xref: /openbmc/linux/drivers/nvme/host/fabrics.c (revision 4da722ca)
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 
62 	list_add_tail(&host->list, &nvmf_hosts);
63 out_unlock:
64 	mutex_unlock(&nvmf_hosts_mutex);
65 	return host;
66 }
67 
68 static struct nvmf_host *nvmf_host_default(void)
69 {
70 	struct nvmf_host *host;
71 
72 	host = kmalloc(sizeof(*host), GFP_KERNEL);
73 	if (!host)
74 		return NULL;
75 
76 	kref_init(&host->ref);
77 	snprintf(host->nqn, NVMF_NQN_SIZE,
78 		"nqn.2014-08.org.nvmexpress:NVMf:uuid:%pUb", &host->id);
79 
80 	mutex_lock(&nvmf_hosts_mutex);
81 	list_add_tail(&host->list, &nvmf_hosts);
82 	mutex_unlock(&nvmf_hosts_mutex);
83 
84 	return host;
85 }
86 
87 static void nvmf_host_destroy(struct kref *ref)
88 {
89 	struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
90 
91 	mutex_lock(&nvmf_hosts_mutex);
92 	list_del(&host->list);
93 	mutex_unlock(&nvmf_hosts_mutex);
94 
95 	kfree(host);
96 }
97 
98 static void nvmf_host_put(struct nvmf_host *host)
99 {
100 	if (host)
101 		kref_put(&host->ref, nvmf_host_destroy);
102 }
103 
104 /**
105  * nvmf_get_address() -  Get address/port
106  * @ctrl:	Host NVMe controller instance which we got the address
107  * @buf:	OUTPUT parameter that will contain the address/port
108  * @size:	buffer size
109  */
110 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
111 {
112 	int len = 0;
113 
114 	if (ctrl->opts->mask & NVMF_OPT_TRADDR)
115 		len += snprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
116 	if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
117 		len += snprintf(buf + len, size - len, "%strsvcid=%s",
118 				(len) ? "," : "", ctrl->opts->trsvcid);
119 	if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
120 		len += snprintf(buf + len, size - len, "%shost_traddr=%s",
121 				(len) ? "," : "", ctrl->opts->host_traddr);
122 	len += snprintf(buf + len, size - len, "\n");
123 
124 	return len;
125 }
126 EXPORT_SYMBOL_GPL(nvmf_get_address);
127 
128 /**
129  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
130  * @ctrl:	Host NVMe controller instance maintaining the admin
131  *		queue used to submit the property read command to
132  *		the allocated NVMe controller resource on the target system.
133  * @off:	Starting offset value of the targeted property
134  *		register (see the fabrics section of the NVMe standard).
135  * @val:	OUTPUT parameter that will contain the value of
136  *		the property after a successful read.
137  *
138  * Used by the host system to retrieve a 32-bit capsule property value
139  * from an NVMe controller on the target system.
140  *
141  * ("Capsule property" is an "PCIe register concept" applied to the
142  * NVMe fabrics space.)
143  *
144  * Return:
145  *	0: successful read
146  *	> 0: NVMe error status code
147  *	< 0: Linux errno error code
148  */
149 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
150 {
151 	struct nvme_command cmd;
152 	union nvme_result res;
153 	int ret;
154 
155 	memset(&cmd, 0, sizeof(cmd));
156 	cmd.prop_get.opcode = nvme_fabrics_command;
157 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
158 	cmd.prop_get.offset = cpu_to_le32(off);
159 
160 	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
161 			NVME_QID_ANY, 0, 0);
162 
163 	if (ret >= 0)
164 		*val = le64_to_cpu(res.u64);
165 	if (unlikely(ret != 0))
166 		dev_err(ctrl->device,
167 			"Property Get error: %d, offset %#x\n",
168 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
169 
170 	return ret;
171 }
172 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
173 
174 /**
175  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
176  * @ctrl:	Host NVMe controller instance maintaining the admin
177  *		queue used to submit the property read command to
178  *		the allocated controller resource on the target system.
179  * @off:	Starting offset value of the targeted property
180  *		register (see the fabrics section of the NVMe standard).
181  * @val:	OUTPUT parameter that will contain the value of
182  *		the property after a successful read.
183  *
184  * Used by the host system to retrieve a 64-bit capsule property value
185  * from an NVMe controller on the target system.
186  *
187  * ("Capsule property" is an "PCIe register concept" applied to the
188  * NVMe fabrics space.)
189  *
190  * Return:
191  *	0: successful read
192  *	> 0: NVMe error status code
193  *	< 0: Linux errno error code
194  */
195 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
196 {
197 	struct nvme_command cmd;
198 	union nvme_result res;
199 	int ret;
200 
201 	memset(&cmd, 0, sizeof(cmd));
202 	cmd.prop_get.opcode = nvme_fabrics_command;
203 	cmd.prop_get.fctype = nvme_fabrics_type_property_get;
204 	cmd.prop_get.attrib = 1;
205 	cmd.prop_get.offset = cpu_to_le32(off);
206 
207 	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
208 			NVME_QID_ANY, 0, 0);
209 
210 	if (ret >= 0)
211 		*val = le64_to_cpu(res.u64);
212 	if (unlikely(ret != 0))
213 		dev_err(ctrl->device,
214 			"Property Get error: %d, offset %#x\n",
215 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
216 	return ret;
217 }
218 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
219 
220 /**
221  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
222  * @ctrl:	Host NVMe controller instance maintaining the admin
223  *		queue used to submit the property read command to
224  *		the allocated NVMe controller resource on the target system.
225  * @off:	Starting offset value of the targeted property
226  *		register (see the fabrics section of the NVMe standard).
227  * @val:	Input parameter that contains the value to be
228  *		written to the property.
229  *
230  * Used by the NVMe host system to write a 32-bit capsule property value
231  * to an NVMe controller on the target system.
232  *
233  * ("Capsule property" is an "PCIe register concept" applied to the
234  * NVMe fabrics space.)
235  *
236  * Return:
237  *	0: successful write
238  *	> 0: NVMe error status code
239  *	< 0: Linux errno error code
240  */
241 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
242 {
243 	struct nvme_command cmd;
244 	int ret;
245 
246 	memset(&cmd, 0, sizeof(cmd));
247 	cmd.prop_set.opcode = nvme_fabrics_command;
248 	cmd.prop_set.fctype = nvme_fabrics_type_property_set;
249 	cmd.prop_set.attrib = 0;
250 	cmd.prop_set.offset = cpu_to_le32(off);
251 	cmd.prop_set.value = cpu_to_le64(val);
252 
253 	ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
254 			NVME_QID_ANY, 0, 0);
255 	if (unlikely(ret))
256 		dev_err(ctrl->device,
257 			"Property Set error: %d, offset %#x\n",
258 			ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
259 	return ret;
260 }
261 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
262 
263 /**
264  * nvmf_log_connect_error() - Error-parsing-diagnostic print
265  * out function for connect() errors.
266  *
267  * @ctrl: the specific /dev/nvmeX device that had the error.
268  *
269  * @errval: Error code to be decoded in a more human-friendly
270  *	    printout.
271  *
272  * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
273  *
274  * @cmd: This is the SQE portion of a submission capsule.
275  *
276  * @data: This is the "Data" portion of a submission capsule.
277  */
278 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
279 		int errval, int offset, struct nvme_command *cmd,
280 		struct nvmf_connect_data *data)
281 {
282 	int err_sctype = errval & (~NVME_SC_DNR);
283 
284 	switch (err_sctype) {
285 
286 	case (NVME_SC_CONNECT_INVALID_PARAM):
287 		if (offset >> 16) {
288 			char *inv_data = "Connect Invalid Data Parameter";
289 
290 			switch (offset & 0xffff) {
291 			case (offsetof(struct nvmf_connect_data, cntlid)):
292 				dev_err(ctrl->device,
293 					"%s, cntlid: %d\n",
294 					inv_data, data->cntlid);
295 				break;
296 			case (offsetof(struct nvmf_connect_data, hostnqn)):
297 				dev_err(ctrl->device,
298 					"%s, hostnqn \"%s\"\n",
299 					inv_data, data->hostnqn);
300 				break;
301 			case (offsetof(struct nvmf_connect_data, subsysnqn)):
302 				dev_err(ctrl->device,
303 					"%s, subsysnqn \"%s\"\n",
304 					inv_data, data->subsysnqn);
305 				break;
306 			default:
307 				dev_err(ctrl->device,
308 					"%s, starting byte offset: %d\n",
309 				       inv_data, offset & 0xffff);
310 				break;
311 			}
312 		} else {
313 			char *inv_sqe = "Connect Invalid SQE Parameter";
314 
315 			switch (offset) {
316 			case (offsetof(struct nvmf_connect_command, qid)):
317 				dev_err(ctrl->device,
318 				       "%s, qid %d\n",
319 					inv_sqe, cmd->connect.qid);
320 				break;
321 			default:
322 				dev_err(ctrl->device,
323 					"%s, starting byte offset: %d\n",
324 					inv_sqe, offset);
325 			}
326 		}
327 		break;
328 
329 	case NVME_SC_CONNECT_INVALID_HOST:
330 		dev_err(ctrl->device,
331 			"Connect for subsystem %s is not allowed, hostnqn: %s\n",
332 			data->subsysnqn, data->hostnqn);
333 		break;
334 
335 	case NVME_SC_CONNECT_CTRL_BUSY:
336 		dev_err(ctrl->device,
337 			"Connect command failed: controller is busy or not available\n");
338 		break;
339 
340 	case NVME_SC_CONNECT_FORMAT:
341 		dev_err(ctrl->device,
342 			"Connect incompatible format: %d",
343 			cmd->connect.recfmt);
344 		break;
345 
346 	default:
347 		dev_err(ctrl->device,
348 			"Connect command failed, error wo/DNR bit: %d\n",
349 			err_sctype);
350 		break;
351 	} /* switch (err_sctype) */
352 }
353 
354 /**
355  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
356  *				API function.
357  * @ctrl:	Host nvme controller instance used to request
358  *              a new NVMe controller allocation on the target
359  *              system and  establish an NVMe Admin connection to
360  *              that controller.
361  *
362  * This function enables an NVMe host device to request a new allocation of
363  * an NVMe controller resource on a target system as well establish a
364  * fabrics-protocol connection of the NVMe Admin queue between the
365  * host system device and the allocated NVMe controller on the
366  * target system via a NVMe Fabrics "Connect" command.
367  *
368  * Return:
369  *	0: success
370  *	> 0: NVMe error status code
371  *	< 0: Linux errno error code
372  *
373  */
374 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
375 {
376 	struct nvme_command cmd;
377 	union nvme_result res;
378 	struct nvmf_connect_data *data;
379 	int ret;
380 
381 	memset(&cmd, 0, sizeof(cmd));
382 	cmd.connect.opcode = nvme_fabrics_command;
383 	cmd.connect.fctype = nvme_fabrics_type_connect;
384 	cmd.connect.qid = 0;
385 	cmd.connect.sqsize = cpu_to_le16(NVME_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 	uuid_copy(&data->hostid, &ctrl->opts->host->id);
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, &res,
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(res.u32),
408 				       &cmd, data);
409 		goto out_free_data;
410 	}
411 
412 	ctrl->cntlid = le16_to_cpu(res.u16);
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 	union nvme_result res;
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 	uuid_copy(&data->hostid, &ctrl->opts->host->id);
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, &res,
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(res.u32),
467 				       &cmd, data);
468 	}
469 	kfree(data);
470 	return ret;
471 }
472 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
473 
474 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
475 {
476 	if (ctrl->opts->max_reconnects != -1 &&
477 	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
478 		return true;
479 
480 	return false;
481 }
482 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
483 
484 /**
485  * nvmf_register_transport() - NVMe Fabrics Library registration function.
486  * @ops:	Transport ops instance to be registered to the
487  *		common fabrics library.
488  *
489  * API function that registers the type of specific transport fabric
490  * being implemented to the common NVMe fabrics library. Part of
491  * the overall init sequence of starting up a fabrics driver.
492  */
493 int nvmf_register_transport(struct nvmf_transport_ops *ops)
494 {
495 	if (!ops->create_ctrl)
496 		return -EINVAL;
497 
498 	mutex_lock(&nvmf_transports_mutex);
499 	list_add_tail(&ops->entry, &nvmf_transports);
500 	mutex_unlock(&nvmf_transports_mutex);
501 
502 	return 0;
503 }
504 EXPORT_SYMBOL_GPL(nvmf_register_transport);
505 
506 /**
507  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
508  * @ops:	Transport ops instance to be unregistered from the
509  *		common fabrics library.
510  *
511  * Fabrics API function that unregisters the type of specific transport
512  * fabric being implemented from the common NVMe fabrics library.
513  * Part of the overall exit sequence of unloading the implemented driver.
514  */
515 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
516 {
517 	mutex_lock(&nvmf_transports_mutex);
518 	list_del(&ops->entry);
519 	mutex_unlock(&nvmf_transports_mutex);
520 }
521 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
522 
523 static struct nvmf_transport_ops *nvmf_lookup_transport(
524 		struct nvmf_ctrl_options *opts)
525 {
526 	struct nvmf_transport_ops *ops;
527 
528 	lockdep_assert_held(&nvmf_transports_mutex);
529 
530 	list_for_each_entry(ops, &nvmf_transports, entry) {
531 		if (strcmp(ops->name, opts->transport) == 0)
532 			return ops;
533 	}
534 
535 	return NULL;
536 }
537 
538 static const match_table_t opt_tokens = {
539 	{ NVMF_OPT_TRANSPORT,		"transport=%s"		},
540 	{ NVMF_OPT_TRADDR,		"traddr=%s"		},
541 	{ NVMF_OPT_TRSVCID,		"trsvcid=%s"		},
542 	{ NVMF_OPT_NQN,			"nqn=%s"		},
543 	{ NVMF_OPT_QUEUE_SIZE,		"queue_size=%d"		},
544 	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
545 	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
546 	{ NVMF_OPT_CTRL_LOSS_TMO,	"ctrl_loss_tmo=%d"	},
547 	{ NVMF_OPT_KATO,		"keep_alive_tmo=%d"	},
548 	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
549 	{ NVMF_OPT_HOST_TRADDR,		"host_traddr=%s"	},
550 	{ NVMF_OPT_HOST_ID,		"hostid=%s"		},
551 	{ NVMF_OPT_ERR,			NULL			}
552 };
553 
554 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
555 		const char *buf)
556 {
557 	substring_t args[MAX_OPT_ARGS];
558 	char *options, *o, *p;
559 	int token, ret = 0;
560 	size_t nqnlen  = 0;
561 	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
562 	uuid_t hostid;
563 
564 	/* Set defaults */
565 	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
566 	opts->nr_io_queues = num_online_cpus();
567 	opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
568 
569 	options = o = kstrdup(buf, GFP_KERNEL);
570 	if (!options)
571 		return -ENOMEM;
572 
573 	uuid_gen(&hostid);
574 
575 	while ((p = strsep(&o, ",\n")) != NULL) {
576 		if (!*p)
577 			continue;
578 
579 		token = match_token(p, opt_tokens, args);
580 		opts->mask |= token;
581 		switch (token) {
582 		case NVMF_OPT_TRANSPORT:
583 			p = match_strdup(args);
584 			if (!p) {
585 				ret = -ENOMEM;
586 				goto out;
587 			}
588 			opts->transport = p;
589 			break;
590 		case NVMF_OPT_NQN:
591 			p = match_strdup(args);
592 			if (!p) {
593 				ret = -ENOMEM;
594 				goto out;
595 			}
596 			opts->subsysnqn = p;
597 			nqnlen = strlen(opts->subsysnqn);
598 			if (nqnlen >= NVMF_NQN_SIZE) {
599 				pr_err("%s needs to be < %d bytes\n",
600 					opts->subsysnqn, NVMF_NQN_SIZE);
601 				ret = -EINVAL;
602 				goto out;
603 			}
604 			opts->discovery_nqn =
605 				!(strcmp(opts->subsysnqn,
606 					 NVME_DISC_SUBSYS_NAME));
607 			if (opts->discovery_nqn)
608 				opts->nr_io_queues = 0;
609 			break;
610 		case NVMF_OPT_TRADDR:
611 			p = match_strdup(args);
612 			if (!p) {
613 				ret = -ENOMEM;
614 				goto out;
615 			}
616 			opts->traddr = p;
617 			break;
618 		case NVMF_OPT_TRSVCID:
619 			p = match_strdup(args);
620 			if (!p) {
621 				ret = -ENOMEM;
622 				goto out;
623 			}
624 			opts->trsvcid = p;
625 			break;
626 		case NVMF_OPT_QUEUE_SIZE:
627 			if (match_int(args, &token)) {
628 				ret = -EINVAL;
629 				goto out;
630 			}
631 			if (token < NVMF_MIN_QUEUE_SIZE ||
632 			    token > NVMF_MAX_QUEUE_SIZE) {
633 				pr_err("Invalid queue_size %d\n", token);
634 				ret = -EINVAL;
635 				goto out;
636 			}
637 			opts->queue_size = token;
638 			break;
639 		case NVMF_OPT_NR_IO_QUEUES:
640 			if (match_int(args, &token)) {
641 				ret = -EINVAL;
642 				goto out;
643 			}
644 			if (token <= 0) {
645 				pr_err("Invalid number of IOQs %d\n", token);
646 				ret = -EINVAL;
647 				goto out;
648 			}
649 			opts->nr_io_queues = min_t(unsigned int,
650 					num_online_cpus(), token);
651 			break;
652 		case NVMF_OPT_KATO:
653 			if (match_int(args, &token)) {
654 				ret = -EINVAL;
655 				goto out;
656 			}
657 
658 			if (opts->discovery_nqn) {
659 				pr_err("Discovery controllers cannot accept keep_alive_tmo != 0\n");
660 				ret = -EINVAL;
661 				goto out;
662 			}
663 
664 			if (token < 0) {
665 				pr_err("Invalid keep_alive_tmo %d\n", token);
666 				ret = -EINVAL;
667 				goto out;
668 			} else if (token == 0) {
669 				/* Allowed for debug */
670 				pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
671 			}
672 			opts->kato = token;
673 			break;
674 		case NVMF_OPT_CTRL_LOSS_TMO:
675 			if (match_int(args, &token)) {
676 				ret = -EINVAL;
677 				goto out;
678 			}
679 
680 			if (token < 0)
681 				pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
682 			ctrl_loss_tmo = token;
683 			break;
684 		case NVMF_OPT_HOSTNQN:
685 			if (opts->host) {
686 				pr_err("hostnqn already user-assigned: %s\n",
687 				       opts->host->nqn);
688 				ret = -EADDRINUSE;
689 				goto out;
690 			}
691 			p = match_strdup(args);
692 			if (!p) {
693 				ret = -ENOMEM;
694 				goto out;
695 			}
696 			nqnlen = strlen(p);
697 			if (nqnlen >= NVMF_NQN_SIZE) {
698 				pr_err("%s needs to be < %d bytes\n",
699 					p, NVMF_NQN_SIZE);
700 				kfree(p);
701 				ret = -EINVAL;
702 				goto out;
703 			}
704 			opts->host = nvmf_host_add(p);
705 			kfree(p);
706 			if (!opts->host) {
707 				ret = -ENOMEM;
708 				goto out;
709 			}
710 			break;
711 		case NVMF_OPT_RECONNECT_DELAY:
712 			if (match_int(args, &token)) {
713 				ret = -EINVAL;
714 				goto out;
715 			}
716 			if (token <= 0) {
717 				pr_err("Invalid reconnect_delay %d\n", token);
718 				ret = -EINVAL;
719 				goto out;
720 			}
721 			opts->reconnect_delay = token;
722 			break;
723 		case NVMF_OPT_HOST_TRADDR:
724 			p = match_strdup(args);
725 			if (!p) {
726 				ret = -ENOMEM;
727 				goto out;
728 			}
729 			opts->host_traddr = p;
730 			break;
731 		case NVMF_OPT_HOST_ID:
732 			p = match_strdup(args);
733 			if (!p) {
734 				ret = -ENOMEM;
735 				goto out;
736 			}
737 			if (uuid_parse(p, &hostid)) {
738 				ret = -EINVAL;
739 				goto out;
740 			}
741 			break;
742 		default:
743 			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
744 				p);
745 			ret = -EINVAL;
746 			goto out;
747 		}
748 	}
749 
750 	if (ctrl_loss_tmo < 0)
751 		opts->max_reconnects = -1;
752 	else
753 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
754 						opts->reconnect_delay);
755 
756 	if (!opts->host) {
757 		kref_get(&nvmf_default_host->ref);
758 		opts->host = nvmf_default_host;
759 	}
760 
761 	uuid_copy(&opts->host->id, &hostid);
762 
763 out:
764 	if (!opts->discovery_nqn && !opts->kato)
765 		opts->kato = NVME_DEFAULT_KATO;
766 	kfree(options);
767 	return ret;
768 }
769 
770 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
771 		unsigned int required_opts)
772 {
773 	if ((opts->mask & required_opts) != required_opts) {
774 		int i;
775 
776 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
777 			if ((opt_tokens[i].token & required_opts) &&
778 			    !(opt_tokens[i].token & opts->mask)) {
779 				pr_warn("missing parameter '%s'\n",
780 					opt_tokens[i].pattern);
781 			}
782 		}
783 
784 		return -EINVAL;
785 	}
786 
787 	return 0;
788 }
789 
790 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
791 		unsigned int allowed_opts)
792 {
793 	if (opts->mask & ~allowed_opts) {
794 		int i;
795 
796 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
797 			if (opt_tokens[i].token & ~allowed_opts) {
798 				pr_warn("invalid parameter '%s'\n",
799 					opt_tokens[i].pattern);
800 			}
801 		}
802 
803 		return -EINVAL;
804 	}
805 
806 	return 0;
807 }
808 
809 void nvmf_free_options(struct nvmf_ctrl_options *opts)
810 {
811 	nvmf_host_put(opts->host);
812 	kfree(opts->transport);
813 	kfree(opts->traddr);
814 	kfree(opts->trsvcid);
815 	kfree(opts->subsysnqn);
816 	kfree(opts->host_traddr);
817 	kfree(opts);
818 }
819 EXPORT_SYMBOL_GPL(nvmf_free_options);
820 
821 #define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
822 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
823 				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
824 				 NVMF_OPT_HOST_ID)
825 
826 static struct nvme_ctrl *
827 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
828 {
829 	struct nvmf_ctrl_options *opts;
830 	struct nvmf_transport_ops *ops;
831 	struct nvme_ctrl *ctrl;
832 	int ret;
833 
834 	opts = kzalloc(sizeof(*opts), GFP_KERNEL);
835 	if (!opts)
836 		return ERR_PTR(-ENOMEM);
837 
838 	ret = nvmf_parse_options(opts, buf);
839 	if (ret)
840 		goto out_free_opts;
841 
842 	/*
843 	 * Check the generic options first as we need a valid transport for
844 	 * the lookup below.  Then clear the generic flags so that transport
845 	 * drivers don't have to care about them.
846 	 */
847 	ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
848 	if (ret)
849 		goto out_free_opts;
850 	opts->mask &= ~NVMF_REQUIRED_OPTS;
851 
852 	mutex_lock(&nvmf_transports_mutex);
853 	ops = nvmf_lookup_transport(opts);
854 	if (!ops) {
855 		pr_info("no handler found for transport %s.\n",
856 			opts->transport);
857 		ret = -EINVAL;
858 		goto out_unlock;
859 	}
860 
861 	ret = nvmf_check_required_opts(opts, ops->required_opts);
862 	if (ret)
863 		goto out_unlock;
864 	ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
865 				ops->allowed_opts | ops->required_opts);
866 	if (ret)
867 		goto out_unlock;
868 
869 	ctrl = ops->create_ctrl(dev, opts);
870 	if (IS_ERR(ctrl)) {
871 		ret = PTR_ERR(ctrl);
872 		goto out_unlock;
873 	}
874 
875 	if (strcmp(ctrl->subnqn, opts->subsysnqn)) {
876 		dev_warn(ctrl->device,
877 			"controller returned incorrect NQN: \"%s\".\n",
878 			ctrl->subnqn);
879 		mutex_unlock(&nvmf_transports_mutex);
880 		ctrl->ops->delete_ctrl(ctrl);
881 		return ERR_PTR(-EINVAL);
882 	}
883 
884 	mutex_unlock(&nvmf_transports_mutex);
885 	return ctrl;
886 
887 out_unlock:
888 	mutex_unlock(&nvmf_transports_mutex);
889 out_free_opts:
890 	nvmf_free_options(opts);
891 	return ERR_PTR(ret);
892 }
893 
894 static struct class *nvmf_class;
895 static struct device *nvmf_device;
896 static DEFINE_MUTEX(nvmf_dev_mutex);
897 
898 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
899 		size_t count, loff_t *pos)
900 {
901 	struct seq_file *seq_file = file->private_data;
902 	struct nvme_ctrl *ctrl;
903 	const char *buf;
904 	int ret = 0;
905 
906 	if (count > PAGE_SIZE)
907 		return -ENOMEM;
908 
909 	buf = memdup_user_nul(ubuf, count);
910 	if (IS_ERR(buf))
911 		return PTR_ERR(buf);
912 
913 	mutex_lock(&nvmf_dev_mutex);
914 	if (seq_file->private) {
915 		ret = -EINVAL;
916 		goto out_unlock;
917 	}
918 
919 	ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
920 	if (IS_ERR(ctrl)) {
921 		ret = PTR_ERR(ctrl);
922 		goto out_unlock;
923 	}
924 
925 	seq_file->private = ctrl;
926 
927 out_unlock:
928 	mutex_unlock(&nvmf_dev_mutex);
929 	kfree(buf);
930 	return ret ? ret : count;
931 }
932 
933 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
934 {
935 	struct nvme_ctrl *ctrl;
936 	int ret = 0;
937 
938 	mutex_lock(&nvmf_dev_mutex);
939 	ctrl = seq_file->private;
940 	if (!ctrl) {
941 		ret = -EINVAL;
942 		goto out_unlock;
943 	}
944 
945 	seq_printf(seq_file, "instance=%d,cntlid=%d\n",
946 			ctrl->instance, ctrl->cntlid);
947 
948 out_unlock:
949 	mutex_unlock(&nvmf_dev_mutex);
950 	return ret;
951 }
952 
953 static int nvmf_dev_open(struct inode *inode, struct file *file)
954 {
955 	/*
956 	 * The miscdevice code initializes file->private_data, but doesn't
957 	 * make use of it later.
958 	 */
959 	file->private_data = NULL;
960 	return single_open(file, nvmf_dev_show, NULL);
961 }
962 
963 static int nvmf_dev_release(struct inode *inode, struct file *file)
964 {
965 	struct seq_file *seq_file = file->private_data;
966 	struct nvme_ctrl *ctrl = seq_file->private;
967 
968 	if (ctrl)
969 		nvme_put_ctrl(ctrl);
970 	return single_release(inode, file);
971 }
972 
973 static const struct file_operations nvmf_dev_fops = {
974 	.owner		= THIS_MODULE,
975 	.write		= nvmf_dev_write,
976 	.read		= seq_read,
977 	.open		= nvmf_dev_open,
978 	.release	= nvmf_dev_release,
979 };
980 
981 static struct miscdevice nvmf_misc = {
982 	.minor		= MISC_DYNAMIC_MINOR,
983 	.name           = "nvme-fabrics",
984 	.fops		= &nvmf_dev_fops,
985 };
986 
987 static int __init nvmf_init(void)
988 {
989 	int ret;
990 
991 	nvmf_default_host = nvmf_host_default();
992 	if (!nvmf_default_host)
993 		return -ENOMEM;
994 
995 	nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
996 	if (IS_ERR(nvmf_class)) {
997 		pr_err("couldn't register class nvme-fabrics\n");
998 		ret = PTR_ERR(nvmf_class);
999 		goto out_free_host;
1000 	}
1001 
1002 	nvmf_device =
1003 		device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1004 	if (IS_ERR(nvmf_device)) {
1005 		pr_err("couldn't create nvme-fabris device!\n");
1006 		ret = PTR_ERR(nvmf_device);
1007 		goto out_destroy_class;
1008 	}
1009 
1010 	ret = misc_register(&nvmf_misc);
1011 	if (ret) {
1012 		pr_err("couldn't register misc device: %d\n", ret);
1013 		goto out_destroy_device;
1014 	}
1015 
1016 	return 0;
1017 
1018 out_destroy_device:
1019 	device_destroy(nvmf_class, MKDEV(0, 0));
1020 out_destroy_class:
1021 	class_destroy(nvmf_class);
1022 out_free_host:
1023 	nvmf_host_put(nvmf_default_host);
1024 	return ret;
1025 }
1026 
1027 static void __exit nvmf_exit(void)
1028 {
1029 	misc_deregister(&nvmf_misc);
1030 	device_destroy(nvmf_class, MKDEV(0, 0));
1031 	class_destroy(nvmf_class);
1032 	nvmf_host_put(nvmf_default_host);
1033 
1034 	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1035 	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1036 	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1037 	BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1038 }
1039 
1040 MODULE_LICENSE("GPL v2");
1041 
1042 module_init(nvmf_init);
1043 module_exit(nvmf_exit);
1044