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 strlcpy(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 switch (err_sctype) { 274 case NVME_SC_CONNECT_INVALID_PARAM: 275 if (offset >> 16) { 276 char *inv_data = "Connect Invalid Data Parameter"; 277 278 switch (offset & 0xffff) { 279 case (offsetof(struct nvmf_connect_data, cntlid)): 280 dev_err(ctrl->device, 281 "%s, cntlid: %d\n", 282 inv_data, data->cntlid); 283 break; 284 case (offsetof(struct nvmf_connect_data, hostnqn)): 285 dev_err(ctrl->device, 286 "%s, hostnqn \"%s\"\n", 287 inv_data, data->hostnqn); 288 break; 289 case (offsetof(struct nvmf_connect_data, subsysnqn)): 290 dev_err(ctrl->device, 291 "%s, subsysnqn \"%s\"\n", 292 inv_data, data->subsysnqn); 293 break; 294 default: 295 dev_err(ctrl->device, 296 "%s, starting byte offset: %d\n", 297 inv_data, offset & 0xffff); 298 break; 299 } 300 } else { 301 char *inv_sqe = "Connect Invalid SQE Parameter"; 302 303 switch (offset) { 304 case (offsetof(struct nvmf_connect_command, qid)): 305 dev_err(ctrl->device, 306 "%s, qid %d\n", 307 inv_sqe, cmd->connect.qid); 308 break; 309 default: 310 dev_err(ctrl->device, 311 "%s, starting byte offset: %d\n", 312 inv_sqe, offset); 313 } 314 } 315 break; 316 case NVME_SC_CONNECT_INVALID_HOST: 317 dev_err(ctrl->device, 318 "Connect for subsystem %s is not allowed, hostnqn: %s\n", 319 data->subsysnqn, data->hostnqn); 320 break; 321 case NVME_SC_CONNECT_CTRL_BUSY: 322 dev_err(ctrl->device, 323 "Connect command failed: controller is busy or not available\n"); 324 break; 325 case NVME_SC_CONNECT_FORMAT: 326 dev_err(ctrl->device, 327 "Connect incompatible format: %d", 328 cmd->connect.recfmt); 329 break; 330 case NVME_SC_HOST_PATH_ERROR: 331 dev_err(ctrl->device, 332 "Connect command failed: host path error\n"); 333 break; 334 case NVME_SC_AUTH_REQUIRED: 335 dev_err(ctrl->device, 336 "Connect command failed: authentication required\n"); 337 break; 338 default: 339 dev_err(ctrl->device, 340 "Connect command failed, error wo/DNR bit: %d\n", 341 err_sctype); 342 break; 343 } 344 } 345 346 /** 347 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect" 348 * API function. 349 * @ctrl: Host nvme controller instance used to request 350 * a new NVMe controller allocation on the target 351 * system and establish an NVMe Admin connection to 352 * that controller. 353 * 354 * This function enables an NVMe host device to request a new allocation of 355 * an NVMe controller resource on a target system as well establish a 356 * fabrics-protocol connection of the NVMe Admin queue between the 357 * host system device and the allocated NVMe controller on the 358 * target system via a NVMe Fabrics "Connect" command. 359 * 360 * Return: 361 * 0: success 362 * > 0: NVMe error status code 363 * < 0: Linux errno error code 364 * 365 */ 366 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl) 367 { 368 struct nvme_command cmd = { }; 369 union nvme_result res; 370 struct nvmf_connect_data *data; 371 int ret; 372 373 cmd.connect.opcode = nvme_fabrics_command; 374 cmd.connect.fctype = nvme_fabrics_type_connect; 375 cmd.connect.qid = 0; 376 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1); 377 378 /* 379 * Set keep-alive timeout in seconds granularity (ms * 1000) 380 */ 381 cmd.connect.kato = cpu_to_le32(ctrl->kato * 1000); 382 383 if (ctrl->opts->disable_sqflow) 384 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW; 385 386 data = kzalloc(sizeof(*data), GFP_KERNEL); 387 if (!data) 388 return -ENOMEM; 389 390 uuid_copy(&data->hostid, &ctrl->opts->host->id); 391 data->cntlid = cpu_to_le16(0xffff); 392 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 393 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 394 395 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, 396 data, sizeof(*data), NVME_QID_ANY, 1, 397 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT); 398 if (ret) { 399 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 400 &cmd, data); 401 goto out_free_data; 402 } 403 404 ctrl->cntlid = le16_to_cpu(res.u16); 405 406 out_free_data: 407 kfree(data); 408 return ret; 409 } 410 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue); 411 412 /** 413 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect" 414 * API function. 415 * @ctrl: Host nvme controller instance used to establish an 416 * NVMe I/O queue connection to the already allocated NVMe 417 * controller on the target system. 418 * @qid: NVMe I/O queue number for the new I/O connection between 419 * host and target (note qid == 0 is illegal as this is 420 * the Admin queue, per NVMe standard). 421 * 422 * This function issues a fabrics-protocol connection 423 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command) 424 * between the host system device and the allocated NVMe controller 425 * on the target system. 426 * 427 * Return: 428 * 0: success 429 * > 0: NVMe error status code 430 * < 0: Linux errno error code 431 */ 432 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid) 433 { 434 struct nvme_command cmd = { }; 435 struct nvmf_connect_data *data; 436 union nvme_result res; 437 int ret; 438 439 cmd.connect.opcode = nvme_fabrics_command; 440 cmd.connect.fctype = nvme_fabrics_type_connect; 441 cmd.connect.qid = cpu_to_le16(qid); 442 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize); 443 444 if (ctrl->opts->disable_sqflow) 445 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW; 446 447 data = kzalloc(sizeof(*data), GFP_KERNEL); 448 if (!data) 449 return -ENOMEM; 450 451 uuid_copy(&data->hostid, &ctrl->opts->host->id); 452 data->cntlid = cpu_to_le16(ctrl->cntlid); 453 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 454 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 455 456 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res, 457 data, sizeof(*data), qid, 1, 458 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT); 459 if (ret) { 460 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 461 &cmd, data); 462 } 463 kfree(data); 464 return ret; 465 } 466 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue); 467 468 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl) 469 { 470 if (ctrl->opts->max_reconnects == -1 || 471 ctrl->nr_reconnects < ctrl->opts->max_reconnects) 472 return true; 473 474 return false; 475 } 476 EXPORT_SYMBOL_GPL(nvmf_should_reconnect); 477 478 /** 479 * nvmf_register_transport() - NVMe Fabrics Library registration function. 480 * @ops: Transport ops instance to be registered to the 481 * common fabrics library. 482 * 483 * API function that registers the type of specific transport fabric 484 * being implemented to the common NVMe fabrics library. Part of 485 * the overall init sequence of starting up a fabrics driver. 486 */ 487 int nvmf_register_transport(struct nvmf_transport_ops *ops) 488 { 489 if (!ops->create_ctrl) 490 return -EINVAL; 491 492 down_write(&nvmf_transports_rwsem); 493 list_add_tail(&ops->entry, &nvmf_transports); 494 up_write(&nvmf_transports_rwsem); 495 496 return 0; 497 } 498 EXPORT_SYMBOL_GPL(nvmf_register_transport); 499 500 /** 501 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function. 502 * @ops: Transport ops instance to be unregistered from the 503 * common fabrics library. 504 * 505 * Fabrics API function that unregisters the type of specific transport 506 * fabric being implemented from the common NVMe fabrics library. 507 * Part of the overall exit sequence of unloading the implemented driver. 508 */ 509 void nvmf_unregister_transport(struct nvmf_transport_ops *ops) 510 { 511 down_write(&nvmf_transports_rwsem); 512 list_del(&ops->entry); 513 up_write(&nvmf_transports_rwsem); 514 } 515 EXPORT_SYMBOL_GPL(nvmf_unregister_transport); 516 517 static struct nvmf_transport_ops *nvmf_lookup_transport( 518 struct nvmf_ctrl_options *opts) 519 { 520 struct nvmf_transport_ops *ops; 521 522 lockdep_assert_held(&nvmf_transports_rwsem); 523 524 list_for_each_entry(ops, &nvmf_transports, entry) { 525 if (strcmp(ops->name, opts->transport) == 0) 526 return ops; 527 } 528 529 return NULL; 530 } 531 532 static const match_table_t opt_tokens = { 533 { NVMF_OPT_TRANSPORT, "transport=%s" }, 534 { NVMF_OPT_TRADDR, "traddr=%s" }, 535 { NVMF_OPT_TRSVCID, "trsvcid=%s" }, 536 { NVMF_OPT_NQN, "nqn=%s" }, 537 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" }, 538 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" }, 539 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" }, 540 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" }, 541 { NVMF_OPT_KATO, "keep_alive_tmo=%d" }, 542 { NVMF_OPT_HOSTNQN, "hostnqn=%s" }, 543 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" }, 544 { NVMF_OPT_HOST_IFACE, "host_iface=%s" }, 545 { NVMF_OPT_HOST_ID, "hostid=%s" }, 546 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" }, 547 { NVMF_OPT_DISABLE_SQFLOW, "disable_sqflow" }, 548 { NVMF_OPT_HDR_DIGEST, "hdr_digest" }, 549 { NVMF_OPT_DATA_DIGEST, "data_digest" }, 550 { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" }, 551 { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" }, 552 { NVMF_OPT_TOS, "tos=%d" }, 553 { NVMF_OPT_FAIL_FAST_TMO, "fast_io_fail_tmo=%d" }, 554 { NVMF_OPT_DISCOVERY, "discovery" }, 555 { NVMF_OPT_ERR, NULL } 556 }; 557 558 static int nvmf_parse_options(struct nvmf_ctrl_options *opts, 559 const char *buf) 560 { 561 substring_t args[MAX_OPT_ARGS]; 562 char *options, *o, *p; 563 int token, ret = 0; 564 size_t nqnlen = 0; 565 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO; 566 uuid_t hostid; 567 568 /* Set defaults */ 569 opts->queue_size = NVMF_DEF_QUEUE_SIZE; 570 opts->nr_io_queues = num_online_cpus(); 571 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY; 572 opts->kato = 0; 573 opts->duplicate_connect = false; 574 opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO; 575 opts->hdr_digest = false; 576 opts->data_digest = false; 577 opts->tos = -1; /* < 0 == use transport default */ 578 579 options = o = kstrdup(buf, GFP_KERNEL); 580 if (!options) 581 return -ENOMEM; 582 583 uuid_gen(&hostid); 584 585 while ((p = strsep(&o, ",\n")) != NULL) { 586 if (!*p) 587 continue; 588 589 token = match_token(p, opt_tokens, args); 590 opts->mask |= token; 591 switch (token) { 592 case NVMF_OPT_TRANSPORT: 593 p = match_strdup(args); 594 if (!p) { 595 ret = -ENOMEM; 596 goto out; 597 } 598 kfree(opts->transport); 599 opts->transport = p; 600 break; 601 case NVMF_OPT_NQN: 602 p = match_strdup(args); 603 if (!p) { 604 ret = -ENOMEM; 605 goto out; 606 } 607 kfree(opts->subsysnqn); 608 opts->subsysnqn = p; 609 nqnlen = strlen(opts->subsysnqn); 610 if (nqnlen >= NVMF_NQN_SIZE) { 611 pr_err("%s needs to be < %d bytes\n", 612 opts->subsysnqn, NVMF_NQN_SIZE); 613 ret = -EINVAL; 614 goto out; 615 } 616 opts->discovery_nqn = 617 !(strcmp(opts->subsysnqn, 618 NVME_DISC_SUBSYS_NAME)); 619 break; 620 case NVMF_OPT_TRADDR: 621 p = match_strdup(args); 622 if (!p) { 623 ret = -ENOMEM; 624 goto out; 625 } 626 kfree(opts->traddr); 627 opts->traddr = p; 628 break; 629 case NVMF_OPT_TRSVCID: 630 p = match_strdup(args); 631 if (!p) { 632 ret = -ENOMEM; 633 goto out; 634 } 635 kfree(opts->trsvcid); 636 opts->trsvcid = p; 637 break; 638 case NVMF_OPT_QUEUE_SIZE: 639 if (match_int(args, &token)) { 640 ret = -EINVAL; 641 goto out; 642 } 643 if (token < NVMF_MIN_QUEUE_SIZE || 644 token > NVMF_MAX_QUEUE_SIZE) { 645 pr_err("Invalid queue_size %d\n", token); 646 ret = -EINVAL; 647 goto out; 648 } 649 opts->queue_size = token; 650 break; 651 case NVMF_OPT_NR_IO_QUEUES: 652 if (match_int(args, &token)) { 653 ret = -EINVAL; 654 goto out; 655 } 656 if (token <= 0) { 657 pr_err("Invalid number of IOQs %d\n", token); 658 ret = -EINVAL; 659 goto out; 660 } 661 if (opts->discovery_nqn) { 662 pr_debug("Ignoring nr_io_queues value for discovery controller\n"); 663 break; 664 } 665 666 opts->nr_io_queues = min_t(unsigned int, 667 num_online_cpus(), token); 668 break; 669 case NVMF_OPT_KATO: 670 if (match_int(args, &token)) { 671 ret = -EINVAL; 672 goto out; 673 } 674 675 if (token < 0) { 676 pr_err("Invalid keep_alive_tmo %d\n", token); 677 ret = -EINVAL; 678 goto out; 679 } else if (token == 0 && !opts->discovery_nqn) { 680 /* Allowed for debug */ 681 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n"); 682 } 683 opts->kato = token; 684 break; 685 case NVMF_OPT_CTRL_LOSS_TMO: 686 if (match_int(args, &token)) { 687 ret = -EINVAL; 688 goto out; 689 } 690 691 if (token < 0) 692 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n"); 693 ctrl_loss_tmo = token; 694 break; 695 case NVMF_OPT_FAIL_FAST_TMO: 696 if (match_int(args, &token)) { 697 ret = -EINVAL; 698 goto out; 699 } 700 701 if (token >= 0) 702 pr_warn("I/O fail on reconnect controller after %d sec\n", 703 token); 704 else 705 token = -1; 706 707 opts->fast_io_fail_tmo = token; 708 break; 709 case NVMF_OPT_HOSTNQN: 710 if (opts->host) { 711 pr_err("hostnqn already user-assigned: %s\n", 712 opts->host->nqn); 713 ret = -EADDRINUSE; 714 goto out; 715 } 716 p = match_strdup(args); 717 if (!p) { 718 ret = -ENOMEM; 719 goto out; 720 } 721 nqnlen = strlen(p); 722 if (nqnlen >= NVMF_NQN_SIZE) { 723 pr_err("%s needs to be < %d bytes\n", 724 p, NVMF_NQN_SIZE); 725 kfree(p); 726 ret = -EINVAL; 727 goto out; 728 } 729 opts->host = nvmf_host_add(p); 730 kfree(p); 731 if (!opts->host) { 732 ret = -ENOMEM; 733 goto out; 734 } 735 break; 736 case NVMF_OPT_RECONNECT_DELAY: 737 if (match_int(args, &token)) { 738 ret = -EINVAL; 739 goto out; 740 } 741 if (token <= 0) { 742 pr_err("Invalid reconnect_delay %d\n", token); 743 ret = -EINVAL; 744 goto out; 745 } 746 opts->reconnect_delay = token; 747 break; 748 case NVMF_OPT_HOST_TRADDR: 749 p = match_strdup(args); 750 if (!p) { 751 ret = -ENOMEM; 752 goto out; 753 } 754 kfree(opts->host_traddr); 755 opts->host_traddr = p; 756 break; 757 case NVMF_OPT_HOST_IFACE: 758 p = match_strdup(args); 759 if (!p) { 760 ret = -ENOMEM; 761 goto out; 762 } 763 kfree(opts->host_iface); 764 opts->host_iface = p; 765 break; 766 case NVMF_OPT_HOST_ID: 767 p = match_strdup(args); 768 if (!p) { 769 ret = -ENOMEM; 770 goto out; 771 } 772 ret = uuid_parse(p, &hostid); 773 if (ret) { 774 pr_err("Invalid hostid %s\n", p); 775 ret = -EINVAL; 776 kfree(p); 777 goto out; 778 } 779 kfree(p); 780 break; 781 case NVMF_OPT_DUP_CONNECT: 782 opts->duplicate_connect = true; 783 break; 784 case NVMF_OPT_DISABLE_SQFLOW: 785 opts->disable_sqflow = true; 786 break; 787 case NVMF_OPT_HDR_DIGEST: 788 opts->hdr_digest = true; 789 break; 790 case NVMF_OPT_DATA_DIGEST: 791 opts->data_digest = true; 792 break; 793 case NVMF_OPT_NR_WRITE_QUEUES: 794 if (match_int(args, &token)) { 795 ret = -EINVAL; 796 goto out; 797 } 798 if (token <= 0) { 799 pr_err("Invalid nr_write_queues %d\n", token); 800 ret = -EINVAL; 801 goto out; 802 } 803 opts->nr_write_queues = token; 804 break; 805 case NVMF_OPT_NR_POLL_QUEUES: 806 if (match_int(args, &token)) { 807 ret = -EINVAL; 808 goto out; 809 } 810 if (token <= 0) { 811 pr_err("Invalid nr_poll_queues %d\n", token); 812 ret = -EINVAL; 813 goto out; 814 } 815 opts->nr_poll_queues = token; 816 break; 817 case NVMF_OPT_TOS: 818 if (match_int(args, &token)) { 819 ret = -EINVAL; 820 goto out; 821 } 822 if (token < 0) { 823 pr_err("Invalid type of service %d\n", token); 824 ret = -EINVAL; 825 goto out; 826 } 827 if (token > 255) { 828 pr_warn("Clamping type of service to 255\n"); 829 token = 255; 830 } 831 opts->tos = token; 832 break; 833 case NVMF_OPT_DISCOVERY: 834 opts->discovery_nqn = true; 835 break; 836 default: 837 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n", 838 p); 839 ret = -EINVAL; 840 goto out; 841 } 842 } 843 844 if (opts->discovery_nqn) { 845 opts->nr_io_queues = 0; 846 opts->nr_write_queues = 0; 847 opts->nr_poll_queues = 0; 848 opts->duplicate_connect = true; 849 } else { 850 if (!opts->kato) 851 opts->kato = NVME_DEFAULT_KATO; 852 } 853 if (ctrl_loss_tmo < 0) { 854 opts->max_reconnects = -1; 855 } else { 856 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo, 857 opts->reconnect_delay); 858 if (ctrl_loss_tmo < opts->fast_io_fail_tmo) 859 pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n", 860 opts->fast_io_fail_tmo, ctrl_loss_tmo); 861 } 862 863 if (!opts->host) { 864 kref_get(&nvmf_default_host->ref); 865 opts->host = nvmf_default_host; 866 } 867 868 uuid_copy(&opts->host->id, &hostid); 869 870 out: 871 kfree(options); 872 return ret; 873 } 874 875 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts, 876 unsigned int required_opts) 877 { 878 if ((opts->mask & required_opts) != required_opts) { 879 unsigned int i; 880 881 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 882 if ((opt_tokens[i].token & required_opts) && 883 !(opt_tokens[i].token & opts->mask)) { 884 pr_warn("missing parameter '%s'\n", 885 opt_tokens[i].pattern); 886 } 887 } 888 889 return -EINVAL; 890 } 891 892 return 0; 893 } 894 895 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl, 896 struct nvmf_ctrl_options *opts) 897 { 898 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) || 899 strcmp(opts->traddr, ctrl->opts->traddr) || 900 strcmp(opts->trsvcid, ctrl->opts->trsvcid)) 901 return false; 902 903 /* 904 * Checking the local address is rough. In most cases, none is specified 905 * and the host port is selected by the stack. 906 * 907 * Assume no match if: 908 * - local address is specified and address is not the same 909 * - local address is not specified but remote is, or vice versa 910 * (admin using specific host_traddr when it matters). 911 */ 912 if ((opts->mask & NVMF_OPT_HOST_TRADDR) && 913 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) { 914 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr)) 915 return false; 916 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) || 917 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) { 918 return false; 919 } 920 921 return true; 922 } 923 EXPORT_SYMBOL_GPL(nvmf_ip_options_match); 924 925 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts, 926 unsigned int allowed_opts) 927 { 928 if (opts->mask & ~allowed_opts) { 929 unsigned int i; 930 931 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 932 if ((opt_tokens[i].token & opts->mask) && 933 (opt_tokens[i].token & ~allowed_opts)) { 934 pr_warn("invalid parameter '%s'\n", 935 opt_tokens[i].pattern); 936 } 937 } 938 939 return -EINVAL; 940 } 941 942 return 0; 943 } 944 945 void nvmf_free_options(struct nvmf_ctrl_options *opts) 946 { 947 nvmf_host_put(opts->host); 948 kfree(opts->transport); 949 kfree(opts->traddr); 950 kfree(opts->trsvcid); 951 kfree(opts->subsysnqn); 952 kfree(opts->host_traddr); 953 kfree(opts->host_iface); 954 kfree(opts); 955 } 956 EXPORT_SYMBOL_GPL(nvmf_free_options); 957 958 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN) 959 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \ 960 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \ 961 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\ 962 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\ 963 NVMF_OPT_FAIL_FAST_TMO) 964 965 static struct nvme_ctrl * 966 nvmf_create_ctrl(struct device *dev, const char *buf) 967 { 968 struct nvmf_ctrl_options *opts; 969 struct nvmf_transport_ops *ops; 970 struct nvme_ctrl *ctrl; 971 int ret; 972 973 opts = kzalloc(sizeof(*opts), GFP_KERNEL); 974 if (!opts) 975 return ERR_PTR(-ENOMEM); 976 977 ret = nvmf_parse_options(opts, buf); 978 if (ret) 979 goto out_free_opts; 980 981 982 request_module("nvme-%s", opts->transport); 983 984 /* 985 * Check the generic options first as we need a valid transport for 986 * the lookup below. Then clear the generic flags so that transport 987 * drivers don't have to care about them. 988 */ 989 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS); 990 if (ret) 991 goto out_free_opts; 992 opts->mask &= ~NVMF_REQUIRED_OPTS; 993 994 down_read(&nvmf_transports_rwsem); 995 ops = nvmf_lookup_transport(opts); 996 if (!ops) { 997 pr_info("no handler found for transport %s.\n", 998 opts->transport); 999 ret = -EINVAL; 1000 goto out_unlock; 1001 } 1002 1003 if (!try_module_get(ops->module)) { 1004 ret = -EBUSY; 1005 goto out_unlock; 1006 } 1007 up_read(&nvmf_transports_rwsem); 1008 1009 ret = nvmf_check_required_opts(opts, ops->required_opts); 1010 if (ret) 1011 goto out_module_put; 1012 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS | 1013 ops->allowed_opts | ops->required_opts); 1014 if (ret) 1015 goto out_module_put; 1016 1017 ctrl = ops->create_ctrl(dev, opts); 1018 if (IS_ERR(ctrl)) { 1019 ret = PTR_ERR(ctrl); 1020 goto out_module_put; 1021 } 1022 1023 module_put(ops->module); 1024 return ctrl; 1025 1026 out_module_put: 1027 module_put(ops->module); 1028 goto out_free_opts; 1029 out_unlock: 1030 up_read(&nvmf_transports_rwsem); 1031 out_free_opts: 1032 nvmf_free_options(opts); 1033 return ERR_PTR(ret); 1034 } 1035 1036 static struct class *nvmf_class; 1037 static struct device *nvmf_device; 1038 static DEFINE_MUTEX(nvmf_dev_mutex); 1039 1040 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf, 1041 size_t count, loff_t *pos) 1042 { 1043 struct seq_file *seq_file = file->private_data; 1044 struct nvme_ctrl *ctrl; 1045 const char *buf; 1046 int ret = 0; 1047 1048 if (count > PAGE_SIZE) 1049 return -ENOMEM; 1050 1051 buf = memdup_user_nul(ubuf, count); 1052 if (IS_ERR(buf)) 1053 return PTR_ERR(buf); 1054 1055 mutex_lock(&nvmf_dev_mutex); 1056 if (seq_file->private) { 1057 ret = -EINVAL; 1058 goto out_unlock; 1059 } 1060 1061 ctrl = nvmf_create_ctrl(nvmf_device, buf); 1062 if (IS_ERR(ctrl)) { 1063 ret = PTR_ERR(ctrl); 1064 goto out_unlock; 1065 } 1066 1067 seq_file->private = ctrl; 1068 1069 out_unlock: 1070 mutex_unlock(&nvmf_dev_mutex); 1071 kfree(buf); 1072 return ret ? ret : count; 1073 } 1074 1075 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file) 1076 { 1077 const struct match_token *tok; 1078 int idx; 1079 1080 /* 1081 * Add dummy entries for instance and cntlid to 1082 * signal an invalid/non-existing controller 1083 */ 1084 seq_puts(seq_file, "instance=-1,cntlid=-1"); 1085 for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) { 1086 tok = &opt_tokens[idx]; 1087 if (tok->token == NVMF_OPT_ERR) 1088 continue; 1089 seq_puts(seq_file, ","); 1090 seq_puts(seq_file, tok->pattern); 1091 } 1092 seq_puts(seq_file, "\n"); 1093 } 1094 1095 static int nvmf_dev_show(struct seq_file *seq_file, void *private) 1096 { 1097 struct nvme_ctrl *ctrl; 1098 1099 mutex_lock(&nvmf_dev_mutex); 1100 ctrl = seq_file->private; 1101 if (!ctrl) { 1102 __nvmf_concat_opt_tokens(seq_file); 1103 goto out_unlock; 1104 } 1105 1106 seq_printf(seq_file, "instance=%d,cntlid=%d\n", 1107 ctrl->instance, ctrl->cntlid); 1108 1109 out_unlock: 1110 mutex_unlock(&nvmf_dev_mutex); 1111 return 0; 1112 } 1113 1114 static int nvmf_dev_open(struct inode *inode, struct file *file) 1115 { 1116 /* 1117 * The miscdevice code initializes file->private_data, but doesn't 1118 * make use of it later. 1119 */ 1120 file->private_data = NULL; 1121 return single_open(file, nvmf_dev_show, NULL); 1122 } 1123 1124 static int nvmf_dev_release(struct inode *inode, struct file *file) 1125 { 1126 struct seq_file *seq_file = file->private_data; 1127 struct nvme_ctrl *ctrl = seq_file->private; 1128 1129 if (ctrl) 1130 nvme_put_ctrl(ctrl); 1131 return single_release(inode, file); 1132 } 1133 1134 static const struct file_operations nvmf_dev_fops = { 1135 .owner = THIS_MODULE, 1136 .write = nvmf_dev_write, 1137 .read = seq_read, 1138 .open = nvmf_dev_open, 1139 .release = nvmf_dev_release, 1140 }; 1141 1142 static struct miscdevice nvmf_misc = { 1143 .minor = MISC_DYNAMIC_MINOR, 1144 .name = "nvme-fabrics", 1145 .fops = &nvmf_dev_fops, 1146 }; 1147 1148 static int __init nvmf_init(void) 1149 { 1150 int ret; 1151 1152 nvmf_default_host = nvmf_host_default(); 1153 if (!nvmf_default_host) 1154 return -ENOMEM; 1155 1156 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics"); 1157 if (IS_ERR(nvmf_class)) { 1158 pr_err("couldn't register class nvme-fabrics\n"); 1159 ret = PTR_ERR(nvmf_class); 1160 goto out_free_host; 1161 } 1162 1163 nvmf_device = 1164 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl"); 1165 if (IS_ERR(nvmf_device)) { 1166 pr_err("couldn't create nvme-fabris device!\n"); 1167 ret = PTR_ERR(nvmf_device); 1168 goto out_destroy_class; 1169 } 1170 1171 ret = misc_register(&nvmf_misc); 1172 if (ret) { 1173 pr_err("couldn't register misc device: %d\n", ret); 1174 goto out_destroy_device; 1175 } 1176 1177 return 0; 1178 1179 out_destroy_device: 1180 device_destroy(nvmf_class, MKDEV(0, 0)); 1181 out_destroy_class: 1182 class_destroy(nvmf_class); 1183 out_free_host: 1184 nvmf_host_put(nvmf_default_host); 1185 return ret; 1186 } 1187 1188 static void __exit nvmf_exit(void) 1189 { 1190 misc_deregister(&nvmf_misc); 1191 device_destroy(nvmf_class, MKDEV(0, 0)); 1192 class_destroy(nvmf_class); 1193 nvmf_host_put(nvmf_default_host); 1194 1195 BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64); 1196 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64); 1197 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64); 1198 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64); 1199 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024); 1200 } 1201 1202 MODULE_LICENSE("GPL v2"); 1203 1204 module_init(nvmf_init); 1205 module_exit(nvmf_exit); 1206