xref: /linux/drivers/nvme/host/fabrics.c (revision 001821b0e79716c4e17c71d8e053a23599a7a508)
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 #include <linux/nvme-keyring.h>
16 
17 static LIST_HEAD(nvmf_transports);
18 static DECLARE_RWSEM(nvmf_transports_rwsem);
19 
20 static LIST_HEAD(nvmf_hosts);
21 static DEFINE_MUTEX(nvmf_hosts_mutex);
22 
23 static struct nvmf_host *nvmf_default_host;
24 
25 static struct nvmf_host *nvmf_host_alloc(const char *hostnqn, uuid_t *id)
26 {
27 	struct nvmf_host *host;
28 
29 	host = kmalloc(sizeof(*host), GFP_KERNEL);
30 	if (!host)
31 		return NULL;
32 
33 	kref_init(&host->ref);
34 	uuid_copy(&host->id, id);
35 	strscpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
36 
37 	return host;
38 }
39 
40 static struct nvmf_host *nvmf_host_add(const char *hostnqn, uuid_t *id)
41 {
42 	struct nvmf_host *host;
43 
44 	mutex_lock(&nvmf_hosts_mutex);
45 
46 	/*
47 	 * We have defined a host as how it is perceived by the target.
48 	 * Therefore, we don't allow different Host NQNs with the same Host ID.
49 	 * Similarly, we do not allow the usage of the same Host NQN with
50 	 * different Host IDs. This'll maintain unambiguous host identification.
51 	 */
52 	list_for_each_entry(host, &nvmf_hosts, list) {
53 		bool same_hostnqn = !strcmp(host->nqn, hostnqn);
54 		bool same_hostid = uuid_equal(&host->id, id);
55 
56 		if (same_hostnqn && same_hostid) {
57 			kref_get(&host->ref);
58 			goto out_unlock;
59 		}
60 		if (same_hostnqn) {
61 			pr_err("found same hostnqn %s but different hostid %pUb\n",
62 			       hostnqn, id);
63 			host = ERR_PTR(-EINVAL);
64 			goto out_unlock;
65 		}
66 		if (same_hostid) {
67 			pr_err("found same hostid %pUb but different hostnqn %s\n",
68 			       id, hostnqn);
69 			host = ERR_PTR(-EINVAL);
70 			goto out_unlock;
71 		}
72 	}
73 
74 	host = nvmf_host_alloc(hostnqn, id);
75 	if (!host) {
76 		host = ERR_PTR(-ENOMEM);
77 		goto out_unlock;
78 	}
79 
80 	list_add_tail(&host->list, &nvmf_hosts);
81 out_unlock:
82 	mutex_unlock(&nvmf_hosts_mutex);
83 	return host;
84 }
85 
86 static struct nvmf_host *nvmf_host_default(void)
87 {
88 	struct nvmf_host *host;
89 	char nqn[NVMF_NQN_SIZE];
90 	uuid_t id;
91 
92 	uuid_gen(&id);
93 	snprintf(nqn, NVMF_NQN_SIZE,
94 		"nqn.2014-08.org.nvmexpress:uuid:%pUb", &id);
95 
96 	host = nvmf_host_alloc(nqn, &id);
97 	if (!host)
98 		return NULL;
99 
100 	mutex_lock(&nvmf_hosts_mutex);
101 	list_add_tail(&host->list, &nvmf_hosts);
102 	mutex_unlock(&nvmf_hosts_mutex);
103 
104 	return host;
105 }
106 
107 static void nvmf_host_destroy(struct kref *ref)
108 {
109 	struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
110 
111 	mutex_lock(&nvmf_hosts_mutex);
112 	list_del(&host->list);
113 	mutex_unlock(&nvmf_hosts_mutex);
114 
115 	kfree(host);
116 }
117 
118 static void nvmf_host_put(struct nvmf_host *host)
119 {
120 	if (host)
121 		kref_put(&host->ref, nvmf_host_destroy);
122 }
123 
124 /**
125  * nvmf_get_address() -  Get address/port
126  * @ctrl:	Host NVMe controller instance which we got the address
127  * @buf:	OUTPUT parameter that will contain the address/port
128  * @size:	buffer size
129  */
130 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
131 {
132 	int len = 0;
133 
134 	if (ctrl->opts->mask & NVMF_OPT_TRADDR)
135 		len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
136 	if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
137 		len += scnprintf(buf + len, size - len, "%strsvcid=%s",
138 				(len) ? "," : "", ctrl->opts->trsvcid);
139 	if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
140 		len += scnprintf(buf + len, size - len, "%shost_traddr=%s",
141 				(len) ? "," : "", ctrl->opts->host_traddr);
142 	if (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)
143 		len += scnprintf(buf + len, size - len, "%shost_iface=%s",
144 				(len) ? "," : "", ctrl->opts->host_iface);
145 	len += scnprintf(buf + len, size - len, "\n");
146 
147 	return len;
148 }
149 EXPORT_SYMBOL_GPL(nvmf_get_address);
150 
151 /**
152  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
153  * @ctrl:	Host NVMe controller instance maintaining the admin
154  *		queue used to submit the property read command to
155  *		the allocated NVMe controller resource on the target system.
156  * @off:	Starting offset value of the targeted property
157  *		register (see the fabrics section of the NVMe standard).
158  * @val:	OUTPUT parameter that will contain the value of
159  *		the property after a successful read.
160  *
161  * Used by the host system to retrieve a 32-bit capsule property value
162  * from an NVMe controller on the target system.
163  *
164  * ("Capsule property" is an "PCIe register concept" applied to the
165  * NVMe fabrics space.)
166  *
167  * Return:
168  *	0: successful read
169  *	> 0: NVMe error status code
170  *	< 0: Linux errno error code
171  */
172 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
173 {
174 	struct nvme_command cmd = { };
175 	union nvme_result res;
176 	int ret;
177 
178 	cmd.prop_get.opcode = nvme_fabrics_command;
179 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
180 	cmd.prop_get.offset = cpu_to_le32(off);
181 
182 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0,
183 			NVME_QID_ANY, 0);
184 
185 	if (ret >= 0)
186 		*val = le64_to_cpu(res.u64);
187 	if (unlikely(ret != 0))
188 		dev_err(ctrl->device,
189 			"Property Get error: %d, offset %#x\n",
190 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
191 
192 	return ret;
193 }
194 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
195 
196 /**
197  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
198  * @ctrl:	Host NVMe controller instance maintaining the admin
199  *		queue used to submit the property read command to
200  *		the allocated controller resource on the target system.
201  * @off:	Starting offset value of the targeted property
202  *		register (see the fabrics section of the NVMe standard).
203  * @val:	OUTPUT parameter that will contain the value of
204  *		the property after a successful read.
205  *
206  * Used by the host system to retrieve a 64-bit capsule property value
207  * from an NVMe controller on the target system.
208  *
209  * ("Capsule property" is an "PCIe register concept" applied to the
210  * NVMe fabrics space.)
211  *
212  * Return:
213  *	0: successful read
214  *	> 0: NVMe error status code
215  *	< 0: Linux errno error code
216  */
217 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
218 {
219 	struct nvme_command cmd = { };
220 	union nvme_result res;
221 	int ret;
222 
223 	cmd.prop_get.opcode = nvme_fabrics_command;
224 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
225 	cmd.prop_get.attrib = 1;
226 	cmd.prop_get.offset = cpu_to_le32(off);
227 
228 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0,
229 			NVME_QID_ANY, 0);
230 
231 	if (ret >= 0)
232 		*val = le64_to_cpu(res.u64);
233 	if (unlikely(ret != 0))
234 		dev_err(ctrl->device,
235 			"Property Get error: %d, offset %#x\n",
236 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
237 	return ret;
238 }
239 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
240 
241 /**
242  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
243  * @ctrl:	Host NVMe controller instance maintaining the admin
244  *		queue used to submit the property read command to
245  *		the allocated NVMe controller resource on the target system.
246  * @off:	Starting offset value of the targeted property
247  *		register (see the fabrics section of the NVMe standard).
248  * @val:	Input parameter that contains the value to be
249  *		written to the property.
250  *
251  * Used by the NVMe host system to write a 32-bit capsule property value
252  * to an NVMe controller on the target system.
253  *
254  * ("Capsule property" is an "PCIe register concept" applied to the
255  * NVMe fabrics space.)
256  *
257  * Return:
258  *	0: successful write
259  *	> 0: NVMe error status code
260  *	< 0: Linux errno error code
261  */
262 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
263 {
264 	struct nvme_command cmd = { };
265 	int ret;
266 
267 	cmd.prop_set.opcode = nvme_fabrics_command;
268 	cmd.prop_set.fctype = nvme_fabrics_type_property_set;
269 	cmd.prop_set.attrib = 0;
270 	cmd.prop_set.offset = cpu_to_le32(off);
271 	cmd.prop_set.value = cpu_to_le64(val);
272 
273 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0,
274 			NVME_QID_ANY, 0);
275 	if (unlikely(ret))
276 		dev_err(ctrl->device,
277 			"Property Set error: %d, offset %#x\n",
278 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
279 	return ret;
280 }
281 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
282 
283 /**
284  * nvmf_log_connect_error() - Error-parsing-diagnostic print out function for
285  * 				connect() errors.
286  * @ctrl:	The specific /dev/nvmeX device that had the error.
287  * @errval:	Error code to be decoded in a more human-friendly
288  * 		printout.
289  * @offset:	For use with the NVMe error code
290  * 		NVME_SC_CONNECT_INVALID_PARAM.
291  * @cmd:	This is the SQE portion of a submission capsule.
292  * @data:	This is the "Data" portion of a submission capsule.
293  */
294 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
295 		int errval, int offset, struct nvme_command *cmd,
296 		struct nvmf_connect_data *data)
297 {
298 	int err_sctype = errval & ~NVME_SC_DNR;
299 
300 	if (errval < 0) {
301 		dev_err(ctrl->device,
302 			"Connect command failed, errno: %d\n", errval);
303 		return;
304 	}
305 
306 	switch (err_sctype) {
307 	case NVME_SC_CONNECT_INVALID_PARAM:
308 		if (offset >> 16) {
309 			char *inv_data = "Connect Invalid Data Parameter";
310 
311 			switch (offset & 0xffff) {
312 			case (offsetof(struct nvmf_connect_data, cntlid)):
313 				dev_err(ctrl->device,
314 					"%s, cntlid: %d\n",
315 					inv_data, data->cntlid);
316 				break;
317 			case (offsetof(struct nvmf_connect_data, hostnqn)):
318 				dev_err(ctrl->device,
319 					"%s, hostnqn \"%s\"\n",
320 					inv_data, data->hostnqn);
321 				break;
322 			case (offsetof(struct nvmf_connect_data, subsysnqn)):
323 				dev_err(ctrl->device,
324 					"%s, subsysnqn \"%s\"\n",
325 					inv_data, data->subsysnqn);
326 				break;
327 			default:
328 				dev_err(ctrl->device,
329 					"%s, starting byte offset: %d\n",
330 				       inv_data, offset & 0xffff);
331 				break;
332 			}
333 		} else {
334 			char *inv_sqe = "Connect Invalid SQE Parameter";
335 
336 			switch (offset) {
337 			case (offsetof(struct nvmf_connect_command, qid)):
338 				dev_err(ctrl->device,
339 				       "%s, qid %d\n",
340 					inv_sqe, cmd->connect.qid);
341 				break;
342 			default:
343 				dev_err(ctrl->device,
344 					"%s, starting byte offset: %d\n",
345 					inv_sqe, offset);
346 			}
347 		}
348 		break;
349 	case NVME_SC_CONNECT_INVALID_HOST:
350 		dev_err(ctrl->device,
351 			"Connect for subsystem %s is not allowed, hostnqn: %s\n",
352 			data->subsysnqn, data->hostnqn);
353 		break;
354 	case NVME_SC_CONNECT_CTRL_BUSY:
355 		dev_err(ctrl->device,
356 			"Connect command failed: controller is busy or not available\n");
357 		break;
358 	case NVME_SC_CONNECT_FORMAT:
359 		dev_err(ctrl->device,
360 			"Connect incompatible format: %d",
361 			cmd->connect.recfmt);
362 		break;
363 	case NVME_SC_HOST_PATH_ERROR:
364 		dev_err(ctrl->device,
365 			"Connect command failed: host path error\n");
366 		break;
367 	case NVME_SC_AUTH_REQUIRED:
368 		dev_err(ctrl->device,
369 			"Connect command failed: authentication required\n");
370 		break;
371 	default:
372 		dev_err(ctrl->device,
373 			"Connect command failed, error wo/DNR bit: %d\n",
374 			err_sctype);
375 		break;
376 	}
377 }
378 
379 static struct nvmf_connect_data *nvmf_connect_data_prep(struct nvme_ctrl *ctrl,
380 		u16 cntlid)
381 {
382 	struct nvmf_connect_data *data;
383 
384 	data = kzalloc(sizeof(*data), GFP_KERNEL);
385 	if (!data)
386 		return NULL;
387 
388 	uuid_copy(&data->hostid, &ctrl->opts->host->id);
389 	data->cntlid = cpu_to_le16(cntlid);
390 	strscpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
391 	strscpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
392 
393 	return data;
394 }
395 
396 static void nvmf_connect_cmd_prep(struct nvme_ctrl *ctrl, u16 qid,
397 		struct nvme_command *cmd)
398 {
399 	cmd->connect.opcode = nvme_fabrics_command;
400 	cmd->connect.fctype = nvme_fabrics_type_connect;
401 	cmd->connect.qid = cpu_to_le16(qid);
402 
403 	if (qid) {
404 		cmd->connect.sqsize = cpu_to_le16(ctrl->sqsize);
405 	} else {
406 		cmd->connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
407 
408 		/*
409 		 * set keep-alive timeout in seconds granularity (ms * 1000)
410 		 */
411 		cmd->connect.kato = cpu_to_le32(ctrl->kato * 1000);
412 	}
413 
414 	if (ctrl->opts->disable_sqflow)
415 		cmd->connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
416 }
417 
418 /**
419  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
420  *				API function.
421  * @ctrl:	Host nvme controller instance used to request
422  *              a new NVMe controller allocation on the target
423  *              system and  establish an NVMe Admin connection to
424  *              that controller.
425  *
426  * This function enables an NVMe host device to request a new allocation of
427  * an NVMe controller resource on a target system as well establish a
428  * fabrics-protocol connection of the NVMe Admin queue between the
429  * host system device and the allocated NVMe controller on the
430  * target system via a NVMe Fabrics "Connect" command.
431  */
432 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
433 {
434 	struct nvme_command cmd = { };
435 	union nvme_result res;
436 	struct nvmf_connect_data *data;
437 	int ret;
438 	u32 result;
439 
440 	nvmf_connect_cmd_prep(ctrl, 0, &cmd);
441 
442 	data = nvmf_connect_data_prep(ctrl, 0xffff);
443 	if (!data)
444 		return -ENOMEM;
445 
446 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
447 			data, sizeof(*data), NVME_QID_ANY,
448 			NVME_SUBMIT_AT_HEAD |
449 			NVME_SUBMIT_NOWAIT |
450 			NVME_SUBMIT_RESERVED);
451 	if (ret) {
452 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
453 				       &cmd, data);
454 		goto out_free_data;
455 	}
456 
457 	result = le32_to_cpu(res.u32);
458 	ctrl->cntlid = result & 0xFFFF;
459 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
460 		/* Secure concatenation is not implemented */
461 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
462 			dev_warn(ctrl->device,
463 				 "qid 0: secure concatenation is not supported\n");
464 			ret = -EOPNOTSUPP;
465 			goto out_free_data;
466 		}
467 		/* Authentication required */
468 		ret = nvme_auth_negotiate(ctrl, 0);
469 		if (ret) {
470 			dev_warn(ctrl->device,
471 				 "qid 0: authentication setup failed\n");
472 			goto out_free_data;
473 		}
474 		ret = nvme_auth_wait(ctrl, 0);
475 		if (ret) {
476 			dev_warn(ctrl->device,
477 				 "qid 0: authentication failed, error %d\n",
478 				 ret);
479 		} else
480 			dev_info(ctrl->device,
481 				 "qid 0: authenticated\n");
482 	}
483 out_free_data:
484 	kfree(data);
485 	return ret;
486 }
487 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
488 
489 /**
490  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
491  *			     API function.
492  * @ctrl:	Host nvme controller instance used to establish an
493  *		NVMe I/O queue connection to the already allocated NVMe
494  *		controller on the target system.
495  * @qid:	NVMe I/O queue number for the new I/O connection between
496  *		host and target (note qid == 0 is illegal as this is
497  *		the Admin queue, per NVMe standard).
498  *
499  * This function issues a fabrics-protocol connection
500  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
501  * between the host system device and the allocated NVMe controller
502  * on the target system.
503  *
504  * Return:
505  *	0: success
506  *	> 0: NVMe error status code
507  *	< 0: Linux errno error code
508  */
509 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
510 {
511 	struct nvme_command cmd = { };
512 	struct nvmf_connect_data *data;
513 	union nvme_result res;
514 	int ret;
515 	u32 result;
516 
517 	nvmf_connect_cmd_prep(ctrl, qid, &cmd);
518 
519 	data = nvmf_connect_data_prep(ctrl, ctrl->cntlid);
520 	if (!data)
521 		return -ENOMEM;
522 
523 	ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
524 			data, sizeof(*data), qid,
525 			NVME_SUBMIT_AT_HEAD |
526 			NVME_SUBMIT_RESERVED |
527 			NVME_SUBMIT_NOWAIT);
528 	if (ret) {
529 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
530 				       &cmd, data);
531 		goto out_free_data;
532 	}
533 	result = le32_to_cpu(res.u32);
534 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
535 		/* Secure concatenation is not implemented */
536 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
537 			dev_warn(ctrl->device,
538 				 "qid 0: secure concatenation is not supported\n");
539 			ret = -EOPNOTSUPP;
540 			goto out_free_data;
541 		}
542 		/* Authentication required */
543 		ret = nvme_auth_negotiate(ctrl, qid);
544 		if (ret) {
545 			dev_warn(ctrl->device,
546 				 "qid %d: authentication setup failed\n", qid);
547 			goto out_free_data;
548 		}
549 		ret = nvme_auth_wait(ctrl, qid);
550 		if (ret) {
551 			dev_warn(ctrl->device,
552 				 "qid %u: authentication failed, error %d\n",
553 				 qid, ret);
554 		}
555 	}
556 out_free_data:
557 	kfree(data);
558 	return ret;
559 }
560 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
561 
562 /*
563  * Evaluate the status information returned by the transport in order to decided
564  * if a reconnect attempt should be scheduled.
565  *
566  * Do not retry when:
567  *
568  * - the DNR bit is set and the specification states no further connect
569  *   attempts with the same set of paramenters should be attempted.
570  *
571  * - when the authentication attempt fails, because the key was invalid.
572  *   This error code is set on the host side.
573  */
574 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl, int status)
575 {
576 	if (status > 0 && (status & NVME_SC_DNR))
577 		return false;
578 
579 	if (status == -EKEYREJECTED)
580 		return false;
581 
582 	if (ctrl->opts->max_reconnects == -1 ||
583 	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
584 		return true;
585 
586 	return false;
587 }
588 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
589 
590 /**
591  * nvmf_register_transport() - NVMe Fabrics Library registration function.
592  * @ops:	Transport ops instance to be registered to the
593  *		common fabrics library.
594  *
595  * API function that registers the type of specific transport fabric
596  * being implemented to the common NVMe fabrics library. Part of
597  * the overall init sequence of starting up a fabrics driver.
598  */
599 int nvmf_register_transport(struct nvmf_transport_ops *ops)
600 {
601 	if (!ops->create_ctrl)
602 		return -EINVAL;
603 
604 	down_write(&nvmf_transports_rwsem);
605 	list_add_tail(&ops->entry, &nvmf_transports);
606 	up_write(&nvmf_transports_rwsem);
607 
608 	return 0;
609 }
610 EXPORT_SYMBOL_GPL(nvmf_register_transport);
611 
612 /**
613  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
614  * @ops:	Transport ops instance to be unregistered from the
615  *		common fabrics library.
616  *
617  * Fabrics API function that unregisters the type of specific transport
618  * fabric being implemented from the common NVMe fabrics library.
619  * Part of the overall exit sequence of unloading the implemented driver.
620  */
621 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
622 {
623 	down_write(&nvmf_transports_rwsem);
624 	list_del(&ops->entry);
625 	up_write(&nvmf_transports_rwsem);
626 }
627 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
628 
629 static struct nvmf_transport_ops *nvmf_lookup_transport(
630 		struct nvmf_ctrl_options *opts)
631 {
632 	struct nvmf_transport_ops *ops;
633 
634 	lockdep_assert_held(&nvmf_transports_rwsem);
635 
636 	list_for_each_entry(ops, &nvmf_transports, entry) {
637 		if (strcmp(ops->name, opts->transport) == 0)
638 			return ops;
639 	}
640 
641 	return NULL;
642 }
643 
644 static struct key *nvmf_parse_key(int key_id)
645 {
646 	struct key *key;
647 
648 	if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
649 		pr_err("TLS is not supported\n");
650 		return ERR_PTR(-EINVAL);
651 	}
652 
653 	key = key_lookup(key_id);
654 	if (IS_ERR(key))
655 		pr_err("key id %08x not found\n", key_id);
656 	else
657 		pr_debug("Using key id %08x\n", key_id);
658 	return key;
659 }
660 
661 static const match_table_t opt_tokens = {
662 	{ NVMF_OPT_TRANSPORT,		"transport=%s"		},
663 	{ NVMF_OPT_TRADDR,		"traddr=%s"		},
664 	{ NVMF_OPT_TRSVCID,		"trsvcid=%s"		},
665 	{ NVMF_OPT_NQN,			"nqn=%s"		},
666 	{ NVMF_OPT_QUEUE_SIZE,		"queue_size=%d"		},
667 	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
668 	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
669 	{ NVMF_OPT_CTRL_LOSS_TMO,	"ctrl_loss_tmo=%d"	},
670 	{ NVMF_OPT_KATO,		"keep_alive_tmo=%d"	},
671 	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
672 	{ NVMF_OPT_HOST_TRADDR,		"host_traddr=%s"	},
673 	{ NVMF_OPT_HOST_IFACE,		"host_iface=%s"		},
674 	{ NVMF_OPT_HOST_ID,		"hostid=%s"		},
675 	{ NVMF_OPT_DUP_CONNECT,		"duplicate_connect"	},
676 	{ NVMF_OPT_DISABLE_SQFLOW,	"disable_sqflow"	},
677 	{ NVMF_OPT_HDR_DIGEST,		"hdr_digest"		},
678 	{ NVMF_OPT_DATA_DIGEST,		"data_digest"		},
679 	{ NVMF_OPT_NR_WRITE_QUEUES,	"nr_write_queues=%d"	},
680 	{ NVMF_OPT_NR_POLL_QUEUES,	"nr_poll_queues=%d"	},
681 	{ NVMF_OPT_TOS,			"tos=%d"		},
682 #ifdef CONFIG_NVME_TCP_TLS
683 	{ NVMF_OPT_KEYRING,		"keyring=%d"		},
684 	{ NVMF_OPT_TLS_KEY,		"tls_key=%d"		},
685 #endif
686 	{ NVMF_OPT_FAIL_FAST_TMO,	"fast_io_fail_tmo=%d"	},
687 	{ NVMF_OPT_DISCOVERY,		"discovery"		},
688 #ifdef CONFIG_NVME_HOST_AUTH
689 	{ NVMF_OPT_DHCHAP_SECRET,	"dhchap_secret=%s"	},
690 	{ NVMF_OPT_DHCHAP_CTRL_SECRET,	"dhchap_ctrl_secret=%s"	},
691 #endif
692 #ifdef CONFIG_NVME_TCP_TLS
693 	{ NVMF_OPT_TLS,			"tls"			},
694 #endif
695 	{ NVMF_OPT_ERR,			NULL			}
696 };
697 
698 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
699 		const char *buf)
700 {
701 	substring_t args[MAX_OPT_ARGS];
702 	char *options, *o, *p;
703 	int token, ret = 0;
704 	size_t nqnlen  = 0;
705 	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO, key_id;
706 	uuid_t hostid;
707 	char hostnqn[NVMF_NQN_SIZE];
708 	struct key *key;
709 
710 	/* Set defaults */
711 	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
712 	opts->nr_io_queues = num_online_cpus();
713 	opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
714 	opts->kato = 0;
715 	opts->duplicate_connect = false;
716 	opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
717 	opts->hdr_digest = false;
718 	opts->data_digest = false;
719 	opts->tos = -1; /* < 0 == use transport default */
720 	opts->tls = false;
721 	opts->tls_key = NULL;
722 	opts->keyring = NULL;
723 
724 	options = o = kstrdup(buf, GFP_KERNEL);
725 	if (!options)
726 		return -ENOMEM;
727 
728 	/* use default host if not given by user space */
729 	uuid_copy(&hostid, &nvmf_default_host->id);
730 	strscpy(hostnqn, nvmf_default_host->nqn, NVMF_NQN_SIZE);
731 
732 	while ((p = strsep(&o, ",\n")) != NULL) {
733 		if (!*p)
734 			continue;
735 
736 		token = match_token(p, opt_tokens, args);
737 		opts->mask |= token;
738 		switch (token) {
739 		case NVMF_OPT_TRANSPORT:
740 			p = match_strdup(args);
741 			if (!p) {
742 				ret = -ENOMEM;
743 				goto out;
744 			}
745 			kfree(opts->transport);
746 			opts->transport = p;
747 			break;
748 		case NVMF_OPT_NQN:
749 			p = match_strdup(args);
750 			if (!p) {
751 				ret = -ENOMEM;
752 				goto out;
753 			}
754 			kfree(opts->subsysnqn);
755 			opts->subsysnqn = p;
756 			nqnlen = strlen(opts->subsysnqn);
757 			if (nqnlen >= NVMF_NQN_SIZE) {
758 				pr_err("%s needs to be < %d bytes\n",
759 					opts->subsysnqn, NVMF_NQN_SIZE);
760 				ret = -EINVAL;
761 				goto out;
762 			}
763 			opts->discovery_nqn =
764 				!(strcmp(opts->subsysnqn,
765 					 NVME_DISC_SUBSYS_NAME));
766 			break;
767 		case NVMF_OPT_TRADDR:
768 			p = match_strdup(args);
769 			if (!p) {
770 				ret = -ENOMEM;
771 				goto out;
772 			}
773 			kfree(opts->traddr);
774 			opts->traddr = p;
775 			break;
776 		case NVMF_OPT_TRSVCID:
777 			p = match_strdup(args);
778 			if (!p) {
779 				ret = -ENOMEM;
780 				goto out;
781 			}
782 			kfree(opts->trsvcid);
783 			opts->trsvcid = p;
784 			break;
785 		case NVMF_OPT_QUEUE_SIZE:
786 			if (match_int(args, &token)) {
787 				ret = -EINVAL;
788 				goto out;
789 			}
790 			if (token < NVMF_MIN_QUEUE_SIZE ||
791 			    token > NVMF_MAX_QUEUE_SIZE) {
792 				pr_err("Invalid queue_size %d\n", token);
793 				ret = -EINVAL;
794 				goto out;
795 			}
796 			opts->queue_size = token;
797 			break;
798 		case NVMF_OPT_NR_IO_QUEUES:
799 			if (match_int(args, &token)) {
800 				ret = -EINVAL;
801 				goto out;
802 			}
803 			if (token <= 0) {
804 				pr_err("Invalid number of IOQs %d\n", token);
805 				ret = -EINVAL;
806 				goto out;
807 			}
808 			if (opts->discovery_nqn) {
809 				pr_debug("Ignoring nr_io_queues value for discovery controller\n");
810 				break;
811 			}
812 
813 			opts->nr_io_queues = min_t(unsigned int,
814 					num_online_cpus(), token);
815 			break;
816 		case NVMF_OPT_KATO:
817 			if (match_int(args, &token)) {
818 				ret = -EINVAL;
819 				goto out;
820 			}
821 
822 			if (token < 0) {
823 				pr_err("Invalid keep_alive_tmo %d\n", token);
824 				ret = -EINVAL;
825 				goto out;
826 			} else if (token == 0 && !opts->discovery_nqn) {
827 				/* Allowed for debug */
828 				pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
829 			}
830 			opts->kato = token;
831 			break;
832 		case NVMF_OPT_CTRL_LOSS_TMO:
833 			if (match_int(args, &token)) {
834 				ret = -EINVAL;
835 				goto out;
836 			}
837 
838 			if (token < 0)
839 				pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
840 			ctrl_loss_tmo = token;
841 			break;
842 		case NVMF_OPT_FAIL_FAST_TMO:
843 			if (match_int(args, &token)) {
844 				ret = -EINVAL;
845 				goto out;
846 			}
847 
848 			if (token >= 0)
849 				pr_warn("I/O fail on reconnect controller after %d sec\n",
850 					token);
851 			else
852 				token = -1;
853 
854 			opts->fast_io_fail_tmo = token;
855 			break;
856 		case NVMF_OPT_HOSTNQN:
857 			if (opts->host) {
858 				pr_err("hostnqn already user-assigned: %s\n",
859 				       opts->host->nqn);
860 				ret = -EADDRINUSE;
861 				goto out;
862 			}
863 			p = match_strdup(args);
864 			if (!p) {
865 				ret = -ENOMEM;
866 				goto out;
867 			}
868 			nqnlen = strlen(p);
869 			if (nqnlen >= NVMF_NQN_SIZE) {
870 				pr_err("%s needs to be < %d bytes\n",
871 					p, NVMF_NQN_SIZE);
872 				kfree(p);
873 				ret = -EINVAL;
874 				goto out;
875 			}
876 			strscpy(hostnqn, p, NVMF_NQN_SIZE);
877 			kfree(p);
878 			break;
879 		case NVMF_OPT_RECONNECT_DELAY:
880 			if (match_int(args, &token)) {
881 				ret = -EINVAL;
882 				goto out;
883 			}
884 			if (token <= 0) {
885 				pr_err("Invalid reconnect_delay %d\n", token);
886 				ret = -EINVAL;
887 				goto out;
888 			}
889 			opts->reconnect_delay = token;
890 			break;
891 		case NVMF_OPT_HOST_TRADDR:
892 			p = match_strdup(args);
893 			if (!p) {
894 				ret = -ENOMEM;
895 				goto out;
896 			}
897 			kfree(opts->host_traddr);
898 			opts->host_traddr = p;
899 			break;
900 		case NVMF_OPT_HOST_IFACE:
901 			p = match_strdup(args);
902 			if (!p) {
903 				ret = -ENOMEM;
904 				goto out;
905 			}
906 			kfree(opts->host_iface);
907 			opts->host_iface = p;
908 			break;
909 		case NVMF_OPT_HOST_ID:
910 			p = match_strdup(args);
911 			if (!p) {
912 				ret = -ENOMEM;
913 				goto out;
914 			}
915 			ret = uuid_parse(p, &hostid);
916 			if (ret) {
917 				pr_err("Invalid hostid %s\n", p);
918 				ret = -EINVAL;
919 				kfree(p);
920 				goto out;
921 			}
922 			kfree(p);
923 			break;
924 		case NVMF_OPT_DUP_CONNECT:
925 			opts->duplicate_connect = true;
926 			break;
927 		case NVMF_OPT_DISABLE_SQFLOW:
928 			opts->disable_sqflow = true;
929 			break;
930 		case NVMF_OPT_HDR_DIGEST:
931 			opts->hdr_digest = true;
932 			break;
933 		case NVMF_OPT_DATA_DIGEST:
934 			opts->data_digest = true;
935 			break;
936 		case NVMF_OPT_NR_WRITE_QUEUES:
937 			if (match_int(args, &token)) {
938 				ret = -EINVAL;
939 				goto out;
940 			}
941 			if (token <= 0) {
942 				pr_err("Invalid nr_write_queues %d\n", token);
943 				ret = -EINVAL;
944 				goto out;
945 			}
946 			opts->nr_write_queues = token;
947 			break;
948 		case NVMF_OPT_NR_POLL_QUEUES:
949 			if (match_int(args, &token)) {
950 				ret = -EINVAL;
951 				goto out;
952 			}
953 			if (token <= 0) {
954 				pr_err("Invalid nr_poll_queues %d\n", token);
955 				ret = -EINVAL;
956 				goto out;
957 			}
958 			opts->nr_poll_queues = token;
959 			break;
960 		case NVMF_OPT_TOS:
961 			if (match_int(args, &token)) {
962 				ret = -EINVAL;
963 				goto out;
964 			}
965 			if (token < 0) {
966 				pr_err("Invalid type of service %d\n", token);
967 				ret = -EINVAL;
968 				goto out;
969 			}
970 			if (token > 255) {
971 				pr_warn("Clamping type of service to 255\n");
972 				token = 255;
973 			}
974 			opts->tos = token;
975 			break;
976 		case NVMF_OPT_KEYRING:
977 			if (match_int(args, &key_id) || key_id <= 0) {
978 				ret = -EINVAL;
979 				goto out;
980 			}
981 			key = nvmf_parse_key(key_id);
982 			if (IS_ERR(key)) {
983 				ret = PTR_ERR(key);
984 				goto out;
985 			}
986 			key_put(opts->keyring);
987 			opts->keyring = key;
988 			break;
989 		case NVMF_OPT_TLS_KEY:
990 			if (match_int(args, &key_id) || key_id <= 0) {
991 				ret = -EINVAL;
992 				goto out;
993 			}
994 			key = nvmf_parse_key(key_id);
995 			if (IS_ERR(key)) {
996 				ret = PTR_ERR(key);
997 				goto out;
998 			}
999 			key_put(opts->tls_key);
1000 			opts->tls_key = key;
1001 			break;
1002 		case NVMF_OPT_DISCOVERY:
1003 			opts->discovery_nqn = true;
1004 			break;
1005 		case NVMF_OPT_DHCHAP_SECRET:
1006 			p = match_strdup(args);
1007 			if (!p) {
1008 				ret = -ENOMEM;
1009 				goto out;
1010 			}
1011 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
1012 				pr_err("Invalid DH-CHAP secret %s\n", p);
1013 				ret = -EINVAL;
1014 				goto out;
1015 			}
1016 			kfree(opts->dhchap_secret);
1017 			opts->dhchap_secret = p;
1018 			break;
1019 		case NVMF_OPT_DHCHAP_CTRL_SECRET:
1020 			p = match_strdup(args);
1021 			if (!p) {
1022 				ret = -ENOMEM;
1023 				goto out;
1024 			}
1025 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
1026 				pr_err("Invalid DH-CHAP secret %s\n", p);
1027 				ret = -EINVAL;
1028 				goto out;
1029 			}
1030 			kfree(opts->dhchap_ctrl_secret);
1031 			opts->dhchap_ctrl_secret = p;
1032 			break;
1033 		case NVMF_OPT_TLS:
1034 			if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
1035 				pr_err("TLS is not supported\n");
1036 				ret = -EINVAL;
1037 				goto out;
1038 			}
1039 			opts->tls = true;
1040 			break;
1041 		default:
1042 			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
1043 				p);
1044 			ret = -EINVAL;
1045 			goto out;
1046 		}
1047 	}
1048 
1049 	if (opts->discovery_nqn) {
1050 		opts->nr_io_queues = 0;
1051 		opts->nr_write_queues = 0;
1052 		opts->nr_poll_queues = 0;
1053 		opts->duplicate_connect = true;
1054 	} else {
1055 		if (!opts->kato)
1056 			opts->kato = NVME_DEFAULT_KATO;
1057 	}
1058 	if (ctrl_loss_tmo < 0) {
1059 		opts->max_reconnects = -1;
1060 	} else {
1061 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
1062 						opts->reconnect_delay);
1063 		if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
1064 			pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
1065 				opts->fast_io_fail_tmo, ctrl_loss_tmo);
1066 	}
1067 
1068 	opts->host = nvmf_host_add(hostnqn, &hostid);
1069 	if (IS_ERR(opts->host)) {
1070 		ret = PTR_ERR(opts->host);
1071 		opts->host = NULL;
1072 		goto out;
1073 	}
1074 
1075 out:
1076 	kfree(options);
1077 	return ret;
1078 }
1079 
1080 void nvmf_set_io_queues(struct nvmf_ctrl_options *opts, u32 nr_io_queues,
1081 			u32 io_queues[HCTX_MAX_TYPES])
1082 {
1083 	if (opts->nr_write_queues && opts->nr_io_queues < nr_io_queues) {
1084 		/*
1085 		 * separate read/write queues
1086 		 * hand out dedicated default queues only after we have
1087 		 * sufficient read queues.
1088 		 */
1089 		io_queues[HCTX_TYPE_READ] = opts->nr_io_queues;
1090 		nr_io_queues -= io_queues[HCTX_TYPE_READ];
1091 		io_queues[HCTX_TYPE_DEFAULT] =
1092 			min(opts->nr_write_queues, nr_io_queues);
1093 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1094 	} else {
1095 		/*
1096 		 * shared read/write queues
1097 		 * either no write queues were requested, or we don't have
1098 		 * sufficient queue count to have dedicated default queues.
1099 		 */
1100 		io_queues[HCTX_TYPE_DEFAULT] =
1101 			min(opts->nr_io_queues, nr_io_queues);
1102 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1103 	}
1104 
1105 	if (opts->nr_poll_queues && nr_io_queues) {
1106 		/* map dedicated poll queues only if we have queues left */
1107 		io_queues[HCTX_TYPE_POLL] =
1108 			min(opts->nr_poll_queues, nr_io_queues);
1109 	}
1110 }
1111 EXPORT_SYMBOL_GPL(nvmf_set_io_queues);
1112 
1113 void nvmf_map_queues(struct blk_mq_tag_set *set, struct nvme_ctrl *ctrl,
1114 		     u32 io_queues[HCTX_MAX_TYPES])
1115 {
1116 	struct nvmf_ctrl_options *opts = ctrl->opts;
1117 
1118 	if (opts->nr_write_queues && io_queues[HCTX_TYPE_READ]) {
1119 		/* separate read/write queues */
1120 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1121 			io_queues[HCTX_TYPE_DEFAULT];
1122 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1123 		set->map[HCTX_TYPE_READ].nr_queues =
1124 			io_queues[HCTX_TYPE_READ];
1125 		set->map[HCTX_TYPE_READ].queue_offset =
1126 			io_queues[HCTX_TYPE_DEFAULT];
1127 	} else {
1128 		/* shared read/write queues */
1129 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1130 			io_queues[HCTX_TYPE_DEFAULT];
1131 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1132 		set->map[HCTX_TYPE_READ].nr_queues =
1133 			io_queues[HCTX_TYPE_DEFAULT];
1134 		set->map[HCTX_TYPE_READ].queue_offset = 0;
1135 	}
1136 
1137 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1138 	blk_mq_map_queues(&set->map[HCTX_TYPE_READ]);
1139 	if (opts->nr_poll_queues && io_queues[HCTX_TYPE_POLL]) {
1140 		/* map dedicated poll queues only if we have queues left */
1141 		set->map[HCTX_TYPE_POLL].nr_queues = io_queues[HCTX_TYPE_POLL];
1142 		set->map[HCTX_TYPE_POLL].queue_offset =
1143 			io_queues[HCTX_TYPE_DEFAULT] +
1144 			io_queues[HCTX_TYPE_READ];
1145 		blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
1146 	}
1147 
1148 	dev_info(ctrl->device,
1149 		"mapped %d/%d/%d default/read/poll queues.\n",
1150 		io_queues[HCTX_TYPE_DEFAULT],
1151 		io_queues[HCTX_TYPE_READ],
1152 		io_queues[HCTX_TYPE_POLL]);
1153 }
1154 EXPORT_SYMBOL_GPL(nvmf_map_queues);
1155 
1156 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
1157 		unsigned int required_opts)
1158 {
1159 	if ((opts->mask & required_opts) != required_opts) {
1160 		unsigned int i;
1161 
1162 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1163 			if ((opt_tokens[i].token & required_opts) &&
1164 			    !(opt_tokens[i].token & opts->mask)) {
1165 				pr_warn("missing parameter '%s'\n",
1166 					opt_tokens[i].pattern);
1167 			}
1168 		}
1169 
1170 		return -EINVAL;
1171 	}
1172 
1173 	return 0;
1174 }
1175 
1176 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
1177 		struct nvmf_ctrl_options *opts)
1178 {
1179 	if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
1180 	    strcmp(opts->traddr, ctrl->opts->traddr) ||
1181 	    strcmp(opts->trsvcid, ctrl->opts->trsvcid))
1182 		return false;
1183 
1184 	/*
1185 	 * Checking the local address or host interfaces is rough.
1186 	 *
1187 	 * In most cases, none is specified and the host port or
1188 	 * host interface is selected by the stack.
1189 	 *
1190 	 * Assume no match if:
1191 	 * -  local address or host interface is specified and address
1192 	 *    or host interface is not the same
1193 	 * -  local address or host interface is not specified but
1194 	 *    remote is, or vice versa (admin using specific
1195 	 *    host_traddr/host_iface when it matters).
1196 	 */
1197 	if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
1198 	    (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1199 		if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
1200 			return false;
1201 	} else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
1202 		   (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1203 		return false;
1204 	}
1205 
1206 	if ((opts->mask & NVMF_OPT_HOST_IFACE) &&
1207 	    (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1208 		if (strcmp(opts->host_iface, ctrl->opts->host_iface))
1209 			return false;
1210 	} else if ((opts->mask & NVMF_OPT_HOST_IFACE) ||
1211 		   (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1212 		return false;
1213 	}
1214 
1215 	return true;
1216 }
1217 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
1218 
1219 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
1220 		unsigned int allowed_opts)
1221 {
1222 	if (opts->mask & ~allowed_opts) {
1223 		unsigned int i;
1224 
1225 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1226 			if ((opt_tokens[i].token & opts->mask) &&
1227 			    (opt_tokens[i].token & ~allowed_opts)) {
1228 				pr_warn("invalid parameter '%s'\n",
1229 					opt_tokens[i].pattern);
1230 			}
1231 		}
1232 
1233 		return -EINVAL;
1234 	}
1235 
1236 	return 0;
1237 }
1238 
1239 void nvmf_free_options(struct nvmf_ctrl_options *opts)
1240 {
1241 	nvmf_host_put(opts->host);
1242 	key_put(opts->keyring);
1243 	key_put(opts->tls_key);
1244 	kfree(opts->transport);
1245 	kfree(opts->traddr);
1246 	kfree(opts->trsvcid);
1247 	kfree(opts->subsysnqn);
1248 	kfree(opts->host_traddr);
1249 	kfree(opts->host_iface);
1250 	kfree(opts->dhchap_secret);
1251 	kfree(opts->dhchap_ctrl_secret);
1252 	kfree(opts);
1253 }
1254 EXPORT_SYMBOL_GPL(nvmf_free_options);
1255 
1256 #define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
1257 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
1258 				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
1259 				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
1260 				 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\
1261 				 NVMF_OPT_FAIL_FAST_TMO | NVMF_OPT_DHCHAP_SECRET |\
1262 				 NVMF_OPT_DHCHAP_CTRL_SECRET)
1263 
1264 static struct nvme_ctrl *
1265 nvmf_create_ctrl(struct device *dev, const char *buf)
1266 {
1267 	struct nvmf_ctrl_options *opts;
1268 	struct nvmf_transport_ops *ops;
1269 	struct nvme_ctrl *ctrl;
1270 	int ret;
1271 
1272 	opts = kzalloc(sizeof(*opts), GFP_KERNEL);
1273 	if (!opts)
1274 		return ERR_PTR(-ENOMEM);
1275 
1276 	ret = nvmf_parse_options(opts, buf);
1277 	if (ret)
1278 		goto out_free_opts;
1279 
1280 
1281 	request_module("nvme-%s", opts->transport);
1282 
1283 	/*
1284 	 * Check the generic options first as we need a valid transport for
1285 	 * the lookup below.  Then clear the generic flags so that transport
1286 	 * drivers don't have to care about them.
1287 	 */
1288 	ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1289 	if (ret)
1290 		goto out_free_opts;
1291 	opts->mask &= ~NVMF_REQUIRED_OPTS;
1292 
1293 	down_read(&nvmf_transports_rwsem);
1294 	ops = nvmf_lookup_transport(opts);
1295 	if (!ops) {
1296 		pr_info("no handler found for transport %s.\n",
1297 			opts->transport);
1298 		ret = -EINVAL;
1299 		goto out_unlock;
1300 	}
1301 
1302 	if (!try_module_get(ops->module)) {
1303 		ret = -EBUSY;
1304 		goto out_unlock;
1305 	}
1306 	up_read(&nvmf_transports_rwsem);
1307 
1308 	ret = nvmf_check_required_opts(opts, ops->required_opts);
1309 	if (ret)
1310 		goto out_module_put;
1311 	ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1312 				ops->allowed_opts | ops->required_opts);
1313 	if (ret)
1314 		goto out_module_put;
1315 
1316 	ctrl = ops->create_ctrl(dev, opts);
1317 	if (IS_ERR(ctrl)) {
1318 		ret = PTR_ERR(ctrl);
1319 		goto out_module_put;
1320 	}
1321 
1322 	module_put(ops->module);
1323 	return ctrl;
1324 
1325 out_module_put:
1326 	module_put(ops->module);
1327 	goto out_free_opts;
1328 out_unlock:
1329 	up_read(&nvmf_transports_rwsem);
1330 out_free_opts:
1331 	nvmf_free_options(opts);
1332 	return ERR_PTR(ret);
1333 }
1334 
1335 static const struct class nvmf_class = {
1336 	.name = "nvme-fabrics",
1337 };
1338 
1339 static struct device *nvmf_device;
1340 static DEFINE_MUTEX(nvmf_dev_mutex);
1341 
1342 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1343 		size_t count, loff_t *pos)
1344 {
1345 	struct seq_file *seq_file = file->private_data;
1346 	struct nvme_ctrl *ctrl;
1347 	const char *buf;
1348 	int ret = 0;
1349 
1350 	if (count > PAGE_SIZE)
1351 		return -ENOMEM;
1352 
1353 	buf = memdup_user_nul(ubuf, count);
1354 	if (IS_ERR(buf))
1355 		return PTR_ERR(buf);
1356 
1357 	mutex_lock(&nvmf_dev_mutex);
1358 	if (seq_file->private) {
1359 		ret = -EINVAL;
1360 		goto out_unlock;
1361 	}
1362 
1363 	ctrl = nvmf_create_ctrl(nvmf_device, buf);
1364 	if (IS_ERR(ctrl)) {
1365 		ret = PTR_ERR(ctrl);
1366 		goto out_unlock;
1367 	}
1368 
1369 	seq_file->private = ctrl;
1370 
1371 out_unlock:
1372 	mutex_unlock(&nvmf_dev_mutex);
1373 	kfree(buf);
1374 	return ret ? ret : count;
1375 }
1376 
1377 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file)
1378 {
1379 	const struct match_token *tok;
1380 	int idx;
1381 
1382 	/*
1383 	 * Add dummy entries for instance and cntlid to
1384 	 * signal an invalid/non-existing controller
1385 	 */
1386 	seq_puts(seq_file, "instance=-1,cntlid=-1");
1387 	for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) {
1388 		tok = &opt_tokens[idx];
1389 		if (tok->token == NVMF_OPT_ERR)
1390 			continue;
1391 		seq_puts(seq_file, ",");
1392 		seq_puts(seq_file, tok->pattern);
1393 	}
1394 	seq_puts(seq_file, "\n");
1395 }
1396 
1397 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1398 {
1399 	struct nvme_ctrl *ctrl;
1400 
1401 	mutex_lock(&nvmf_dev_mutex);
1402 	ctrl = seq_file->private;
1403 	if (!ctrl) {
1404 		__nvmf_concat_opt_tokens(seq_file);
1405 		goto out_unlock;
1406 	}
1407 
1408 	seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1409 			ctrl->instance, ctrl->cntlid);
1410 
1411 out_unlock:
1412 	mutex_unlock(&nvmf_dev_mutex);
1413 	return 0;
1414 }
1415 
1416 static int nvmf_dev_open(struct inode *inode, struct file *file)
1417 {
1418 	/*
1419 	 * The miscdevice code initializes file->private_data, but doesn't
1420 	 * make use of it later.
1421 	 */
1422 	file->private_data = NULL;
1423 	return single_open(file, nvmf_dev_show, NULL);
1424 }
1425 
1426 static int nvmf_dev_release(struct inode *inode, struct file *file)
1427 {
1428 	struct seq_file *seq_file = file->private_data;
1429 	struct nvme_ctrl *ctrl = seq_file->private;
1430 
1431 	if (ctrl)
1432 		nvme_put_ctrl(ctrl);
1433 	return single_release(inode, file);
1434 }
1435 
1436 static const struct file_operations nvmf_dev_fops = {
1437 	.owner		= THIS_MODULE,
1438 	.write		= nvmf_dev_write,
1439 	.read		= seq_read,
1440 	.open		= nvmf_dev_open,
1441 	.release	= nvmf_dev_release,
1442 };
1443 
1444 static struct miscdevice nvmf_misc = {
1445 	.minor		= MISC_DYNAMIC_MINOR,
1446 	.name           = "nvme-fabrics",
1447 	.fops		= &nvmf_dev_fops,
1448 };
1449 
1450 static int __init nvmf_init(void)
1451 {
1452 	int ret;
1453 
1454 	nvmf_default_host = nvmf_host_default();
1455 	if (!nvmf_default_host)
1456 		return -ENOMEM;
1457 
1458 	ret = class_register(&nvmf_class);
1459 	if (ret) {
1460 		pr_err("couldn't register class nvme-fabrics\n");
1461 		goto out_free_host;
1462 	}
1463 
1464 	nvmf_device =
1465 		device_create(&nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1466 	if (IS_ERR(nvmf_device)) {
1467 		pr_err("couldn't create nvme-fabrics device!\n");
1468 		ret = PTR_ERR(nvmf_device);
1469 		goto out_destroy_class;
1470 	}
1471 
1472 	ret = misc_register(&nvmf_misc);
1473 	if (ret) {
1474 		pr_err("couldn't register misc device: %d\n", ret);
1475 		goto out_destroy_device;
1476 	}
1477 
1478 	return 0;
1479 
1480 out_destroy_device:
1481 	device_destroy(&nvmf_class, MKDEV(0, 0));
1482 out_destroy_class:
1483 	class_unregister(&nvmf_class);
1484 out_free_host:
1485 	nvmf_host_put(nvmf_default_host);
1486 	return ret;
1487 }
1488 
1489 static void __exit nvmf_exit(void)
1490 {
1491 	misc_deregister(&nvmf_misc);
1492 	device_destroy(&nvmf_class, MKDEV(0, 0));
1493 	class_unregister(&nvmf_class);
1494 	nvmf_host_put(nvmf_default_host);
1495 
1496 	BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1497 	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1498 	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1499 	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1500 	BUILD_BUG_ON(sizeof(struct nvmf_auth_send_command) != 64);
1501 	BUILD_BUG_ON(sizeof(struct nvmf_auth_receive_command) != 64);
1502 	BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1503 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_negotiate_data) != 8);
1504 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_challenge_data) != 16);
1505 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_reply_data) != 16);
1506 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success1_data) != 16);
1507 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success2_data) != 16);
1508 }
1509 
1510 MODULE_LICENSE("GPL v2");
1511 MODULE_DESCRIPTION("NVMe host fabrics library");
1512 
1513 module_init(nvmf_init);
1514 module_exit(nvmf_exit);
1515