xref: /openbmc/linux/drivers/nvme/host/fabrics.c (revision 2d96b44f)
1 /*
2  * NVMe over Fabrics common host code.
3  * Copyright (c) 2015-2016 HGST, a Western Digital Company.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms and conditions of the GNU General Public License,
7  * version 2, as published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope it will be useful, but WITHOUT
10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
12  * more details.
13  */
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/init.h>
16 #include <linux/miscdevice.h>
17 #include <linux/module.h>
18 #include <linux/mutex.h>
19 #include <linux/parser.h>
20 #include <linux/seq_file.h>
21 #include "nvme.h"
22 #include "fabrics.h"
23 
24 static LIST_HEAD(nvmf_transports);
25 static DEFINE_MUTEX(nvmf_transports_mutex);
26 
27 static LIST_HEAD(nvmf_hosts);
28 static DEFINE_MUTEX(nvmf_hosts_mutex);
29 
30 static struct nvmf_host *nvmf_default_host;
31 
32 static struct nvmf_host *__nvmf_host_find(const char *hostnqn)
33 {
34 	struct nvmf_host *host;
35 
36 	list_for_each_entry(host, &nvmf_hosts, list) {
37 		if (!strcmp(host->nqn, hostnqn))
38 			return host;
39 	}
40 
41 	return NULL;
42 }
43 
44 static struct nvmf_host *nvmf_host_add(const char *hostnqn)
45 {
46 	struct nvmf_host *host;
47 
48 	mutex_lock(&nvmf_hosts_mutex);
49 	host = __nvmf_host_find(hostnqn);
50 	if (host) {
51 		kref_get(&host->ref);
52 		goto out_unlock;
53 	}
54 
55 	host = kmalloc(sizeof(*host), GFP_KERNEL);
56 	if (!host)
57 		goto out_unlock;
58 
59 	kref_init(&host->ref);
60 	memcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
61 	uuid_be_gen(&host->id);
62 
63 	list_add_tail(&host->list, &nvmf_hosts);
64 out_unlock:
65 	mutex_unlock(&nvmf_hosts_mutex);
66 	return host;
67 }
68 
69 static struct nvmf_host *nvmf_host_default(void)
70 {
71 	struct nvmf_host *host;
72 
73 	host = kmalloc(sizeof(*host), GFP_KERNEL);
74 	if (!host)
75 		return NULL;
76 
77 	kref_init(&host->ref);
78 	uuid_be_gen(&host->id);
79 	snprintf(host->nqn, NVMF_NQN_SIZE,
80 		"nqn.2014-08.org.nvmexpress:NVMf:uuid:%pUb", &host->id);
81 
82 	mutex_lock(&nvmf_hosts_mutex);
83 	list_add_tail(&host->list, &nvmf_hosts);
84 	mutex_unlock(&nvmf_hosts_mutex);
85 
86 	return host;
87 }
88 
89 static void nvmf_host_destroy(struct kref *ref)
90 {
91 	struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
92 
93 	mutex_lock(&nvmf_hosts_mutex);
94 	list_del(&host->list);
95 	mutex_unlock(&nvmf_hosts_mutex);
96 
97 	kfree(host);
98 }
99 
100 static void nvmf_host_put(struct nvmf_host *host)
101 {
102 	if (host)
103 		kref_put(&host->ref, nvmf_host_destroy);
104 }
105 
106 /**
107  * nvmf_get_address() -  Get address/port
108  * @ctrl:	Host NVMe controller instance which we got the address
109  * @buf:	OUTPUT parameter that will contain the address/port
110  * @size:	buffer size
111  */
112 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
113 {
114 	int len = 0;
115 
116 	if (ctrl->opts->mask & NVMF_OPT_TRADDR)
117 		len += snprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
118 	if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
119 		len += snprintf(buf + len, size - len, "%strsvcid=%s",
120 				(len) ? "," : "", ctrl->opts->trsvcid);
121 	if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
122 		len += snprintf(buf + len, size - len, "%shost_traddr=%s",
123 				(len) ? "," : "", ctrl->opts->host_traddr);
124 	len += snprintf(buf + len, size - len, "\n");
125 
126 	return len;
127 }
128 EXPORT_SYMBOL_GPL(nvmf_get_address);
129 
130 /**
131  * nvmf_get_subsysnqn() - Get subsystem NQN
132  * @ctrl:	Host NVMe controller instance which we got the NQN
133  */
134 const char *nvmf_get_subsysnqn(struct nvme_ctrl *ctrl)
135 {
136 	return ctrl->opts->subsysnqn;
137 }
138 EXPORT_SYMBOL_GPL(nvmf_get_subsysnqn);
139 
140 /**
141  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
142  * @ctrl:	Host NVMe controller instance maintaining the admin
143  *		queue used to submit the property read command to
144  *		the allocated NVMe controller resource on the target system.
145  * @off:	Starting offset value of the targeted property
146  *		register (see the fabrics section of the NVMe standard).
147  * @val:	OUTPUT parameter that will contain the value of
148  *		the property after a successful read.
149  *
150  * Used by the host system to retrieve a 32-bit capsule property value
151  * from an NVMe controller on the target system.
152  *
153  * ("Capsule property" is an "PCIe register concept" applied to the
154  * NVMe fabrics space.)
155  *
156  * Return:
157  *	0: successful read
158  *	> 0: NVMe error status code
159  *	< 0: Linux errno error code
160  */
161 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
162 {
163 	struct nvme_command cmd;
164 	struct nvme_completion cqe;
165 	int ret;
166 
167 	memset(&cmd, 0, sizeof(cmd));
168 	cmd.prop_get.opcode = nvme_fabrics_command;
169 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
170 	cmd.prop_get.offset = cpu_to_le32(off);
171 
172 	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe, NULL, 0, 0,
173 			NVME_QID_ANY, 0, 0);
174 
175 	if (ret >= 0)
176 		*val = le64_to_cpu(cqe.result64);
177 	if (unlikely(ret != 0))
178 		dev_err(ctrl->device,
179 			"Property Get error: %d, offset %#x\n",
180 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
181 
182 	return ret;
183 }
184 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
185 
186 /**
187  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
188  * @ctrl:	Host NVMe controller instance maintaining the admin
189  *		queue used to submit the property read command to
190  *		the allocated controller resource on the target system.
191  * @off:	Starting offset value of the targeted property
192  *		register (see the fabrics section of the NVMe standard).
193  * @val:	OUTPUT parameter that will contain the value of
194  *		the property after a successful read.
195  *
196  * Used by the host system to retrieve a 64-bit capsule property value
197  * from an NVMe controller on the target system.
198  *
199  * ("Capsule property" is an "PCIe register concept" applied to the
200  * NVMe fabrics space.)
201  *
202  * Return:
203  *	0: successful read
204  *	> 0: NVMe error status code
205  *	< 0: Linux errno error code
206  */
207 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
208 {
209 	struct nvme_command cmd;
210 	struct nvme_completion cqe;
211 	int ret;
212 
213 	memset(&cmd, 0, sizeof(cmd));
214 	cmd.prop_get.opcode = nvme_fabrics_command;
215 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
216 	cmd.prop_get.attrib = 1;
217 	cmd.prop_get.offset = cpu_to_le32(off);
218 
219 	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe, NULL, 0, 0,
220 			NVME_QID_ANY, 0, 0);
221 
222 	if (ret >= 0)
223 		*val = le64_to_cpu(cqe.result64);
224 	if (unlikely(ret != 0))
225 		dev_err(ctrl->device,
226 			"Property Get error: %d, offset %#x\n",
227 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
228 	return ret;
229 }
230 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
231 
232 /**
233  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
234  * @ctrl:	Host NVMe controller instance maintaining the admin
235  *		queue used to submit the property read command to
236  *		the allocated NVMe controller resource on the target system.
237  * @off:	Starting offset value of the targeted property
238  *		register (see the fabrics section of the NVMe standard).
239  * @val:	Input parameter that contains the value to be
240  *		written to the property.
241  *
242  * Used by the NVMe host system to write a 32-bit capsule property value
243  * to an NVMe controller on the target system.
244  *
245  * ("Capsule property" is an "PCIe register concept" applied to the
246  * NVMe fabrics space.)
247  *
248  * Return:
249  *	0: successful write
250  *	> 0: NVMe error status code
251  *	< 0: Linux errno error code
252  */
253 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
254 {
255 	struct nvme_command cmd;
256 	int ret;
257 
258 	memset(&cmd, 0, sizeof(cmd));
259 	cmd.prop_set.opcode = nvme_fabrics_command;
260 	cmd.prop_set.fctype = nvme_fabrics_type_property_set;
261 	cmd.prop_set.attrib = 0;
262 	cmd.prop_set.offset = cpu_to_le32(off);
263 	cmd.prop_set.value = cpu_to_le64(val);
264 
265 	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
266 			NVME_QID_ANY, 0, 0);
267 	if (unlikely(ret))
268 		dev_err(ctrl->device,
269 			"Property Set error: %d, offset %#x\n",
270 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
271 	return ret;
272 }
273 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
274 
275 /**
276  * nvmf_log_connect_error() - Error-parsing-diagnostic print
277  * out function for connect() errors.
278  *
279  * @ctrl: the specific /dev/nvmeX device that had the error.
280  *
281  * @errval: Error code to be decoded in a more human-friendly
282  *	    printout.
283  *
284  * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
285  *
286  * @cmd: This is the SQE portion of a submission capsule.
287  *
288  * @data: This is the "Data" portion of a submission capsule.
289  */
290 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
291 		int errval, int offset, struct nvme_command *cmd,
292 		struct nvmf_connect_data *data)
293 {
294 	int err_sctype = errval & (~NVME_SC_DNR);
295 
296 	switch (err_sctype) {
297 
298 	case (NVME_SC_CONNECT_INVALID_PARAM):
299 		if (offset >> 16) {
300 			char *inv_data = "Connect Invalid Data Parameter";
301 
302 			switch (offset & 0xffff) {
303 			case (offsetof(struct nvmf_connect_data, cntlid)):
304 				dev_err(ctrl->device,
305 					"%s, cntlid: %d\n",
306 					inv_data, data->cntlid);
307 				break;
308 			case (offsetof(struct nvmf_connect_data, hostnqn)):
309 				dev_err(ctrl->device,
310 					"%s, hostnqn \"%s\"\n",
311 					inv_data, data->hostnqn);
312 				break;
313 			case (offsetof(struct nvmf_connect_data, subsysnqn)):
314 				dev_err(ctrl->device,
315 					"%s, subsysnqn \"%s\"\n",
316 					inv_data, data->subsysnqn);
317 				break;
318 			default:
319 				dev_err(ctrl->device,
320 					"%s, starting byte offset: %d\n",
321 				       inv_data, offset & 0xffff);
322 				break;
323 			}
324 		} else {
325 			char *inv_sqe = "Connect Invalid SQE Parameter";
326 
327 			switch (offset) {
328 			case (offsetof(struct nvmf_connect_command, qid)):
329 				dev_err(ctrl->device,
330 				       "%s, qid %d\n",
331 					inv_sqe, cmd->connect.qid);
332 				break;
333 			default:
334 				dev_err(ctrl->device,
335 					"%s, starting byte offset: %d\n",
336 					inv_sqe, offset);
337 			}
338 		}
339 		break;
340 	default:
341 		dev_err(ctrl->device,
342 			"Connect command failed, error wo/DNR bit: %d\n",
343 			err_sctype);
344 		break;
345 	} /* switch (err_sctype) */
346 }
347 
348 /**
349  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
350  *				API function.
351  * @ctrl:	Host nvme controller instance used to request
352  *              a new NVMe controller allocation on the target
353  *              system and  establish an NVMe Admin connection to
354  *              that controller.
355  *
356  * This function enables an NVMe host device to request a new allocation of
357  * an NVMe controller resource on a target system as well establish a
358  * fabrics-protocol connection of the NVMe Admin queue between the
359  * host system device and the allocated NVMe controller on the
360  * target system via a NVMe Fabrics "Connect" command.
361  *
362  * Return:
363  *	0: success
364  *	> 0: NVMe error status code
365  *	< 0: Linux errno error code
366  *
367  */
368 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
369 {
370 	struct nvme_command cmd;
371 	struct nvme_completion cqe;
372 	struct nvmf_connect_data *data;
373 	int ret;
374 
375 	memset(&cmd, 0, sizeof(cmd));
376 	cmd.connect.opcode = nvme_fabrics_command;
377 	cmd.connect.fctype = nvme_fabrics_type_connect;
378 	cmd.connect.qid = 0;
379 
380 	/*
381 	 * fabrics spec sets a minimum of depth 32 for admin queue,
382 	 * so set the queue with this depth always until
383 	 * justification otherwise.
384 	 */
385 	cmd.connect.sqsize = cpu_to_le16(NVMF_AQ_DEPTH - 1);
386 
387 	/*
388 	 * Set keep-alive timeout in seconds granularity (ms * 1000)
389 	 * and add a grace period for controller kato enforcement
390 	 */
391 	cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
392 		cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
393 
394 	data = kzalloc(sizeof(*data), GFP_KERNEL);
395 	if (!data)
396 		return -ENOMEM;
397 
398 	memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_be));
399 	data->cntlid = cpu_to_le16(0xffff);
400 	strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
401 	strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
402 
403 	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &cqe,
404 			data, sizeof(*data), 0, NVME_QID_ANY, 1,
405 			BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
406 	if (ret) {
407 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(cqe.result),
408 				       &cmd, data);
409 		goto out_free_data;
410 	}
411 
412 	ctrl->cntlid = le16_to_cpu(cqe.result16);
413 
414 out_free_data:
415 	kfree(data);
416 	return ret;
417 }
418 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
419 
420 /**
421  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
422  *			     API function.
423  * @ctrl:	Host nvme controller instance used to establish an
424  *		NVMe I/O queue connection to the already allocated NVMe
425  *		controller on the target system.
426  * @qid:	NVMe I/O queue number for the new I/O connection between
427  *		host and target (note qid == 0 is illegal as this is
428  *		the Admin queue, per NVMe standard).
429  *
430  * This function issues a fabrics-protocol connection
431  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
432  * between the host system device and the allocated NVMe controller
433  * on the target system.
434  *
435  * Return:
436  *	0: success
437  *	> 0: NVMe error status code
438  *	< 0: Linux errno error code
439  */
440 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
441 {
442 	struct nvme_command cmd;
443 	struct nvmf_connect_data *data;
444 	struct nvme_completion cqe;
445 	int ret;
446 
447 	memset(&cmd, 0, sizeof(cmd));
448 	cmd.connect.opcode = nvme_fabrics_command;
449 	cmd.connect.fctype = nvme_fabrics_type_connect;
450 	cmd.connect.qid = cpu_to_le16(qid);
451 	cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
452 
453 	data = kzalloc(sizeof(*data), GFP_KERNEL);
454 	if (!data)
455 		return -ENOMEM;
456 
457 	memcpy(&data->hostid, &ctrl->opts->host->id, sizeof(uuid_be));
458 	data->cntlid = cpu_to_le16(ctrl->cntlid);
459 	strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
460 	strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
461 
462 	ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &cqe,
463 			data, sizeof(*data), 0, qid, 1,
464 			BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
465 	if (ret) {
466 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(cqe.result),
467 				       &cmd, data);
468 	}
469 	kfree(data);
470 	return ret;
471 }
472 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
473 
474 /**
475  * nvmf_register_transport() - NVMe Fabrics Library registration function.
476  * @ops:	Transport ops instance to be registered to the
477  *		common fabrics library.
478  *
479  * API function that registers the type of specific transport fabric
480  * being implemented to the common NVMe fabrics library. Part of
481  * the overall init sequence of starting up a fabrics driver.
482  */
483 void nvmf_register_transport(struct nvmf_transport_ops *ops)
484 {
485 	mutex_lock(&nvmf_transports_mutex);
486 	list_add_tail(&ops->entry, &nvmf_transports);
487 	mutex_unlock(&nvmf_transports_mutex);
488 }
489 EXPORT_SYMBOL_GPL(nvmf_register_transport);
490 
491 /**
492  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
493  * @ops:	Transport ops instance to be unregistered from the
494  *		common fabrics library.
495  *
496  * Fabrics API function that unregisters the type of specific transport
497  * fabric being implemented from the common NVMe fabrics library.
498  * Part of the overall exit sequence of unloading the implemented driver.
499  */
500 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
501 {
502 	mutex_lock(&nvmf_transports_mutex);
503 	list_del(&ops->entry);
504 	mutex_unlock(&nvmf_transports_mutex);
505 }
506 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
507 
508 static struct nvmf_transport_ops *nvmf_lookup_transport(
509 		struct nvmf_ctrl_options *opts)
510 {
511 	struct nvmf_transport_ops *ops;
512 
513 	lockdep_assert_held(&nvmf_transports_mutex);
514 
515 	list_for_each_entry(ops, &nvmf_transports, entry) {
516 		if (strcmp(ops->name, opts->transport) == 0)
517 			return ops;
518 	}
519 
520 	return NULL;
521 }
522 
523 static const match_table_t opt_tokens = {
524 	{ NVMF_OPT_TRANSPORT,		"transport=%s"		},
525 	{ NVMF_OPT_TRADDR,		"traddr=%s"		},
526 	{ NVMF_OPT_TRSVCID,		"trsvcid=%s"		},
527 	{ NVMF_OPT_NQN,			"nqn=%s"		},
528 	{ NVMF_OPT_QUEUE_SIZE,		"queue_size=%d"		},
529 	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
530 	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
531 	{ NVMF_OPT_KATO,		"keep_alive_tmo=%d"	},
532 	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
533 	{ NVMF_OPT_HOST_TRADDR,		"host_traddr=%s"	},
534 	{ NVMF_OPT_ERR,			NULL			}
535 };
536 
537 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
538 		const char *buf)
539 {
540 	substring_t args[MAX_OPT_ARGS];
541 	char *options, *o, *p;
542 	int token, ret = 0;
543 	size_t nqnlen  = 0;
544 
545 	/* Set defaults */
546 	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
547 	opts->nr_io_queues = num_online_cpus();
548 	opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
549 
550 	options = o = kstrdup(buf, GFP_KERNEL);
551 	if (!options)
552 		return -ENOMEM;
553 
554 	while ((p = strsep(&o, ",\n")) != NULL) {
555 		if (!*p)
556 			continue;
557 
558 		token = match_token(p, opt_tokens, args);
559 		opts->mask |= token;
560 		switch (token) {
561 		case NVMF_OPT_TRANSPORT:
562 			p = match_strdup(args);
563 			if (!p) {
564 				ret = -ENOMEM;
565 				goto out;
566 			}
567 			opts->transport = p;
568 			break;
569 		case NVMF_OPT_NQN:
570 			p = match_strdup(args);
571 			if (!p) {
572 				ret = -ENOMEM;
573 				goto out;
574 			}
575 			opts->subsysnqn = p;
576 			nqnlen = strlen(opts->subsysnqn);
577 			if (nqnlen >= NVMF_NQN_SIZE) {
578 				pr_err("%s needs to be < %d bytes\n",
579 				opts->subsysnqn, NVMF_NQN_SIZE);
580 				ret = -EINVAL;
581 				goto out;
582 			}
583 			opts->discovery_nqn =
584 				!(strcmp(opts->subsysnqn,
585 					 NVME_DISC_SUBSYS_NAME));
586 			if (opts->discovery_nqn)
587 				opts->nr_io_queues = 0;
588 			break;
589 		case NVMF_OPT_TRADDR:
590 			p = match_strdup(args);
591 			if (!p) {
592 				ret = -ENOMEM;
593 				goto out;
594 			}
595 			opts->traddr = p;
596 			break;
597 		case NVMF_OPT_TRSVCID:
598 			p = match_strdup(args);
599 			if (!p) {
600 				ret = -ENOMEM;
601 				goto out;
602 			}
603 			opts->trsvcid = p;
604 			break;
605 		case NVMF_OPT_QUEUE_SIZE:
606 			if (match_int(args, &token)) {
607 				ret = -EINVAL;
608 				goto out;
609 			}
610 			if (token < NVMF_MIN_QUEUE_SIZE ||
611 			    token > NVMF_MAX_QUEUE_SIZE) {
612 				pr_err("Invalid queue_size %d\n", token);
613 				ret = -EINVAL;
614 				goto out;
615 			}
616 			opts->queue_size = token;
617 			break;
618 		case NVMF_OPT_NR_IO_QUEUES:
619 			if (match_int(args, &token)) {
620 				ret = -EINVAL;
621 				goto out;
622 			}
623 			if (token <= 0) {
624 				pr_err("Invalid number of IOQs %d\n", token);
625 				ret = -EINVAL;
626 				goto out;
627 			}
628 			opts->nr_io_queues = min_t(unsigned int,
629 					num_online_cpus(), token);
630 			break;
631 		case NVMF_OPT_KATO:
632 			if (match_int(args, &token)) {
633 				ret = -EINVAL;
634 				goto out;
635 			}
636 
637 			if (opts->discovery_nqn) {
638 				pr_err("Discovery controllers cannot accept keep_alive_tmo != 0\n");
639 				ret = -EINVAL;
640 				goto out;
641 			}
642 
643 			if (token < 0) {
644 				pr_err("Invalid keep_alive_tmo %d\n", token);
645 				ret = -EINVAL;
646 				goto out;
647 			} else if (token == 0) {
648 				/* Allowed for debug */
649 				pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
650 			}
651 			opts->kato = token;
652 			break;
653 		case NVMF_OPT_HOSTNQN:
654 			if (opts->host) {
655 				pr_err("hostnqn already user-assigned: %s\n",
656 				       opts->host->nqn);
657 				ret = -EADDRINUSE;
658 				goto out;
659 			}
660 			p = match_strdup(args);
661 			if (!p) {
662 				ret = -ENOMEM;
663 				goto out;
664 			}
665 			nqnlen = strlen(p);
666 			if (nqnlen >= NVMF_NQN_SIZE) {
667 				pr_err("%s needs to be < %d bytes\n",
668 					p, NVMF_NQN_SIZE);
669 				ret = -EINVAL;
670 				goto out;
671 			}
672 			opts->host = nvmf_host_add(p);
673 			if (!opts->host) {
674 				ret = -ENOMEM;
675 				goto out;
676 			}
677 			break;
678 		case NVMF_OPT_RECONNECT_DELAY:
679 			if (match_int(args, &token)) {
680 				ret = -EINVAL;
681 				goto out;
682 			}
683 			if (token <= 0) {
684 				pr_err("Invalid reconnect_delay %d\n", token);
685 				ret = -EINVAL;
686 				goto out;
687 			}
688 			opts->reconnect_delay = token;
689 			break;
690 		case NVMF_OPT_HOST_TRADDR:
691 			p = match_strdup(args);
692 			if (!p) {
693 				ret = -ENOMEM;
694 				goto out;
695 			}
696 			opts->host_traddr = p;
697 			break;
698 		default:
699 			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
700 				p);
701 			ret = -EINVAL;
702 			goto out;
703 		}
704 	}
705 
706 	if (!opts->host) {
707 		kref_get(&nvmf_default_host->ref);
708 		opts->host = nvmf_default_host;
709 	}
710 
711 out:
712 	if (!opts->discovery_nqn && !opts->kato)
713 		opts->kato = NVME_DEFAULT_KATO;
714 	kfree(options);
715 	return ret;
716 }
717 
718 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
719 		unsigned int required_opts)
720 {
721 	if ((opts->mask & required_opts) != required_opts) {
722 		int i;
723 
724 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
725 			if ((opt_tokens[i].token & required_opts) &&
726 			    !(opt_tokens[i].token & opts->mask)) {
727 				pr_warn("missing parameter '%s'\n",
728 					opt_tokens[i].pattern);
729 			}
730 		}
731 
732 		return -EINVAL;
733 	}
734 
735 	return 0;
736 }
737 
738 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
739 		unsigned int allowed_opts)
740 {
741 	if (opts->mask & ~allowed_opts) {
742 		int i;
743 
744 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
745 			if (opt_tokens[i].token & ~allowed_opts) {
746 				pr_warn("invalid parameter '%s'\n",
747 					opt_tokens[i].pattern);
748 			}
749 		}
750 
751 		return -EINVAL;
752 	}
753 
754 	return 0;
755 }
756 
757 void nvmf_free_options(struct nvmf_ctrl_options *opts)
758 {
759 	nvmf_host_put(opts->host);
760 	kfree(opts->transport);
761 	kfree(opts->traddr);
762 	kfree(opts->trsvcid);
763 	kfree(opts->subsysnqn);
764 	kfree(opts->host_traddr);
765 	kfree(opts);
766 }
767 EXPORT_SYMBOL_GPL(nvmf_free_options);
768 
769 #define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
770 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
771 				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN)
772 
773 static struct nvme_ctrl *
774 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
775 {
776 	struct nvmf_ctrl_options *opts;
777 	struct nvmf_transport_ops *ops;
778 	struct nvme_ctrl *ctrl;
779 	int ret;
780 
781 	opts = kzalloc(sizeof(*opts), GFP_KERNEL);
782 	if (!opts)
783 		return ERR_PTR(-ENOMEM);
784 
785 	ret = nvmf_parse_options(opts, buf);
786 	if (ret)
787 		goto out_free_opts;
788 
789 	/*
790 	 * Check the generic options first as we need a valid transport for
791 	 * the lookup below.  Then clear the generic flags so that transport
792 	 * drivers don't have to care about them.
793 	 */
794 	ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
795 	if (ret)
796 		goto out_free_opts;
797 	opts->mask &= ~NVMF_REQUIRED_OPTS;
798 
799 	mutex_lock(&nvmf_transports_mutex);
800 	ops = nvmf_lookup_transport(opts);
801 	if (!ops) {
802 		pr_info("no handler found for transport %s.\n",
803 			opts->transport);
804 		ret = -EINVAL;
805 		goto out_unlock;
806 	}
807 
808 	ret = nvmf_check_required_opts(opts, ops->required_opts);
809 	if (ret)
810 		goto out_unlock;
811 	ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
812 				ops->allowed_opts | ops->required_opts);
813 	if (ret)
814 		goto out_unlock;
815 
816 	ctrl = ops->create_ctrl(dev, opts);
817 	if (IS_ERR(ctrl)) {
818 		ret = PTR_ERR(ctrl);
819 		goto out_unlock;
820 	}
821 
822 	mutex_unlock(&nvmf_transports_mutex);
823 	return ctrl;
824 
825 out_unlock:
826 	mutex_unlock(&nvmf_transports_mutex);
827 out_free_opts:
828 	nvmf_host_put(opts->host);
829 	kfree(opts);
830 	return ERR_PTR(ret);
831 }
832 
833 static struct class *nvmf_class;
834 static struct device *nvmf_device;
835 static DEFINE_MUTEX(nvmf_dev_mutex);
836 
837 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
838 		size_t count, loff_t *pos)
839 {
840 	struct seq_file *seq_file = file->private_data;
841 	struct nvme_ctrl *ctrl;
842 	const char *buf;
843 	int ret = 0;
844 
845 	if (count > PAGE_SIZE)
846 		return -ENOMEM;
847 
848 	buf = memdup_user_nul(ubuf, count);
849 	if (IS_ERR(buf))
850 		return PTR_ERR(buf);
851 
852 	mutex_lock(&nvmf_dev_mutex);
853 	if (seq_file->private) {
854 		ret = -EINVAL;
855 		goto out_unlock;
856 	}
857 
858 	ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
859 	if (IS_ERR(ctrl)) {
860 		ret = PTR_ERR(ctrl);
861 		goto out_unlock;
862 	}
863 
864 	seq_file->private = ctrl;
865 
866 out_unlock:
867 	mutex_unlock(&nvmf_dev_mutex);
868 	kfree(buf);
869 	return ret ? ret : count;
870 }
871 
872 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
873 {
874 	struct nvme_ctrl *ctrl;
875 	int ret = 0;
876 
877 	mutex_lock(&nvmf_dev_mutex);
878 	ctrl = seq_file->private;
879 	if (!ctrl) {
880 		ret = -EINVAL;
881 		goto out_unlock;
882 	}
883 
884 	seq_printf(seq_file, "instance=%d,cntlid=%d\n",
885 			ctrl->instance, ctrl->cntlid);
886 
887 out_unlock:
888 	mutex_unlock(&nvmf_dev_mutex);
889 	return ret;
890 }
891 
892 static int nvmf_dev_open(struct inode *inode, struct file *file)
893 {
894 	/*
895 	 * The miscdevice code initializes file->private_data, but doesn't
896 	 * make use of it later.
897 	 */
898 	file->private_data = NULL;
899 	return single_open(file, nvmf_dev_show, NULL);
900 }
901 
902 static int nvmf_dev_release(struct inode *inode, struct file *file)
903 {
904 	struct seq_file *seq_file = file->private_data;
905 	struct nvme_ctrl *ctrl = seq_file->private;
906 
907 	if (ctrl)
908 		nvme_put_ctrl(ctrl);
909 	return single_release(inode, file);
910 }
911 
912 static const struct file_operations nvmf_dev_fops = {
913 	.owner		= THIS_MODULE,
914 	.write		= nvmf_dev_write,
915 	.read		= seq_read,
916 	.open		= nvmf_dev_open,
917 	.release	= nvmf_dev_release,
918 };
919 
920 static struct miscdevice nvmf_misc = {
921 	.minor		= MISC_DYNAMIC_MINOR,
922 	.name           = "nvme-fabrics",
923 	.fops		= &nvmf_dev_fops,
924 };
925 
926 static int __init nvmf_init(void)
927 {
928 	int ret;
929 
930 	nvmf_default_host = nvmf_host_default();
931 	if (!nvmf_default_host)
932 		return -ENOMEM;
933 
934 	nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
935 	if (IS_ERR(nvmf_class)) {
936 		pr_err("couldn't register class nvme-fabrics\n");
937 		ret = PTR_ERR(nvmf_class);
938 		goto out_free_host;
939 	}
940 
941 	nvmf_device =
942 		device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
943 	if (IS_ERR(nvmf_device)) {
944 		pr_err("couldn't create nvme-fabris device!\n");
945 		ret = PTR_ERR(nvmf_device);
946 		goto out_destroy_class;
947 	}
948 
949 	ret = misc_register(&nvmf_misc);
950 	if (ret) {
951 		pr_err("couldn't register misc device: %d\n", ret);
952 		goto out_destroy_device;
953 	}
954 
955 	return 0;
956 
957 out_destroy_device:
958 	device_destroy(nvmf_class, MKDEV(0, 0));
959 out_destroy_class:
960 	class_destroy(nvmf_class);
961 out_free_host:
962 	nvmf_host_put(nvmf_default_host);
963 	return ret;
964 }
965 
966 static void __exit nvmf_exit(void)
967 {
968 	misc_deregister(&nvmf_misc);
969 	device_destroy(nvmf_class, MKDEV(0, 0));
970 	class_destroy(nvmf_class);
971 	nvmf_host_put(nvmf_default_host);
972 
973 	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
974 	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
975 	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
976 	BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
977 }
978 
979 MODULE_LICENSE("GPL v2");
980 
981 module_init(nvmf_init);
982 module_exit(nvmf_exit);
983