xref: /linux/drivers/nvme/host/fabrics.c (revision 26bb0d3f38a764b743a3ad5c8b6e5b5044d7ceb4)
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 
nvmf_host_alloc(const char * hostnqn,uuid_t * id)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 
nvmf_host_add(const char * hostnqn,uuid_t * id)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 
nvmf_host_default(void)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 
nvmf_host_destroy(struct kref * ref)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 
nvmf_host_put(struct nvmf_host * host)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  */
nvmf_get_address(struct nvme_ctrl * ctrl,char * buf,int size)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  */
nvmf_reg_read32(struct nvme_ctrl * ctrl,u32 off,u32 * val)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, NVME_SUBMIT_RESERVED);
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_STATUS_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  */
nvmf_reg_read64(struct nvme_ctrl * ctrl,u32 off,u64 * val)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, NVME_SUBMIT_RESERVED);
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_STATUS_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  */
nvmf_reg_write32(struct nvme_ctrl * ctrl,u32 off,u32 val)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, NVME_SUBMIT_RESERVED);
275 	if (unlikely(ret))
276 		dev_err(ctrl->device,
277 			"Property Set error: %d, offset %#x\n",
278 			ret > 0 ? ret & ~NVME_STATUS_DNR : ret, off);
279 	return ret;
280 }
281 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
282 
nvmf_subsystem_reset(struct nvme_ctrl * ctrl)283 int nvmf_subsystem_reset(struct nvme_ctrl *ctrl)
284 {
285 	int ret;
286 
287 	if (!nvme_wait_reset(ctrl))
288 		return -EBUSY;
289 
290 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, NVME_SUBSYS_RESET);
291 	if (ret)
292 		return ret;
293 
294 	return nvme_try_sched_reset(ctrl);
295 }
296 EXPORT_SYMBOL_GPL(nvmf_subsystem_reset);
297 
298 /**
299  * nvmf_log_connect_error() - Error-parsing-diagnostic print out function for
300  * 				connect() errors.
301  * @ctrl:	The specific /dev/nvmeX device that had the error.
302  * @errval:	Error code to be decoded in a more human-friendly
303  * 		printout.
304  * @offset:	For use with the NVMe error code
305  * 		NVME_SC_CONNECT_INVALID_PARAM.
306  * @cmd:	This is the SQE portion of a submission capsule.
307  * @data:	This is the "Data" portion of a submission capsule.
308  */
nvmf_log_connect_error(struct nvme_ctrl * ctrl,int errval,int offset,struct nvme_command * cmd,struct nvmf_connect_data * data)309 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
310 		int errval, int offset, struct nvme_command *cmd,
311 		struct nvmf_connect_data *data)
312 {
313 	int err_sctype = errval & ~NVME_STATUS_DNR;
314 
315 	if (errval < 0) {
316 		dev_err(ctrl->device,
317 			"Connect command failed, errno: %d\n", errval);
318 		return;
319 	}
320 
321 	switch (err_sctype) {
322 	case NVME_SC_CONNECT_INVALID_PARAM:
323 		if (offset >> 16) {
324 			char *inv_data = "Connect Invalid Data Parameter";
325 
326 			switch (offset & 0xffff) {
327 			case (offsetof(struct nvmf_connect_data, cntlid)):
328 				dev_err(ctrl->device,
329 					"%s, cntlid: %d\n",
330 					inv_data, data->cntlid);
331 				break;
332 			case (offsetof(struct nvmf_connect_data, hostnqn)):
333 				dev_err(ctrl->device,
334 					"%s, hostnqn \"%s\"\n",
335 					inv_data, data->hostnqn);
336 				break;
337 			case (offsetof(struct nvmf_connect_data, subsysnqn)):
338 				dev_err(ctrl->device,
339 					"%s, subsysnqn \"%s\"\n",
340 					inv_data, data->subsysnqn);
341 				break;
342 			default:
343 				dev_err(ctrl->device,
344 					"%s, starting byte offset: %d\n",
345 				       inv_data, offset & 0xffff);
346 				break;
347 			}
348 		} else {
349 			char *inv_sqe = "Connect Invalid SQE Parameter";
350 
351 			switch (offset) {
352 			case (offsetof(struct nvmf_connect_command, qid)):
353 				dev_err(ctrl->device,
354 				       "%s, qid %d\n",
355 					inv_sqe, cmd->connect.qid);
356 				break;
357 			default:
358 				dev_err(ctrl->device,
359 					"%s, starting byte offset: %d\n",
360 					inv_sqe, offset);
361 			}
362 		}
363 		break;
364 	case NVME_SC_CONNECT_INVALID_HOST:
365 		dev_err(ctrl->device,
366 			"Connect for subsystem %s is not allowed, hostnqn: %s\n",
367 			data->subsysnqn, data->hostnqn);
368 		break;
369 	case NVME_SC_CONNECT_CTRL_BUSY:
370 		dev_err(ctrl->device,
371 			"Connect command failed: controller is busy or not available\n");
372 		break;
373 	case NVME_SC_CONNECT_FORMAT:
374 		dev_err(ctrl->device,
375 			"Connect incompatible format: %d",
376 			cmd->connect.recfmt);
377 		break;
378 	case NVME_SC_HOST_PATH_ERROR:
379 		dev_err(ctrl->device,
380 			"Connect command failed: host path error\n");
381 		break;
382 	case NVME_SC_AUTH_REQUIRED:
383 		dev_err(ctrl->device,
384 			"Connect command failed: authentication required\n");
385 		break;
386 	default:
387 		dev_err(ctrl->device,
388 			"Connect command failed, error wo/DNR bit: %d\n",
389 			err_sctype);
390 		break;
391 	}
392 }
393 
nvmf_connect_data_prep(struct nvme_ctrl * ctrl,u16 cntlid)394 static struct nvmf_connect_data *nvmf_connect_data_prep(struct nvme_ctrl *ctrl,
395 		u16 cntlid)
396 {
397 	struct nvmf_connect_data *data;
398 
399 	data = kzalloc(sizeof(*data), GFP_KERNEL);
400 	if (!data)
401 		return NULL;
402 
403 	uuid_copy(&data->hostid, &ctrl->opts->host->id);
404 	data->cntlid = cpu_to_le16(cntlid);
405 	strscpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
406 	strscpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
407 
408 	return data;
409 }
410 
nvmf_connect_cmd_prep(struct nvme_ctrl * ctrl,u16 qid,struct nvme_command * cmd)411 static void nvmf_connect_cmd_prep(struct nvme_ctrl *ctrl, u16 qid,
412 		struct nvme_command *cmd)
413 {
414 	cmd->connect.opcode = nvme_fabrics_command;
415 	cmd->connect.fctype = nvme_fabrics_type_connect;
416 	cmd->connect.qid = cpu_to_le16(qid);
417 
418 	if (qid) {
419 		cmd->connect.sqsize = cpu_to_le16(ctrl->sqsize);
420 	} else {
421 		cmd->connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
422 
423 		/*
424 		 * set keep-alive timeout in seconds granularity (ms * 1000)
425 		 */
426 		cmd->connect.kato = cpu_to_le32(ctrl->kato * 1000);
427 	}
428 
429 	if (ctrl->opts->disable_sqflow)
430 		cmd->connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW;
431 }
432 
433 /**
434  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
435  *				API function.
436  * @ctrl:	Host nvme controller instance used to request
437  *              a new NVMe controller allocation on the target
438  *              system and  establish an NVMe Admin connection to
439  *              that controller.
440  *
441  * This function enables an NVMe host device to request a new allocation of
442  * an NVMe controller resource on a target system as well establish a
443  * fabrics-protocol connection of the NVMe Admin queue between the
444  * host system device and the allocated NVMe controller on the
445  * target system via a NVMe Fabrics "Connect" command.
446  */
nvmf_connect_admin_queue(struct nvme_ctrl * ctrl)447 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
448 {
449 	struct nvme_command cmd = { };
450 	union nvme_result res;
451 	struct nvmf_connect_data *data;
452 	int ret;
453 	u32 result;
454 
455 	nvmf_connect_cmd_prep(ctrl, 0, &cmd);
456 
457 	data = nvmf_connect_data_prep(ctrl, 0xffff);
458 	if (!data)
459 		return -ENOMEM;
460 
461 	ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res,
462 			data, sizeof(*data), NVME_QID_ANY,
463 			NVME_SUBMIT_AT_HEAD |
464 			NVME_SUBMIT_NOWAIT |
465 			NVME_SUBMIT_RESERVED);
466 	if (ret) {
467 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
468 				       &cmd, data);
469 		goto out_free_data;
470 	}
471 
472 	result = le32_to_cpu(res.u32);
473 	ctrl->cntlid = result & 0xFFFF;
474 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
475 		/* Secure concatenation is not implemented */
476 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
477 			dev_warn(ctrl->device,
478 				 "qid 0: secure concatenation is not supported\n");
479 			ret = -EOPNOTSUPP;
480 			goto out_free_data;
481 		}
482 		/* Authentication required */
483 		ret = nvme_auth_negotiate(ctrl, 0);
484 		if (ret) {
485 			dev_warn(ctrl->device,
486 				 "qid 0: authentication setup failed\n");
487 			goto out_free_data;
488 		}
489 		ret = nvme_auth_wait(ctrl, 0);
490 		if (ret) {
491 			dev_warn(ctrl->device,
492 				 "qid 0: authentication failed, error %d\n",
493 				 ret);
494 		} else
495 			dev_info(ctrl->device,
496 				 "qid 0: authenticated\n");
497 	}
498 out_free_data:
499 	kfree(data);
500 	return ret;
501 }
502 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
503 
504 /**
505  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
506  *			     API function.
507  * @ctrl:	Host nvme controller instance used to establish an
508  *		NVMe I/O queue connection to the already allocated NVMe
509  *		controller on the target system.
510  * @qid:	NVMe I/O queue number for the new I/O connection between
511  *		host and target (note qid == 0 is illegal as this is
512  *		the Admin queue, per NVMe standard).
513  *
514  * This function issues a fabrics-protocol connection
515  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
516  * between the host system device and the allocated NVMe controller
517  * on the target system.
518  *
519  * Return:
520  *	0: success
521  *	> 0: NVMe error status code
522  *	< 0: Linux errno error code
523  */
nvmf_connect_io_queue(struct nvme_ctrl * ctrl,u16 qid)524 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
525 {
526 	struct nvme_command cmd = { };
527 	struct nvmf_connect_data *data;
528 	union nvme_result res;
529 	int ret;
530 	u32 result;
531 
532 	nvmf_connect_cmd_prep(ctrl, qid, &cmd);
533 
534 	data = nvmf_connect_data_prep(ctrl, ctrl->cntlid);
535 	if (!data)
536 		return -ENOMEM;
537 
538 	ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
539 			data, sizeof(*data), qid,
540 			NVME_SUBMIT_AT_HEAD |
541 			NVME_SUBMIT_RESERVED |
542 			NVME_SUBMIT_NOWAIT);
543 	if (ret) {
544 		nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
545 				       &cmd, data);
546 		goto out_free_data;
547 	}
548 	result = le32_to_cpu(res.u32);
549 	if (result & (NVME_CONNECT_AUTHREQ_ATR | NVME_CONNECT_AUTHREQ_ASCR)) {
550 		/* Secure concatenation is not implemented */
551 		if (result & NVME_CONNECT_AUTHREQ_ASCR) {
552 			dev_warn(ctrl->device,
553 				 "qid 0: secure concatenation is not supported\n");
554 			ret = -EOPNOTSUPP;
555 			goto out_free_data;
556 		}
557 		/* Authentication required */
558 		ret = nvme_auth_negotiate(ctrl, qid);
559 		if (ret) {
560 			dev_warn(ctrl->device,
561 				 "qid %d: authentication setup failed\n", qid);
562 			goto out_free_data;
563 		}
564 		ret = nvme_auth_wait(ctrl, qid);
565 		if (ret) {
566 			dev_warn(ctrl->device,
567 				 "qid %u: authentication failed, error %d\n",
568 				 qid, ret);
569 		}
570 	}
571 out_free_data:
572 	kfree(data);
573 	return ret;
574 }
575 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
576 
577 /*
578  * Evaluate the status information returned by the transport in order to decided
579  * if a reconnect attempt should be scheduled.
580  *
581  * Do not retry when:
582  *
583  * - the DNR bit is set and the specification states no further connect
584  *   attempts with the same set of paramenters should be attempted.
585  *
586  * - when the authentication attempt fails, because the key was invalid.
587  *   This error code is set on the host side.
588  */
nvmf_should_reconnect(struct nvme_ctrl * ctrl,int status)589 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl, int status)
590 {
591 	if (status > 0 && (status & NVME_STATUS_DNR))
592 		return false;
593 
594 	if (status == -EKEYREJECTED)
595 		return false;
596 
597 	if (ctrl->opts->max_reconnects == -1 ||
598 	    ctrl->nr_reconnects < ctrl->opts->max_reconnects)
599 		return true;
600 
601 	return false;
602 }
603 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
604 
605 /**
606  * nvmf_register_transport() - NVMe Fabrics Library registration function.
607  * @ops:	Transport ops instance to be registered to the
608  *		common fabrics library.
609  *
610  * API function that registers the type of specific transport fabric
611  * being implemented to the common NVMe fabrics library. Part of
612  * the overall init sequence of starting up a fabrics driver.
613  */
nvmf_register_transport(struct nvmf_transport_ops * ops)614 int nvmf_register_transport(struct nvmf_transport_ops *ops)
615 {
616 	if (!ops->create_ctrl)
617 		return -EINVAL;
618 
619 	down_write(&nvmf_transports_rwsem);
620 	list_add_tail(&ops->entry, &nvmf_transports);
621 	up_write(&nvmf_transports_rwsem);
622 
623 	return 0;
624 }
625 EXPORT_SYMBOL_GPL(nvmf_register_transport);
626 
627 /**
628  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
629  * @ops:	Transport ops instance to be unregistered from the
630  *		common fabrics library.
631  *
632  * Fabrics API function that unregisters the type of specific transport
633  * fabric being implemented from the common NVMe fabrics library.
634  * Part of the overall exit sequence of unloading the implemented driver.
635  */
nvmf_unregister_transport(struct nvmf_transport_ops * ops)636 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
637 {
638 	down_write(&nvmf_transports_rwsem);
639 	list_del(&ops->entry);
640 	up_write(&nvmf_transports_rwsem);
641 }
642 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
643 
nvmf_lookup_transport(struct nvmf_ctrl_options * opts)644 static struct nvmf_transport_ops *nvmf_lookup_transport(
645 		struct nvmf_ctrl_options *opts)
646 {
647 	struct nvmf_transport_ops *ops;
648 
649 	lockdep_assert_held(&nvmf_transports_rwsem);
650 
651 	list_for_each_entry(ops, &nvmf_transports, entry) {
652 		if (strcmp(ops->name, opts->transport) == 0)
653 			return ops;
654 	}
655 
656 	return NULL;
657 }
658 
nvmf_parse_key(int key_id)659 static struct key *nvmf_parse_key(int key_id)
660 {
661 	struct key *key;
662 
663 	if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
664 		pr_err("TLS is not supported\n");
665 		return ERR_PTR(-EINVAL);
666 	}
667 
668 	key = nvme_tls_key_lookup(key_id);
669 	if (IS_ERR(key))
670 		pr_err("key id %08x not found\n", key_id);
671 	else
672 		pr_debug("Using key id %08x\n", key_id);
673 	return key;
674 }
675 
676 static const match_table_t opt_tokens = {
677 	{ NVMF_OPT_TRANSPORT,		"transport=%s"		},
678 	{ NVMF_OPT_TRADDR,		"traddr=%s"		},
679 	{ NVMF_OPT_TRSVCID,		"trsvcid=%s"		},
680 	{ NVMF_OPT_NQN,			"nqn=%s"		},
681 	{ NVMF_OPT_QUEUE_SIZE,		"queue_size=%d"		},
682 	{ NVMF_OPT_NR_IO_QUEUES,	"nr_io_queues=%d"	},
683 	{ NVMF_OPT_RECONNECT_DELAY,	"reconnect_delay=%d"	},
684 	{ NVMF_OPT_CTRL_LOSS_TMO,	"ctrl_loss_tmo=%d"	},
685 	{ NVMF_OPT_KATO,		"keep_alive_tmo=%d"	},
686 	{ NVMF_OPT_HOSTNQN,		"hostnqn=%s"		},
687 	{ NVMF_OPT_HOST_TRADDR,		"host_traddr=%s"	},
688 	{ NVMF_OPT_HOST_IFACE,		"host_iface=%s"		},
689 	{ NVMF_OPT_HOST_ID,		"hostid=%s"		},
690 	{ NVMF_OPT_DUP_CONNECT,		"duplicate_connect"	},
691 	{ NVMF_OPT_DISABLE_SQFLOW,	"disable_sqflow"	},
692 	{ NVMF_OPT_HDR_DIGEST,		"hdr_digest"		},
693 	{ NVMF_OPT_DATA_DIGEST,		"data_digest"		},
694 	{ NVMF_OPT_NR_WRITE_QUEUES,	"nr_write_queues=%d"	},
695 	{ NVMF_OPT_NR_POLL_QUEUES,	"nr_poll_queues=%d"	},
696 	{ NVMF_OPT_TOS,			"tos=%d"		},
697 #ifdef CONFIG_NVME_TCP_TLS
698 	{ NVMF_OPT_KEYRING,		"keyring=%d"		},
699 	{ NVMF_OPT_TLS_KEY,		"tls_key=%d"		},
700 #endif
701 	{ NVMF_OPT_FAIL_FAST_TMO,	"fast_io_fail_tmo=%d"	},
702 	{ NVMF_OPT_DISCOVERY,		"discovery"		},
703 #ifdef CONFIG_NVME_HOST_AUTH
704 	{ NVMF_OPT_DHCHAP_SECRET,	"dhchap_secret=%s"	},
705 	{ NVMF_OPT_DHCHAP_CTRL_SECRET,	"dhchap_ctrl_secret=%s"	},
706 #endif
707 #ifdef CONFIG_NVME_TCP_TLS
708 	{ NVMF_OPT_TLS,			"tls"			},
709 #endif
710 	{ NVMF_OPT_ERR,			NULL			}
711 };
712 
nvmf_parse_options(struct nvmf_ctrl_options * opts,const char * buf)713 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
714 		const char *buf)
715 {
716 	substring_t args[MAX_OPT_ARGS];
717 	char *options, *o, *p;
718 	int token, ret = 0;
719 	size_t nqnlen  = 0;
720 	int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO, key_id;
721 	uuid_t hostid;
722 	char hostnqn[NVMF_NQN_SIZE];
723 	struct key *key;
724 
725 	/* Set defaults */
726 	opts->queue_size = NVMF_DEF_QUEUE_SIZE;
727 	opts->nr_io_queues = num_online_cpus();
728 	opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
729 	opts->kato = 0;
730 	opts->duplicate_connect = false;
731 	opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO;
732 	opts->hdr_digest = false;
733 	opts->data_digest = false;
734 	opts->tos = -1; /* < 0 == use transport default */
735 	opts->tls = false;
736 	opts->tls_key = NULL;
737 	opts->keyring = NULL;
738 
739 	options = o = kstrdup(buf, GFP_KERNEL);
740 	if (!options)
741 		return -ENOMEM;
742 
743 	/* use default host if not given by user space */
744 	uuid_copy(&hostid, &nvmf_default_host->id);
745 	strscpy(hostnqn, nvmf_default_host->nqn, NVMF_NQN_SIZE);
746 
747 	while ((p = strsep(&o, ",\n")) != NULL) {
748 		if (!*p)
749 			continue;
750 
751 		token = match_token(p, opt_tokens, args);
752 		opts->mask |= token;
753 		switch (token) {
754 		case NVMF_OPT_TRANSPORT:
755 			p = match_strdup(args);
756 			if (!p) {
757 				ret = -ENOMEM;
758 				goto out;
759 			}
760 			kfree(opts->transport);
761 			opts->transport = p;
762 			break;
763 		case NVMF_OPT_NQN:
764 			p = match_strdup(args);
765 			if (!p) {
766 				ret = -ENOMEM;
767 				goto out;
768 			}
769 			kfree(opts->subsysnqn);
770 			opts->subsysnqn = p;
771 			nqnlen = strlen(opts->subsysnqn);
772 			if (nqnlen >= NVMF_NQN_SIZE) {
773 				pr_err("%s needs to be < %d bytes\n",
774 					opts->subsysnqn, NVMF_NQN_SIZE);
775 				ret = -EINVAL;
776 				goto out;
777 			}
778 			opts->discovery_nqn =
779 				!(strcmp(opts->subsysnqn,
780 					 NVME_DISC_SUBSYS_NAME));
781 			break;
782 		case NVMF_OPT_TRADDR:
783 			p = match_strdup(args);
784 			if (!p) {
785 				ret = -ENOMEM;
786 				goto out;
787 			}
788 			kfree(opts->traddr);
789 			opts->traddr = p;
790 			break;
791 		case NVMF_OPT_TRSVCID:
792 			p = match_strdup(args);
793 			if (!p) {
794 				ret = -ENOMEM;
795 				goto out;
796 			}
797 			kfree(opts->trsvcid);
798 			opts->trsvcid = p;
799 			break;
800 		case NVMF_OPT_QUEUE_SIZE:
801 			if (match_int(args, &token)) {
802 				ret = -EINVAL;
803 				goto out;
804 			}
805 			if (token < NVMF_MIN_QUEUE_SIZE ||
806 			    token > NVMF_MAX_QUEUE_SIZE) {
807 				pr_err("Invalid queue_size %d\n", token);
808 				ret = -EINVAL;
809 				goto out;
810 			}
811 			opts->queue_size = token;
812 			break;
813 		case NVMF_OPT_NR_IO_QUEUES:
814 			if (match_int(args, &token)) {
815 				ret = -EINVAL;
816 				goto out;
817 			}
818 			if (token <= 0) {
819 				pr_err("Invalid number of IOQs %d\n", token);
820 				ret = -EINVAL;
821 				goto out;
822 			}
823 			if (opts->discovery_nqn) {
824 				pr_debug("Ignoring nr_io_queues value for discovery controller\n");
825 				break;
826 			}
827 
828 			opts->nr_io_queues = min_t(unsigned int,
829 					num_online_cpus(), token);
830 			break;
831 		case NVMF_OPT_KATO:
832 			if (match_int(args, &token)) {
833 				ret = -EINVAL;
834 				goto out;
835 			}
836 
837 			if (token < 0) {
838 				pr_err("Invalid keep_alive_tmo %d\n", token);
839 				ret = -EINVAL;
840 				goto out;
841 			} else if (token == 0 && !opts->discovery_nqn) {
842 				/* Allowed for debug */
843 				pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
844 			}
845 			opts->kato = token;
846 			break;
847 		case NVMF_OPT_CTRL_LOSS_TMO:
848 			if (match_int(args, &token)) {
849 				ret = -EINVAL;
850 				goto out;
851 			}
852 
853 			if (token < 0)
854 				pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
855 			ctrl_loss_tmo = token;
856 			break;
857 		case NVMF_OPT_FAIL_FAST_TMO:
858 			if (match_int(args, &token)) {
859 				ret = -EINVAL;
860 				goto out;
861 			}
862 
863 			if (token >= 0)
864 				pr_warn("I/O fail on reconnect controller after %d sec\n",
865 					token);
866 			else
867 				token = -1;
868 
869 			opts->fast_io_fail_tmo = token;
870 			break;
871 		case NVMF_OPT_HOSTNQN:
872 			if (opts->host) {
873 				pr_err("hostnqn already user-assigned: %s\n",
874 				       opts->host->nqn);
875 				ret = -EADDRINUSE;
876 				goto out;
877 			}
878 			p = match_strdup(args);
879 			if (!p) {
880 				ret = -ENOMEM;
881 				goto out;
882 			}
883 			nqnlen = strlen(p);
884 			if (nqnlen >= NVMF_NQN_SIZE) {
885 				pr_err("%s needs to be < %d bytes\n",
886 					p, NVMF_NQN_SIZE);
887 				kfree(p);
888 				ret = -EINVAL;
889 				goto out;
890 			}
891 			strscpy(hostnqn, p, NVMF_NQN_SIZE);
892 			kfree(p);
893 			break;
894 		case NVMF_OPT_RECONNECT_DELAY:
895 			if (match_int(args, &token)) {
896 				ret = -EINVAL;
897 				goto out;
898 			}
899 			if (token <= 0) {
900 				pr_err("Invalid reconnect_delay %d\n", token);
901 				ret = -EINVAL;
902 				goto out;
903 			}
904 			opts->reconnect_delay = token;
905 			break;
906 		case NVMF_OPT_HOST_TRADDR:
907 			p = match_strdup(args);
908 			if (!p) {
909 				ret = -ENOMEM;
910 				goto out;
911 			}
912 			kfree(opts->host_traddr);
913 			opts->host_traddr = p;
914 			break;
915 		case NVMF_OPT_HOST_IFACE:
916 			p = match_strdup(args);
917 			if (!p) {
918 				ret = -ENOMEM;
919 				goto out;
920 			}
921 			kfree(opts->host_iface);
922 			opts->host_iface = p;
923 			break;
924 		case NVMF_OPT_HOST_ID:
925 			p = match_strdup(args);
926 			if (!p) {
927 				ret = -ENOMEM;
928 				goto out;
929 			}
930 			ret = uuid_parse(p, &hostid);
931 			if (ret) {
932 				pr_err("Invalid hostid %s\n", p);
933 				ret = -EINVAL;
934 				kfree(p);
935 				goto out;
936 			}
937 			kfree(p);
938 			break;
939 		case NVMF_OPT_DUP_CONNECT:
940 			opts->duplicate_connect = true;
941 			break;
942 		case NVMF_OPT_DISABLE_SQFLOW:
943 			opts->disable_sqflow = true;
944 			break;
945 		case NVMF_OPT_HDR_DIGEST:
946 			opts->hdr_digest = true;
947 			break;
948 		case NVMF_OPT_DATA_DIGEST:
949 			opts->data_digest = true;
950 			break;
951 		case NVMF_OPT_NR_WRITE_QUEUES:
952 			if (match_int(args, &token)) {
953 				ret = -EINVAL;
954 				goto out;
955 			}
956 			if (token <= 0) {
957 				pr_err("Invalid nr_write_queues %d\n", token);
958 				ret = -EINVAL;
959 				goto out;
960 			}
961 			opts->nr_write_queues = token;
962 			break;
963 		case NVMF_OPT_NR_POLL_QUEUES:
964 			if (match_int(args, &token)) {
965 				ret = -EINVAL;
966 				goto out;
967 			}
968 			if (token <= 0) {
969 				pr_err("Invalid nr_poll_queues %d\n", token);
970 				ret = -EINVAL;
971 				goto out;
972 			}
973 			opts->nr_poll_queues = token;
974 			break;
975 		case NVMF_OPT_TOS:
976 			if (match_int(args, &token)) {
977 				ret = -EINVAL;
978 				goto out;
979 			}
980 			if (token < 0) {
981 				pr_err("Invalid type of service %d\n", token);
982 				ret = -EINVAL;
983 				goto out;
984 			}
985 			if (token > 255) {
986 				pr_warn("Clamping type of service to 255\n");
987 				token = 255;
988 			}
989 			opts->tos = token;
990 			break;
991 		case NVMF_OPT_KEYRING:
992 			if (match_int(args, &key_id) || key_id <= 0) {
993 				ret = -EINVAL;
994 				goto out;
995 			}
996 			key = nvmf_parse_key(key_id);
997 			if (IS_ERR(key)) {
998 				ret = PTR_ERR(key);
999 				goto out;
1000 			}
1001 			key_put(opts->keyring);
1002 			opts->keyring = key;
1003 			break;
1004 		case NVMF_OPT_TLS_KEY:
1005 			if (match_int(args, &key_id) || key_id <= 0) {
1006 				ret = -EINVAL;
1007 				goto out;
1008 			}
1009 			key = nvmf_parse_key(key_id);
1010 			if (IS_ERR(key)) {
1011 				ret = PTR_ERR(key);
1012 				goto out;
1013 			}
1014 			key_put(opts->tls_key);
1015 			opts->tls_key = key;
1016 			break;
1017 		case NVMF_OPT_DISCOVERY:
1018 			opts->discovery_nqn = true;
1019 			break;
1020 		case NVMF_OPT_DHCHAP_SECRET:
1021 			p = match_strdup(args);
1022 			if (!p) {
1023 				ret = -ENOMEM;
1024 				goto out;
1025 			}
1026 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
1027 				pr_err("Invalid DH-CHAP secret %s\n", p);
1028 				ret = -EINVAL;
1029 				goto out;
1030 			}
1031 			kfree(opts->dhchap_secret);
1032 			opts->dhchap_secret = p;
1033 			break;
1034 		case NVMF_OPT_DHCHAP_CTRL_SECRET:
1035 			p = match_strdup(args);
1036 			if (!p) {
1037 				ret = -ENOMEM;
1038 				goto out;
1039 			}
1040 			if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) {
1041 				pr_err("Invalid DH-CHAP secret %s\n", p);
1042 				ret = -EINVAL;
1043 				goto out;
1044 			}
1045 			kfree(opts->dhchap_ctrl_secret);
1046 			opts->dhchap_ctrl_secret = p;
1047 			break;
1048 		case NVMF_OPT_TLS:
1049 			if (!IS_ENABLED(CONFIG_NVME_TCP_TLS)) {
1050 				pr_err("TLS is not supported\n");
1051 				ret = -EINVAL;
1052 				goto out;
1053 			}
1054 			opts->tls = true;
1055 			break;
1056 		default:
1057 			pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
1058 				p);
1059 			ret = -EINVAL;
1060 			goto out;
1061 		}
1062 	}
1063 
1064 	if (opts->discovery_nqn) {
1065 		opts->nr_io_queues = 0;
1066 		opts->nr_write_queues = 0;
1067 		opts->nr_poll_queues = 0;
1068 		opts->duplicate_connect = true;
1069 	} else {
1070 		if (!opts->kato)
1071 			opts->kato = NVME_DEFAULT_KATO;
1072 	}
1073 	if (ctrl_loss_tmo < 0) {
1074 		opts->max_reconnects = -1;
1075 	} else {
1076 		opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
1077 						opts->reconnect_delay);
1078 		if (ctrl_loss_tmo < opts->fast_io_fail_tmo)
1079 			pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n",
1080 				opts->fast_io_fail_tmo, ctrl_loss_tmo);
1081 	}
1082 
1083 	opts->host = nvmf_host_add(hostnqn, &hostid);
1084 	if (IS_ERR(opts->host)) {
1085 		ret = PTR_ERR(opts->host);
1086 		opts->host = NULL;
1087 		goto out;
1088 	}
1089 
1090 out:
1091 	kfree(options);
1092 	return ret;
1093 }
1094 
nvmf_set_io_queues(struct nvmf_ctrl_options * opts,u32 nr_io_queues,u32 io_queues[HCTX_MAX_TYPES])1095 void nvmf_set_io_queues(struct nvmf_ctrl_options *opts, u32 nr_io_queues,
1096 			u32 io_queues[HCTX_MAX_TYPES])
1097 {
1098 	if (opts->nr_write_queues && opts->nr_io_queues < nr_io_queues) {
1099 		/*
1100 		 * separate read/write queues
1101 		 * hand out dedicated default queues only after we have
1102 		 * sufficient read queues.
1103 		 */
1104 		io_queues[HCTX_TYPE_READ] = opts->nr_io_queues;
1105 		nr_io_queues -= io_queues[HCTX_TYPE_READ];
1106 		io_queues[HCTX_TYPE_DEFAULT] =
1107 			min(opts->nr_write_queues, nr_io_queues);
1108 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1109 	} else {
1110 		/*
1111 		 * shared read/write queues
1112 		 * either no write queues were requested, or we don't have
1113 		 * sufficient queue count to have dedicated default queues.
1114 		 */
1115 		io_queues[HCTX_TYPE_DEFAULT] =
1116 			min(opts->nr_io_queues, nr_io_queues);
1117 		nr_io_queues -= io_queues[HCTX_TYPE_DEFAULT];
1118 	}
1119 
1120 	if (opts->nr_poll_queues && nr_io_queues) {
1121 		/* map dedicated poll queues only if we have queues left */
1122 		io_queues[HCTX_TYPE_POLL] =
1123 			min(opts->nr_poll_queues, nr_io_queues);
1124 	}
1125 }
1126 EXPORT_SYMBOL_GPL(nvmf_set_io_queues);
1127 
nvmf_map_queues(struct blk_mq_tag_set * set,struct nvme_ctrl * ctrl,u32 io_queues[HCTX_MAX_TYPES])1128 void nvmf_map_queues(struct blk_mq_tag_set *set, struct nvme_ctrl *ctrl,
1129 		     u32 io_queues[HCTX_MAX_TYPES])
1130 {
1131 	struct nvmf_ctrl_options *opts = ctrl->opts;
1132 
1133 	if (opts->nr_write_queues && io_queues[HCTX_TYPE_READ]) {
1134 		/* separate read/write queues */
1135 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1136 			io_queues[HCTX_TYPE_DEFAULT];
1137 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1138 		set->map[HCTX_TYPE_READ].nr_queues =
1139 			io_queues[HCTX_TYPE_READ];
1140 		set->map[HCTX_TYPE_READ].queue_offset =
1141 			io_queues[HCTX_TYPE_DEFAULT];
1142 	} else {
1143 		/* shared read/write queues */
1144 		set->map[HCTX_TYPE_DEFAULT].nr_queues =
1145 			io_queues[HCTX_TYPE_DEFAULT];
1146 		set->map[HCTX_TYPE_DEFAULT].queue_offset = 0;
1147 		set->map[HCTX_TYPE_READ].nr_queues =
1148 			io_queues[HCTX_TYPE_DEFAULT];
1149 		set->map[HCTX_TYPE_READ].queue_offset = 0;
1150 	}
1151 
1152 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1153 	blk_mq_map_queues(&set->map[HCTX_TYPE_READ]);
1154 	if (opts->nr_poll_queues && io_queues[HCTX_TYPE_POLL]) {
1155 		/* map dedicated poll queues only if we have queues left */
1156 		set->map[HCTX_TYPE_POLL].nr_queues = io_queues[HCTX_TYPE_POLL];
1157 		set->map[HCTX_TYPE_POLL].queue_offset =
1158 			io_queues[HCTX_TYPE_DEFAULT] +
1159 			io_queues[HCTX_TYPE_READ];
1160 		blk_mq_map_queues(&set->map[HCTX_TYPE_POLL]);
1161 	}
1162 
1163 	dev_info(ctrl->device,
1164 		"mapped %d/%d/%d default/read/poll queues.\n",
1165 		io_queues[HCTX_TYPE_DEFAULT],
1166 		io_queues[HCTX_TYPE_READ],
1167 		io_queues[HCTX_TYPE_POLL]);
1168 }
1169 EXPORT_SYMBOL_GPL(nvmf_map_queues);
1170 
nvmf_check_required_opts(struct nvmf_ctrl_options * opts,unsigned int required_opts)1171 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
1172 		unsigned int required_opts)
1173 {
1174 	if ((opts->mask & required_opts) != required_opts) {
1175 		unsigned int i;
1176 
1177 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1178 			if ((opt_tokens[i].token & required_opts) &&
1179 			    !(opt_tokens[i].token & opts->mask)) {
1180 				pr_warn("missing parameter '%s'\n",
1181 					opt_tokens[i].pattern);
1182 			}
1183 		}
1184 
1185 		return -EINVAL;
1186 	}
1187 
1188 	return 0;
1189 }
1190 
nvmf_ip_options_match(struct nvme_ctrl * ctrl,struct nvmf_ctrl_options * opts)1191 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl,
1192 		struct nvmf_ctrl_options *opts)
1193 {
1194 	if (!nvmf_ctlr_matches_baseopts(ctrl, opts) ||
1195 	    strcmp(opts->traddr, ctrl->opts->traddr) ||
1196 	    strcmp(opts->trsvcid, ctrl->opts->trsvcid))
1197 		return false;
1198 
1199 	/*
1200 	 * Checking the local address or host interfaces is rough.
1201 	 *
1202 	 * In most cases, none is specified and the host port or
1203 	 * host interface is selected by the stack.
1204 	 *
1205 	 * Assume no match if:
1206 	 * -  local address or host interface is specified and address
1207 	 *    or host interface is not the same
1208 	 * -  local address or host interface is not specified but
1209 	 *    remote is, or vice versa (admin using specific
1210 	 *    host_traddr/host_iface when it matters).
1211 	 */
1212 	if ((opts->mask & NVMF_OPT_HOST_TRADDR) &&
1213 	    (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1214 		if (strcmp(opts->host_traddr, ctrl->opts->host_traddr))
1215 			return false;
1216 	} else if ((opts->mask & NVMF_OPT_HOST_TRADDR) ||
1217 		   (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) {
1218 		return false;
1219 	}
1220 
1221 	if ((opts->mask & NVMF_OPT_HOST_IFACE) &&
1222 	    (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1223 		if (strcmp(opts->host_iface, ctrl->opts->host_iface))
1224 			return false;
1225 	} else if ((opts->mask & NVMF_OPT_HOST_IFACE) ||
1226 		   (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) {
1227 		return false;
1228 	}
1229 
1230 	return true;
1231 }
1232 EXPORT_SYMBOL_GPL(nvmf_ip_options_match);
1233 
nvmf_check_allowed_opts(struct nvmf_ctrl_options * opts,unsigned int allowed_opts)1234 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
1235 		unsigned int allowed_opts)
1236 {
1237 	if (opts->mask & ~allowed_opts) {
1238 		unsigned int i;
1239 
1240 		for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
1241 			if ((opt_tokens[i].token & opts->mask) &&
1242 			    (opt_tokens[i].token & ~allowed_opts)) {
1243 				pr_warn("invalid parameter '%s'\n",
1244 					opt_tokens[i].pattern);
1245 			}
1246 		}
1247 
1248 		return -EINVAL;
1249 	}
1250 
1251 	return 0;
1252 }
1253 
nvmf_free_options(struct nvmf_ctrl_options * opts)1254 void nvmf_free_options(struct nvmf_ctrl_options *opts)
1255 {
1256 	nvmf_host_put(opts->host);
1257 	key_put(opts->keyring);
1258 	key_put(opts->tls_key);
1259 	kfree(opts->transport);
1260 	kfree(opts->traddr);
1261 	kfree(opts->trsvcid);
1262 	kfree(opts->subsysnqn);
1263 	kfree(opts->host_traddr);
1264 	kfree(opts->host_iface);
1265 	kfree(opts->dhchap_secret);
1266 	kfree(opts->dhchap_ctrl_secret);
1267 	kfree(opts);
1268 }
1269 EXPORT_SYMBOL_GPL(nvmf_free_options);
1270 
1271 #define NVMF_REQUIRED_OPTS	(NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
1272 #define NVMF_ALLOWED_OPTS	(NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
1273 				 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
1274 				 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\
1275 				 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\
1276 				 NVMF_OPT_FAIL_FAST_TMO | NVMF_OPT_DHCHAP_SECRET |\
1277 				 NVMF_OPT_DHCHAP_CTRL_SECRET)
1278 
1279 static struct nvme_ctrl *
nvmf_create_ctrl(struct device * dev,const char * buf)1280 nvmf_create_ctrl(struct device *dev, const char *buf)
1281 {
1282 	struct nvmf_ctrl_options *opts;
1283 	struct nvmf_transport_ops *ops;
1284 	struct nvme_ctrl *ctrl;
1285 	int ret;
1286 
1287 	opts = kzalloc(sizeof(*opts), GFP_KERNEL);
1288 	if (!opts)
1289 		return ERR_PTR(-ENOMEM);
1290 
1291 	ret = nvmf_parse_options(opts, buf);
1292 	if (ret)
1293 		goto out_free_opts;
1294 
1295 
1296 	request_module("nvme-%s", opts->transport);
1297 
1298 	/*
1299 	 * Check the generic options first as we need a valid transport for
1300 	 * the lookup below.  Then clear the generic flags so that transport
1301 	 * drivers don't have to care about them.
1302 	 */
1303 	ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
1304 	if (ret)
1305 		goto out_free_opts;
1306 	opts->mask &= ~NVMF_REQUIRED_OPTS;
1307 
1308 	down_read(&nvmf_transports_rwsem);
1309 	ops = nvmf_lookup_transport(opts);
1310 	if (!ops) {
1311 		pr_info("no handler found for transport %s.\n",
1312 			opts->transport);
1313 		ret = -EINVAL;
1314 		goto out_unlock;
1315 	}
1316 
1317 	if (!try_module_get(ops->module)) {
1318 		ret = -EBUSY;
1319 		goto out_unlock;
1320 	}
1321 	up_read(&nvmf_transports_rwsem);
1322 
1323 	ret = nvmf_check_required_opts(opts, ops->required_opts);
1324 	if (ret)
1325 		goto out_module_put;
1326 	ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
1327 				ops->allowed_opts | ops->required_opts);
1328 	if (ret)
1329 		goto out_module_put;
1330 
1331 	ctrl = ops->create_ctrl(dev, opts);
1332 	if (IS_ERR(ctrl)) {
1333 		ret = PTR_ERR(ctrl);
1334 		goto out_module_put;
1335 	}
1336 
1337 	module_put(ops->module);
1338 	return ctrl;
1339 
1340 out_module_put:
1341 	module_put(ops->module);
1342 	goto out_free_opts;
1343 out_unlock:
1344 	up_read(&nvmf_transports_rwsem);
1345 out_free_opts:
1346 	nvmf_free_options(opts);
1347 	return ERR_PTR(ret);
1348 }
1349 
1350 static const struct class nvmf_class = {
1351 	.name = "nvme-fabrics",
1352 };
1353 
1354 static struct device *nvmf_device;
1355 static DEFINE_MUTEX(nvmf_dev_mutex);
1356 
nvmf_dev_write(struct file * file,const char __user * ubuf,size_t count,loff_t * pos)1357 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
1358 		size_t count, loff_t *pos)
1359 {
1360 	struct seq_file *seq_file = file->private_data;
1361 	struct nvme_ctrl *ctrl;
1362 	const char *buf;
1363 	int ret = 0;
1364 
1365 	if (count > PAGE_SIZE)
1366 		return -ENOMEM;
1367 
1368 	buf = memdup_user_nul(ubuf, count);
1369 	if (IS_ERR(buf))
1370 		return PTR_ERR(buf);
1371 
1372 	mutex_lock(&nvmf_dev_mutex);
1373 	if (seq_file->private) {
1374 		ret = -EINVAL;
1375 		goto out_unlock;
1376 	}
1377 
1378 	ctrl = nvmf_create_ctrl(nvmf_device, buf);
1379 	if (IS_ERR(ctrl)) {
1380 		ret = PTR_ERR(ctrl);
1381 		goto out_unlock;
1382 	}
1383 
1384 	seq_file->private = ctrl;
1385 
1386 out_unlock:
1387 	mutex_unlock(&nvmf_dev_mutex);
1388 	kfree(buf);
1389 	return ret ? ret : count;
1390 }
1391 
__nvmf_concat_opt_tokens(struct seq_file * seq_file)1392 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file)
1393 {
1394 	const struct match_token *tok;
1395 	int idx;
1396 
1397 	/*
1398 	 * Add dummy entries for instance and cntlid to
1399 	 * signal an invalid/non-existing controller
1400 	 */
1401 	seq_puts(seq_file, "instance=-1,cntlid=-1");
1402 	for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) {
1403 		tok = &opt_tokens[idx];
1404 		if (tok->token == NVMF_OPT_ERR)
1405 			continue;
1406 		seq_putc(seq_file, ',');
1407 		seq_puts(seq_file, tok->pattern);
1408 	}
1409 	seq_putc(seq_file, '\n');
1410 }
1411 
nvmf_dev_show(struct seq_file * seq_file,void * private)1412 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1413 {
1414 	struct nvme_ctrl *ctrl;
1415 
1416 	mutex_lock(&nvmf_dev_mutex);
1417 	ctrl = seq_file->private;
1418 	if (!ctrl) {
1419 		__nvmf_concat_opt_tokens(seq_file);
1420 		goto out_unlock;
1421 	}
1422 
1423 	seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1424 			ctrl->instance, ctrl->cntlid);
1425 
1426 out_unlock:
1427 	mutex_unlock(&nvmf_dev_mutex);
1428 	return 0;
1429 }
1430 
nvmf_dev_open(struct inode * inode,struct file * file)1431 static int nvmf_dev_open(struct inode *inode, struct file *file)
1432 {
1433 	/*
1434 	 * The miscdevice code initializes file->private_data, but doesn't
1435 	 * make use of it later.
1436 	 */
1437 	file->private_data = NULL;
1438 	return single_open(file, nvmf_dev_show, NULL);
1439 }
1440 
nvmf_dev_release(struct inode * inode,struct file * file)1441 static int nvmf_dev_release(struct inode *inode, struct file *file)
1442 {
1443 	struct seq_file *seq_file = file->private_data;
1444 	struct nvme_ctrl *ctrl = seq_file->private;
1445 
1446 	if (ctrl)
1447 		nvme_put_ctrl(ctrl);
1448 	return single_release(inode, file);
1449 }
1450 
1451 static const struct file_operations nvmf_dev_fops = {
1452 	.owner		= THIS_MODULE,
1453 	.write		= nvmf_dev_write,
1454 	.read		= seq_read,
1455 	.open		= nvmf_dev_open,
1456 	.release	= nvmf_dev_release,
1457 };
1458 
1459 static struct miscdevice nvmf_misc = {
1460 	.minor		= MISC_DYNAMIC_MINOR,
1461 	.name           = "nvme-fabrics",
1462 	.fops		= &nvmf_dev_fops,
1463 };
1464 
nvmf_init(void)1465 static int __init nvmf_init(void)
1466 {
1467 	int ret;
1468 
1469 	nvmf_default_host = nvmf_host_default();
1470 	if (!nvmf_default_host)
1471 		return -ENOMEM;
1472 
1473 	ret = class_register(&nvmf_class);
1474 	if (ret) {
1475 		pr_err("couldn't register class nvme-fabrics\n");
1476 		goto out_free_host;
1477 	}
1478 
1479 	nvmf_device =
1480 		device_create(&nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1481 	if (IS_ERR(nvmf_device)) {
1482 		pr_err("couldn't create nvme-fabrics device!\n");
1483 		ret = PTR_ERR(nvmf_device);
1484 		goto out_destroy_class;
1485 	}
1486 
1487 	ret = misc_register(&nvmf_misc);
1488 	if (ret) {
1489 		pr_err("couldn't register misc device: %d\n", ret);
1490 		goto out_destroy_device;
1491 	}
1492 
1493 	return 0;
1494 
1495 out_destroy_device:
1496 	device_destroy(&nvmf_class, MKDEV(0, 0));
1497 out_destroy_class:
1498 	class_unregister(&nvmf_class);
1499 out_free_host:
1500 	nvmf_host_put(nvmf_default_host);
1501 	return ret;
1502 }
1503 
nvmf_exit(void)1504 static void __exit nvmf_exit(void)
1505 {
1506 	misc_deregister(&nvmf_misc);
1507 	device_destroy(&nvmf_class, MKDEV(0, 0));
1508 	class_unregister(&nvmf_class);
1509 	nvmf_host_put(nvmf_default_host);
1510 
1511 	BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64);
1512 	BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1513 	BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1514 	BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1515 	BUILD_BUG_ON(sizeof(struct nvmf_auth_send_command) != 64);
1516 	BUILD_BUG_ON(sizeof(struct nvmf_auth_receive_command) != 64);
1517 	BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1518 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_negotiate_data) != 8);
1519 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_challenge_data) != 16);
1520 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_reply_data) != 16);
1521 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success1_data) != 16);
1522 	BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success2_data) != 16);
1523 }
1524 
1525 MODULE_LICENSE("GPL v2");
1526 MODULE_DESCRIPTION("NVMe host fabrics library");
1527 
1528 module_init(nvmf_init);
1529 module_exit(nvmf_exit);
1530