1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * NVMe over Fabrics common host code. 4 * Copyright (c) 2015-2016 HGST, a Western Digital Company. 5 */ 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 #include <linux/init.h> 8 #include <linux/miscdevice.h> 9 #include <linux/module.h> 10 #include <linux/mutex.h> 11 #include <linux/parser.h> 12 #include <linux/seq_file.h> 13 #include "nvme.h" 14 #include "fabrics.h" 15 16 static LIST_HEAD(nvmf_transports); 17 static DECLARE_RWSEM(nvmf_transports_rwsem); 18 19 static LIST_HEAD(nvmf_hosts); 20 static DEFINE_MUTEX(nvmf_hosts_mutex); 21 22 static struct nvmf_host *nvmf_default_host; 23 24 static struct nvmf_host *__nvmf_host_find(const char *hostnqn) 25 { 26 struct nvmf_host *host; 27 28 list_for_each_entry(host, &nvmf_hosts, list) { 29 if (!strcmp(host->nqn, hostnqn)) 30 return host; 31 } 32 33 return NULL; 34 } 35 36 static struct nvmf_host *nvmf_host_add(const char *hostnqn) 37 { 38 struct nvmf_host *host; 39 40 mutex_lock(&nvmf_hosts_mutex); 41 host = __nvmf_host_find(hostnqn); 42 if (host) { 43 kref_get(&host->ref); 44 goto out_unlock; 45 } 46 47 host = kmalloc(sizeof(*host), GFP_KERNEL); 48 if (!host) 49 goto out_unlock; 50 51 kref_init(&host->ref); 52 strscpy(host->nqn, hostnqn, NVMF_NQN_SIZE); 53 54 list_add_tail(&host->list, &nvmf_hosts); 55 out_unlock: 56 mutex_unlock(&nvmf_hosts_mutex); 57 return host; 58 } 59 60 static struct nvmf_host *nvmf_host_default(void) 61 { 62 struct nvmf_host *host; 63 64 host = kmalloc(sizeof(*host), GFP_KERNEL); 65 if (!host) 66 return NULL; 67 68 kref_init(&host->ref); 69 uuid_gen(&host->id); 70 snprintf(host->nqn, NVMF_NQN_SIZE, 71 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id); 72 73 mutex_lock(&nvmf_hosts_mutex); 74 list_add_tail(&host->list, &nvmf_hosts); 75 mutex_unlock(&nvmf_hosts_mutex); 76 77 return host; 78 } 79 80 static void nvmf_host_destroy(struct kref *ref) 81 { 82 struct nvmf_host *host = container_of(ref, struct nvmf_host, ref); 83 84 mutex_lock(&nvmf_hosts_mutex); 85 list_del(&host->list); 86 mutex_unlock(&nvmf_hosts_mutex); 87 88 kfree(host); 89 } 90 91 static void nvmf_host_put(struct nvmf_host *host) 92 { 93 if (host) 94 kref_put(&host->ref, nvmf_host_destroy); 95 } 96 97 /** 98 * nvmf_get_address() - Get address/port 99 * @ctrl: Host NVMe controller instance which we got the address 100 * @buf: OUTPUT parameter that will contain the address/port 101 * @size: buffer size 102 */ 103 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size) 104 { 105 int len = 0; 106 107 if (ctrl->opts->mask & NVMF_OPT_TRADDR) 108 len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr); 109 if (ctrl->opts->mask & NVMF_OPT_TRSVCID) 110 len += scnprintf(buf + len, size - len, "%strsvcid=%s", 111 (len) ? "," : "", ctrl->opts->trsvcid); 112 if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR) 113 len += scnprintf(buf + len, size - len, "%shost_traddr=%s", 114 (len) ? "," : "", ctrl->opts->host_traddr); 115 if (ctrl->opts->mask & NVMF_OPT_HOST_IFACE) 116 len += scnprintf(buf + len, size - len, "%shost_iface=%s", 117 (len) ? "," : "", ctrl->opts->host_iface); 118 len += scnprintf(buf + len, size - len, "\n"); 119 120 return len; 121 } 122 EXPORT_SYMBOL_GPL(nvmf_get_address); 123 124 /** 125 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function. 126 * @ctrl: Host NVMe controller instance maintaining the admin 127 * queue used to submit the property read command to 128 * the allocated NVMe controller resource on the target system. 129 * @off: Starting offset value of the targeted property 130 * register (see the fabrics section of the NVMe standard). 131 * @val: OUTPUT parameter that will contain the value of 132 * the property after a successful read. 133 * 134 * Used by the host system to retrieve a 32-bit capsule property value 135 * from an NVMe controller on the target system. 136 * 137 * ("Capsule property" is an "PCIe register concept" applied to the 138 * NVMe fabrics space.) 139 * 140 * Return: 141 * 0: successful read 142 * > 0: NVMe error status code 143 * < 0: Linux errno error code 144 */ 145 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) 146 { 147 struct nvme_command cmd = { }; 148 union nvme_result res; 149 int ret; 150 151 cmd.prop_get.opcode = nvme_fabrics_command; 152 cmd.prop_get.fctype = nvme_fabrics_type_property_get; 153 cmd.prop_get.offset = cpu_to_le32(off); 154 155 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 156 NVME_QID_ANY, 0, 0); 157 158 if (ret >= 0) 159 *val = le64_to_cpu(res.u64); 160 if (unlikely(ret != 0)) 161 dev_err(ctrl->device, 162 "Property Get error: %d, offset %#x\n", 163 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 164 165 return ret; 166 } 167 EXPORT_SYMBOL_GPL(nvmf_reg_read32); 168 169 /** 170 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function. 171 * @ctrl: Host NVMe controller instance maintaining the admin 172 * queue used to submit the property read command to 173 * the allocated controller resource on the target system. 174 * @off: Starting offset value of the targeted property 175 * register (see the fabrics section of the NVMe standard). 176 * @val: OUTPUT parameter that will contain the value of 177 * the property after a successful read. 178 * 179 * Used by the host system to retrieve a 64-bit capsule property value 180 * from an NVMe controller on the target system. 181 * 182 * ("Capsule property" is an "PCIe register concept" applied to the 183 * NVMe fabrics space.) 184 * 185 * Return: 186 * 0: successful read 187 * > 0: NVMe error status code 188 * < 0: Linux errno error code 189 */ 190 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) 191 { 192 struct nvme_command cmd = { }; 193 union nvme_result res; 194 int ret; 195 196 cmd.prop_get.opcode = nvme_fabrics_command; 197 cmd.prop_get.fctype = nvme_fabrics_type_property_get; 198 cmd.prop_get.attrib = 1; 199 cmd.prop_get.offset = cpu_to_le32(off); 200 201 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 202 NVME_QID_ANY, 0, 0); 203 204 if (ret >= 0) 205 *val = le64_to_cpu(res.u64); 206 if (unlikely(ret != 0)) 207 dev_err(ctrl->device, 208 "Property Get error: %d, offset %#x\n", 209 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 210 return ret; 211 } 212 EXPORT_SYMBOL_GPL(nvmf_reg_read64); 213 214 /** 215 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function. 216 * @ctrl: Host NVMe controller instance maintaining the admin 217 * queue used to submit the property read command to 218 * the allocated NVMe controller resource on the target system. 219 * @off: Starting offset value of the targeted property 220 * register (see the fabrics section of the NVMe standard). 221 * @val: Input parameter that contains the value to be 222 * written to the property. 223 * 224 * Used by the NVMe host system to write a 32-bit capsule property value 225 * to an NVMe controller on the target system. 226 * 227 * ("Capsule property" is an "PCIe register concept" applied to the 228 * NVMe fabrics space.) 229 * 230 * Return: 231 * 0: successful write 232 * > 0: NVMe error status code 233 * < 0: Linux errno error code 234 */ 235 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) 236 { 237 struct nvme_command cmd = { }; 238 int ret; 239 240 cmd.prop_set.opcode = nvme_fabrics_command; 241 cmd.prop_set.fctype = nvme_fabrics_type_property_set; 242 cmd.prop_set.attrib = 0; 243 cmd.prop_set.offset = cpu_to_le32(off); 244 cmd.prop_set.value = cpu_to_le64(val); 245 246 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0, 247 NVME_QID_ANY, 0, 0); 248 if (unlikely(ret)) 249 dev_err(ctrl->device, 250 "Property Set error: %d, offset %#x\n", 251 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 252 return ret; 253 } 254 EXPORT_SYMBOL_GPL(nvmf_reg_write32); 255 256 /** 257 * nvmf_log_connect_error() - Error-parsing-diagnostic print out function for 258 * connect() errors. 259 * @ctrl: The specific /dev/nvmeX device that had the error. 260 * @errval: Error code to be decoded in a more human-friendly 261 * printout. 262 * @offset: For use with the NVMe error code 263 * NVME_SC_CONNECT_INVALID_PARAM. 264 * @cmd: This is the SQE portion of a submission capsule. 265 * @data: This is the "Data" portion of a submission capsule. 266 */ 267 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl, 268 int errval, int offset, struct nvme_command *cmd, 269 struct nvmf_connect_data *data) 270 { 271 int err_sctype = errval & ~NVME_SC_DNR; 272 273 if (errval < 0) { 274 dev_err(ctrl->device, 275 "Connect command failed, errno: %d\n", errval); 276 return; 277 } 278 279 switch (err_sctype) { 280 case NVME_SC_CONNECT_INVALID_PARAM: 281 if (offset >> 16) { 282 char *inv_data = "Connect Invalid Data Parameter"; 283 284 switch (offset & 0xffff) { 285 case (offsetof(struct nvmf_connect_data, cntlid)): 286 dev_err(ctrl->device, 287 "%s, cntlid: %d\n", 288 inv_data, data->cntlid); 289 break; 290 case (offsetof(struct nvmf_connect_data, hostnqn)): 291 dev_err(ctrl->device, 292 "%s, hostnqn \"%s\"\n", 293 inv_data, data->hostnqn); 294 break; 295 case (offsetof(struct nvmf_connect_data, subsysnqn)): 296 dev_err(ctrl->device, 297 "%s, subsysnqn \"%s\"\n", 298 inv_data, data->subsysnqn); 299 break; 300 default: 301 dev_err(ctrl->device, 302 "%s, starting byte offset: %d\n", 303 inv_data, offset & 0xffff); 304 break; 305 } 306 } else { 307 char *inv_sqe = "Connect Invalid SQE Parameter"; 308 309 switch (offset) { 310 case (offsetof(struct nvmf_connect_command, qid)): 311 dev_err(ctrl->device, 312 "%s, qid %d\n", 313 inv_sqe, cmd->connect.qid); 314 break; 315 default: 316 dev_err(ctrl->device, 317 "%s, starting byte offset: %d\n", 318 inv_sqe, offset); 319 } 320 } 321 break; 322 case NVME_SC_CONNECT_INVALID_HOST: 323 dev_err(ctrl->device, 324 "Connect for subsystem %s is not allowed, hostnqn: %s\n", 325 data->subsysnqn, data->hostnqn); 326 break; 327 case NVME_SC_CONNECT_CTRL_BUSY: 328 dev_err(ctrl->device, 329 "Connect command failed: controller is busy or not available\n"); 330 break; 331 case NVME_SC_CONNECT_FORMAT: 332 dev_err(ctrl->device, 333 "Connect incompatible format: %d", 334 cmd->connect.recfmt); 335 break; 336 case NVME_SC_HOST_PATH_ERROR: 337 dev_err(ctrl->device, 338 "Connect command failed: host path error\n"); 339 break; 340 case NVME_SC_AUTH_REQUIRED: 341 dev_err(ctrl->device, 342 "Connect command failed: authentication required\n"); 343 break; 344 default: 345 dev_err(ctrl->device, 346 "Connect command failed, error wo/DNR bit: %d\n", 347 err_sctype); 348 break; 349 } 350 } 351 352 /** 353 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect" 354 * API function. 355 * @ctrl: Host nvme controller instance used to request 356 * a new NVMe controller allocation on the target 357 * system and establish an NVMe Admin connection to 358 * that controller. 359 * 360 * This function enables an NVMe host device to request a new allocation of 361 * an NVMe controller resource on a target system as well establish a 362 * fabrics-protocol connection of the NVMe Admin queue between the 363 * host system device and the allocated NVMe controller on the 364 * target system via a NVMe Fabrics "Connect" command. 365 * 366 * Return: 367 * 0: success 368 * > 0: NVMe error status code 369 * < 0: Linux errno error code 370 * 371 */ 372 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl) 373 { 374 struct nvme_command cmd = { }; 375 union nvme_result res; 376 struct nvmf_connect_data *data; 377 int ret; 378 u32 result; 379 380 cmd.connect.opcode = nvme_fabrics_command; 381 cmd.connect.fctype = nvme_fabrics_type_connect; 382 cmd.connect.qid = 0; 383 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1); 384 385 /* 386 * Set keep-alive timeout in seconds granularity (ms * 1000) 387 */ 388 cmd.connect.kato = cpu_to_le32(ctrl->kato * 1000); 389 390 if (ctrl->opts->disable_sqflow) 391 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW; 392 393 data = kzalloc(sizeof(*data), GFP_KERNEL); 394 if (!data) 395 return -ENOMEM; 396 397 uuid_copy(&data->hostid, &ctrl->opts->host->id); 398 data->cntlid = cpu_to_le16(0xffff); 399 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 400 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 401 402 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, 403 data, sizeof(*data), NVME_QID_ANY, 1, 404 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT); 405 if (ret) { 406 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 407 &cmd, data); 408 goto out_free_data; 409 } 410 411 result = le32_to_cpu(res.u32); 412 ctrl->cntlid = result & 0xFFFF; 413 if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) { 414 /* Secure concatenation is not implemented */ 415 if (result & NVME_CONNECT_AUTHREQ_ASCR) { 416 dev_warn(ctrl->device, 417 "qid 0: secure concatenation is not supported\n"); 418 ret = NVME_SC_AUTH_REQUIRED; 419 goto out_free_data; 420 } 421 /* Authentication required */ 422 ret = nvme_auth_negotiate(ctrl, 0); 423 if (ret) { 424 dev_warn(ctrl->device, 425 "qid 0: authentication setup failed\n"); 426 ret = NVME_SC_AUTH_REQUIRED; 427 goto out_free_data; 428 } 429 ret = nvme_auth_wait(ctrl, 0); 430 if (ret) 431 dev_warn(ctrl->device, 432 "qid 0: authentication failed\n"); 433 else 434 dev_info(ctrl->device, 435 "qid 0: authenticated\n"); 436 } 437 out_free_data: 438 kfree(data); 439 return ret; 440 } 441 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue); 442 443 /** 444 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect" 445 * API function. 446 * @ctrl: Host nvme controller instance used to establish an 447 * NVMe I/O queue connection to the already allocated NVMe 448 * controller on the target system. 449 * @qid: NVMe I/O queue number for the new I/O connection between 450 * host and target (note qid == 0 is illegal as this is 451 * the Admin queue, per NVMe standard). 452 * 453 * This function issues a fabrics-protocol connection 454 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command) 455 * between the host system device and the allocated NVMe controller 456 * on the target system. 457 * 458 * Return: 459 * 0: success 460 * > 0: NVMe error status code 461 * < 0: Linux errno error code 462 */ 463 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid) 464 { 465 struct nvme_command cmd = { }; 466 struct nvmf_connect_data *data; 467 union nvme_result res; 468 int ret; 469 u32 result; 470 471 cmd.connect.opcode = nvme_fabrics_command; 472 cmd.connect.fctype = nvme_fabrics_type_connect; 473 cmd.connect.qid = cpu_to_le16(qid); 474 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize); 475 476 if (ctrl->opts->disable_sqflow) 477 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW; 478 479 data = kzalloc(sizeof(*data), GFP_KERNEL); 480 if (!data) 481 return -ENOMEM; 482 483 uuid_copy(&data->hostid, &ctrl->opts->host->id); 484 data->cntlid = cpu_to_le16(ctrl->cntlid); 485 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 486 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 487 488 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res, 489 data, sizeof(*data), qid, 1, 490 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT); 491 if (ret) { 492 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 493 &cmd, data); 494 } 495 result = le32_to_cpu(res.u32); 496 if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) { 497 /* Secure concatenation is not implemented */ 498 if (result & NVME_CONNECT_AUTHREQ_ASCR) { 499 dev_warn(ctrl->device, 500 "qid 0: secure concatenation is not supported\n"); 501 ret = NVME_SC_AUTH_REQUIRED; 502 goto out_free_data; 503 } 504 /* Authentication required */ 505 ret = nvme_auth_negotiate(ctrl, qid); 506 if (ret) { 507 dev_warn(ctrl->device, 508 "qid %d: authentication setup failed\n", qid); 509 ret = NVME_SC_AUTH_REQUIRED; 510 } else { 511 ret = nvme_auth_wait(ctrl, qid); 512 if (ret) 513 dev_warn(ctrl->device, 514 "qid %u: authentication failed\n", qid); 515 } 516 } 517 out_free_data: 518 kfree(data); 519 return ret; 520 } 521 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue); 522 523 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl) 524 { 525 if (ctrl->opts->max_reconnects == -1 || 526 ctrl->nr_reconnects < ctrl->opts->max_reconnects) 527 return true; 528 529 return false; 530 } 531 EXPORT_SYMBOL_GPL(nvmf_should_reconnect); 532 533 /** 534 * nvmf_register_transport() - NVMe Fabrics Library registration function. 535 * @ops: Transport ops instance to be registered to the 536 * common fabrics library. 537 * 538 * API function that registers the type of specific transport fabric 539 * being implemented to the common NVMe fabrics library. Part of 540 * the overall init sequence of starting up a fabrics driver. 541 */ 542 int nvmf_register_transport(struct nvmf_transport_ops *ops) 543 { 544 if (!ops->create_ctrl) 545 return -EINVAL; 546 547 down_write(&nvmf_transports_rwsem); 548 list_add_tail(&ops->entry, &nvmf_transports); 549 up_write(&nvmf_transports_rwsem); 550 551 return 0; 552 } 553 EXPORT_SYMBOL_GPL(nvmf_register_transport); 554 555 /** 556 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function. 557 * @ops: Transport ops instance to be unregistered from the 558 * common fabrics library. 559 * 560 * Fabrics API function that unregisters the type of specific transport 561 * fabric being implemented from the common NVMe fabrics library. 562 * Part of the overall exit sequence of unloading the implemented driver. 563 */ 564 void nvmf_unregister_transport(struct nvmf_transport_ops *ops) 565 { 566 down_write(&nvmf_transports_rwsem); 567 list_del(&ops->entry); 568 up_write(&nvmf_transports_rwsem); 569 } 570 EXPORT_SYMBOL_GPL(nvmf_unregister_transport); 571 572 static struct nvmf_transport_ops *nvmf_lookup_transport( 573 struct nvmf_ctrl_options *opts) 574 { 575 struct nvmf_transport_ops *ops; 576 577 lockdep_assert_held(&nvmf_transports_rwsem); 578 579 list_for_each_entry(ops, &nvmf_transports, entry) { 580 if (strcmp(ops->name, opts->transport) == 0) 581 return ops; 582 } 583 584 return NULL; 585 } 586 587 static const match_table_t opt_tokens = { 588 { NVMF_OPT_TRANSPORT, "transport=%s" }, 589 { NVMF_OPT_TRADDR, "traddr=%s" }, 590 { NVMF_OPT_TRSVCID, "trsvcid=%s" }, 591 { NVMF_OPT_NQN, "nqn=%s" }, 592 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" }, 593 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" }, 594 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" }, 595 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" }, 596 { NVMF_OPT_KATO, "keep_alive_tmo=%d" }, 597 { NVMF_OPT_HOSTNQN, "hostnqn=%s" }, 598 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" }, 599 { NVMF_OPT_HOST_IFACE, "host_iface=%s" }, 600 { NVMF_OPT_HOST_ID, "hostid=%s" }, 601 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" }, 602 { NVMF_OPT_DISABLE_SQFLOW, "disable_sqflow" }, 603 { NVMF_OPT_HDR_DIGEST, "hdr_digest" }, 604 { NVMF_OPT_DATA_DIGEST, "data_digest" }, 605 { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" }, 606 { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" }, 607 { NVMF_OPT_TOS, "tos=%d" }, 608 { NVMF_OPT_FAIL_FAST_TMO, "fast_io_fail_tmo=%d" }, 609 { NVMF_OPT_DISCOVERY, "discovery" }, 610 { NVMF_OPT_DHCHAP_SECRET, "dhchap_secret=%s" }, 611 { NVMF_OPT_DHCHAP_CTRL_SECRET, "dhchap_ctrl_secret=%s" }, 612 { NVMF_OPT_ERR, NULL } 613 }; 614 615 static int nvmf_parse_options(struct nvmf_ctrl_options *opts, 616 const char *buf) 617 { 618 substring_t args[MAX_OPT_ARGS]; 619 char *options, *o, *p; 620 int token, ret = 0; 621 size_t nqnlen = 0; 622 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO; 623 uuid_t hostid; 624 625 /* Set defaults */ 626 opts->queue_size = NVMF_DEF_QUEUE_SIZE; 627 opts->nr_io_queues = num_online_cpus(); 628 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY; 629 opts->kato = 0; 630 opts->duplicate_connect = false; 631 opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO; 632 opts->hdr_digest = false; 633 opts->data_digest = false; 634 opts->tos = -1; /* < 0 == use transport default */ 635 636 options = o = kstrdup(buf, GFP_KERNEL); 637 if (!options) 638 return -ENOMEM; 639 640 uuid_gen(&hostid); 641 642 while ((p = strsep(&o, ",\n")) != NULL) { 643 if (!*p) 644 continue; 645 646 token = match_token(p, opt_tokens, args); 647 opts->mask |= token; 648 switch (token) { 649 case NVMF_OPT_TRANSPORT: 650 p = match_strdup(args); 651 if (!p) { 652 ret = -ENOMEM; 653 goto out; 654 } 655 kfree(opts->transport); 656 opts->transport = p; 657 break; 658 case NVMF_OPT_NQN: 659 p = match_strdup(args); 660 if (!p) { 661 ret = -ENOMEM; 662 goto out; 663 } 664 kfree(opts->subsysnqn); 665 opts->subsysnqn = p; 666 nqnlen = strlen(opts->subsysnqn); 667 if (nqnlen >= NVMF_NQN_SIZE) { 668 pr_err("%s needs to be < %d bytes\n", 669 opts->subsysnqn, NVMF_NQN_SIZE); 670 ret = -EINVAL; 671 goto out; 672 } 673 opts->discovery_nqn = 674 !(strcmp(opts->subsysnqn, 675 NVME_DISC_SUBSYS_NAME)); 676 break; 677 case NVMF_OPT_TRADDR: 678 p = match_strdup(args); 679 if (!p) { 680 ret = -ENOMEM; 681 goto out; 682 } 683 kfree(opts->traddr); 684 opts->traddr = p; 685 break; 686 case NVMF_OPT_TRSVCID: 687 p = match_strdup(args); 688 if (!p) { 689 ret = -ENOMEM; 690 goto out; 691 } 692 kfree(opts->trsvcid); 693 opts->trsvcid = p; 694 break; 695 case NVMF_OPT_QUEUE_SIZE: 696 if (match_int(args, &token)) { 697 ret = -EINVAL; 698 goto out; 699 } 700 if (token < NVMF_MIN_QUEUE_SIZE || 701 token > NVMF_MAX_QUEUE_SIZE) { 702 pr_err("Invalid queue_size %d\n", token); 703 ret = -EINVAL; 704 goto out; 705 } 706 opts->queue_size = token; 707 break; 708 case NVMF_OPT_NR_IO_QUEUES: 709 if (match_int(args, &token)) { 710 ret = -EINVAL; 711 goto out; 712 } 713 if (token <= 0) { 714 pr_err("Invalid number of IOQs %d\n", token); 715 ret = -EINVAL; 716 goto out; 717 } 718 if (opts->discovery_nqn) { 719 pr_debug("Ignoring nr_io_queues value for discovery controller\n"); 720 break; 721 } 722 723 opts->nr_io_queues = min_t(unsigned int, 724 num_online_cpus(), token); 725 break; 726 case NVMF_OPT_KATO: 727 if (match_int(args, &token)) { 728 ret = -EINVAL; 729 goto out; 730 } 731 732 if (token < 0) { 733 pr_err("Invalid keep_alive_tmo %d\n", token); 734 ret = -EINVAL; 735 goto out; 736 } else if (token == 0 && !opts->discovery_nqn) { 737 /* Allowed for debug */ 738 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n"); 739 } 740 opts->kato = token; 741 break; 742 case NVMF_OPT_CTRL_LOSS_TMO: 743 if (match_int(args, &token)) { 744 ret = -EINVAL; 745 goto out; 746 } 747 748 if (token < 0) 749 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n"); 750 ctrl_loss_tmo = token; 751 break; 752 case NVMF_OPT_FAIL_FAST_TMO: 753 if (match_int(args, &token)) { 754 ret = -EINVAL; 755 goto out; 756 } 757 758 if (token >= 0) 759 pr_warn("I/O fail on reconnect controller after %d sec\n", 760 token); 761 else 762 token = -1; 763 764 opts->fast_io_fail_tmo = token; 765 break; 766 case NVMF_OPT_HOSTNQN: 767 if (opts->host) { 768 pr_err("hostnqn already user-assigned: %s\n", 769 opts->host->nqn); 770 ret = -EADDRINUSE; 771 goto out; 772 } 773 p = match_strdup(args); 774 if (!p) { 775 ret = -ENOMEM; 776 goto out; 777 } 778 nqnlen = strlen(p); 779 if (nqnlen >= NVMF_NQN_SIZE) { 780 pr_err("%s needs to be < %d bytes\n", 781 p, NVMF_NQN_SIZE); 782 kfree(p); 783 ret = -EINVAL; 784 goto out; 785 } 786 opts->host = nvmf_host_add(p); 787 kfree(p); 788 if (!opts->host) { 789 ret = -ENOMEM; 790 goto out; 791 } 792 break; 793 case NVMF_OPT_RECONNECT_DELAY: 794 if (match_int(args, &token)) { 795 ret = -EINVAL; 796 goto out; 797 } 798 if (token <= 0) { 799 pr_err("Invalid reconnect_delay %d\n", token); 800 ret = -EINVAL; 801 goto out; 802 } 803 opts->reconnect_delay = token; 804 break; 805 case NVMF_OPT_HOST_TRADDR: 806 p = match_strdup(args); 807 if (!p) { 808 ret = -ENOMEM; 809 goto out; 810 } 811 kfree(opts->host_traddr); 812 opts->host_traddr = p; 813 break; 814 case NVMF_OPT_HOST_IFACE: 815 p = match_strdup(args); 816 if (!p) { 817 ret = -ENOMEM; 818 goto out; 819 } 820 kfree(opts->host_iface); 821 opts->host_iface = p; 822 break; 823 case NVMF_OPT_HOST_ID: 824 p = match_strdup(args); 825 if (!p) { 826 ret = -ENOMEM; 827 goto out; 828 } 829 ret = uuid_parse(p, &hostid); 830 if (ret) { 831 pr_err("Invalid hostid %s\n", p); 832 ret = -EINVAL; 833 kfree(p); 834 goto out; 835 } 836 kfree(p); 837 break; 838 case NVMF_OPT_DUP_CONNECT: 839 opts->duplicate_connect = true; 840 break; 841 case NVMF_OPT_DISABLE_SQFLOW: 842 opts->disable_sqflow = true; 843 break; 844 case NVMF_OPT_HDR_DIGEST: 845 opts->hdr_digest = true; 846 break; 847 case NVMF_OPT_DATA_DIGEST: 848 opts->data_digest = true; 849 break; 850 case NVMF_OPT_NR_WRITE_QUEUES: 851 if (match_int(args, &token)) { 852 ret = -EINVAL; 853 goto out; 854 } 855 if (token <= 0) { 856 pr_err("Invalid nr_write_queues %d\n", token); 857 ret = -EINVAL; 858 goto out; 859 } 860 opts->nr_write_queues = token; 861 break; 862 case NVMF_OPT_NR_POLL_QUEUES: 863 if (match_int(args, &token)) { 864 ret = -EINVAL; 865 goto out; 866 } 867 if (token <= 0) { 868 pr_err("Invalid nr_poll_queues %d\n", token); 869 ret = -EINVAL; 870 goto out; 871 } 872 opts->nr_poll_queues = token; 873 break; 874 case NVMF_OPT_TOS: 875 if (match_int(args, &token)) { 876 ret = -EINVAL; 877 goto out; 878 } 879 if (token < 0) { 880 pr_err("Invalid type of service %d\n", token); 881 ret = -EINVAL; 882 goto out; 883 } 884 if (token > 255) { 885 pr_warn("Clamping type of service to 255\n"); 886 token = 255; 887 } 888 opts->tos = token; 889 break; 890 case NVMF_OPT_DISCOVERY: 891 opts->discovery_nqn = true; 892 break; 893 case NVMF_OPT_DHCHAP_SECRET: 894 p = match_strdup(args); 895 if (!p) { 896 ret = -ENOMEM; 897 goto out; 898 } 899 if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) { 900 pr_err("Invalid DH-CHAP secret %s\n", p); 901 ret = -EINVAL; 902 goto out; 903 } 904 kfree(opts->dhchap_secret); 905 opts->dhchap_secret = p; 906 break; 907 case NVMF_OPT_DHCHAP_CTRL_SECRET: 908 p = match_strdup(args); 909 if (!p) { 910 ret = -ENOMEM; 911 goto out; 912 } 913 if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) { 914 pr_err("Invalid DH-CHAP secret %s\n", p); 915 ret = -EINVAL; 916 goto out; 917 } 918 kfree(opts->dhchap_ctrl_secret); 919 opts->dhchap_ctrl_secret = p; 920 break; 921 default: 922 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n", 923 p); 924 ret = -EINVAL; 925 goto out; 926 } 927 } 928 929 if (opts->discovery_nqn) { 930 opts->nr_io_queues = 0; 931 opts->nr_write_queues = 0; 932 opts->nr_poll_queues = 0; 933 opts->duplicate_connect = true; 934 } else { 935 if (!opts->kato) 936 opts->kato = NVME_DEFAULT_KATO; 937 } 938 if (ctrl_loss_tmo < 0) { 939 opts->max_reconnects = -1; 940 } else { 941 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo, 942 opts->reconnect_delay); 943 if (ctrl_loss_tmo < opts->fast_io_fail_tmo) 944 pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n", 945 opts->fast_io_fail_tmo, ctrl_loss_tmo); 946 } 947 948 if (!opts->host) { 949 kref_get(&nvmf_default_host->ref); 950 opts->host = nvmf_default_host; 951 } 952 953 uuid_copy(&opts->host->id, &hostid); 954 955 out: 956 kfree(options); 957 return ret; 958 } 959 960 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts, 961 unsigned int required_opts) 962 { 963 if ((opts->mask & required_opts) != required_opts) { 964 unsigned int i; 965 966 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 967 if ((opt_tokens[i].token & required_opts) && 968 !(opt_tokens[i].token & opts->mask)) { 969 pr_warn("missing parameter '%s'\n", 970 opt_tokens[i].pattern); 971 } 972 } 973 974 return -EINVAL; 975 } 976 977 return 0; 978 } 979 980 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl, 981 struct nvmf_ctrl_options *opts) 982 { 983 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) || 984 strcmp(opts->traddr, ctrl->opts->traddr) || 985 strcmp(opts->trsvcid, ctrl->opts->trsvcid)) 986 return false; 987 988 /* 989 * Checking the local address or host interfaces is rough. 990 * 991 * In most cases, none is specified and the host port or 992 * host interface is selected by the stack. 993 * 994 * Assume no match if: 995 * - local address or host interface is specified and address 996 * or host interface is not the same 997 * - local address or host interface is not specified but 998 * remote is, or vice versa (admin using specific 999 * host_traddr/host_iface when it matters). 1000 */ 1001 if ((opts->mask & NVMF_OPT_HOST_TRADDR) && 1002 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) { 1003 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr)) 1004 return false; 1005 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) || 1006 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) { 1007 return false; 1008 } 1009 1010 if ((opts->mask & NVMF_OPT_HOST_IFACE) && 1011 (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) { 1012 if (strcmp(opts->host_iface, ctrl->opts->host_iface)) 1013 return false; 1014 } else if ((opts->mask & NVMF_OPT_HOST_IFACE) || 1015 (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) { 1016 return false; 1017 } 1018 1019 return true; 1020 } 1021 EXPORT_SYMBOL_GPL(nvmf_ip_options_match); 1022 1023 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts, 1024 unsigned int allowed_opts) 1025 { 1026 if (opts->mask & ~allowed_opts) { 1027 unsigned int i; 1028 1029 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 1030 if ((opt_tokens[i].token & opts->mask) && 1031 (opt_tokens[i].token & ~allowed_opts)) { 1032 pr_warn("invalid parameter '%s'\n", 1033 opt_tokens[i].pattern); 1034 } 1035 } 1036 1037 return -EINVAL; 1038 } 1039 1040 return 0; 1041 } 1042 1043 void nvmf_free_options(struct nvmf_ctrl_options *opts) 1044 { 1045 nvmf_host_put(opts->host); 1046 kfree(opts->transport); 1047 kfree(opts->traddr); 1048 kfree(opts->trsvcid); 1049 kfree(opts->subsysnqn); 1050 kfree(opts->host_traddr); 1051 kfree(opts->host_iface); 1052 kfree(opts->dhchap_secret); 1053 kfree(opts->dhchap_ctrl_secret); 1054 kfree(opts); 1055 } 1056 EXPORT_SYMBOL_GPL(nvmf_free_options); 1057 1058 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN) 1059 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \ 1060 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \ 1061 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\ 1062 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\ 1063 NVMF_OPT_FAIL_FAST_TMO | NVMF_OPT_DHCHAP_SECRET |\ 1064 NVMF_OPT_DHCHAP_CTRL_SECRET) 1065 1066 static struct nvme_ctrl * 1067 nvmf_create_ctrl(struct device *dev, const char *buf) 1068 { 1069 struct nvmf_ctrl_options *opts; 1070 struct nvmf_transport_ops *ops; 1071 struct nvme_ctrl *ctrl; 1072 int ret; 1073 1074 opts = kzalloc(sizeof(*opts), GFP_KERNEL); 1075 if (!opts) 1076 return ERR_PTR(-ENOMEM); 1077 1078 ret = nvmf_parse_options(opts, buf); 1079 if (ret) 1080 goto out_free_opts; 1081 1082 1083 request_module("nvme-%s", opts->transport); 1084 1085 /* 1086 * Check the generic options first as we need a valid transport for 1087 * the lookup below. Then clear the generic flags so that transport 1088 * drivers don't have to care about them. 1089 */ 1090 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS); 1091 if (ret) 1092 goto out_free_opts; 1093 opts->mask &= ~NVMF_REQUIRED_OPTS; 1094 1095 down_read(&nvmf_transports_rwsem); 1096 ops = nvmf_lookup_transport(opts); 1097 if (!ops) { 1098 pr_info("no handler found for transport %s.\n", 1099 opts->transport); 1100 ret = -EINVAL; 1101 goto out_unlock; 1102 } 1103 1104 if (!try_module_get(ops->module)) { 1105 ret = -EBUSY; 1106 goto out_unlock; 1107 } 1108 up_read(&nvmf_transports_rwsem); 1109 1110 ret = nvmf_check_required_opts(opts, ops->required_opts); 1111 if (ret) 1112 goto out_module_put; 1113 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS | 1114 ops->allowed_opts | ops->required_opts); 1115 if (ret) 1116 goto out_module_put; 1117 1118 ctrl = ops->create_ctrl(dev, opts); 1119 if (IS_ERR(ctrl)) { 1120 ret = PTR_ERR(ctrl); 1121 goto out_module_put; 1122 } 1123 1124 module_put(ops->module); 1125 return ctrl; 1126 1127 out_module_put: 1128 module_put(ops->module); 1129 goto out_free_opts; 1130 out_unlock: 1131 up_read(&nvmf_transports_rwsem); 1132 out_free_opts: 1133 nvmf_free_options(opts); 1134 return ERR_PTR(ret); 1135 } 1136 1137 static struct class *nvmf_class; 1138 static struct device *nvmf_device; 1139 static DEFINE_MUTEX(nvmf_dev_mutex); 1140 1141 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf, 1142 size_t count, loff_t *pos) 1143 { 1144 struct seq_file *seq_file = file->private_data; 1145 struct nvme_ctrl *ctrl; 1146 const char *buf; 1147 int ret = 0; 1148 1149 if (count > PAGE_SIZE) 1150 return -ENOMEM; 1151 1152 buf = memdup_user_nul(ubuf, count); 1153 if (IS_ERR(buf)) 1154 return PTR_ERR(buf); 1155 1156 mutex_lock(&nvmf_dev_mutex); 1157 if (seq_file->private) { 1158 ret = -EINVAL; 1159 goto out_unlock; 1160 } 1161 1162 ctrl = nvmf_create_ctrl(nvmf_device, buf); 1163 if (IS_ERR(ctrl)) { 1164 ret = PTR_ERR(ctrl); 1165 goto out_unlock; 1166 } 1167 1168 seq_file->private = ctrl; 1169 1170 out_unlock: 1171 mutex_unlock(&nvmf_dev_mutex); 1172 kfree(buf); 1173 return ret ? ret : count; 1174 } 1175 1176 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file) 1177 { 1178 const struct match_token *tok; 1179 int idx; 1180 1181 /* 1182 * Add dummy entries for instance and cntlid to 1183 * signal an invalid/non-existing controller 1184 */ 1185 seq_puts(seq_file, "instance=-1,cntlid=-1"); 1186 for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) { 1187 tok = &opt_tokens[idx]; 1188 if (tok->token == NVMF_OPT_ERR) 1189 continue; 1190 seq_puts(seq_file, ","); 1191 seq_puts(seq_file, tok->pattern); 1192 } 1193 seq_puts(seq_file, "\n"); 1194 } 1195 1196 static int nvmf_dev_show(struct seq_file *seq_file, void *private) 1197 { 1198 struct nvme_ctrl *ctrl; 1199 1200 mutex_lock(&nvmf_dev_mutex); 1201 ctrl = seq_file->private; 1202 if (!ctrl) { 1203 __nvmf_concat_opt_tokens(seq_file); 1204 goto out_unlock; 1205 } 1206 1207 seq_printf(seq_file, "instance=%d,cntlid=%d\n", 1208 ctrl->instance, ctrl->cntlid); 1209 1210 out_unlock: 1211 mutex_unlock(&nvmf_dev_mutex); 1212 return 0; 1213 } 1214 1215 static int nvmf_dev_open(struct inode *inode, struct file *file) 1216 { 1217 /* 1218 * The miscdevice code initializes file->private_data, but doesn't 1219 * make use of it later. 1220 */ 1221 file->private_data = NULL; 1222 return single_open(file, nvmf_dev_show, NULL); 1223 } 1224 1225 static int nvmf_dev_release(struct inode *inode, struct file *file) 1226 { 1227 struct seq_file *seq_file = file->private_data; 1228 struct nvme_ctrl *ctrl = seq_file->private; 1229 1230 if (ctrl) 1231 nvme_put_ctrl(ctrl); 1232 return single_release(inode, file); 1233 } 1234 1235 static const struct file_operations nvmf_dev_fops = { 1236 .owner = THIS_MODULE, 1237 .write = nvmf_dev_write, 1238 .read = seq_read, 1239 .open = nvmf_dev_open, 1240 .release = nvmf_dev_release, 1241 }; 1242 1243 static struct miscdevice nvmf_misc = { 1244 .minor = MISC_DYNAMIC_MINOR, 1245 .name = "nvme-fabrics", 1246 .fops = &nvmf_dev_fops, 1247 }; 1248 1249 static int __init nvmf_init(void) 1250 { 1251 int ret; 1252 1253 nvmf_default_host = nvmf_host_default(); 1254 if (!nvmf_default_host) 1255 return -ENOMEM; 1256 1257 nvmf_class = class_create("nvme-fabrics"); 1258 if (IS_ERR(nvmf_class)) { 1259 pr_err("couldn't register class nvme-fabrics\n"); 1260 ret = PTR_ERR(nvmf_class); 1261 goto out_free_host; 1262 } 1263 1264 nvmf_device = 1265 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl"); 1266 if (IS_ERR(nvmf_device)) { 1267 pr_err("couldn't create nvme-fabrics device!\n"); 1268 ret = PTR_ERR(nvmf_device); 1269 goto out_destroy_class; 1270 } 1271 1272 ret = misc_register(&nvmf_misc); 1273 if (ret) { 1274 pr_err("couldn't register misc device: %d\n", ret); 1275 goto out_destroy_device; 1276 } 1277 1278 return 0; 1279 1280 out_destroy_device: 1281 device_destroy(nvmf_class, MKDEV(0, 0)); 1282 out_destroy_class: 1283 class_destroy(nvmf_class); 1284 out_free_host: 1285 nvmf_host_put(nvmf_default_host); 1286 return ret; 1287 } 1288 1289 static void __exit nvmf_exit(void) 1290 { 1291 misc_deregister(&nvmf_misc); 1292 device_destroy(nvmf_class, MKDEV(0, 0)); 1293 class_destroy(nvmf_class); 1294 nvmf_host_put(nvmf_default_host); 1295 1296 BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64); 1297 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64); 1298 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64); 1299 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64); 1300 BUILD_BUG_ON(sizeof(struct nvmf_auth_send_command) != 64); 1301 BUILD_BUG_ON(sizeof(struct nvmf_auth_receive_command) != 64); 1302 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024); 1303 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_negotiate_data) != 8); 1304 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_challenge_data) != 16); 1305 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_reply_data) != 16); 1306 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success1_data) != 16); 1307 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success2_data) != 16); 1308 } 1309 1310 MODULE_LICENSE("GPL v2"); 1311 1312 module_init(nvmf_init); 1313 module_exit(nvmf_exit); 1314