Admin queue between the 365 * host system device and the allocated NVMe controller on the 366 * target system via a NVMe Fabrics "Connect" command. 367 * 368 * Return: 369 * 0: success 370 * > 0: NVMe error status code 371 * < 0: Linux errno error code 372 * 373 */ 374 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl) 375 { 376 struct nvme_command cmd; 377 union nvme_result res; 378 struct nvmf_connect_data *data; 379 int ret; 380 381 memset(&cmd, 0, sizeof(cmd)); 382 cmd.connect.opcode = nvme_fabrics_command; 383 cmd.connect.fctype = nvme_fabrics_type_connect; 384 cmd.connect.qid = 0; 385 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1); 386 387 /* 388 * Set keep-alive timeout in seconds granularity (ms * 1000) 389 * and add a grace period for controller kato enforcement 390 */ 391 cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 : 392 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000); 393 394 data = kzalloc(sizeof(*data), GFP_KERNEL); 395 if (!data) 396 return -ENOMEM; 397 398 uuid_copy(&data->hostid, &ctrl->opts->host->id); 399 data->cntlid = cpu_to_le16(0xffff); 400 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 401 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 402 403 ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, 404 data, sizeof(*data), 0, NVME_QID_ANY, 1, 405 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT); 406 if (ret) { 407 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 408 &cmd, data); 409 goto out_free_data; 410 } 411 412 ctrl->cntlid = le16_to_cpu(res.u16); 413 414 out_free_data: 415 kfree(data); 416 return ret; 417 } 418 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue); 419 420 /** 421 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect" 422 * API function. 423 * @ctrl: Host nvme controller instance used to establish an 424 * NVMe I/O queue connection to the already allocated NVMe 425 * controller on the target system. 426 * @qid: NVMe I/O queue number for the new I/O connection between 427 * host and target (note qid == 0 is illegal as this is 428 * the Admin queue, per NVMe standard). 429 * 430 * This function issues a fabrics-protocol connection 431 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command) 432 * between the host system device and the allocated NVMe controller 433 * on the target system. 434 * 435 * Return: 436 * 0: success 437 * > 0: NVMe error status code 438 * < 0: Linux errno error code 439 */ 440 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid) 441 { 442 struct nvme_command cmd; 443 struct nvmf_connect_data *data; 444 union nvme_result res; 445 int ret; 446 447 memset(&cmd, 0, sizeof(cmd)); 448 cmd.connect.opcode = nvme_fabrics_command; 449 cmd.connect.fctype = nvme_fabrics_type_connect; 450 cmd.connect.qid = cpu_to_le16(qid); 451 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize); 452 453 data = kzalloc(sizeof(*data), GFP_KERNEL); 454 if (!data) 455 return -ENOMEM; 456 457 uuid_copy(&data->hostid, &ctrl->opts->host->id); 458 data->cntlid = cpu_to_le16(ctrl->cntlid); 459 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 460 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 461 462 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res, 463 data, sizeof(*data), 0, qid, 1, 464 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT); 465 if (ret) { 466 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 467 &cmd, data); 468 } 469 kfree(data); 470 return ret; 471 } 472 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue); 473 474 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl) 475 { 476 if (ctrl->opts->max_reconnects != -1 && 477 ctrl->nr_reconnects < ctrl->opts->max_reconnects) 478 return true; 479 480 return false; 481 } 482 EXPORT_SYMBOL_GPL(nvmf_should_reconnect); 483 484 /** 485 * nvmf_register_transport() - NVMe Fabrics Library registration function. 486 * @ops: Transport ops instance to be registered to the 487 * common fabrics library. 488 * 489 * API function that registers the type of specific transport fabric 490 * being implemented to the common NVMe fabrics library. Part of 491 * the overall init sequence of starting up a fabrics driver. 492 */ 493 int nvmf_register_transport(struct nvmf_transport_ops *ops) 494 { 495 if (!ops->create_ctrl) 496 return -EINVAL; 497 498 down_write(&nvmf_transports_rwsem); 499 list_add_tail(&ops->entry, &nvmf_transports); 500 up_write(&nvmf_transports_rwsem); 501 502 return 0; 503 } 504 EXPORT_SYMBOL_GPL(nvmf_register_transport); 505 506 /** 507 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function. 508 * @ops: Transport ops instance to be unregistered from the 509 * common fabrics library. 510 * 511 * Fabrics API function that unregisters the type of specific transport 512 * fabric being implemented from the common NVMe fabrics library. 513 * Part of the overall exit sequence of unloading the implemented driver. 514 */ 515 void nvmf_unregister_transport(struct nvmf_transport_ops *ops) 516 { 517 down_write(&nvmf_transports_rwsem); 518 list_del(&ops->entry); 519 up_write(&nvmf_transports_rwsem); 520 } 521 EXPORT_SYMBOL_GPL(nvmf_unregister_transport); 522 523 static struct nvmf_transport_ops *nvmf_lookup_transport( 524 struct nvmf_ctrl_options *opts) 525 { 526 struct nvmf_transport_ops *ops; 527 528 lockdep_assert_held(&nvmf_transports_rwsem); 529 530 list_for_each_entry(ops, &nvmf_transports, entry) { 531 if (strcmp(ops->name, opts->transport) == 0) 532 return ops; 533 } 534 535 return NULL; 536 } 537 538 static const match_table_t opt_tokens = { 539 { NVMF_OPT_TRANSPORT, "transport=%s" }, 540 { NVMF_OPT_TRADDR, "traddr=%s" }, 541 { NVMF_OPT_TRSVCID, "trsvcid=%s" }, 542 { NVMF_OPT_NQN, "nqn=%s" }, 543 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" }, 544 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" }, 545 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" }, 546 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" }, 547 { NVMF_OPT_KATO, "keep_alive_tmo=%d" }, 548 { NVMF_OPT_HOSTNQN, "hostnqn=%s" }, 549 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" }, 550 { NVMF_OPT_HOST_ID, "hostid=%s" }, 551 { NVMF_OPT_ERR, NULL } 552 }; 553 554 static int nvmf_parse_options(struct nvmf_ctrl_options *opts, 555 const char *buf) 556 { 557 substring_t args[MAX_OPT_ARGS]; 558 char *options, *o, *p; 559 int token, ret = 0; 560 size_t nqnlen = 0; 561 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO; 562 uuid_t hostid; 563 564 /* Set defaults */ 565 opts->queue_size = NVMF_DEF_QUEUE_SIZE; 566 opts->nr_io_queues = num_online_cpus(); 567 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY; 568 opts->kato = NVME_DEFAULT_KATO; 569 570 options = o = kstrdup(buf, GFP_KERNEL); 571 if (!options) 572 return -ENOMEM; 573 574 uuid_gen(&hostid); 575 576 while ((p = strsep(&o, ",\n")) != NULL) { 577 if (!*p) 578 continue; 579 580 token = match_token(p, opt_tokens, args); 581 opts->mask |= token; 582 switch (token) { 583 case NVMF_OPT_TRANSPORT: 584 p = match_strdup(args); 585 if (!p) { 586 ret = -ENOMEM; 587 goto out; 588 } 589 opts->transport = p; 590 break; 591 case NVMF_OPT_NQN: 592 p = match_strdup(args); 593 if (!p) { 594 ret = -ENOMEM; 595 goto out; 596 } 597 opts->subsysnqn = p; 598 nqnlen = strlen(opts->subsysnqn); 599 if (nqnlen >= NVMF_NQN_SIZE) { 600 pr_err("%s needs to be < %d bytes\n", 601 opts->subsysnqn, NVMF_NQN_SIZE); 602 ret = -EINVAL; 603 goto out; 604 } 605 opts->discovery_nqn = 606 !(strcmp(opts->subsysnqn, 607 NVME_DISC_SUBSYS_NAME)); 608 if (opts->discovery_nqn) 609 opts->nr_io_queues = 0; 610 break; 611 case NVMF_OPT_TRADDR: 612 p = match_strdup(args); 613 if (!p) { 614 ret = -ENOMEM; 615 goto out; 616 } 617 opts->traddr = p; 618 break; 619 case NVMF_OPT_TRSVCID: 620 p = match_strdup(args); 621 if (!p) { 622 ret = -ENOMEM; 623 goto out; 624 } 625 opts->trsvcid = p; 626 break; 627 case NVMF_OPT_QUEUE_SIZE: 628 if (match_int(args, &token)) { 629 ret = -EINVAL; 630 goto out; 631 } 632 if (token < NVMF_MIN_QUEUE_SIZE || 633 token > NVMF_MAX_QUEUE_SIZE) { 634 pr_err("Invalid queue_size %d\n", token); 635 ret = -EINVAL; 636 goto out; 637 } 638 opts->queue_size = token; 639 break; 640 case NVMF_OPT_NR_IO_QUEUES: 641 if (match_int(args, &token)) { 642 ret = -EINVAL; 643 goto out; 644 } 645 if (token <= 0) { 646 pr_err("Invalid number of IOQs %d\n", token); 647 ret = -EINVAL; 648 goto out; 649 } 650 opts->nr_io_queues = min_t(unsigned int, 651 num_online_cpus(), token); 652 break; 653 case NVMF_OPT_KATO: 654 if (match_int(args, &token)) { 655 ret = -EINVAL; 656 goto out; 657 } 658 659 if (token < 0) { 660 pr_err("Invalid keep_alive_tmo %d\n", token); 661 ret = -EINVAL; 662 goto out; 663 } else if (token == 0 && !opts->discovery_nqn) { 664 /* Allowed for debug */ 665 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n"); 666 } 667 opts->kato = token; 668 669 if (opts->discovery_nqn && opts->kato) { 670 pr_err("Discovery controllers cannot accept KATO != 0\n"); 671 ret = -EINVAL; 672 goto out; 673 } 674 675 break; 676 case NVMF_OPT_CTRL_LOSS_TMO: 677 if (match_int(args, &token)) { 678 ret = -EINVAL; 679 goto out; 680 } 681 682 if (token < 0) 683 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n"); 684 ctrl_loss_tmo = token; 685 break; 686 case NVMF_OPT_HOSTNQN: 687 if (opts->host) { 688 pr_err("hostnqn already user-assigned: %s\n", 689 opts->host->nqn); 690 ret = -EADDRINUSE; 691 goto out; 692 } 693 p = match_strdup(args); 694 if (!p) { 695 ret = -ENOMEM; 696 goto out; 697 } 698 nqnlen = strlen(p); 699 if (nqnlen >= NVMF_NQN_SIZE) { 700 pr_err("%s needs to be < %d bytes\n", 701 p, NVMF_NQN_SIZE); 702 kfree(p); 703 ret = -EINVAL; 704 goto out; 705 } 706 opts->host = nvmf_host_add(p); 707 kfree(p); 708 if (!opts->host) { 709 ret = -ENOMEM; 710 goto out; 711 } 712 break; 713 case NVMF_OPT_RECONNECT_DELAY: 714 if (match_int(args, &token)) { 715 ret = -EINVAL; 716 goto out; 717 } 718 if (token <= 0) { 719 pr_err("Invalid reconnect_delay %d\n", token); 720 ret = -EINVAL; 721 goto out; 722 } 723 opts->reconnect_delay = token; 724 break; 725 case NVMF_OPT_HOST_TRADDR: 726 p = match_strdup(args); 727 if (!p) { 728 ret = -ENOMEM; 729 goto out; 730 } 731 opts->host_traddr = p; 732 break; 733 case NVMF_OPT_HOST_ID: 734 p = match_strdup(args); 735 if (!p) { 736 ret = -ENOMEM; 737 goto out; 738 } 739 if (uuid_parse(p, &hostid)) { 740 pr_err("Invalid hostid %s\n", p); 741 ret = -EINVAL; 742 goto out; 743 } 744 break; 745 default: 746 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n", 747 p); 748 ret = -EINVAL; 749 goto out; 750 } 751 } 752 753 if (ctrl_loss_tmo < 0) 754 opts->max_reconnects = -1; 755 else 756 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo, 757 opts->reconnect_delay); 758 759 if (!opts->host) { 760 kref_get(&nvmf_default_host->ref); 761 opts->host = nvmf_default_host; 762 } 763 764 uuid_copy(&opts->host->id, &hostid); 765 766 out: 767 kfree(options); 768 return ret; 769 } 770 771 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts, 772 unsigned int required_opts) 773 { 774 if ((opts->mask & required_opts) != required_opts) { 775 int i; 776 777 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 778 if ((opt_tokens[i].token & required_opts) && 779 !(opt_tokens[i].token & opts->mask)) { 780 pr_warn("missing parameter '%s'\n", 781 opt_tokens[i].pattern); 782 } 783 } 784 785 return -EINVAL; 786 } 787 788 return 0; 789 } 790 791 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts, 792 unsigned int allowed_opts) 793 { 794 if (opts->mask & ~allowed_opts) { 795 int i; 796 797 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 798 if ((opt_tokens[i].token & opts->mask) && 799 (opt_tokens[i].token & ~allowed_opts)) { 800 pr_warn("invalid parameter '%s'\n", 801 opt_tokens[i].pattern); 802 } 803 } 804 805 return -EINVAL; 806 } 807 808 return 0; 809 } 810 811 void nvmf_free_options(struct nvmf_ctrl_options *opts) 812 { 813 nvmf_host_put(opts->host); 814 kfree(opts->transport); 815 kfree(opts->traddr); 816 kfree(opts->trsvcid); 817 kfree(opts->subsysnqn); 818 kfree(opts->host_traddr); 819 kfree(opts); 820 } 821 EXPORT_SYMBOL_GPL(nvmf_free_options); 822 823 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN) 824 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \ 825 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \ 826 NVMF_OPT_HOST_ID) 827 828 static struct nvme_ctrl * 829 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count) 830 { 831 struct nvmf_ctrl_options *opts; 832 struct nvmf_transport_ops *ops; 833 struct nvme_ctrl *ctrl; 834 int ret; 835 836 opts = kzalloc(sizeof(*opts), GFP_KERNEL); 837 if (!opts) 838 return ERR_PTR(-ENOMEM); 839 840 ret = nvmf_parse_options(opts, buf); 841 if (ret) 842 goto out_free_opts; 843 844 /* 845 * Check the generic options first as we need a valid transport for 846 * the lookup below. Then clear the generic flags so that transport 847 * drivers don't have to care about them. 848 */ 849 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS); 850 if (ret) 851 goto out_free_opts; 852 opts->mask &= ~NVMF_REQUIRED_OPTS; 853 854 down_read(&nvmf_transports_rwsem); 855 ops = nvmf_lookup_transport(opts); 856 if (!ops) { 857 pr_info("no handler found for transport %s.\n", 858 opts->transport); 859 ret = -EINVAL; 860 goto out_unlock; 861 } 862 863 ret = nvmf_check_required_opts(opts, ops->required_opts); 864 if (ret) 865 goto out_unlock; 866 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS | 867 ops->allowed_opts | ops->required_opts); 868 if (ret) 869 goto out_unlock; 870 871 ctrl = ops->create_ctrl(dev, opts); 872 if (IS_ERR(ctrl)) { 873 ret = PTR_ERR(ctrl); 874 goto out_unlock; 875 } 876 877 if (strcmp(ctrl->subnqn, opts->subsysnqn)) { 878 dev_warn(ctrl->device, 879 "controller returned incorrect NQN: \"%s\".\n", 880 ctrl->subnqn); 881 up_read(&nvmf_transports_rwsem); 882 ctrl->ops->delete_ctrl(ctrl); 883 return ERR_PTR(-EINVAL); 884 } 885 886 up_read(&nvmf_transports_rwsem); 887 return ctrl; 888 889 out_unlock: 890 up_read(&nvmf_transports_rwsem); 891 out_free_opts: 892 nvmf_free_options(opts); 893 return ERR_PTR(ret); 894 } 895 896 static struct class *nvmf_class; 897 static struct device *nvmf_device; 898 static DEFINE_MUTEX(nvmf_dev_mutex); 899 900 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf, 901 size_t count, loff_t *pos) 902 { 903 struct seq_file *seq_file = file->private_data; 904 struct nvme_ctrl *ctrl; 905 const char *buf; 906 int ret = 0; 907 908 if (count > PAGE_SIZE) 909 return -ENOMEM; 910 911 buf = memdup_user_nul(ubuf, count); 912 if (IS_ERR(buf)) 913 return PTR_ERR(buf); 914 915 mutex_lock(&nvmf_dev_mutex); 916 if (seq_file->private) { 917 ret = -EINVAL; 918 goto out_unlock; 919 } 920 921 ctrl = nvmf_create_ctrl(nvmf_device, buf, count); 922 if (IS_ERR(ctrl)) { 923 ret = PTR_ERR(ctrl); 924 goto out_unlock; 925 } 926 927 seq_file->private = ctrl; 928 929 out_unlock: 930 mutex_unlock(&nvmf_dev_mutex); 931 kfree(buf); 932 return ret ? ret : count; 933 } 934 935 static int nvmf_dev_show(struct seq_file *seq_file, void *private) 936 { 937 struct nvme_ctrl *ctrl; 938 int ret = 0; 939 940 mutex_lock(&nvmf_dev_mutex); 941 ctrl = seq_file->private; 942 if (!ctrl) { 943 ret = -EINVAL; 944 goto out_unlock; 945 } 946 947 seq_printf(seq_file, "instance=%d,cntlid=%d\n", 948 ctrl->instance, ctrl->cntlid); 949 950 out_unlock: 951 mutex_unlock(&nvmf_dev_mutex); 952 return ret; 953 } 954 955 static int nvmf_dev_open(struct inode *inode, struct file *file) 956 { 957 /* 958 * The miscdevice code initializes file->private_data, but doesn't 959 * make use of it later. 960 */ 961 file->private_data = NULL; 962 return single_open(file, nvmf_dev_show, NULL); 963 } 964 965 static int nvmf_dev_release(struct inode *inode, struct file *file) 966 { 967 struct seq_file *seq_file = file->private_data; 968 struct nvme_ctrl *ctrl = seq_file->private; 969 970 if (ctrl) 971 nvme_put_ctrl(ctrl); 972 return single_release(inode, file); 973 } 974 975 static const struct file_operations nvmf_dev_fops = { 976 .owner = THIS_MODULE, 977 .write = nvmf_dev_write, 978 .read = seq_read, 979 .open = nvmf_dev_open, 980 .release = nvmf_dev_release, 981 }; 982 983 static struct miscdevice nvmf_misc = { 984 .minor = MISC_DYNAMIC_MINOR, 985 .name = "nvme-fabrics", 986 .fops = &nvmf_dev_fops, 987 }; 988 989 static int __init nvmf_init(void) 990 { 991 int ret; 992 993 nvmf_default_host = nvmf_host_default(); 994 if (!nvmf_default_host) 995 return -ENOMEM; 996 997 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics"); 998 if (IS_ERR(nvmf_class)) { 999 pr_err("couldn't register class nvme-fabrics\n"); 1000 ret = PTR_ERR(nvmf_class); 1001 goto out_free_host; 1002 } 1003 1004 nvmf_device = 1005 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl"); 1006 if (IS_ERR(nvmf_device)) { 1007 pr_err("couldn't create nvme-fabris device!\n"); 1008 ret = PTR_ERR(nvmf_device); 1009 goto out_destroy_class; 1010 } 1011 1012 ret = misc_register(&nvmf_misc); 1013 if (ret) { 1014 pr_err("couldn't register misc device: %d\n", ret); 1015 goto out_destroy_device; 1016 } 1017 1018 return 0; 1019 1020 out_destroy_device: 1021 device_destroy(nvmf_class, MKDEV(0, 0)); 1022 out_destroy_class: 1023 class_destroy(nvmf_class); 1024 out_free_host: 1025 nvmf_host_put(nvmf_default_host); 1026 return ret; 1027 } 1028 1029 static void __exit nvmf_exit(void) 1030 { 1031 misc_deregister(&nvmf_misc); 1032 device_destroy(nvmf_class, MKDEV(0, 0)); 1033 class_destroy(nvmf_class); 1034 nvmf_host_put(nvmf_default_host); 1035 1036 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64); 1037 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64); 1038 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64); 1039 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024); 1040 } 1041 1042 MODULE_LICENSE("GPL v2"); 1043 1044 module_init(nvmf_init); 1045 module_exit(nvmf_exit); 1046