xref: /linux/drivers/nvme/host/fabrics.c (revision f82811e22b480a203a438d8e1f29af9c93ccbb0c)
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  * Return:
433  *	0: success
434  *	> 0: NVMe error status code
435  *	< 0: Linux errno error code
436  *
437  */
438 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
439 {
440 	struct nvme_command cmd = { };
441 	union nvme_result res;
442 	struct nvmf_connect_data *data;
443 	int ret;
444 	u32 result;
445 
446 	nvmf_connect_cmd_prep(ctrl, 0, &cmd);
447 
448 	data = nvmf_connect_data_prep(ctrl, 0xffff);
449 	if (!data)
450 		return -ENOMEM;
451 
452 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
453 			data, sizeof(*data), NVME_QID_ANY,
454 			NVME_SUBMIT_AT_HEAD |
455 			NVME_SUBMIT_NOWAIT |
456 			NVME_SUBMIT_RESERVED);
457 	if (ret) {
458 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
459 				       &cmd, data);
460 		goto out_free_data;
461 	}
462 
463 	result = le32_to_cpu(res.u32);
464 	ctrl->cntlid = result & 0xFFFF;
465 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
466 		/* Secure concatenation is not implemented */
467 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
468 			dev_warn(ctrl->device,
469 				 "qid 0: secure concatenation is not supported\n");
470 			ret = NVME_SC_AUTH_REQUIRED;
471 			goto out_free_data;
472 		}
473 		/* Authentication required */
474 		ret = nvme_auth_negotiate(ctrl, 0);
475 		if (ret) {
476 			dev_warn(ctrl->device,
477 				 "qid 0: authentication setup failed\n");
478 			ret = NVME_SC_AUTH_REQUIRED;
479 			goto out_free_data;
480 		}
481 		ret = nvme_auth_wait(ctrl, 0);
482 		if (ret)
483 			dev_warn(ctrl->device,
484 				 "qid 0: authentication failed\n");
485 		else
486 			dev_info(ctrl->device,
487 				 "qid 0: authenticated\n");
488 	}
489 out_free_data:
490 	kfree(data);
491 	return ret;
492 }
493 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
494 
495 /**
496  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
497  *			     API function.
498  * @ctrl:	Host nvme controller instance used to establish an
499  *		NVMe I/O queue connection to the already allocated NVMe
500  *		controller on the target system.
501  * @qid:	NVMe I/O queue number for the new I/O connection between
502  *		host and target (note qid == 0 is illegal as this is
503  *		the Admin queue, per NVMe standard).
504  *
505  * This function issues a fabrics-protocol connection
506  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
507  * between the host system device and the allocated NVMe controller
508  * on the target system.
509  *
510  * Return:
511  *	0: success
512  *	> 0: NVMe error status code
513  *	< 0: Linux errno error code
514  */
515 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
516 {
517 	struct nvme_command cmd = { };
518 	struct nvmf_connect_data *data;
519 	union nvme_result res;
520 	int ret;
521 	u32 result;
522 
523 	nvmf_connect_cmd_prep(ctrl, qid, &cmd);
524 
525 	data = nvmf_connect_data_prep(ctrl, ctrl->cntlid);
526 	if (!data)
527 		return -ENOMEM;
528 
529 	ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
530 			data, sizeof(*data), qid,
531 			NVME_SUBMIT_AT_HEAD |
532 			NVME_SUBMIT_RESERVED |
533 			NVME_SUBMIT_NOWAIT);
534 	if (ret) {
535 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
536 				       &cmd, data);
537 	}
538 	result = le32_to_cpu(res.u32);
539 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
540 		/* Secure concatenation is not implemented */
541 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
542 			dev_warn(ctrl->device,
543 				 "qid 0: secure concatenation is not supported\n");
544 			ret = NVME_SC_AUTH_REQUIRED;
545 			goto out_free_data;
546 		}
547 		/* Authentication required */
548 		ret = nvme_auth_negotiate(ctrl, qid);
549 		if (ret) {
550 			dev_warn(ctrl->device,
551 				 "qid %d: authentication setup failed\n", qid);
552 			ret = NVME_SC_AUTH_REQUIRED;
553 		} else {
554 			ret = nvme_auth_wait(ctrl, qid);
555 			if (ret)
556 				dev_warn(ctrl->device,
557 					 "qid %u: authentication failed\n", qid);
558 		}
559 	}
560 out_free_data:
561 	kfree(data);
562 	return ret;
563 }
564 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
565 
566 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
567 {
568 	if (ctrl->opts->max_reconnects == -1 ||
569 	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
570 		return true;
571 
572 	return false;
573 }
574 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
575 
576 /**
577  * nvmf_register_transport() - NVMe Fabrics Library registration function.
578  * @ops:	Transport ops instance to be registered to the
579  *		common fabrics library.
580  *
581  * API function that registers the type of specific transport fabric
582  * being implemented to the common NVMe fabrics library. Part of
583  * the overall init sequence of starting up a fabrics driver.
584  */
585 int nvmf_register_transport(struct nvmf_transport_ops *ops)
586 {
587 	if (!ops->create_ctrl)
588 		return -EINVAL;
589 
590 	down_write(&nvmf_transports_rwsem);
591 	list_add_tail(&ops->entry, &nvmf_transports);
592 	up_write(&nvmf_transports_rwsem);
593 
594 	return 0;
595 }
596 EXPORT_SYMBOL_GPL(nvmf_register_transport);
597 
598 /**
599  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
600  * @ops:	Transport ops instance to be unregistered from the
601  *		common fabrics library.
602  *
603  * Fabrics API function that unregisters the type of specific transport
604  * fabric being implemented from the common NVMe fabrics library.
605  * Part of the overall exit sequence of unloading the implemented driver.
606  */
607 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
608 {
609 	down_write(&nvmf_transports_rwsem);
610 	list_del(&ops->entry);
611 	up_write(&nvmf_transports_rwsem);
612 }
613 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
614 
615 static struct nvmf_transport_ops *nvmf_lookup_transport(
616 		struct nvmf_ctrl_options *opts)
617 {
618 	struct nvmf_transport_ops *ops;
619 
620 	lockdep_assert_held(&nvmf_transports_rwsem);
621 
622 	list_for_each_entry(ops, &nvmf_transports, entry) {
623 		if (strcmp(ops->name, opts->transport) == 0)
624 			return ops;
625 	}
626 
627 	return NULL;
628 }
629 
630 static struct key *nvmf_parse_key(int key_id)
631 {
632 	struct key *key;
633 
634 	if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
635 		pr_err("TLS is not supported\n");
636 		return ERR_PTR(-EINVAL);
637 	}
638 
639 	key = key_lookup(key_id);
640 	if (!IS_ERR(key))
641 		pr_err("key id %08x not found\n", key_id);
642 	else
643 		pr_debug("Using key id %08x\n", key_id);
644 	return key;
645 }
646 
647 static const match_table_t opt_tokens = {
648 	{ NVMF_OPT_TRANSPORT,		"transport=%s"		},
649 	{ NVMF_OPT_TRADDR,		"traddr=%s"		},
650 	{ NVMF_OPT_TRSVCID,		"trsvcid=%s"		},
651 	{ NVMF_OPT_NQN,			"nqn=%s"		},
652 	{ NVMF_OPT_QUEUE_SIZE,		"queue_size=%d"		},
653 	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
654 	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
655 	{ NVMF_OPT_CTRL_LOSS_TMO,	"ctrl_loss_tmo=%d"	},
656 	{ NVMF_OPT_KATO,		"keep_alive_tmo=%d"	},
657 	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
658 	{ NVMF_OPT_HOST_TRADDR,		"host_traddr=%s"	},
659 	{ NVMF_OPT_HOST_IFACE,		"host_iface=%s"		},
660 	{ NVMF_OPT_HOST_ID,		"hostid=%s"		},
661 	{ NVMF_OPT_DUP_CONNECT,		"duplicate_connect"	},
662 	{ NVMF_OPT_DISABLE_SQFLOW,	"disable_sqflow"	},
663 	{ NVMF_OPT_HDR_DIGEST,		"hdr_digest"		},
664 	{ NVMF_OPT_DATA_DIGEST,		"data_digest"		},
665 	{ NVMF_OPT_NR_WRITE_QUEUES,	"nr_write_queues=%d"	},
666 	{ NVMF_OPT_NR_POLL_QUEUES,	"nr_poll_queues=%d"	},
667 	{ NVMF_OPT_TOS,			"tos=%d"		},
668 #ifdef CONFIG_NVME_TCP_TLS
669 	{ NVMF_OPT_KEYRING,		"keyring=%d"		},
670 	{ NVMF_OPT_TLS_KEY,		"tls_key=%d"		},
671 #endif
672 	{ NVMF_OPT_FAIL_FAST_TMO,	"fast_io_fail_tmo=%d"	},
673 	{ NVMF_OPT_DISCOVERY,		"discovery"		},
674 #ifdef CONFIG_NVME_HOST_AUTH
675 	{ NVMF_OPT_DHCHAP_SECRET,	"dhchap_secret=%s"	},
676 	{ NVMF_OPT_DHCHAP_CTRL_SECRET,	"dhchap_ctrl_secret=%s"	},
677 #endif
678 #ifdef CONFIG_NVME_TCP_TLS
679 	{ NVMF_OPT_TLS,			"tls"			},
680 #endif
681 	{ NVMF_OPT_ERR,			NULL			}
682 };
683 
684 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
685 		const char *buf)
686 {
687 	substring_t args[MAX_OPT_ARGS];
688 	char *options, *o, *p;
689 	int token, ret = 0;
690 	size_t nqnlen  = 0;
691 	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO, key_id;
692 	uuid_t hostid;
693 	char hostnqn[NVMF_NQN_SIZE];
694 	struct key *key;
695 
696 	/* Set defaults */
697 	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
698 	opts->nr_io_queues = num_online_cpus();
699 	opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
700 	opts->kato = 0;
701 	opts->duplicate_connect = false;
702 	opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
703 	opts->hdr_digest = false;
704 	opts->data_digest = false;
705 	opts->tos = -1; /* < 0 == use transport default */
706 	opts->tls = false;
707 	opts->tls_key = NULL;
708 	opts->keyring = NULL;
709 
710 	options = o = kstrdup(buf, GFP_KERNEL);
711 	if (!options)
712 		return -ENOMEM;
713 
714 	/* use default host if not given by user space */
715 	uuid_copy(&hostid, &nvmf_default_host->id);
716 	strscpy(hostnqn, nvmf_default_host->nqn, NVMF_NQN_SIZE);
717 
718 	while ((p = strsep(&o, ",\n")) != NULL) {
719 		if (!*p)
720 			continue;
721 
722 		token = match_token(p, opt_tokens, args);
723 		opts->mask |= token;
724 		switch (token) {
725 		case NVMF_OPT_TRANSPORT:
726 			p = match_strdup(args);
727 			if (!p) {
728 				ret = -ENOMEM;
729 				goto out;
730 			}
731 			kfree(opts->transport);
732 			opts->transport = p;
733 			break;
734 		case NVMF_OPT_NQN:
735 			p = match_strdup(args);
736 			if (!p) {
737 				ret = -ENOMEM;
738 				goto out;
739 			}
740 			kfree(opts->subsysnqn);
741 			opts->subsysnqn = p;
742 			nqnlen = strlen(opts->subsysnqn);
743 			if (nqnlen >= NVMF_NQN_SIZE) {
744 				pr_err("%s needs to be < %d bytes\n",
745 					opts->subsysnqn, NVMF_NQN_SIZE);
746 				ret = -EINVAL;
747 				goto out;
748 			}
749 			opts->discovery_nqn =
750 				!(strcmp(opts->subsysnqn,
751 					 NVME_DISC_SUBSYS_NAME));
752 			break;
753 		case NVMF_OPT_TRADDR:
754 			p = match_strdup(args);
755 			if (!p) {
756 				ret = -ENOMEM;
757 				goto out;
758 			}
759 			kfree(opts->traddr);
760 			opts->traddr = p;
761 			break;
762 		case NVMF_OPT_TRSVCID:
763 			p = match_strdup(args);
764 			if (!p) {
765 				ret = -ENOMEM;
766 				goto out;
767 			}
768 			kfree(opts->trsvcid);
769 			opts->trsvcid = p;
770 			break;
771 		case NVMF_OPT_QUEUE_SIZE:
772 			if (match_int(args, &token)) {
773 				ret = -EINVAL;
774 				goto out;
775 			}
776 			if (token < NVMF_MIN_QUEUE_SIZE ||
777 			    token > NVMF_MAX_QUEUE_SIZE) {
778 				pr_err("Invalid queue_size %d\n", token);
779 				ret = -EINVAL;
780 				goto out;
781 			}
782 			opts->queue_size = token;
783 			break;
784 		case NVMF_OPT_NR_IO_QUEUES:
785 			if (match_int(args, &token)) {
786 				ret = -EINVAL;
787 				goto out;
788 			}
789 			if (token <= 0) {
790 				pr_err("Invalid number of IOQs %d\n", token);
791 				ret = -EINVAL;
792 				goto out;
793 			}
794 			if (opts->discovery_nqn) {
795 				pr_debug("Ignoring nr_io_queues value for discovery controller\n");
796 				break;
797 			}
798 
799 			opts->nr_io_queues = min_t(unsigned int,
800 					num_online_cpus(), token);
801 			break;
802 		case NVMF_OPT_KATO:
803 			if (match_int(args, &token)) {
804 				ret = -EINVAL;
805 				goto out;
806 			}
807 
808 			if (token < 0) {
809 				pr_err("Invalid keep_alive_tmo %d\n", token);
810 				ret = -EINVAL;
811 				goto out;
812 			} else if (token == 0 && !opts->discovery_nqn) {
813 				/* Allowed for debug */
814 				pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
815 			}
816 			opts->kato = token;
817 			break;
818 		case NVMF_OPT_CTRL_LOSS_TMO:
819 			if (match_int(args, &token)) {
820 				ret = -EINVAL;
821 				goto out;
822 			}
823 
824 			if (token < 0)
825 				pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
826 			ctrl_loss_tmo = token;
827 			break;
828 		case NVMF_OPT_FAIL_FAST_TMO:
829 			if (match_int(args, &token)) {
830 				ret = -EINVAL;
831 				goto out;
832 			}
833 
834 			if (token >= 0)
835 				pr_warn("I/O fail on reconnect controller after %d sec\n",
836 					token);
837 			else
838 				token = -1;
839 
840 			opts->fast_io_fail_tmo = token;
841 			break;
842 		case NVMF_OPT_HOSTNQN:
843 			if (opts->host) {
844 				pr_err("hostnqn already user-assigned: %s\n",
845 				       opts->host->nqn);
846 				ret = -EADDRINUSE;
847 				goto out;
848 			}
849 			p = match_strdup(args);
850 			if (!p) {
851 				ret = -ENOMEM;
852 				goto out;
853 			}
854 			nqnlen = strlen(p);
855 			if (nqnlen >= NVMF_NQN_SIZE) {
856 				pr_err("%s needs to be < %d bytes\n",
857 					p, NVMF_NQN_SIZE);
858 				kfree(p);
859 				ret = -EINVAL;
860 				goto out;
861 			}
862 			strscpy(hostnqn, p, NVMF_NQN_SIZE);
863 			kfree(p);
864 			break;
865 		case NVMF_OPT_RECONNECT_DELAY:
866 			if (match_int(args, &token)) {
867 				ret = -EINVAL;
868 				goto out;
869 			}
870 			if (token <= 0) {
871 				pr_err("Invalid reconnect_delay %d\n", token);
872 				ret = -EINVAL;
873 				goto out;
874 			}
875 			opts->reconnect_delay = token;
876 			break;
877 		case NVMF_OPT_HOST_TRADDR:
878 			p = match_strdup(args);
879 			if (!p) {
880 				ret = -ENOMEM;
881 				goto out;
882 			}
883 			kfree(opts->host_traddr);
884 			opts->host_traddr = p;
885 			break;
886 		case NVMF_OPT_HOST_IFACE:
887 			p = match_strdup(args);
888 			if (!p) {
889 				ret = -ENOMEM;
890 				goto out;
891 			}
892 			kfree(opts->host_iface);
893 			opts->host_iface = p;
894 			break;
895 		case NVMF_OPT_HOST_ID:
896 			p = match_strdup(args);
897 			if (!p) {
898 				ret = -ENOMEM;
899 				goto out;
900 			}
901 			ret = uuid_parse(p, &hostid);
902 			if (ret) {
903 				pr_err("Invalid hostid %s\n", p);
904 				ret = -EINVAL;
905 				kfree(p);
906 				goto out;
907 			}
908 			kfree(p);
909 			break;
910 		case NVMF_OPT_DUP_CONNECT:
911 			opts->duplicate_connect = true;
912 			break;
913 		case NVMF_OPT_DISABLE_SQFLOW:
914 			opts->disable_sqflow = true;
915 			break;
916 		case NVMF_OPT_HDR_DIGEST:
917 			opts->hdr_digest = true;
918 			break;
919 		case NVMF_OPT_DATA_DIGEST:
920 			opts->data_digest = true;
921 			break;
922 		case NVMF_OPT_NR_WRITE_QUEUES:
923 			if (match_int(args, &token)) {
924 				ret = -EINVAL;
925 				goto out;
926 			}
927 			if (token <= 0) {
928 				pr_err("Invalid nr_write_queues %d\n", token);
929 				ret = -EINVAL;
930 				goto out;
931 			}
932 			opts->nr_write_queues = token;
933 			break;
934 		case NVMF_OPT_NR_POLL_QUEUES:
935 			if (match_int(args, &token)) {
936 				ret = -EINVAL;
937 				goto out;
938 			}
939 			if (token <= 0) {
940 				pr_err("Invalid nr_poll_queues %d\n", token);
941 				ret = -EINVAL;
942 				goto out;
943 			}
944 			opts->nr_poll_queues = token;
945 			break;
946 		case NVMF_OPT_TOS:
947 			if (match_int(args, &token)) {
948 				ret = -EINVAL;
949 				goto out;
950 			}
951 			if (token < 0) {
952 				pr_err("Invalid type of service %d\n", token);
953 				ret = -EINVAL;
954 				goto out;
955 			}
956 			if (token > 255) {
957 				pr_warn("Clamping type of service to 255\n");
958 				token = 255;
959 			}
960 			opts->tos = token;
961 			break;
962 		case NVMF_OPT_KEYRING:
963 			if (match_int(args, &key_id) || key_id <= 0) {
964 				ret = -EINVAL;
965 				goto out;
966 			}
967 			key = nvmf_parse_key(key_id);
968 			if (IS_ERR(key)) {
969 				ret = PTR_ERR(key);
970 				goto out;
971 			}
972 			key_put(opts->keyring);
973 			opts->keyring = key;
974 			break;
975 		case NVMF_OPT_TLS_KEY:
976 			if (match_int(args, &key_id) || key_id <= 0) {
977 				ret = -EINVAL;
978 				goto out;
979 			}
980 			key = nvmf_parse_key(key_id);
981 			if (IS_ERR(key)) {
982 				ret = PTR_ERR(key);
983 				goto out;
984 			}
985 			key_put(opts->tls_key);
986 			opts->tls_key = key;
987 			break;
988 		case NVMF_OPT_DISCOVERY:
989 			opts->discovery_nqn = true;
990 			break;
991 		case NVMF_OPT_DHCHAP_SECRET:
992 			p = match_strdup(args);
993 			if (!p) {
994 				ret = -ENOMEM;
995 				goto out;
996 			}
997 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
998 				pr_err("Invalid DH-CHAP secret %s\n", p);
999 				ret = -EINVAL;
1000 				goto out;
1001 			}
1002 			kfree(opts->dhchap_secret);
1003 			opts->dhchap_secret = p;
1004 			break;
1005 		case NVMF_OPT_DHCHAP_CTRL_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_ctrl_secret);
1017 			opts->dhchap_ctrl_secret = p;
1018 			break;
1019 		case NVMF_OPT_TLS:
1020 			if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
1021 				pr_err("TLS is not supported\n");
1022 				ret = -EINVAL;
1023 				goto out;
1024 			}
1025 			opts->tls = true;
1026 			break;
1027 		default:
1028 			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
1029 				p);
1030 			ret = -EINVAL;
1031 			goto out;
1032 		}
1033 	}
1034 
1035 	if (opts->discovery_nqn) {
1036 		opts->nr_io_queues = 0;
1037 		opts->nr_write_queues = 0;
1038 		opts->nr_poll_queues = 0;
1039 		opts->duplicate_connect = true;
1040 	} else {
1041 		if (!opts->kato)
1042 			opts->kato = NVME_DEFAULT_KATO;
1043 	}
1044 	if (ctrl_loss_tmo < 0) {
1045 		opts->max_reconnects = -1;
1046 	} else {
1047 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
1048 						opts->reconnect_delay);
1049 		if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
1050 			pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
1051 				opts->fast_io_fail_tmo, ctrl_loss_tmo);
1052 	}
1053 
1054 	opts->host = nvmf_host_add(hostnqn, &hostid);
1055 	if (IS_ERR(opts->host)) {
1056 		ret = PTR_ERR(opts->host);
1057 		opts->host = NULL;
1058 		goto out;
1059 	}
1060 
1061 out:
1062 	kfree(options);
1063 	return ret;
1064 }
1065 
1066 void nvmf_set_io_queues(struct nvmf_ctrl_options *opts, u32 nr_io_queues,
1067 			u32 io_queues[HCTX_MAX_TYPES])
1068 {
1069 	if (opts->nr_write_queues && opts->nr_io_queues < nr_io_queues) {
1070 		/*
1071 		 * separate read/write queues
1072 		 * hand out dedicated default queues only after we have
1073 		 * sufficient read queues.
1074 		 */
1075 		io_queues[HCTX_TYPE_READ] = opts->nr_io_queues;
1076 		nr_io_queues -= io_queues[HCTX_TYPE_READ];
1077 		io_queues[HCTX_TYPE_DEFAULT] =
1078 			min(opts->nr_write_queues, nr_io_queues);
1079 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1080 	} else {
1081 		/*
1082 		 * shared read/write queues
1083 		 * either no write queues were requested, or we don't have
1084 		 * sufficient queue count to have dedicated default queues.
1085 		 */
1086 		io_queues[HCTX_TYPE_DEFAULT] =
1087 			min(opts->nr_io_queues, nr_io_queues);
1088 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1089 	}
1090 
1091 	if (opts->nr_poll_queues && nr_io_queues) {
1092 		/* map dedicated poll queues only if we have queues left */
1093 		io_queues[HCTX_TYPE_POLL] =
1094 			min(opts->nr_poll_queues, nr_io_queues);
1095 	}
1096 }
1097 EXPORT_SYMBOL_GPL(nvmf_set_io_queues);
1098 
1099 void nvmf_map_queues(struct blk_mq_tag_set *set, struct nvme_ctrl *ctrl,
1100 		     u32 io_queues[HCTX_MAX_TYPES])
1101 {
1102 	struct nvmf_ctrl_options *opts = ctrl->opts;
1103 
1104 	if (opts->nr_write_queues && io_queues[HCTX_TYPE_READ]) {
1105 		/* separate read/write queues */
1106 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1107 			io_queues[HCTX_TYPE_DEFAULT];
1108 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1109 		set->map[HCTX_TYPE_READ].nr_queues =
1110 			io_queues[HCTX_TYPE_READ];
1111 		set->map[HCTX_TYPE_READ].queue_offset =
1112 			io_queues[HCTX_TYPE_DEFAULT];
1113 	} else {
1114 		/* shared read/write queues */
1115 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1116 			io_queues[HCTX_TYPE_DEFAULT];
1117 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1118 		set->map[HCTX_TYPE_READ].nr_queues =
1119 			io_queues[HCTX_TYPE_DEFAULT];
1120 		set->map[HCTX_TYPE_READ].queue_offset = 0;
1121 	}
1122 
1123 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1124 	blk_mq_map_queues(&set->map[HCTX_TYPE_READ]);
1125 	if (opts->nr_poll_queues && io_queues[HCTX_TYPE_POLL]) {
1126 		/* map dedicated poll queues only if we have queues left */
1127 		set->map[HCTX_TYPE_POLL].nr_queues = io_queues[HCTX_TYPE_POLL];
1128 		set->map[HCTX_TYPE_POLL].queue_offset =
1129 			io_queues[HCTX_TYPE_DEFAULT] +
1130 			io_queues[HCTX_TYPE_READ];
1131 		blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
1132 	}
1133 
1134 	dev_info(ctrl->device,
1135 		"mapped %d/%d/%d default/read/poll queues.\n",
1136 		io_queues[HCTX_TYPE_DEFAULT],
1137 		io_queues[HCTX_TYPE_READ],
1138 		io_queues[HCTX_TYPE_POLL]);
1139 }
1140 EXPORT_SYMBOL_GPL(nvmf_map_queues);
1141 
1142 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
1143 		unsigned int required_opts)
1144 {
1145 	if ((opts->mask & required_opts) != required_opts) {
1146 		unsigned int i;
1147 
1148 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1149 			if ((opt_tokens[i].token & required_opts) &&
1150 			    !(opt_tokens[i].token & opts->mask)) {
1151 				pr_warn("missing parameter '%s'\n",
1152 					opt_tokens[i].pattern);
1153 			}
1154 		}
1155 
1156 		return -EINVAL;
1157 	}
1158 
1159 	return 0;
1160 }
1161 
1162 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
1163 		struct nvmf_ctrl_options *opts)
1164 {
1165 	if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
1166 	    strcmp(opts->traddr, ctrl->opts->traddr) ||
1167 	    strcmp(opts->trsvcid, ctrl->opts->trsvcid))
1168 		return false;
1169 
1170 	/*
1171 	 * Checking the local address or host interfaces is rough.
1172 	 *
1173 	 * In most cases, none is specified and the host port or
1174 	 * host interface is selected by the stack.
1175 	 *
1176 	 * Assume no match if:
1177 	 * -  local address or host interface is specified and address
1178 	 *    or host interface is not the same
1179 	 * -  local address or host interface is not specified but
1180 	 *    remote is, or vice versa (admin using specific
1181 	 *    host_traddr/host_iface when it matters).
1182 	 */
1183 	if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
1184 	    (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1185 		if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
1186 			return false;
1187 	} else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
1188 		   (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1189 		return false;
1190 	}
1191 
1192 	if ((opts->mask & NVMF_OPT_HOST_IFACE) &&
1193 	    (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1194 		if (strcmp(opts->host_iface, ctrl->opts->host_iface))
1195 			return false;
1196 	} else if ((opts->mask & NVMF_OPT_HOST_IFACE) ||
1197 		   (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1198 		return false;
1199 	}
1200 
1201 	return true;
1202 }
1203 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
1204 
1205 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
1206 		unsigned int allowed_opts)
1207 {
1208 	if (opts->mask & ~allowed_opts) {
1209 		unsigned int i;
1210 
1211 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1212 			if ((opt_tokens[i].token & opts->mask) &&
1213 			    (opt_tokens[i].token & ~allowed_opts)) {
1214 				pr_warn("invalid parameter '%s'\n",
1215 					opt_tokens[i].pattern);
1216 			}
1217 		}
1218 
1219 		return -EINVAL;
1220 	}
1221 
1222 	return 0;
1223 }
1224 
1225 void nvmf_free_options(struct nvmf_ctrl_options *opts)
1226 {
1227 	nvmf_host_put(opts->host);
1228 	key_put(opts->keyring);
1229 	key_put(opts->tls_key);
1230 	kfree(opts->transport);
1231 	kfree(opts->traddr);
1232 	kfree(opts->trsvcid);
1233 	kfree(opts->subsysnqn);
1234 	kfree(opts->host_traddr);
1235 	kfree(opts->host_iface);
1236 	kfree(opts->dhchap_secret);
1237 	kfree(opts->dhchap_ctrl_secret);
1238 	kfree(opts);
1239 }
1240 EXPORT_SYMBOL_GPL(nvmf_free_options);
1241 
1242 #define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
1243 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
1244 				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
1245 				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
1246 				 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\
1247 				 NVMF_OPT_FAIL_FAST_TMO | NVMF_OPT_DHCHAP_SECRET |\
1248 				 NVMF_OPT_DHCHAP_CTRL_SECRET)
1249 
1250 static struct nvme_ctrl *
1251 nvmf_create_ctrl(struct device *dev, const char *buf)
1252 {
1253 	struct nvmf_ctrl_options *opts;
1254 	struct nvmf_transport_ops *ops;
1255 	struct nvme_ctrl *ctrl;
1256 	int ret;
1257 
1258 	opts = kzalloc(sizeof(*opts), GFP_KERNEL);
1259 	if (!opts)
1260 		return ERR_PTR(-ENOMEM);
1261 
1262 	ret = nvmf_parse_options(opts, buf);
1263 	if (ret)
1264 		goto out_free_opts;
1265 
1266 
1267 	request_module("nvme-%s", opts->transport);
1268 
1269 	/*
1270 	 * Check the generic options first as we need a valid transport for
1271 	 * the lookup below.  Then clear the generic flags so that transport
1272 	 * drivers don't have to care about them.
1273 	 */
1274 	ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1275 	if (ret)
1276 		goto out_free_opts;
1277 	opts->mask &= ~NVMF_REQUIRED_OPTS;
1278 
1279 	down_read(&nvmf_transports_rwsem);
1280 	ops = nvmf_lookup_transport(opts);
1281 	if (!ops) {
1282 		pr_info("no handler found for transport %s.\n",
1283 			opts->transport);
1284 		ret = -EINVAL;
1285 		goto out_unlock;
1286 	}
1287 
1288 	if (!try_module_get(ops->module)) {
1289 		ret = -EBUSY;
1290 		goto out_unlock;
1291 	}
1292 	up_read(&nvmf_transports_rwsem);
1293 
1294 	ret = nvmf_check_required_opts(opts, ops->required_opts);
1295 	if (ret)
1296 		goto out_module_put;
1297 	ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1298 				ops->allowed_opts | ops->required_opts);
1299 	if (ret)
1300 		goto out_module_put;
1301 
1302 	ctrl = ops->create_ctrl(dev, opts);
1303 	if (IS_ERR(ctrl)) {
1304 		ret = PTR_ERR(ctrl);
1305 		goto out_module_put;
1306 	}
1307 
1308 	module_put(ops->module);
1309 	return ctrl;
1310 
1311 out_module_put:
1312 	module_put(ops->module);
1313 	goto out_free_opts;
1314 out_unlock:
1315 	up_read(&nvmf_transports_rwsem);
1316 out_free_opts:
1317 	nvmf_free_options(opts);
1318 	return ERR_PTR(ret);
1319 }
1320 
1321 static struct class *nvmf_class;
1322 static struct device *nvmf_device;
1323 static DEFINE_MUTEX(nvmf_dev_mutex);
1324 
1325 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1326 		size_t count, loff_t *pos)
1327 {
1328 	struct seq_file *seq_file = file->private_data;
1329 	struct nvme_ctrl *ctrl;
1330 	const char *buf;
1331 	int ret = 0;
1332 
1333 	if (count > PAGE_SIZE)
1334 		return -ENOMEM;
1335 
1336 	buf = memdup_user_nul(ubuf, count);
1337 	if (IS_ERR(buf))
1338 		return PTR_ERR(buf);
1339 
1340 	mutex_lock(&nvmf_dev_mutex);
1341 	if (seq_file->private) {
1342 		ret = -EINVAL;
1343 		goto out_unlock;
1344 	}
1345 
1346 	ctrl = nvmf_create_ctrl(nvmf_device, buf);
1347 	if (IS_ERR(ctrl)) {
1348 		ret = PTR_ERR(ctrl);
1349 		goto out_unlock;
1350 	}
1351 
1352 	seq_file->private = ctrl;
1353 
1354 out_unlock:
1355 	mutex_unlock(&nvmf_dev_mutex);
1356 	kfree(buf);
1357 	return ret ? ret : count;
1358 }
1359 
1360 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file)
1361 {
1362 	const struct match_token *tok;
1363 	int idx;
1364 
1365 	/*
1366 	 * Add dummy entries for instance and cntlid to
1367 	 * signal an invalid/non-existing controller
1368 	 */
1369 	seq_puts(seq_file, "instance=-1,cntlid=-1");
1370 	for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) {
1371 		tok = &opt_tokens[idx];
1372 		if (tok->token == NVMF_OPT_ERR)
1373 			continue;
1374 		seq_puts(seq_file, ",");
1375 		seq_puts(seq_file, tok->pattern);
1376 	}
1377 	seq_puts(seq_file, "\n");
1378 }
1379 
1380 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1381 {
1382 	struct nvme_ctrl *ctrl;
1383 
1384 	mutex_lock(&nvmf_dev_mutex);
1385 	ctrl = seq_file->private;
1386 	if (!ctrl) {
1387 		__nvmf_concat_opt_tokens(seq_file);
1388 		goto out_unlock;
1389 	}
1390 
1391 	seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1392 			ctrl->instance, ctrl->cntlid);
1393 
1394 out_unlock:
1395 	mutex_unlock(&nvmf_dev_mutex);
1396 	return 0;
1397 }
1398 
1399 static int nvmf_dev_open(struct inode *inode, struct file *file)
1400 {
1401 	/*
1402 	 * The miscdevice code initializes file->private_data, but doesn't
1403 	 * make use of it later.
1404 	 */
1405 	file->private_data = NULL;
1406 	return single_open(file, nvmf_dev_show, NULL);
1407 }
1408 
1409 static int nvmf_dev_release(struct inode *inode, struct file *file)
1410 {
1411 	struct seq_file *seq_file = file->private_data;
1412 	struct nvme_ctrl *ctrl = seq_file->private;
1413 
1414 	if (ctrl)
1415 		nvme_put_ctrl(ctrl);
1416 	return single_release(inode, file);
1417 }
1418 
1419 static const struct file_operations nvmf_dev_fops = {
1420 	.owner		= THIS_MODULE,
1421 	.write		= nvmf_dev_write,
1422 	.read		= seq_read,
1423 	.open		= nvmf_dev_open,
1424 	.release	= nvmf_dev_release,
1425 };
1426 
1427 static struct miscdevice nvmf_misc = {
1428 	.minor		= MISC_DYNAMIC_MINOR,
1429 	.name           = "nvme-fabrics",
1430 	.fops		= &nvmf_dev_fops,
1431 };
1432 
1433 static int __init nvmf_init(void)
1434 {
1435 	int ret;
1436 
1437 	nvmf_default_host = nvmf_host_default();
1438 	if (!nvmf_default_host)
1439 		return -ENOMEM;
1440 
1441 	nvmf_class = class_create("nvme-fabrics");
1442 	if (IS_ERR(nvmf_class)) {
1443 		pr_err("couldn't register class nvme-fabrics\n");
1444 		ret = PTR_ERR(nvmf_class);
1445 		goto out_free_host;
1446 	}
1447 
1448 	nvmf_device =
1449 		device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1450 	if (IS_ERR(nvmf_device)) {
1451 		pr_err("couldn't create nvme-fabrics device!\n");
1452 		ret = PTR_ERR(nvmf_device);
1453 		goto out_destroy_class;
1454 	}
1455 
1456 	ret = misc_register(&nvmf_misc);
1457 	if (ret) {
1458 		pr_err("couldn't register misc device: %d\n", ret);
1459 		goto out_destroy_device;
1460 	}
1461 
1462 	return 0;
1463 
1464 out_destroy_device:
1465 	device_destroy(nvmf_class, MKDEV(0, 0));
1466 out_destroy_class:
1467 	class_destroy(nvmf_class);
1468 out_free_host:
1469 	nvmf_host_put(nvmf_default_host);
1470 	return ret;
1471 }
1472 
1473 static void __exit nvmf_exit(void)
1474 {
1475 	misc_deregister(&nvmf_misc);
1476 	device_destroy(nvmf_class, MKDEV(0, 0));
1477 	class_destroy(nvmf_class);
1478 	nvmf_host_put(nvmf_default_host);
1479 
1480 	BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1481 	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1482 	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1483 	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1484 	BUILD_BUG_ON(sizeof(struct nvmf_auth_send_command) != 64);
1485 	BUILD_BUG_ON(sizeof(struct nvmf_auth_receive_command) != 64);
1486 	BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1487 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_negotiate_data) != 8);
1488 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_challenge_data) != 16);
1489 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_reply_data) != 16);
1490 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success1_data) != 16);
1491 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success2_data) != 16);
1492 }
1493 
1494 MODULE_LICENSE("GPL v2");
1495 MODULE_DESCRIPTION("NVMe host fabrics library");
1496 
1497 module_init(nvmf_init);
1498 module_exit(nvmf_exit);
1499